USF Libraries
USF Digital Collections

Association among neonatal mortality, weekend or nighttime admissions and staffing in a Neonatal Intensive Care Unit

MISSING IMAGE

Material Information

Title:
Association among neonatal mortality, weekend or nighttime admissions and staffing in a Neonatal Intensive Care Unit
Physical Description:
Book
Language:
English
Creator:
Stanley, Leisa J
Publisher:
University of South Florida
Place of Publication:
Tampa, Fla.
Publication Date:

Subjects

Subjects / Keywords:
Infant death
Acuity
Nurse staffing
Adverse events
Obstetrical interventions
Dissertations, Academic -- Public Health -- Doctoral -- USF   ( lcsh )
Genre:
bibliography   ( marcgt )
theses   ( marcgt )
non-fiction   ( marcgt )

Notes

Abstract:
ABSTRACT: The purpose of this study was to investigate the time of admission to a Neonatal Intensive Care Unit (NICU) and its association with in-hospital mortality among a cohort of neonates at a regional perinatal center. Two different time points were considered: admissions on the weekend versus the weekday and admissions during the nighttime shift versus the day shift. The secondary purpose of the study was to investigate if registered nurse staffing affected this association between NICU admission day or admission time and in-hospital death. Three separate databases were used which contained information on NICU admissions, hospital deliveries and nurse staffing. These databases were linked resulting in data for each individual mother-infant pair for each separate admission to the NICU. Readmissions to the NICU, NICU admissions which could not be linked with the delivery data, admissions from the Newborn Nursery and transfers from other hospitals were excluded from the study.^ The final study population consisted of 1,846 admissions from October 1, 2001 through December 31, 2006. Weekend admissions were lower than weekday admissions (29.6% versus 70.4%) and nighttime admissions were lower than day admissions (43.2% versus 56.8%). Infants admitted at nighttime were more likely to be low birth weight, have lower Apgar scores and less likely to be delivered by cesarean section. Weekend admissions did not differ significantly from weekday admissions, except weekend admissions were more likely to be Black (33.6% versus 28.6%, p=.30). After adjusting for infant's acuity and other covariates using multivariate logistic regression, the odds of dying on the weekend was not significantly different than weekday admissions (AOR=1.06, 95% CI=.653-1.721) and were not significantly different for nighttime admissions (AOR=1.14, 95% CI=.722-1.79). Nurse staffing was not a significant covariate.^ Covariates which were significant risk factors for death prior to discharge were non-Black race of the infant, Apgar score of less than 7 at five minutes, presence of a fetal anomaly, and use of ventilation during the stay. Infant's birth weight was a significant protective factor.
Thesis:
Dissertation (Ph.D.)--University of South Florida, 2008.
Bibliography:
Includes bibliographical references.
System Details:
System requirements: World Wide Web browser and PDF reader.
System Details:
Mode of access: World Wide Web.
Statement of Responsibility:
by Leisa J. Stanley.
General Note:
Title from PDF of title page.
General Note:
Document formatted into pages; contains 154 pages.
General Note:
Includes vita.

Record Information

Source Institution:
University of South Florida Library
Holding Location:
University of South Florida
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
aleph - 001970181
oclc - 276123180
usfldc doi - E14-SFE0002421
usfldc handle - e14.2421
System ID:
SFS0026739:00001


This item is only available as the following downloads:


Full Text

PAGE 1

Association among Neonatal Mortalit y, Weekend or Nighttime Admissions And Staffing in a Neonatal Intensive Care Unit by Leisa J. Stanley A dissertation submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy Department of Community and Family Health College of Public Health University of South Florida Major Professor: Kare n (Kay) Perrin, Ph.D. Jeffrey Jensen, M.D. Jeffrey D. Kromrey, Ph.D. Charles S. Mahan, M.D. Robert M. Nelson, M.D. Date of Approval: April 4, 2008 Keywords: infant death, acuity, nurse staffing, adverse events, obstetrical interventions Copyright 2008, Leisa J. Stanley

PAGE 2

Dedication This dissertation is dedicated to my beloved Granny whose unconditional love for me and unwavering belief in me gave me th e courage and strength to pursue my dreams.

PAGE 3

Preface There are many individuals who I need to thank who helped me to achieve this dream. My major professor Dr. Kay Perrin provided her guidance, practical advice, encouragement and support. I would not have achieved this accomplishment without her. She was always there with an answer gentle shove or needed praise. Dr. Robert Nelson gave me a wonderful opportunity to conduc t this research project. He helped to pave the way with Tampa General Hospital and was always available to answer questions about neonatal medicine or TGH. Thanks for your courage in having this study conducted. Dr. Charles Mahan has been a mentor and trusted advisor over the past 15 years. He provided his insight thr oughout this research. Thanks for your wisdom and passion regarding obstetrical care and practice. As always, thanks for your leadersh ip on behalf of moms and babies. Dr. Jeffrey Kromrey provided practical guid ance making the statistical aspect of this research understandable and fun. Thanks for your time spent reviewing descriptive statistics, logistic regres sion models, ROC curves and answering my questions. Dr. Jensen stepped in at a crucial time to serve on my committee. Thanks for agreeing to serve and providing your own insights into obstetrical care.

PAGE 4

Jane Murphy gave her full support to this endeavor and gave me the time off I needed to conduct my analysis and write. Kate Booth made sure necessary work was completed at the office even as she finished her own Masters in Public Health. Rebecca Bell, Kim Ziprik, and Kim Houser are friends who never forgot to encourage me and celebrate every milestone with me. Judith Sharp Rose gave me her wise counsel and guidance which helped me through the difficult times. Lo Berry, Dr. Deborah Austin and Vanessa Miskit at the Central Hillsborough Healthy Start Project prayed for me and never forgot to encourage me. Karen Fugate and Leah Godfrey at TGH a ssisted me with getting the databases I needed. Dr. David Darr secured the space I n eeded on the USF Health Science Center server and provided guidance during the proposal development phase. Dee Jeffers secured an office at the Chiles Center for me Jason Salemi gave me his SAS expertise when SPSS would not do what I needed. Finally I want to thank my parents, Bill and Sara Stanley, who gave their enduring love, support and encouragement throughout th e pursuit of my dreams and goals. Mom, thanks for reading to me every day when I was a child. You gave me my first library filled with books like Dr. Seuss Curious George and the Little Red Rabbit Who Wanted Red Wings You started my love of reading and learning. Well, I finally got my red wings and I can still fit into the house!

PAGE 5

Table of Contents List of Tables....................................................................................................................v List of Figures................................................................................................................ vii Abstract....................................................................................................................... .....ix Chapter One: Introduction................................................................................................1 Factors Related to Infant Mortality.......................................................................1 Individual-Level Risk Factors...................................................................2 Contextual-Level Risk Factors.................................................................3 Hospital-Level Risk Factors.....................................................................5 Impact of Community-Based Interven tions to Reduce Infant Mortality..............5 Purpose of the Study.............................................................................................7 Statement of the Problem..........................................................................7 Research Questions...................................................................................8 Significance of the Study..........................................................................9 Definition of Terms...................................................................................9 Chapter Two: Literature Review....................................................................................13 Weekend or Nighttime Events and Infant Mortality...........................................13 Weekend Births and Neonatal Mortality................................................14 Nighttime Births and Neonatal Mortality...............................................19 Nighttime NICU Admissions and Neonatal Mortality...........................20 Weekend/Evening Admissions and In-Hospital Mortality in Other Adult Acute Care Units................................................22 Infant Characteristics and Neonatal Mortality....................................................22 Acuity .....................................................................................................23 Birth Weight and Gestation........................................................25 Apgar Scores...............................................................................26 Physiological Measures..............................................................29 Congenital Anomalies.............................................................................31 Multiple Births........................................................................................31 Gender and Race.....................................................................................32 Hospital Characteristic s and Neonatal Mortality................................................33 Level of Care...........................................................................................34 Volume of NICU Admissions.................................................................35 Obstetrical Interventions and Role in Birth Patterns..............................37 Relationship of Nurse Staffing to In-Hospital Mortality........................39 i

PAGE 6

Skill Mix.....................................................................................40 Nurse to Patient Ratio.................................................................41 RN Nursing Hours......................................................................44 Relationship of Physician Sta ffing to In-Hospital Mortality..................45 Summary of Research.........................................................................................49 Chapter Three: Methods.................................................................................................50 Research Design..................................................................................................50 Study Setting...........................................................................................51 HIPAA and the Protect ion of Human Subjects.......................................52 Study Population.....................................................................................53 Sampling Framework..............................................................................53 Data Collection...................................................................................................54 Obstetrical Database...............................................................................54 Neonatal Intensive Care Unit Database..................................................54 Nurse Staffing Database.........................................................................55 Data Linking Procedures.........................................................................56 Validation and Reliability of Databases.................................................56 Data Analysis .....................................................................................................57 Data Assumptions...................................................................................57 Study Variables.......................................................................................58 Outcome Variables......................................................................58 Exposure Variables.....................................................................58 Covariates...................................................................................59 Descriptive Statistics...............................................................................62 Multivariate Models................................................................................63 Research Questions.................................................................................65 Research Question 1...................................................................65 Research Question 2...................................................................66 Research Question 3...................................................................66 Research Question 4...................................................................66 Sample Size Calculation.....................................................................................67 Chapter Four: Results .....................................................................................................68 Study Population.................................................................................................68 Data Linking Procedures.....................................................................................72 Deterministic Linking.............................................................................72 Probabilistic Linking...............................................................................72 Manual Linking.......................................................................................73 Unlinked Tampa General Hospital Births...............................................74 Data Preparation..................................................................................................75 NICU and OOIS Databases....................................................................75 RN Database...........................................................................................76 Descriptive Analysis...........................................................................................76 Weekend Admissions Compar ed to Weekday Admissions....................76 Nighttime Admissions Comp ared to Day Admissions...........................78 ii

PAGE 7

Bivariate Correlations.............................................................................80 Multivarite Analysis............................................................................................82 Bivariate Relationships for Exposur es and Covariates with Outcome...83 Research Question 1...............................................................................85 Comparison of full-risk adju stment model with reduced-risk adjustment model............................................................87 TGH NICU admissions compared to transfers...........................89 Research Question 2...............................................................................91 Research Question 3...............................................................................93 Research Question 4...............................................................................95 Summary of the Research.................................................................................101 Chapter Five: Discussion..............................................................................................102 Main Study Findings.........................................................................................102 Association of Day or Time of NICU Admission with Mortality........102 Association of Nurse Staffing...............................................................105 Limitations of the Study....................................................................................107 Sample Size...........................................................................................107 Study Setting.........................................................................................108 Missing Data.........................................................................................109 Nurse Staffing.......................................................................................110 Physician Staffing.................................................................................111 Definition of Exposure Variables.........................................................112 Public Health Implications................................................................................113 Future Research Implications...........................................................................115 References.....................................................................................................................118 Appendices....................................................................................................................127 Appendix A: Tampa General Hospital Office of Clinical Research Approval...................................................................................128 Appendix B: Institutional Re view Board Approval, USF..............................129 Appendix C: T-Test Results for Li nked Births with Unlinked Births............131 Appendix D: T-Test Results NICU Admissions with Newborn Nursery Admissions................................................................................132 Appendix E: Descriptive Statisti cs, Weekend to Weekday Admissions.......133 Appendix F: Descriptive Statis tics, Nighttime to Day Admissions...............134 Appendix G: Logistic Regression for Model 1...............................................135 Appendix H: Logistic Regre ssion for Reduced Risk Model..........................137 Appendix I: Receiver Operating Characteristic Curves for ModelFA and ModelRA....................................................................................138 Appendix J: T-Test Results for In-Born and Transfers.................................140 Appendix K: Logistic Regression for Model 2 ..............................................141 Appendix L: Logistic Regression for Model 3...............................................143 Appendix M: Logistic Regression for Model 4A............................................145 Appendix N: Logistic Regression for Model 4B............................................147 iii

PAGE 8

Appendix O: Logistic Regression for Model 4C............................................149 Appendix P: Receiver Operating Characteristic Curves, Model 1................152 About the Author................................................................................................End Page iv

PAGE 9

Table of Tables Table 1. Apgar Scoring System............................................................................27 Table 2. Incidence of Neonata l Death by Gestational Age and Apgar Score............................................................................................28 Table 3. Comparison of NICU Births Linked to OOIS with NICU Births Not Linked to OOIS (N=1905)...............................................................69 Table 4. Comparison of Infants Admitted to NICU First with Infants Admitted To Newborn Nursery First......................................................................71 Table 5. Weekend Admissions Compared to Weekday Admissions (N=1846)...77 Table 6. T-Test Results for Weekend Admissions Compared to Weekday Admissions (N=1846).............................................................................78 Table 7. Nighttime Admissions Compared to Day Admissions (N=1846)..........79 Table 8. T-Test Results for Nighttime Admissions Compared to Day Admissions (N=1846).............................................................................80 Table 9. Bivariate Correla tion Coefficients with Birth Weight and HospitalLevel Variables (N=1846)......................................................................81 Table 10. Bivariate Coorelations with Hospital-Level Variables (N=1846)..........81 Table 11. Bivariate Relationships for Exposures and Covariates with Death Before Discharge (n=1837)....................................................................84 Table 12. Logistic Regression Model 1 for Weekend Exposure with Death Before Discharge (n=1837)....................................................................86 Table 13. -2 Log Likelihood Tests fo r Model 1 with Weekend Exposure.............86 Table 14. Logistic Regression ModelRA for Weekend Exposure with Death Before Discharge (n=1837)....................................................................88 Table 15. -2 Log Likelihood Tests for ModelRA with Weekend Exposure.............88 v

PAGE 10

Table 16. Comparison of In-Born NICU Admissions with Transfer NICU Admissions (N=2572).............................................................................90 Table 17. Logistic Regression Model 2 for Nighttime Exposure with Death Before Discharge (n=1837)....................................................................92 Table 18. -2 Log Likelihood Tests fo r Model 2 with Nighttime Exposure............93 Table 19. Logistic Regression Model 3 for Effect Modification with Death Before Discharge....................................................................................94 Table 20. -2 Log Likelihood Tests for Model 3 with Effect Modification.............95 Table 21. Logistic Regression Model 4A for Weekend Exposure with Hospital Level Covariates Included with..............................................................97 Table 22. -2 Log Likelihood Tests for Model 4A with Weekend Exposure With Hospital-Level Covariates ............................................................98 Table 23. Logistic Regression Model 4B for Nighttime Exposure with Hospital Level Covariates Included with..............................................................99 Table 24. -2 Log Likelihood Tests for Model 4B with Nighttime Exposure With Hospital-Level Covariates ............................................................99 Table 25. Logistic Regression Model 4C for Effect Modification with Hospital Level Covariates Included with............................................................100 Table 26. -2 Log Likelihood Tests for Model 4C with Effect Modification With Hospital-Level Covariates ..........................................................101 Table C1. T-Test Results for NICU Birt hs Linked to OOIS with NICU Births Not Linked to OOIS (N=1905).............................................................131 Table D1. T-Test Results of Infants Admitted to NICU First with Infants Admitted to Newborn Nursery..............................................................132 Table E1. Descriptive Statistics for W eekend Admissions Compared to Weekday Admissions (n=1846)............................................................................133 Table F1. Descriptive Statistics for Nighttime Admissions Compared to Day Admissions............................................................................................134 Table G1. Complete Logistic Regressi on Models for Weekend Exposure with Death Before Discharge (n=1837)........................................................135 vi

PAGE 11

Table H1. Complete Logistic Regre ssion Model for Weekend Exposure with Death Before Discharge........................................................................137 Table J1. T-Test Results of In-Born NICU Admissions with Transfer NICU Admissions (N=2572)...........................................................................140 Table K1. Complete Logistic Regression Models for Nighttime Exposure with Death Before Discharge (n=1837)........................................................141 Table L1. Complete Logistic Regression Models for Effect Modification with Death Before Discharge (n=1837)........................................................143 Table M1. Complete Logistic Regressi on Models 4A for Weekend Exposure with Hospital-Level Covariates............................................................145 Table N1. Complete Logistic Regression Models 4B for Nighttime Exposure....147 Table O1. Complete Logistic Regression Models 4C for Effect Modification.....149 vii

PAGE 12

List of Figures Figure 1. ROC for Full-Risk Adjustment Model.................................................138 Figure 2. ROC for Reduced-Risk Adjustment Model..........................................139 Figure 3. ROC for Anomaly Removed, Model 1.................................................152 Figure 4. ROC for Ventilation Removed, Model 1..............................................153 Figure 5. ROC for Anomaly and Ventilation Removed, Model 1.......................154 viii

PAGE 13

Association among Neonatal Mortality, Week end or Nighttime Admi ssions and Staffing in a Neonatal Intensive Care Unit Leisa J. Stanley ABSTRACT The purpose of this study was to investig ate the time of admission to a Neonatal Intensive Care Unit (NICU) and its associa tion with in-hospital mortality among a cohort of neonates at a regional perinatal center. Two different time points were considered: admissions on the weekend versus the weekda y and admissions during the nighttime shift versus the day shift. The secondary purpose of the study was to inves tigate if registered nurse staffing affected this association between NICU admission day or admission time and in-hospital death. Three separate databases were used which contained information on NICU admissions, hospital deliveries and nurse sta ffing. These databases were linked resulting in data for each individual mother-infant pair for each separate admission to the NICU. Readmissions to the NICU, NICU admissions which could not be linked with the delivery data, admissions from the Newborn Nu rsery and transfers from other hospitals were excluded from the study. The final st udy population consisted of 1,846 admissions from October 1, 2001 through December 31, 2006. Weekend admissions were lower than weekday admissions (29.6% versus 70.4%) and nighttime admissions were lower than da y admissions (43.2% versus 56.8%). Infants admitted at nighttime were more likely to be low birth weight, have lower Apgar scores and less likely to be delivered by cesarean section. Weekend admissions did not differ significantly from weekday admissions, except weekend admissions were more likely to be Black (33.6% versus 28.6%, p=.30). ix

PAGE 14

After adjusting for infants acuity and other covariates using multivariate logistic regression, the odds of dying on the weekend was not significantly different than weekday admissions (AOR=1.06, 95% CI=.653 -1.721) and were not significantly different for nighttime admissions (AOR =1.14, 95% CI=.722-1.79). Nurse staffing was not a significant covariate. C ovariates which were significant risk factors for death prior to discharge were non-Black race of the infant, Apgar score of less than 7 at five minutes, presence of a fetal anomaly, and use of ven tilation during the stay. Infants birth weight was a significant protective factor. x

PAGE 15

1 Chapter One: Introduction The purpose of this study was to investi gate the time of admission to a Neonatal Intensive Care Unit (NICU) and its associati on with in-hospital mortality among a cohort of neonates at a regi onal perinatal center. Two different time point s were considered. The first time point was admission on the weekend versus the weekday. The second time point was admission during the nighttime shift versus the day shift. The secondary purpose of the study was to investigate if registered nurse staffing affected this association between NICU admission day or admission time and in-hospital death. Factors Related to Infant Mortality Infant mortality is a multi-factorial problem. Research into infant mortality, as well as interventions aimed at reducing it, ha ve primarily focused on risk factors at the individual level and how those factors impact infant deaths. These individual level risk factors can occur with th e infant or the mother or in co mbination. This line of research has not fully explained the underlying reasons for infant deaths. Research on infant mortality has begun to explore th e association with contextual level variables that occur at the community level or system level va riables that occur at the hospital level (O'Campo, Xue, Wang, & Caughy, 1997).

PAGE 16

2 Individual-Level Risk Factors Birth weight and gestation are the two most important pr edictors of an infants survival at birth. Survival is positively asso ciated with both measures. Gender and race are also strongly related to survival. Female infants have a higher probability of survival at lower birth weights and gestational age than male infants. Black infants have a higher probability of survival at lower birth weights and gestational age than white infants. The presence of a major birth defect is another risk factor predictive of death in the first year of life (Alexander, Kogan, Himes, & Golde nberg, 1999; Alexander, Tompkins, Allen, & Hulsey, 1999; Ingemarsson, 2003; Mathew s, Menacker, & MacDorman, 2004). Individual-level risk factors related to the in fant are discussed in greater detail in the Literature Review under Infant Characteristics and Neonatal Mortality. Maternal demographic factors such as age, education a nd race have all been found to be independently associated with infa nt mortality and morbidity (O'Campo et al., 1997). Maternal age, both less than 18 years ol d and greater than 34 years old, are related to increased risk of infant death (Khos hnood, Wall, & Lee, 2005; Phipps & Sowers, 2002). Maternal race of Black has been shown to be an independent risk factor for infant death and morbidity, even among Black wome n in higher socioeconomic groups (Adams, Read, & Rawlings, 1993; Goldenberg, Cliver, & Mulvihill, 1996; Healy, Malone, & LM, 2006; McGrady, Sun, Rowley, & Hogue, 1992; Schoendorf, Hogue, Kleinman, & Rowley, 1992). Maternal behavior during pregnancy is also related to birth outcomes. Women who smoke, drink alcohol or us e drugs during pregnancy have a higher risk of delivering

PAGE 17

3 a preterm and/or low birth weight infant (Lundsberg, Bracken, & Saftlas, 1997; Visscher, Feder, Burns, Brady, & Bray, 2003; Windham, Hopkins, Fenster, & Swan, 2000). Both short interpregnancy interval s (< 6 months) and long inte rpregnancy intervals (> 60 months) have been shown to increase the risk of poor birth outcomes (Zhu & Le, 2003; Zhu, Rolfs, Nangle, & Horan, 1999). Contextual-Level Risk Factors Risk factors which assess the effect of macro-level variables are referred to as contextual-level risk factors since they meas ure the effect of the context in which an individual lives. The study of community le vel variables and their impact on infant mortality and morbidity has been utilized by social epidemiology in examining maternal and child health outcomes. Social risk has been conceptualized by measuring different variables related to income disparity at the community level (O'Campo et al., 1997; Pickett & Pearl, 2001; Rajaratnam, Burke, & O'Campo, 2006). In a review of studies which focused on contextual-level variables and maternal and child health outcomes, Rajaratnam, Burke and OCampo (2006) reviewed five studies which focused on low birth weight a nd seven different nei ghborhood constructs. All of these studies used some measure of wealth or soci oeconomic status while there was little agreement on the other construc ts used. These included measures of employment, family structure, population composition, housing, community mobility, education level of residents, occupation of residents, social resources and violence and crime. Research results were mixed with respect to the impact of neighborhood-level

PAGE 18

4 characteristics and the outcome of low birth weight. While income was found to be significant in two of the studi es, population characteristics were not significant in any. Buka and colleagues (2003) found a 13.1 gram decrease in mean birth weight for each increase in one standard deviation of economic disadvantage for Black women. Economic disadvantage was measured using th e United States Census data on poverty, receipt of public assistance and unempl oyment. For White women, the neighborhood effect of economic disadvantage was not significant. Instead, it was the provision of perceived social support that wa s associated with a slight increase in mean birth weight. The increase was small, 17.5 grams. For both gr oups, the mean increase in birth weight may not be clinically signifi cant. Pearl, Braveman, and Abrams (2001) also found that low birth weight increased as the percentage of reside nts living below 200% of the Federal Poverty line increased. Robert s (1997) found that neighborhood economic hardship (as measured by the unemployment ra te and percentage of families in poverty) was associated with an increase in low bi rth weight. However, the neighborhood-level variables socioeconomic status, percentage of Black males, percentage of young residents and crowded housing rates were all inversely associated with low birth weight. OCampo and colleagues did find that pe r capita income of less than $8,000 was associated with an 11% increased risk of lo w birth weight, after adjustment for individual level risk factors. This was one of the initial studies to explore this relationship and was not included in the previous study (O'Campo et al., 1997). Community-level risk factors can also modify the relationship between infant outcomes and individual-level risk s. The protective effect of the receipt of prenatal care

PAGE 19

5 has been shown to vary by the neighborhood in which the woman lives. For women who live in areas with lower unemployment rates, receipt of prenatal care is more protective against low birth weight than women who receive the same level of pr enatal care but live in poorer areas defined by higher rates of unemployment (O'Campo et al., 1997). Hospital-Level Risk Factors Variables at the hospital level that ha ve been studied as impacting patient mortality include the level of care provided, staffing patterns of registered nurses and physicians and volume of admissions. Studies have shown that increased numbers of registered nurses and increased nurse to pati ent ratios, as well as increased physician to patient ratios, have reduced th e occurrence of adverse events. Patients who receive care at tertiary care centers also have fewer a dverse outcomes after adjustment for illness severity. These variables are discussed in detail in the Literature Review under Hospital Characteristics and Neonatal Mortality. Impact of Community-Based Interven tions to Reduce Infant Mortality Research on the effectiveness of community-based interventions aimed at reducing infant mortality has been mixed, not always showing the anticipated positive impact on reducing infant mortality and morbidity. These types of interventions usually involve some form of social support aimed at reducing the mothers unhealthy behavior, improving access to needed services or reduc ing social isolation and stress. Social support normally involves the provision of emotional support, instrumental support or

PAGE 20

6 informational support (Dunkel-Schetter, Sagr estano, Feldman and Killingsowrth, 1996). A recent systematic review by the Cochrane Collaborative did not find any significant association between social support progr ams and improved birth outcomes such as prematurity, low birth weight or perinatal mo rtality. Sixteen randomized trials, involving almost 14,000 pregnant women were incl uded in the meta -analysis (Hodnett & Fredericks, 2003). Evaluations of programs using a non-randomized design have shown mixed results of these interventions. Improved outcomes were for women in specific subgroups, such as those with previous poor outcome s, Black women, teen mothers (Norbeck & Anderson, 1989; Bryce, Stanley & Gamer, 1991; Rogers, 1996; Flynn 1999). Others have failed to show an improvement in birth wei ght or preterm delivery (Spencer, Thomas and Morris, 1989; Oakley, 1990; Villar, 1992; Langer, 1996). Feldman (2000) found that social support improved birth outcomes thr ough a reduction in fetal growth restriction. A recent evaluation of the Florida Healthy Start program did find an improvement in birth weight for high-risk women who r eceived services compared to those who did not. The 2005 evaluation analyzed outcomes for women whose delivery was paid for by Medicaid and compared the 2000 Medicaid bi rth cohort (n=84,785) to the 2002 Medicaid birth cohort (n=. 87,017). Women receiving pren atal care through a regional center for high-risk obstetric patients had a significant reduction in preterm bi rths and very low and low birth weight infants after the implementation of the Medicaid Waiver which provided funding for Healthy Start services. The comp arison group of women in RPICCs who did not receive Healthy Start servi ces had an increase in these rates. However, the reduction

PAGE 21

7 in preterm births and low birt h weight births did not transl ate into reductions in infant deaths for the group of women receiving services (Darr, 2005). Purpose of the Study The purpose of this study is to investig ate the time of admission to a Neonatal Intensive Care Unit (NICU) and its associa tion with in-hospital mortality among a cohort of neonates at a regional perinatal center. Two different time points will be considered. The first time point is admission on the w eekend versus the weekday. The second time point is admission during the ni ghttime shift versus the day shift. The secondary purpose of the study is to investigate if registered nurse staffing af fects this association between NICU admission day or admission time and in-hospital death. Statement of the Problem The association between infant mortality a nd both infant and maternal risk factors has been well studied. However, many interven tions aimed at the individual level, both maternal and infant, have not continued to improve the rates of both prematurity and low birth weight (Alexander, Kogan et al., 1999; Alexander, Tomp kins et al., 1999). Much of the improvement in infant mortality during the nineties occurred due to medical interventions which improved th e survival of preterm and low birth weight infants such as the use of surfactant therapy, antenata l steroids and mechanical ventilation (Sappenfield, 2007). In 2002, the United States infant mortal ity rate increased for the

PAGE 22

8 first time in forty years. Infants who die dur ing the first month of life comprise 67% all infant deaths (Mathews et al., 2004). Given the recent rise in the United States in fant mortality rate and the inability to reduce both preterm and low birt h weight births, it is impor tant to understand what other factors influence infant deaths. Previous re search has shown a higher risk of death for both nighttime and weekend hospi tal admissions. This higher risk was related to patient acuity and/or hospital staffi ng of registered nurses a nd physicians depending on the study. Within this area of research, studies have focused on the association between the day of birth or time of birth and subsequent ne onatal death (the first 27 days of life). This is the time when hospital level variables are mo re likely to affect neonatal deaths due to issues related to quality of care. Research Questions The research questions unde r investigation are: (1) Is there an association between the day of admission (weekday versus weekend) to the NICU and the infants outcome? (2) Is there an association between the time of admission (day versus nighttime) to the NICU and the infants outcome? (3) Is there effect modification betwee n day of admission and time of admission? (4) Does staffing of registered nurses in the NICU mediate the association between day or time of admission to th e NICU and the infants outcome?

PAGE 23

9 Significance of the Study Reductions in the infant mortality ra te which occurred during the 1990s have stalled and the rate increased nationally in 2002 for the first time in forty years (Mathews et al., 2004). Governments at the federal, state and local le vels as well as private organizations have made significant investme nts in community-based interventions to reduce infant mortality. These interventi ons are predominately focused on changing individual-level risk factors. The research has been mixed on their effectiveness. However, there are few studies that have examined both day and time of birth and hospital variables, such as staffing, and thei r impact on neonatal deaths. Regardless of the effectiveness of interventions aimed at changi ng individual behaviors, the impact of these interventions may be negated or reduced by th e impact of system level variables at the hospital level. Definition of Terms The definition of key terms is given below. Time of NICU admission will be defined according to the shifts at study setting hospital. There are two shifts at this hospital. The fi rst shift begins at 7:00 am and ends at 7:00 pm a nd will be defined as day shift The second shift begins at 7:00 pm and ends at 7:00 am and will be defined as the nighttime shift For staff who work a shift that enco mpasses both these shifts (such as 3 pm to 11 pm), their time will be divide d between the two main shifts defined above.

PAGE 24

10 Weekdays will be defined as Monday, beginning at 7:01 am, through Friday at 7:00 pm. Weekends will be defined as Friday beginning at 7:01 pm through Monday at 7:00 am. A neonatal death will be defined as an infant born alive, admitted to the NICU, who died prior to di scharge. This is not the traditional definition of a neonatal death which is an infant born a live who dies within the first 27 days of life. This study will not be able to track infants who were admitted to the NICU and were discharged prior to the first 27 days of life and subsequently died. The purpose of this study is to investigate th e timing of NICU admissions and subsequent deaths and whether or not they are related to hospital-level variables. Therefore, deaths prior to discharge are more appropriate for the research questions under investigation. Low birth weight is defined as an infant bor n alive weighing less than 2,500 grams (5.5 pounds) and very low birth wei ght is defined as an infant born alive weighing less than 1,500 grams (3.5 pounds). Preterm birth is defined as an infant born alive at less than 37 completed weeks of gestation. In-born status refers to infants born at th e study setting hospital who are admitted to the NICU. Out-born status refers to infants born at another hospital who are transferre d to the study setting hosp ital and admitted to the NICU.

PAGE 25

11 Nursing Skill Mix is the ratio of registered nur ses (RN) to licensed practical nurses (LPN) and is defined as total nu mber of RNs in the NICU on the shift divided by the total number of RNs plus LPNs on the shift. RN hours is the total number of hours of RN care per shift or per day and is defined as the total number of RN hours per 12-hour shift or per 24-hour day. Nurse to Patient Ratio is defined as the total number of RNs on the shift at admission to the total number of infants present in the NICU during that shift. The staffing database does not record act ual number of infants per nurse. This will serve as a measure of the average workload per nurse. Capacity is based on the census in the NI CU at midnight each day and is defined as the total number of occupi ed beds per day divided by the total number of available beds per shift. SNAP is the Score for Neonatal Acute P hysiology. The SNAP collects seven physiological measures, after NICU admission, on all birth weight categories of neonates in order to provide an index of neonatal acuity and subsequent likelihood of neonatal death. CRIB is the Clinical Risk Index for Ba bies. The CRIB collects six measures after NICU admission on neonates less than 1,500 grams. The CRIB provides an index of neonatal acuity and subs equent likelihood of neonatal death. APACHE III is the Acute Physiology and Ch ronic Health Evaluation. This index is a measure of patient acuity in adult populations as the SNAP and CRIB are acuity measures in neonates.

PAGE 26

12 Vermont Oxford Network (VON) is a network of 353 hospitals with NICUs located in 49 states and 22 foreign c ountries which systematically collects data from each participating institutio n on infants admitted to the NICUs and their outcomes.

PAGE 27

13 Chapter Two: Literature Review The purpose of this study was to investi gate the time of admission to a Neonatal Intensive Care Unit (NICU) and its associati on with in-hospital mortality among a cohort of neonates at a regi onal perinatal center. Two different time point s were considered. The first time point was admission on the weekend versus the weekday. The second time point was admission during the nighttime shift versus the day shift. The secondary purpose of the study was to investigate if registered nurse staffing affected this association between NICU admission day or admission time and in-hospital death. Weekend or Nighttime Events and Infant Mortality Previous research has investigated wh ether being born on the weekend increases an infants risk of dying during the neonatal period. These studies used birth data files linked to infant death files and investigated death within the first month of life, regardless of whether that death occurred in the hospita l or not. More recent st udies have focused on nighttime admissions to the Neonatal In tensive Care Unit and whether nighttime admissions increase the likelihood of an infant dying prior to discharge.

PAGE 28

14 Weekend Births and Neonatal Mortality Research during the past thirty years has examined the association between weekend births and an increased incidence of neonatal deaths. These studies primarily used linked data sets of birth records and infa nt death records. For t hose studies that had a statistically significant associ ation, Sunday births had the hi ghest rate of mortality. In the first study examining date of birth and subsequent neonatal death, MacFarlane (1978) found that perinatal mortality was 14% higher on Sundays when compared to the highest rate during the week (21.59 per 1,000 liv e births v. 18.99) and 23% higher when compared to the total pe rinatal rate for the cohort (21.59 v. 17.61). This trend in higher neonatal mortality on the weekend held true for neonatal deaths on the first day of life and those during the first week of life. However, MacFarlanes study did not adjust for the influence of illness severity. The influence of birth weight or gestation may have increased the likelihood of death regardless of when the infant was born. Therefore, case mix could explain this higher rate. This study showed that births were lo wer on the weekend than on the weekday possibly due to obstetrical in terventions during the week. Ob stetrical interventions during the week may have influenced the case mix on the weekend resulting in a higher proportion of low birth weight infants being born, but not a higher number Hendry (1981) found that both Sundays and Mondays had more deaths than other days of the week. The Sunday rate was twice the daily average rate. These deaths were largely due to premature births. Sundays al so had the lowest percentage of births

PAGE 29

15 compared to the other days of the week wh ich may have lead to a higher proportion of small, preterm infants on the that day. The first study in the United States to examine this relationship used the 1974-75 Arkansas live birth cohort (N=66,056). Sunda y births also had the highest neonatal mortality rate (13.3). Compared to weekdays, the Sunday rate was 27% to 43% higher depending on the weekday used for comparison. Mangold (1981) also examined the disparity between Nonwhite and White neonat al deaths. For deaths to infants born on Sunday, the Nonwhite neonatal mo rtality rate was 70% higher than the White rate (18.9 v. 11.1). For Nonwhite infants who were low bi rth weight, there was a daily increase in the rate from Monday until Sunday where the rate was the highest. For normal birth weight infants, Nonwhite infants still have a higher mortality rate on Sunday, but this did not hold true for White infants who were norma l birth weight. The au thor suggested that the increase in mortality for Nonwhite in fants was due to a lack of obstetrical interventions and access to appropriate obstetrical care for Nonwhite mothers. Mangold found that births on Sundays were 10% lower than would have been expected given the distribution of births throughout the week. This was more pronounced for White women than for Nonwhite women. Th is resulted in Sundays having the highest proportion of low birth weight infants compared to other days of the week thereby increasing the neonatal mortality rate. Mangol d also found that Sunday deaths due to septicemia and immaturity were significantly higher than the weekday rates while deaths due to birth defects did not show a pattern of higher mortality on the weekends. Deaths due to septicemia and immaturity are more susceptible to quality of clinical care.

PAGE 30

16 In 1983, Mathers examined the distribution of both births and perinatal deaths by day of week. He examined almost 900,000 bi rths in Australia from 1976-79 and found a similar pattern of births by da y of week and infant deaths by day of week of the birth. The weekends had 21% fewer births than the weekday. This resulted in a higher proportion of low birth weight births on the weekend (6% versus 5.1% on the weekday) and subsequently a higher neonatal mortality rate on the weekend (9.87 v. 7.67). This rate was higher on the weekend for low birth wei ght infants, normal birth weight infants, early neonatal deaths and late neonatal deat hs. The higher rate for low birth weight infants persisted when infants with birth defects were excluded from the analyses. Mathers suggested that the higher proportion of low birth weight births on the weekends was due to obstetrical interventions such as inductions and c-sections during the week. The higher weekend mortality rate for all in fants was linked to lower numbers of staff and subsequent quality of care in hospitals on the weekends. Mathers examined if the day of death had an impact on the weekend rate and found no association (Mathers, 1983). Dowding, Duignan, Henry, & MacDonald (1987) did not focus on higher weekend mortality but instead looked at vari ability across each day of the week. Their study adjusted for infant birth weight, whethe r the birth was electiv e or spontaneous and whether the infant was normally formed or had a lethal birth defect. They did not adjust for gestational age. They found a similar pattern of births with the weekend having fewer births and elective deliveries being higher on the weekdays. Low birth weight proportions were the lowest on Mondays (3.8%) and the highest on Saturdays (6.8%). Sundays and Wednesdays had the next highest lo w birth weight rate at 5.6%.

PAGE 31

17 When analyzing deaths by weekend versus weekday, the perinatal mortality rate was 17% higher on the weekends (14.9 v. 12.7). By day of week, the overall perinatal mortality rate was the highest on Wednesday and Saturday while Sundays rate was close to the average rate for the study cohort. Th ere was no association in day of birth and perinatal mortality for spontaneous deliveries or infants without a birth defect. However, for induced births, the weekend perinatal mo rtality rate was almost 80% higher (30.9 v. 17.2). Induced births represented high-risk pr egnancies and lethal anomalies. This is supported by the fact that Wednesday had the highest number of inductions for high-risk indicators and the highest perinatal mortalit y rate. The higher mortality on Saturday was due to the case mix with a higher proportion of high-risk cases among elective deliveries. A study of 1.6 million live births in Ca lifornia using the 1995-1997 linked birth and infant death file found a higher crude odds ratio for weekend births compared to weekday births (OR=1.12, 95% CI=1.05-1.19). The higher rate was evident for both vaginal and c-section deliveri es. This apparent difference between weekend and weekday mortality disappeared after adjustment for birth weight (O R=1.01, 95% CI=.95-1.08). This was true for both vaginal and c-sec tion deliveries. The increase in weekend mortality for larger birt h weight births (births 4500 grams) disappeared after controlling for birth defects. (Gould, Qin, Marks, & Chavez, 2003). Luo, Liu, Wilkins, & Kramer (2004) examined whether stillbirth or neonatal death varied by day of the week of birth. Their study adjusted for the effects of gestation, birth weight and cause of d eath and included both spontaneous and induced births. Births occurring on each day of the week followed the pattern found in other research already

PAGE 32

18 cited. Weekend births were 25% lower than weekday births with Sunday having the lowest proportion of births. The case mix fo r weekend births showed a higher proportion of high-risk infants with a 14% higher pret erm birth rate and a 7% higher low birth weight rate than weekday births. The crude relative risk (RR) was 6% higher on the weekends (6.0 v. 5.7) for stillbirths. Asphyxia deaths were the highest on weekends with deaths due to birth defects lower on the weekends. The crude RR for neonatal mortality was 11% higher on the weekends (3.8 v. 3.4). For weekend deaths, bo th deaths due to asphyxia and immaturity were higher than weekday deaths However, after adjusting for gestational age, there was no difference in the RR between weekend and weekday births. Further adjustment for maternal demographic factors did not cha nge the RR. The excess risk of death for weekend births appeared to be due to a hi gher proportion of preterm, low birth weight infants on the weekends. The shift in cas e mix was most likely due to elective interventions such as inductions and c-sections during the week. Hong et al. (2006) found that perinatal mortality was higher for births that occurred during holidays (Sunday, holidays) than those on the weekday. This study was conducted in Korea where the regular work w eek includes Saturday. After adjusting for infant birth weight, gender, number of birt hs and birth defects, the odds of dying for weekend births was 20% higher than for holiday births (OR=1.2; 95% CI=1.1-1.3). The higher odds of death for holiday births wa s observed for both low birth weight and normal birth weight infants. In adjusted m odels, the odds ratio for both groups of infants was 1.2 with normal birth weight infants having a confidence limit just below the level of

PAGE 33

19 significance (95% CI=1.0-1.4). As in other studies, Sunday had the lowest number of births of any day of the week. Nighttime Births and Neonatal Mortality Heller, Misselwitz and Schmidt (2000) analyzed data on 380,930 births in Germany by comparing infants born at ni ght with infants born during the day and whether death occurred within the first seven days of birth. Infants born at night had an 86% higher likelihood of death within the first seven days after birth (OR=1.86, 95% CI=1.10-3.13). For infants born at night there wa s almost a four-fold increase in the risk of death due to asphyxia (OR=3.89, 95% CI=1.51-10.03) compared to infants born during the day. However, there were very few deaths due to asphyxia (n=21) which resulted in a wide confidence limit. This study excluded preterm infants, infants with a birth defect, infants born by caesarean delivery and fetal deaths. The higher likelihood of death persisted even without th e inclusion of infants with ri sk factors for neonatal death. A 2006 study found an association between infant death resulting from fetal injuries and nighttime admissions in the state of Florida. Investig ators used the NICA (Birth-Related Neurological Injury Compensa tion Association) database to review the records of infants who had died as a result of injuries sustained during birth. This database only includes infants who weighed at least 2,500 grams. Infants with a brain or spinal cord injury suffered at birth, who had subsequently died, were compared to infants with a brain or spinal cord injury who had not died. The injury had to be the result of a lack of oxygen or mechanical injury which occurred during birth. Nighttime was defined

PAGE 34

20 as the period of 11:00 pm to 8:00 am. Infants w ho died were twice as likely to have been born during the nighttime period as the control group (OR=2.09; 95% CI=1.29-3.40) (Urato, Craigo, Chelmow, & O'Brien, 2006). Nighttime NICU Admissions and Neonatal Mortality Researchers have begun to focus on ne onatal intensive care unit (NICU) admissions and if time of admission is relate d to neonatal mortalit y. Lee and colleagues (2003) examined daytime versus nighttime admi ssions to the neonatal intensive care unit and subsequent neonatal death. This study examined almost 5,200 infants between 24-32 weeks gestation. Infants with lethal anoma lies, those less than 23 weeks gestation and those on life support were excluded from the study. The logistic regression models adjusted for sex, gestation and size for gesta tion, birth defects, Apgar score at 5 minutes, outborn status, antenatal treatment with steroids, multiple birth, cesarean delivery, SNAPII score, neonatal nurse practitioner use, in-house presence of a neonatalologist. Daytime births were those births occurri ng from 8 am 5 pm and nighttime births were those births occurring from 5 pm 8 am. Of the patients admitted to the NICU, 60% were admitted during the night. Compared to the night admissions, day admissions were more likely to be outborn, be multiple s, delivered by c-section, and have a lower SNAP-II score. Nighttime admissions had a si gnificantly higher mortality rate than daytime admissions (5.4% versus 4.0%, p<.05). The odds ratio for early evening deaths (5 pm to midnight) was 1.3 (95% CI=0.9-2.1) and for the late night pe riod it increased to 1.6 (95% CI=1.1-2.4) (midnight-8 am). Only th e odds ratio for late night admissions was

PAGE 35

21 significant. For outborn infants, night admissi on was not predictive of neonatal death. For inborn infants, the presence of an in-house physician was protective reducing the odds of death by 40% (OR=0.6, 95% CI=0.4 -0.9) (Lee et al., 2003). However, a recent study in Australia found that night admissions to a regional network of NICUs (N=10) did not increase th e likelihood of death prior to discharge. This study focused on infants who were born at less than 32 weeks gestation. The study grouped holiday, weekend and night admissions together and compared infants admitted during these days/times to infants admitted during the weekday. Daytime was defined as 8:00 am 6:00 pm, Monday Friday. Nighttime was defined as public holidays, weekends (Saturday or Sunday), and after hours (6:01 pm 7:59 am). NICUs in this network are tertiary car e centers (Abdel-Latif Bajuk, & Lui, 2006). There was no significant association betw een admission during night time versus daytime hours (OR=1.069; 95% CI=0.881-1.297). Th is was true for each of three separate periods of gestation: 22-26 weeks, 27-29 weeks, 30-31 weeks. In multivariate analysis, items that were significantly associ ated with early neonata l death (first seven days after birth) were the lack of treatment with antenatal corticosteriods within seven days of delivery, small for gestational age infant, gestations of 22-26 weeks and 27-29 weeks compared to 30-31 weeks and male ge nder. Outborn status of the infant and emergency caesarean delivery were not signifi cantly associated with early neonatal death.

PAGE 36

22 Weekend/Evening Admissions and In-Hospital Mo rtality in Other Adult Acute Care Units Bell and Redelmeier (2001) found that patients admitted to the emergency department on the weekends had a higher mortality than patients admitted during the weekday after adjusting for the patients level of risk, gender and age. Their study investigated the mortality rates for patients admitted with three acute conditions: ruptured abdominal aortic aneurysm, acute epiglotti tis and pulmonary embolism. Adjusted odds ratios for mortality ranged from 1.1 9 to 5.28 (95% CI=1.03-1.36 and 1.01-27.5) for patients admitted during the weekend comp ared with weekday. When the authors analyzed those deaths that were within tw o days after admission, the likelihood of death was even higher for weekend admissions than weekday admissions. However, higher odds ratios were associated with wider c onfidence limits and less precise estimates. The difference between the unadjusted and adjusted odds ratios was small leading the authors to conclude that patients adm itted on the weekend were not significantly more ill than those admitted during the weekday. They attributed the higher mortality to lower staffing levels, staff with less seniorit y, the use of pool workers (non-emergency department workers) and fewer supervisors on the weekends than during the weekdays. They also concluded that patients with medi cal conditions with high case fatality rates had a higher odds of death if admitted dur ing the weekend (Bell & Redelmeier, 2001). Infant Characteristics and Neonatal Mortality Infant characteristics and how they are re lated to neonatal mortality are discussed in this section. Those character istics include birth weight, ge station, Apgar scores, lethal

PAGE 37

23 congenial anomalies, multiple births, race and gender. In addition, since many of these variables are used to adjust for illness seve rity, a discussion of other risk adjustment methods using physiological measures, a nd how they compare to infant level characteristics, is provided. In order to compare outcomes between di fferent groups of ne onates or different clinical settings, it is necessary to adjust for the patients le vel of risk. Risk adjustment separates patients into different levels or strata of risk based on certain measures of acuity or illness severity (Richardson, Tarn ow-Mordi, & Shoo, 1999). There are various methods to adjust for patient acuity. Studies ha ve used infant characteristics at admission to the NICU such as birth weight, gestati on and gender. Specific measures have been developed which use the infants physiology wi thin a certain time period after admission to the NICU. Others have included obstetric practices, such as ante natal steroid use, or delivery room measures such as the Apgar score. Acuity Several studies have found that the patie nts acuity or illness severity was the strongest predicator of mortality These studies used different risk-adjustment methods to control for acuity while investigating the e ffects of different hosp ital-level variables on mortality. This finding was cons istent across different type s of hospitals, study designs and countries.

PAGE 38

24 In a multilevel analysis of hospitals in Sweden, larger regional hospitals with NICUs had a higher risk of neona tal death in high-risk deliveries. After adjusting for case mix or illness severity, this higher risk wa s no longer significant (Merlo et al., 2005). In a study of quality of care for very pr eterm infants (27 and 28 weeks gestation), risk adjustment was calculated using infants characteristics prior to NICU admission. These included male gender, small for gestational age (< 5th percentile), and infants respiration and heart rate at five minutes. Since this study was interested in the quality of care and its impact on mortality, clinical measur es of acuity, such as the CRIB, were not used. These measures use physiological measur es collected within the first 12 hours after admission to the NICU and therefore incl ude the impact of early patient care. The study found that quality of care did not impact survival when resuscitation and surfactant therapy were considered. Howe ver, there was an increase in mortality related to poor care regard ing control of the infants temperature (AOR=1.71, 95% CI=1.21-2.43), appropriate ve ntilation monitored through blood gas measurements and adjustments (AOR=3.29, 95% CI =1.97-5.49) and provision of cardiovascular life support monitored through blood pressure measurements and adjustments (AOR=2.37, 95% CI=1.36-4.13) (Acolet et al., 2005). A 2003 study in Thailand found that the strongest predicator of mortality in adult patients in both medical and surgical units wa s the patients acuity or illness severity. Using the APACHE III scoring system for cr itically ill patients, patients with higher scores were almost seven times more likely to die prior to discharge (AOR=6.8, 95% CI= 5.087-9.179) in multivariate models (Sasichay-Akkadeshanunt, Scalzi, & Jawad, 2003).

PAGE 39

25 Birth weight and gestation. Low birth weight is a lead ing cause of infant death and childhood disability (Childrens Defense Fund, 1992). An infants bi rth weight and gestational age are both independent predicators of infant mortality a nd considered to be the two most important predicators of an infants subsequent hea lth and survival (Mat hews et al., 2004). Previous research has shown that a bi rth weight of less than 2,500 grams is predictive of death within the first m onth of life (Mugford, Szczepura, Lodwick, & Stilwell, 1988). Horbar, Badger, Lewit, Rogowski, & Shiono (1997) found that for each 100 gram increase in birth weight (only 5011,500 gram infants included), the odds of neonatal death after NICU admission dec lined by 36% (AOR= 0.64, 95% CI=0.60-0.67). Validation studies of physiological measures of illness severity have found that birth weight and gestation are inde pendent predicators of neona tal death (Cockburn et al., 1993). In the absence of physiological measur es, many studies have used low birth weight or very low birth wei ght as a measure for acuity (M ugford et al., 1988; Phibbs, Bronstein, Buxton, & Phibbs, 1996) or used bi rth weight with a combination of NICU preadmission characteristic s (Horbar et al., 1997). The Vermont Oxford Network (VON) consis ts of 353 hospitals with NICUs in 49 states and 22 foreign counties. The risk -adjusted model used by VON consists of nonphysiological measures which includes gesta tional age, small for gestational age (less than 10th percentile for gestation by race and gend er), Apgar score at 1 minute, multiple birth, race, gender, major birth defect presen t, prenatal care and vaginal delivery. This

PAGE 40

26 risk-adjusted model compares favorably wi th the Score for Neona tal Acute Physiology which uses physiological measures to asse ss acuity. The score for the area under the curve (AUC) for the VON risk-adjustment m odel is 0.89. The AUC measures a models sensitivity and specificity in accurately predic ting a specific outcome with a score of 1.0 indicating the model is 100% accurate in pred icting the outcome of interest. This model was developed for infants who are less than 1,500 grams (Rogowski, Horbar et al., 2004; Rogowski, Staiger, & Horbar, 2004). Richardson and colleagues found that incr easing gestational age was associated with a decline in illness severity as defi ned by the Score for Ne onatal Acute Physiology. They used gestation instead of birth weight, since it is a m easure of fetal maturity. They added SGA to add the residual effect of birth weight not meas ured using gestation alone. Both measures added to the SNAP score with increasing gestation reducing the score by 1.5 points and SGA increasing the score by 3.6 points (Richardson, Shah et al., 1999). Apgar scores. The Apgar score, developed by Virginia Apgar and published in 1953, evaluates the infants heart rate, respiration, muscle tone, response to st imuli or reflex irritability and color at birth. The infant is rated at one minutes, fi ve minutes and ten minutes. A score of seven to ten is Excellent ; a score of four to six is Fair and a score less than four is Poor Condition with zero indicating death. The fo llowing table presents the scoring system and interpretation of the Apgar sc ore (American Academy of Pediatrics &

PAGE 41

27 American College of Obstetricians and Gyn ecologists, 2006; Baskett, 2000; Finster & Wood, 2005). Table 1 Apgar Scoring System Score Sign 0 1 2 A ppearance Color Blue or Pale Body Pink; Extremities Blue Completely Pink P ulse Heart Rate Absent Slow (< 100 beats/minute) Greater than 100 Beats per Minute G rimace Reflex Irritability No Response Grimace; Some Motion Cry or Active Withdrawal A ctivity Muscle Tone Limp Some Flexion of Extremities Well Flexed; Active motion R espiration Respiratory Effort Absent Slow, Irregular; Weak Cry; Hypoventilation Good, Strong Crying Source: American Academy of Pediat rics & American College of Obstet ricians and Gynecologists, 2006; Baskett, 2000 and Finster & Wood, 2005. In 2006, the American Academy of Pediat rics and the American College of Obstetricians and Gynecologist s published a joint policy statem ent on the Apgar score, its appropriate use and utility 53 years afte r it was published. The Apgar score was considered to be an expression of the infa nts physiological c ondition, has a limited time frame, and includes subjective components a nd was intended to be used to assess a neonates condition only at a sp ecific point in time, shortly after birth. The five minute score of 7-10 was considered normal and th e authors acknowledged a correlation between a score of 0-3 at five minutes and subsequent neonatal death. The authors also cited the study by Casey, McIntire and Leveno (discussed below) which found that a low Apgar score at 5-minutes increased the odds of neona tal death in both preterm and term infants

PAGE 42

28 (American Academy of Pediatrics & Am erican College of Obstetricians and Gynecologists, 2006). Thorngern-Jerneck and Herbst (2001) showed an increased neonatal mortality for term, normal birth weight infants with Apgar scores less than 7 at five minutes. These infants had an increased risk of death 14 times higher than term, normal birth weight infants with Apgar scores 7 or above (OR 14.4, 95% CI=12.5, 16.5). Nighttime births (5:00 pm to 6:59 am), had a 21% higher risk of having a low Apgar score (OR=1.21, 95% CI=1.14-1.29) compared to day births (8:30 am 3:59 pm). The odds of a higher Apgar score showed a curvilinea r relationship with birth weight being higher at both the low birth weight (< 2,500 grams) and higher birth weight (> 3,500) ranges. Casey, McIntire and Leveno (2001) found th at the incidence of neonatal death increased as the 5-minute Apgar score d ecreased. In a study of over 150,000 singleton infants, this negative relati onship was true for both preterm and term infants. The neonatal mortality rate per 1,000 live births for both preterm and term infants by Apgar score is given in the table below. Table 2 Incidence of Neonatal Death by Gestational Age and Apgar Score Neonatal Mortality Rate per 1,000 Live Births Apgar Score at 5-Minutes Preterm Infants Term Infants 0-3 315.0 244.0 4-5 72.0 9.0 7-10 5.0 0.2 Source: Casey, McIntire and Leveno, 2001. When infants with an Apgar score of 710 were the reference group, the adjusted odds ratio for neonatal death for preterm infa nts with a score of 0-3 was 59 (95% CI=40-

PAGE 43

29 87) and for term infants with a score of 0-3 was 1460 (95% CI=835-2555). Due to the small numbers of neonatal deaths in th e cohort, confidence intervals are wide. Physiological measures. There are two measures of illness severity developed for use with neonates. These measures were developed to allow research ers to control for acuity while studying the quality of care across differen t neonatal intensive care units or within one unit over time (Cockburn et al., 1993). The Clinical Risk Index for Babies ( CRIB) was developed using a cohort of infants who were very low birth weight in fants (< 1,500 grams) and preterm (< 31 weeks gestation). This scoring system was developed with a cohort of infants at tertiary care hospitals in the United Kingdom. Six variab les are measured within 12 hours of admission to the NICU: birth weight, ge station, presence of a birth defect, maximum/minimum fraction of inspired oxygen and maximum base excess. The area under the receiver operating cu rve for the CRIB was .90 ( SE =0.05) in the validation cohort. The ROC measures the sensitivity and specificity or the accuracy of prediction. This compares to the ROC for birth weight alone which was .78 ( SE =0.03) (Cockburn et al., 1993). Kaaresen, Dohlen, Fundingsrud and Dahl (1998) used the CRIB to assess the quality of care in one neonatal intensive ca re unit over time. The ROC for the CRIB was 0.88 ( SE =0.02) compared to 0.72 ( SE =0.04) for birth weight alone and 0.71 ( SE =0.03) for gestation alone. Other variables indepe ndently associated with survival, after

PAGE 44

30 controlling for acuity using the CRIB, were female gender (AOR 2.4, 95% CI=1.2-4.9) and surfactant treatment (AOR 2.9, 95% CI= 1.1-7.5). Infants treated with antenatal steroids had an 80% higher odds of survival than infants not treate d, after controlling for birth weight, gestation and gender. However, when the CRIB was entered into the model, treatment with antenatal ster oids was no longer significant ( p =0.397). Another scoring system for mortality risk in neonates is the Score for Neonatal Acute Physiology (SNAP). The SNAP was validat ed for infants of all birth weights. The index has been modified over time with later versions incl uding non-physiological measures (SNAP-II, SNAP-PE and SNAPPE-II). The later versions collect data in the first 12 hours after NICU admission, compared to the first 24 hours of the SNAP. Physiological measures include: mean blood pressure, lowest temperature, fraction of inspired oxygen, lowest serum pH, presence of multiple seizures, and urine output. The later two versions include perinatal extensi on measures such as birth weight, small for gestational age and low Apgar score at five minutes (< 7). To control for the high correlation between birth weight and gestational age, five birth weight categories were used and small for gestational age was a dded to allow for any residual effects of gestation. Only birth weight categorie s of < 750 grams and 750-999 grams were significant predicators of mortality. The RO C for the SNAPPE-II for all birth weights was 0.91 ( SE =0.01) compared to 0.78 ( SE =0.01) for birth weight alone. When birth weight, SGA and Apgar score were used, the ROC was 0.84 (Richardson, Corcoran, Escobar, & Shoo, 2001).

PAGE 45

31 Both the CRIB and SNAP are used along with other measures of acuity such as birth weight, gestation, APGAR sc ores and presence of birth defects in order to adjust for total risk of mortality ((Richardson, Tarnow-M ordi et al., 1999). Therefore, physiological measures are part of an overa ll risk adjustment methodology. Congenital Anomalies Congenital anomalies were the leading cause of death in the first year of life in the most recent national data (Mathews et al., 2004). Anomalies, controlled for in other studies due to their higher mortality rate as compared to other anomalies, are anencephaly, renal agenesis, trisomy 13 and 18 (Cockburn et al., 1993; Sanderson, Sappenfield, Jespersen, Liu, & Baker, 2000). In the research on ne onatal mortality and day of birth or time of birth, st udies used the presence of a bi rth defect as part of their risk-adjustment methodology. In some of these studies, infants with malformations that were considered lethal were excluded fr om the study (Gould et al., 2003; Hamilton & Restrepo, 2003; Heller et al., 2000; Hong et al., 2006; Lee et al., 2003; Stephansson, Dickman, Johansson, Kieler & Cnattingius, 2003). Multiple Births Multiple births (twins, triplets or higher) have a higher incidence of prematurity, low birth weight and subseque nt infant death. Data from the 2002 linked United States birth and infant death certificates showed that the infant mortality rate was five times higher for multiple than for singleton births (32.3 per 1000 live births versus 6.1). This

PAGE 46

32 higher rate was constant across all racial and ethnic groups. As the number of fetuses increases, so does the risk of infant death. Triplets and quadruplets were 10 to 26 times more likely to die in the first year of life than singleton births (Mathews et al., 2004). Gender and Race Studies have found both female gender and Black race as independent predictors of survival with female Black infants having the highest survival advantage and male White infants having the lowest (Mathews et al., 2004; Morse et al., 2006). Black infants are born premature and low birth weight more often than White infants and have twice the rate of extremely preterm (< 33 weeks) and very low birth weight (<1,500 grams) infants than White women. White women consis tently deliver higher birth weight infants even at the same gestational age and level of risk as Black women. However, preterm and low birth weight Black infant s have a higher rate of surv ival during the neonatal period than their White or Hispanic counterparts and that survival advantage seems highest at the lower gestational ages. While at birth weights of less than 500 grams, Black and White infants have similar rates of mortality, Black infants have from 62% to 70% of the birth weight and gestational ag e-specific neonatal mortality rates of White infants when birth weights between 500 grams and 2,500 gram s are considered (Alexander et al., 2003; Alexander, Kogan et al., 1999). Even with a better survival advantage at lower gestational ages, the overall neonatal mortality rate for Black infants is still higher than for White infants. The gap between the two groups appears to be widening even though the overall neonatal

PAGE 47

33 mortality rate is declining. Th is widening disparity may be due to differential responses of Black and White infants to high-risk obstetrical and neonatal care such as surfactant treatment (Alexander, Tompkins et al., 1999; Allen, Alexander, Tompkins, & Hulsey, 2000; Mathews et al., 2004). In a study of Florida infants who weighe d less than 1,000 grams, both Black race and female gender were predictive of higher odds of survival. Black infants were 30% more likely to survive compared to White infants (OR=1.3; 95% CI=1.1-1.5) and female infants had a 70% increased survival advant age over male infants (OR=1.7, 95% CI=1.51.9) (Morse et al., 2006). International studies also show an increased survival advantage for female infants. A Swedish study of over 175,000 births found that male infants had a higher neonatal mortality rate than female infants. This hi gher rate may be due to the higher rate of preterm births among male infants (Ingemar sson, 2003). This may lead to decreased lung maturity. Elsmen, Pupp and Hellstrom-Westas (2004) found that male infants required more mechanical ventilation in the first six hours of life than female infants (60.8% versus 46.2%, p=0.026) even though a higher percenta ge of their mothers received antenatal steroid treatment. Hospital Characteristic s and Neonatal Mortality Hospital characteristics associated with neonatal mortality include the level of care provided by the neonatal intensive care unit, volume of NICU admissions, obstetrical interventions, nurse staffing patte rns and physician staffing patterns. Each of

PAGE 48

34 these is discussed in detail in this section in cluding their relationshi p with neonatal deaths or other adverse patient outcomes, a nd their measurement in other studies. Level of Care There are three levels of care specified for the care of neonates. Level I facilities provide health care for newborns who are he althy or have no ma jor medical problems. These facilities do not have neonatal intensiv e care units (NICU). NI CUs are divided into Level II and Level III facilities. Level II NICUs provide care for neonates who are moderately sick but do not need greater than four hours of assisted ventilation. Level III centers are also called regional or tertiary car e centers. They provide the most advanced and comprehensive care for ill neonates (P hibbs et al., 1996). The designation of a facility also includes a weight criterion. Level III facilitie s in Florida are equipped and staffed to care for neonates who are less than 1,000 grams (R. Nelson, personnel communication, 2007). Infants who are born in a hospital with a neonatal intensive care unit have a higher survival than those infants who are transferred from another hospital without a Level II or Level III NICU (Mugford et al., 1988). After control ling for patient acuity using the CRIB, researchers from the International Neonatal Network in the United Kingdom found that non-tertiary NICUs had a ri sk-adjusted odds of mortality twice as high as tertiary care centers (AOR=2.12, 95% CI= .39-3.24). This study included both inborn patients and transfers (Cockburn et al., 1993). Others have found no association in multivariate models between out-born status and neonatal deaths (Horbar et al., 1997).

PAGE 49

35 A 1998 study investigating reasons for the de clines in neonatal deaths attributed two-thirds of the decline in mortality for very low birth weight infants to the better care hypothesis defined as advances in neonatal care. The authors attribute therapeutic interventions such as the use of surfactants and new respiratory s upport technologies such as Continuous Positive Airway Pressure and High-Frequency Ventilation as underlying reasons for better care and improved surviv al among these infants (Richardson et al., 1998). This type of care is f ound in all Level III NICUs. In a 2000 study of neonatal death and leve l of hospital of delivery, neonatal mortality was higher in Level I or Level II hosp itals compared to Level III hospitals with 24 hour physician coverage for infants wei ghing between 500-1,499 grams (Sanderson et al., 2000). A study in Swedish hospitals af ter regionalization s howed that higher mortality levels in large, tertiary care centers was completely explained by case mix. Only high-risk infants were considered in th is model. When all patients were considered, regional centers with access to all neonatal services had the lowest neonatal mortality compared to smaller hospitals without neonatal care (Merlo et al., 2005). Volume of NICU Admissions Phibbs, Bronstein, Buxton and Phibbs ( 1996) found that Level III NICUs with high volume (at least 15 patients) had the lowe st mortality in risk adjusted models. Mortality was almost 40% lower for patients in tertiary care centers with high volume compared to hospitals with lower volumes and lower levels of care (AOR=0.62, 95% CI=0.46-0.82). When Level III units with higher volume were the referent group,

PAGE 50

36 mortality was 42% to 60% higher for hospitals with Level I, Level II or Level II+ units with both low (< 5 pa tients) and high ( 15) volume. However, 10% of the unmatched cases in the study were low birth weight comp ared to 5% of the matched cases being low birth weight. If low birth weight is dispropo rtionately related to large, tertiary care centers, then the results would be biased toward lower mortality in those centers. In a study conducted by researchers at RAND Corporation, the volume of admissions to NICUs was found to be a poor pr edicator of mortality. Using the Vermont Oxford Database to assess variations in mortality at 332 NICUs in the United States, volume of admission explained only nine percen t of the variation in these rates. There was a threshold of at least 50 very low birt h weight admissions per year where mortality was lower for those NICUs that exceeded this limit. The historical mortality rates at these NICUs was the best predicator of future mo rtality rates (Rogowski, Staiger et al., 2004). However, a 2002 study by the UK Neona tal Staffing Study Group found that patient volume, as measured by maximum occupancy (number of infants in unit on day of birth divided by the maximum occupancy of the unit during th e study period) was related to mortality in a study of 186 NICU s in the United Kingdom. The odds of dying on the day of birth increased by 9% for every 10% increase in maximum occupancy (AOR=1.09, 95% CI=1.01-1.18). Almost every NICU in this study had higher occupancy than the established number of beds. Therefore, these results may relate more to crowding than to volume. Patient volume, as measured by the number of <1,500 gram infant admissions per year, was not a signi ficant predictor in any of the models.

PAGE 51

37 Obstetrical Interventions and Role in Birth Patterns The case mix or illness severity of infants admitted to the NICU is dependent upon obstetrical practices and care provided prio r to delivery and in the delivery room (Richardson, Shah et al., 1999). Research th at has examined the relationship between the timing of birth and subsequent neonatal deat h has observed a pattern of a higher number of births during the week than the weekend. MacFarlane (1978) f ound that the ratio of births (average number of births on each day of the week to the average number of births per day over an entire year) were the lo west on Sunday with a ratio of .77 in 1976. A ratio of one indicates equal bi rth patterns across each day of the week. A ratio less than one indicates fewer births than expected while a ratio of more than one indicates more births than expected. Researchers have obser ved that this type of pattern is due to obstetrical interventi ons during the week. Joyce, Webb and Peacock (2004) found a negative relationship between obstetrical interventions, the availability of obstetricians and stillbirth rates. With 65 English hospitals included in the study, those with hi gher levels of intrapartum interventions ( = -0.21, p=0.003) and higher numbers of obstetricians available for consultations ( = -055, p =0.23) had stillbirth rates that we re lower. This relationship did not hold true for the neonatal mortality rate. Th is reduced stillbirth rate would lead to an increase in live-born infants who may need NICU care. Richardson and colleagues found that the recent reductions in neonatal mortality were due, in part, to obstetrical interventions such as using tocolytics to delay labor and using antenatal corticosterioids to improve lung function prior to birth. They attribute

PAGE 52

38 one-third of the reduction in neonatal deaths to the better babies hypothesis. Therefore, infants arrive in the NICU in better conditi on with improved survival (Richardson et al., 1998). In another study, the use of prenatal st eroids provided a protective effect on the Agpar scores of newborns (AOR=0.50, CI= 0.37-0.68) (Richardson, Shah et al., 1999). Recent studies have explored the outcomes for term infants born by cesarean versus vaginal birth. Kolas, Saugstad, Daitvei t and Nilsen (2006) f ound that infants born by planned cesarean had higher NICU admissions and neonatal mortality rates than infants born through a planned vaginal deli very. Using almost 19,000 term deliveries from 24 obstetric units in Norway, the aut hors found that term infants born by planned cesarean had a 74% higher odds of admission to the NICU than term infants born by planned vaginal delivery (AOR=1.74, 95% CI=1.38, 2.18). When women having a pregnancy that could be consid ered high-risk were excluded fr om the analysis, the results did not change. These infants had twice the risk of having a pulmonary disorder until 39 weeks gestation when the increased risk was no longer significant. MacDorman, Declercq, Menacker and Ma lloy (2006) found that term infants born through cesarean had higher neona tal mortality rates compared to vaginal births. They used a cohort of almost 5.8 million US births limiting the analysis to women with no indicated risk. These are women having singl eton, term infants in vertex presentation without medical complications or risk factors. Infants bor n by cesarean delivery had a neonatal mortality rate almost 3 times highe r than those born by vaginal delivery (1.77 per 1,000 live births versus 0.62 per 1,000 live bi rths). The disparity in these two rates was higher for infants of multiparous women where the neonatal mortality rate was 3.7

PAGE 53

39 times higher. For infants of primiparous wo men, it was 2.2 times higher if the infant was born by cesarean versus vaginal delivery. The authors conducted multivariate analys is to understand the difference in neonatal deaths between these two groups of infants (cesarean versus vaginal delivery). After adjusting for infants birth weight and gestation, mothers age, race/ethnicity, smoking behavior, educational level and parit y, term infants born by cesarean had at least 2 times the odds of dying during the neonatal period than term infants born by vaginal delivery. For the entire cohor t, the odds ratio was 2.71 ( 95% CI=2.43-3.02). When infants with congenital anomalies were excluded from the analysis, the odds ratio decreased to 2.63 (95% CI=2.23-3.10). When infants with congenital anomalies and low Apgar scores (<4) were excluded from the analysis, th e odds ratio decreased to 2.02 (95% CI=1.602.55). The later two models indicaed that the increased risk of mortality was due, in part, to birth defects and poor condition at birth. Howe ver, the infants cond ition at birth only explained 25% of the variance in neonatal mortality between the two groups of infants. These obstetrical interventions affect the case mix in the NICU by improving survival in the delivery room thereby increa sing NICU admissions of infants who would have previously died. These st udies do not provide evidence of which infants would have died due to stillbirth or which infants di ed because they were induced too early. Relationship of Nurse Staffing Patterns to In-Hospital Mortality With changes in the healthcare system including the shortage of qualified nursing personnel in hospitals, an incr ease in the acuity of hosp italized patients and hospital

PAGE 54

40 reengineering efforts resulting in changes in clinical staffing patterns, many studies have investigated the role of skill ed nursing in reducing adverse patient outcomes, including mortality. Studies have focused on defining the important role of nursing by measuring the skill mix of nursing staff (registered to non-registered nursing personnel), the workload of the registered nursing staff a nd the hours of patient care given by RNs. Skill mix. Skill mix includes the proportion of regist ered nurses to other nursing personnel in the unit, the educational level of nurse s in the unit and the use of nursing positions from outside the unit under st udy. The body of research in th is field indicates that a higher skill mix of nurses, reduced patient to nurse ratios and higher skilled nursing hours are negatively associated with adverse pati ent events including death before hospital discharge (Heinz, 2004; Lang, Hodge, Olson, Ro mano, & Kravitz, 2004). In a review of 22 studies on nurse staffing and patient outcomes, Lankshear, Sheldon and Maynard (2005) found that more registered nurses as a proportion of all nursing personnel and more registered nurses per patient resulted in reduced mortality. Tourangeau, Cranley and Jeffs (2006) found that seven out of the ten studies reviewed found a significant association between nursing skill mix and lo wer patient mortality. The most common measure used was nursing skill mix calculated as the number of registered nurses divided by all nursing staff. Aiken and colleagues (2003) found that the proportion of at-leas t Bachelor-level prepared registered nurses in surgical units was inversely related to mortality within 30

PAGE 55

41 days after admission. Patient mortality decr eased by 5% for each 10% increase in the proportion of nurses with at-leas t a Bachelors degree after ad justing for both patient and hospital characteristics (AOR=0.95, 95% CI =0.91-0.99). Nursing experience was not related to patient mortality in this study. Using hierarchical linear regressi on, Estabrooks and colleagues (2005) demonstrated that each measure of skill mix was significantly related to patient mortality. A higher proportion of RNs to other nursing personnel resulted in a 17% reduction in patient mortality (AOR=0.83, 95% CI=0.73-0.96). There was a 19% reduction in patient mortality when there was a higher propor tion of RNs with a Bachelors Degree (AOR=0.81, 0.68, 0.96). There was a positive rela tionship between a higher proportion of non-unit specific nursing personnel and patient mortality (AOR=1.26, 95% CI=1.091.47) resulting in an increase in odds of mortality by 26%. These results were from a fixed-effects model for patient level characteristics. When patient level effects were allowed to vary across hospitals, the c onclusions stated above did not change. Nurse to patient ratio. Nurse to patient ratio has been measur ed directly as the number of patients assigned to each registered nurse or indirectly using the tota l number of patients or beds in a unit divided by the total number of regist ered nurses in the unit for the shift (Carmel & Rowan, 2001). In a 2003 study in Australia found that survival was improved by 82% with the highest infant to nurse ra tio of 1.71-1.97 (AOR= 0.18, 95% CI=0.06-0.50). Similar improvements were found for the medium (1.59-1.70) staffing ratios with a

PAGE 56

42 reduction in mortality of 16% (AOR= 0.84, 95% CI=0.42-1.66). This study examined mortality within the first three days after admission to the NICU for very low birth weight infants while controlling for acuity and unit workload. Acuity was measured using the CRIB. Although 17% of the infants in th e study were transferred to the unit after birth, outborn status was not a significant covariate (Calla ghan, Cartwright, O'Rourke, & Davies, 2003). This study used the number of patients for each nurse instead of the number of nurses for each patient. Callaghan and colleagues (2003) were unable to measure individual infant to nurse workloads during the infants hospital st ay, measured only for the first three days after admission. Instead, their measure was pooled for each infant over each nursing shift during the inpatient stay and measured all nu rses on duty in the unit for each shift and divided that by all infants in th e unit for that particular shift. They then used the CRIB to determine acuity and calculate dependency sc ores by pooling these scores over the shifts for the first three days after admission for each infant. Therefore, the individual needs of each infant during the time period of the study could not be directly calculated. Further, they were not able to measure nursing e xperience or the use of non-NICU nurses. Tucker and colleagues (2002) were unabl e to show an a ssociation between neonatal mortality and nurse to infant ratios. They measured patient workload by dividing the total number of nurses on duty at two time points and dividing that number by the number recommended based on national standard s in the UK. These standards specify the number of nurses recommended per patient ba sed on whether the patient is in intensive care or special care.

PAGE 57

43 In a 2004 study of a pediatric intensive care unit, unit occupancy and patient dependency were associated with an increase in adverse events. Patient dependency was measured as the number of nurses required per bed in four categorie s: non-ventilated or stable patients, mechanically ventilated pati ents, patients needing advanced organ support such as high-frequency oscillatory ventila tion and patients on extracorporeal membrane oxygenation (ECMO). When bed occupancy and patient dependency were considered high, the odds of an adverse event in creased 63% (AOR=1.63, 95% CI=1.03-2.59) (Tibby, Correa-West, Durward, Ferguson, & Murdoch, 2004). Tamow-Mordi, Warden and Shearer (2000) found that when nursing work load increased so did patient mortality in the intensive care unit. Nursing work load was measured by the nursing requirement per bed for each shift and the highest occupancy during the patients stay. These variables were categorized accordi ng to whether or not the occupancy of the ICU reached its maximum capacity and whether the average number of nurses was less than or great er than the required number per bed. At intermediate to high workloads, the odds of death increased by two to three times compared to patients where there was a moderate workload. A 2003 study in Thailand found that the nurse to patient ratio was the only significant predicator of mo rtality among the nurse staffi ng measures studied. Other nursing measures studied and found to be nonsignificant were skill mix, RN experience and proportion of RNs that were Bachelors pr epared. The analysis showed that there was an inverse relationship between the nurse to patient ratio and mortality in both medical and surgical units ( = -1.298, p<.01) (Sasichay-Akka deshanunt et al., 2003).

PAGE 58

44 Aiken, Clarke, Douglas, Sochalski and Silber (2002) found that for each additional patient in the average workload per nurse, the odds of death 30 days post admission increased by 7% (AOR=1.07, 95% CI =1.03-1.12) after controlling for patient and hospital characteristics. This study focu sed on surgical patients and controlled for patient acuity using Diagnostic Related Groups for identifying co-morbid conditions of the patients. Only nurses involved in direct patient care we re included in the study. The addition of other nursing personnel (licensed practical nurses and assistants) did not reduce the odds of death within 30 days after admission. When nursing education was used as a control variable, this relationship between volume and mortality was still significant. The adjusted odds ratio declin ed to 1.06 (95% CI=1.01-1.10) showing a 6% increase in mortality for each additional patient per nurse (Aiken et al., 2003). RN nursing hours. None of the studies reviewed that us ed RN nursing hours as an independent variable were conducted in neonatal or pediatric intensive car e units. Studies presented in this section were done using adult patient populations. Studies have found that an increase in both nursing hours and RN nursing hours has resulted in reduced length of stay and adverse outcomes. There is weaker evidence related to nur sing hours and patient mortality. Lichtig, Knauf, and Miholland ( 1999) found an inverse relationship between total RN nursing hours and patients length of stay. Cho, Ketefiann, Barkauskas and Smith (2003) found that an increase in both nursing hours and the RN hours compared to other nursing personnel resulted in a reduction of pneumonia for patients in hospitals. This

PAGE 59

45 relationship was not significant for other adverse events studied such as pressure ulcers, falls, urinary tract infections, and sepsis. Nu rsing measures were pooled in this study over three separate hospital units: intensive care, coronary care and acute care. In a study of 799 hospitals in 11 states, th ere was a negative relationship between registered nursing hours to patient adverse events such as length of staff, urinary tract infections and hospital-acqui red pneumonia. This was for both the proportion of registered nursing hours to non-registered nu rsing personnel and for the number of RN hours provided per patient day. However, an increase in the proportion of RN hours and an increase in RN hours per patient day did not have a significant impact on patient mortality (Needleman, Buerhaus, Mattke, Stewart, & Zelevinsky, 2002). Tourangeau, Giovannetti, Tu and Wood (2002) investigated total RN nursing hours as a proportion of all nurs ing hours and 30-day patient mo rtality rates. In addition to RN nursing hours being significant, they also found nursing experience and the number of shifts missed were al so important predictors of pa tient mortality. The effect of RN nursing hours was consistent across all t ypes of hospitals in the study. As the proportion of RN hours to all nursing hours increased by 10%, the number of patient deaths declined by five for every 1,000 discharged patients. Relationship of Physician Staffing Patterns to In-Hospital Mortality In addition to the impact of nurse sta ffing patterns on patient outcomes, studies have investigated the impact of physician st affing on the same outcomes. Measures have included the presence of physicians onsite 24 hours per day, seven days per week; the

PAGE 60

46 different levels of physicians (attending, fello w and resident) in a unit (intensive care, surgical); and the ratio of physicians to patien ts. In a review of th e literature on the issue of hospital staffing and mortality, the most common measure used for physician impact was the number of board certified physicians divided by the average daily patient census. Three of five studies found an inverse rela tionship where the higher the percentage of physicians to patients, the lower the mortality. In a review of the research on physicia n staffing and outcomes of both pediatric and adult intensive care unit patients, Pr onovost and colleagues found that 14 of the 15 studies reviewed found a signifi cant association. In units with high intensity staffing of board certified physicians in th e specialty care of cr itically ill patient s, there was a 40% reduction in mortality risk us ing a random-effects, pooled un adjusted relative risk ratio (RR=0.61, 95% CI= 0.50-0.75) (Pronovost et al ., 2002). Two of these studies were related to PICUs and are cited later in this section. Stilwell, Szczepura, & Mugford (1988) found an inverse relationship between the level of pediatricians and mortality of low birth weight infa nts during the perinatal period (fetal deaths > 28 weeks and neonatal deaths during first seven days after birth). Based on their linear regression analysis, the rate of perinatal mortality (8 per 1,000 live births) decreased by .22 for every unit increase in the ratio of pediatri cians per 10,000 births. However, for very low birth weight infants, only birth weight was a significant predicator of perinatal death. When the ratio of pediatricians per the to tal low birth weight infants in the unit was used as an independent variable to expl ain first week mortality, it explained 60% of

PAGE 61

47 the variance in those deaths. When birth weight as a ratio of very lo w birth weight or low birth weight to all births in the unit was added to the equation, it was not significant. Staffing patterns related to nursing were not significant. In a follow-up study, they failed to find pediatrican staffing levels as a significant predicator of mortality after regionalization of neonatal care had occurred in England. The authors theorized that after regionalization women experien cing high-risk pregnancies were more likely to be transferred to hospitals with NICUs prior to the delivery, thereby reducing the effect of those units with higher pediatric st affing levels (Mugford et al., 1988). In a study that investigat ed over 5,000 admissions to 16 different pediatric intensive care units (PICU) in the United States, the authors found that the presence of a pediatric intensivist in th e PICU decreased the odds of mortality by 35% (AOR=0.65, 95% CI=0.44-0.95). There was also a higher odds ratio of death for children in PICUs at teaching hospitals (AOR=1.79, 95% CI=1.23-2.61) However, this was not significant when resident care was included in the model ( p=.39) indicating that the higher risk of death was due to care provided by resident s, not due to receiving care in a teaching hospital. To further investigat e this finding, the authors stratif ied the analysis by year of residency and month of the year. They found increased odds of mortality when care was provided by 1st and 2nd year residents. The risk of mo rtality was also higher when care was provided by residents during the first three months (July-September) of the academic year compared to the last three months (April-June). In the study, 50% of teaching hospitals had resident coverage but no inte nsivist coverage. The volume in the PICU and the transfer status of the pa tient were not significant indi cators of mortality. The authors

PAGE 62

48 used the Pediatric Risk of Mortality (PRISM) score for patient risk-adjustment (Pollack et al., 1994). Goh and colleagues (2001) found that 24 hour coverage in the Pediatric Intensive Care Unit (PICU) by an ICU physician result ed in a reduction in the standardized mortality ratio from 1.57 ( 95% CI=1.251.95) without th is coverage to 0.88 (95% CI=0.63-1.19) when it was present. Patient acu ity was controlled for using the PRISM II score. Standardized mortality ratios were calculated by dividing the observed deaths in the unit by the number of deaths predicted through the PRISM II. Tenner, Dibrell and Taylor (2003) conduc ted a retrospective cohort study of the impact of in-house hospitalists on mortality in the PICU compared to care provided by residents in another hospital. In both un its, the same pediatric group provided the coverage and an intensivist was availabl e after hours for cons ultation. The odds of survival were 2.8 ( p=.013) when a hospitalist was in-hous e compared to the presence of a resident. This significant a ssociation remained even af ter controlling for potential confounding variables such as patient acuity using the PRISM II and various patient diagnoses associated with higher mortality. However, a 1997 study using the Vermont Ox ford Network database did not show a reduction in neonatal mortality when there was a pediatric residency program present (AOR=1.18, 95% CI=.94-1.47). Ra ther, it found that patient-level variables had a much greater effect on mortality. These variables in cluded birth weight, birth defects, Apgar scores, male gender, antenatal steroid treatm ent, multiple gestation, vaginal delivery, race of Black. This study was limited to infant s between 500 and 1,499 grams (Horbar et al.,

PAGE 63

49 1997). It measured the presence of a residency program but did not measure physician staffing patterns onsite at the NICUs in the study. Further, the Vermont Oxford Network Database is voluntary and does not include all NICUs in the United States. This study included small samples for both very small a nd very large NICUs (Horbar et al., 1997). Summary of the Research This study builds on the existing resear ch by examining both the individual effects and effect modification of day of admission and time of admission to a NICU and subsequent neonatal death prior to discharge. Few studies have uti lized NICU data and those studies have addressed the issue of time of admission. This study will also address the issue of nurse staffing as it relates to these deaths. Studies that have analyzed nurse or physician staffing in either the NICU or PI CU have not simultaneously examined the issue of day or time of admission. This study will be able to adjust for case mix and staffing patterns, both mentioned as possible explanations for the higher neonatal death rate on weekends and during nighttime hours.

PAGE 64

50 Chapter Three: Methods The purpose of this study was to investi gate the time of admission to a Neonatal Intensive Care Unit (NICU) and its associati on with in-hospital mortality among a cohort of neonates at a regi onal perinatal center. Two different time point s were considered. The first time point was admission on the weekend versus the weekday. The second time point was admission during the nighttime shift versus the day shift. The secondary purpose of the study was to investigate if registered nurse staffing affected this association between NICU admission day or admission time and in-hospital death. Research Design This study was a retrospective cohort study of admissions to the Neonatal Intensive Care Unit (NICU) of a tertiary ca re hospital with a Le vel III NICU. The study population included all admissions to the NICU from October 1, 2001-December 31, 2006. Infants were excluded from the study if they were admitted to the newborn nursery before admission to the NICU, we re not discharged at the time of the study or the infants record could not be linked with the mothers record. Infant NICU records, maternal obstetric records and nurse and physic ian staffing records were used.

PAGE 65

51 Study Setting Tampa General Hospital is a tertiary car e hospital with 818 beds for acute care and 59 beds for rehabilitation care. The hospi tal serves as the regional trauma and burn center for West Central Florida. TGH is a private hospital governe d by its own Board of Directors. It serves as the teaching hospita l for the University of South Florida (USF) College of Medicine and provides clinical experience for students of nursing at USF, University of Florida, University of Ta mpa, Hillsborough Community College and St. Petersburg College (Tampa General Hospital, 2007). TGH is one of eleven Regional Perinata l Intensive Care Centers (RIPCC) in Florida. It is one of only si x hospitals in Florida which provides Extra Corpeal Membrane Oxygenation (ECMO) treatment. The hospital is a Level III NICU with a total of 42 beds, 2 of which are for ECMO (Tampa General Hospital, 2001, 2007). The NICU is also a Closed Unit, meaning that a single group pr ovides the medical care (Pronovost et al., 2002). For TGH it is board certified neonatologi sts from the USF College of Medicine. In 2006, TGH received the status of Magnet Hospital by the American Academy of Nursing (Tampa General Hospital, 2007) In a study by Aiken, Havens & Sloane (2000) on the characteristics of Magnet Hospitals, they found hospitals with this designation had higher nurse to patient ratios and 50% of the RNs on staff had at least a Bachelors in Nursing degree compared to 34% for non-magnet hospitals.

PAGE 66

52 HIPAA and the Protection of Human Subjects An application was submitted to the Institutional Review Board of the University of South Florida and TGH Office of Clinical Research for approval. The request for access to TGH databases was approved by TGH on April 20, 2007 (Appendix A) and USF under an Expedited Review on May 18, 2007 (Appendix B). IRB approval for the study ends on May 16, 2008. Data from TGH was placed on the secure server at the University of South Florida Health Science Center. This server is housed at NOC (Network Operations Center) on the USF Health campus. All data transfer protoc ols are Health Insurance Portability and Accountability Action (HIPAA) compliant: all data are encrypted using a 256-bit encryption. Data was housed on the LAWTON di rectory within the HSC server. This directory is for the Lawton & Rhea Chiles Ce nter, Department of Family and Community Health, College of Public Health at USF. W ithin this directory a folder, NICU Study, was created which was password protected. Only the Principal Investigator and a faculty member of the Dissertation Advisory Comm ittee had permission to access this folder. Data from TGH was placed on a CD-ROM or 3.5 diskette and hand carried to the USF Chiles Center and uploaded onto the NICU St udy folder on the day it was obtained. The CD-ROMs were kept in a locked file cabinet at the Chiles Center until the study was completed and were destroyed at the end of the IRB approval period. The individual databases which contained identifying informa tion were deleted from the server after the linked data sets were verified as bein g accurately linked mother-infant pairs.

PAGE 67

53 Once all the databases were linked, the iden tifying information on each infant and mother pair was deleted, leaving only the uni que identifier created for the record. This identifier did not use any protected health information but was a sequential number of all records in the database beginning with the numbe r one. The de-identified data sets will be maintained according to USF IRB policy and then they will be destroyed. Study Population Infants available for inclusion in this st udy are those infants who were admitted to the TGH NICU between October 1, 2001 and December 31, 2006. Infants were excluded from the study if they were admitted to the newborn nursery before admission to the NICU, were not discharged at the time of th e study or the infants record could not be linked with the mothers record. Infant NICU records, maternal obstetric records and nurse and physician staffi ng records were used. Sampling Framework Admissions were selected based on data contained in the Tampa General Healthcare Neonatal Intens ive Care Unit Database. All infants admitted to the TGH NICU from October 1, 2001 through December 3 1, 2006 were eligible to be included in the study, unless they met exclusionary criteria.

PAGE 68

54 Data Collection This section describes the three main da tabases used in this study: the Tampa General Healthcare Obstetrics Outcomes Information System, The Tampa General Healthcare Neonatal Intensive Care Unit Data base and the Nurse Staffing Database. The procedures used to link these databases are then described. Finally, there is a discussion of the validity and reliability of these databases. Obstetrical Database The Tampa General Healthcare Obstetrics Outcomes Information System(OOIS) was used to provide relevant information rega rding the infants moth er. The data utilized from this database were: Patient Name, Pa tient Complete Address, Maternal Race, Maternal Age, Prenatal Care, Maternal Transport, Multiple Gestation, Fetal Complications Fetal Anomaly, Primary Type of Delivery, Cesarean Delivery Indication, Oxytocin Induc tion, Oxytocin Augmentation, Oxytocics, Neonatal Information (Delivery Day and Time, Total Nu mber of Fetuses, Gender, Birth Weight, Apgar Score at 5 Minutes, Di sposition of Neonate). Data was pulled from Date of Delivery for the time period October 1, 2001 through December 31, 2006. This information was obtained June 2007. Neonatal Intensive Care Unit Database The Tampa General Healthcare Neonatal Intensive Care Unit Database (NICUD) was used to provide relevant informati on regarding the infant. This is the main database to which the others were linke d. Where data from the OOIS and the NICUD

PAGE 69

55 contained equivalent information (e.g. infant bi rth weight, infant gender, etc.), the data in the NICUD was considered the gold standard and that data was used in the study. This database has been maintained by one i ndividual since its creation, October 1, 2001. The data utilized from this database were: Patient Name, Patient Complete Address, Date of Birth and Date of Admi ssion, Gender, Race, Mothers Age, Prenatal Care, Location of Birth, Gestational Ag e, Birth Weight, Pulmonary Morbidity Information (mechanical ventilation), ECMO use, Transferred From, Discharge Date, NICU Discharge Destination, NICU Di scharge Status. The NICUD contained information on all admissions to the NICU beginning on October 2001. Each infant will be the unit of measure. Data was pulled using Date of Admission for the time period October 1, 2001 to December 31, 2006. This information was obtained June 2007. Nursing Staffing Database The Automated Nursing Staffing Office System ( ANSOS One-Staff) from McKesson provided the information needed on staffing patterns of nurses by day and shift. The ANSOS database contained inform ation on nursing staff by unit. It included information on the type of credential (RN, LPN), unit assignment (NICU), shift worked by day of week and time of day. Specificall y, the Joint Commission for the Accreditation of Healthcare Organizations (JCAHO) report which provided a report by unit, day and shift on the number of Registered Nurses, Licensed Practical Nurses and Technicians was used. This report was exported to individual te xt files. This report contained no employee

PAGE 70

56 names or numbers. Reports on staffing fo r the NICU from October 1, 2001 through December 31, 2006 were used for the study. This information was obtained August 2007. Data Linking Procedures The OOIS and NICUD databases were linke d using identifying data in both the databases that pertained to the mother and infant. The nurse staffing database was linked to the NICUD-OOIS linked database using Date of Admission. After linking all databases, Date of Admission was recoded to Day of Week. Time of Admission was recoded to the shift: Day or Night. The Nurs e Staffing Database was the report used for JAHCO. This report contained no identifying in formation on staff or patients. This final database was stripped of all identifying info rmation: patient names, patient admission and discharge dates, patient addresses, patient bi rth dates and dates of death. Maternal data not linked to NICU admissions was deleted from the final database. Validation and Reliability of the Databases The OOIS and NICUD are maintained by one individual. The databases maintain internal codes to ensure data integrity is maintained. The data is obtained from actual patient registration forms completed by hospita l personnel. Data from these databases is used for the hospitals annual reports to Childrens Medical Services for the statewide Florida Regional Perinatal Intens ive Care Center Annual Report This data is verified by CMS through the comparison of a selected num ber of patient medical records with the information contained in the databases every two years.

PAGE 71

57 The ANSOS database is also maintain ed by one individual. This database serves as the primary scheduler of nursing pe rsonnel for the entire hospital. The integrity of the staffing databases must be obtained in order to ensure proper scheduling of personnel in a cri tical care unit. Data Analysis This section describes the data assumpti ons used during the analysis phase. The study variables are then defined. How the data will be analyzed using descriptive statistics, bivariate correlations and multivariate models is given. Data Assumptions The time of admission is not r ecorded in the NICUD. Therefore, if an infant was admitted to the NICU immediately after deliv ery, the time of admission to the NICU will be considered to be the time of birth record ed in the OOIS. If the infant was admitted to the newborn nursery first, they will be excluded from the study. If the infant was transferred from another hospital, they will be excluded from the analysis of the research questions regarding time of ad mission or effect modification with day of admission and the subsequent impact on infant mortality. If the time of delivery is at the change of a shift, then the time of admission to the NICU was assumed to be the next shift. Numbers of nurses were assumed to be present during the entire shift for which they are scheduled.

PAGE 72

58 Study Variables Study variables include the outcome vari able, the exposure variables and the covariates. The coding of each of these va riables is provided in this section. Outcome variable. The outcome variable was death of an in fant before discharge from the NICU. This was a nominal level variable which wa s dummy-coded with death coded as a 1 and survival coded as a 0 (DEATH). The NICUD reco rds the discharge status of each infant. Exposure variables. The exposure variables were either Weekend Admission or Nighttime Admission. Models were developed for each and effect modification between the two was tested for significance. These are both nominal level va riables. They were dummy-coded with a 1 indicating Weekend Admission or Nighttim e Admission and a 0 indicating Weekday Admission or Daytime Admission. The referent group was coded as 0. Staffing variables, in which staff worked during th e evening shift, 3:00 pm to 11:00 pm, were split between the Day Time and Nighttime Shifts. Weekend Admission (WA) was defined as an infant admitted to the NICU between 7:01 pm Friday and 7:00 am Monday. Nighttime Admission (NA) was defined as an infant admitted to the NICU between 7:01 pm and 7:00 am from Labor & Delivery.

PAGE 73

59 Covariates. The covariates used in the study were in fants acuity, gender and race, type of delivery, induction, fetal anomaly, multiple bi rth, RN hours of patient care, nurse to infant ratio, and NICU capacity. When the vari able was nominal level, the referent group is always coded a 0. Infants Acuity was based on the following variables: o Birth Weight (BW) was a ratio level variab le and was defined as the infants birth weight in gram s. According to the most recent annual report of the Regional Peri natal Intensive Care Centers, birth weight of infants in the RPICC NICUs has a linear relationship with survival up to the 751-1000 gram category. After that birth weight range, survival is higher than 91% and the curve begins to flatten out (Children' s Medical Services, 2006). The type of curve in the dataset used for this study was tested and determined to be close to a norma l distribution for the entire study population so no term was added to the models. Birth weight was defined as a change in 100 grams of birth weight for the multivariate analyses to adjust for the low beta coefficient for a one gram change in birth weight. o Small for Gestational Age (SGA) was a nominal level variable with a 1 coded if the infant was SGA and a 0 coded if the infant was not SGA. SGA was defined according to the work done by

PAGE 74

60 Alexander, Himes, Kaufman, Mor and Kogan (1996) in defining SGA categories by race. o 5minute Apgar (APGAR) was a nominal level variable and was coded a 1 if APGAR is < 7 at five minutes and 0 if Apgar is 7 or above at five minutes. o Vent was a ratio level variable with number of days on the ventilator. This variab le was coded as a dich otomous variable with 1 indicating days on a ventilator and 0 indicating no days on a ventilator. Infants Race (RACE) was a nominal level variable. Categories under this variable were Black and Other. The referent category was Black coded as a 0. Infants Gender (GENDER) was a nominal level variable. Categories under this variable were Female an d Male. The referent category was Female coded as a 0. Fetal Anomaly (ANOMALY) was a nominal level variable. Categories under this variable were Anomaly Pres ent or Not Present. The referent category was Not Present and coded a 0. This will serve as a proxy for presence of a birth defect. This vari able does not contain any information on the type of fetal anomaly, only its presence or absence.

PAGE 75

61 Multiple Birth (Multiple) was a nominal level variable. Categories under this variable were Multiple Birth or Singleton Birth. The referent category was Singleton birth coded a 0. Type of Delivery (TD) was a nominal level variable. Categories under this variable were Spontaneous Vaginal Delivery (SVD) or C-Section (CS). Infant who had both codes listed were coded as a C-Section. The referent category was SVD and was coded a 0. Induction (IND) was a nominal level variable. Categories under this variable were Induction or No Induction with the later as the referent category coded a 0. NICU Capacity (CAP) was an interval level vari able. It was defined as the total number of infants in the NICU (census) per day divided by the total number of beds in the NICU (N=42) This variable was computed from the admission and discharge data in the NICUD. The census number was obtained from TGH for September 30, 2001. Census was defined as the total number of infants in the uni t at midnight of each day. It was computed as (ending census previous day + admissions for that day) discharges for that day. Cens us is not kept by shift. Nurse to Infant Ratio (NIRDAY) was an interval level variable. It was defined as the ratio of nurses by day to the total number of infants in the NICU at midnight for that day. It wa s computed as the number of RNs by 24-hour day/number of infants at midnight for each day.

PAGE 76

62 Nurse to Infant Ratio (NIRSHIFT) was an interval level variable. It was defined as the ratio of nurses by shift to the total number of infants in the NICU at midnight for that day on which the shift occurred. It was computed as the number of RNs by 12-hour shift/number of infants at midnight for each day on which the shift occurred. RN Nursing Hours for Day (RNHR) was a ratio level variable. It was defined as the total number of RN nursing hours per 24-hour day. RN Nursing Hours for Shift (SHIFTHR) was a ratio level variable. It was defined as the total number of RN nursing hours per 12-hour shift. Descriptive Statistics Descriptive statistics were used to de termine the differences between the two cohorts of infants in the study: infants adm itted to the NICU on the weekday compared to infants admitted on the weekend; infants ad mitted to the NICU during the day shift compared to infants admitted during the ni ght shift. Frequencies were utilized to determine difference in demographic inform ation between the two groups. For interval level data, the difference in distributions between the cohor ts was measured using the mean, skew, kurtosis, standard deviati on, minimum and maximum values. Tests for significant differences used either a chi-squa re test or t-test depending on the level of measurement of the variables being analyzed. These chi-square and t-tests used an alpha level at .05 as the threshold needed for sta tistical significance. Bivariate correlations

PAGE 77

63 between the linear covariates were tested to assess multicolinarity and determine if interaction terms were needed. Descriptive analyses were also comple ted on infants admitted to the Newborn Nursery before NICU admission and infants bor n at TGH who could not be linked to the OOIS to determine how their exclusion from the study affected the study results. The Statistical Package for the Social Scien ces (SPSS) Version 15.0 for Windows was used for data analyses. Multivariate Models Multivariate logistic regression analyses were used to test for a significant association between the exposures and out come while controlling for confounding. The log likelihood ratio statistic was used to test for significance or fit between the full or saturated model and reduced models. The diffe rence between the log likelihood statistics for the full model and reduced model multiplied by two approximates a chi-square distribution. Significance for this test will be at the .05 level. A significant chi-square test indicates that the saturated model provides more explanation than the reduced model and the saturated model is maintained. A non-signifi cant chi-square test indicates the reduced model is not significantly different from the saturated model. Therefore, the reduced model provides as much explanatory value as the saturated model and is a more parsimonious model so it is maintained (Agresti, 2002; Tabachnick & Fidell, 2001). The Wald statistic and Confidence Interv als were used to test for significant variables. The Wald statistic is the model co efficient divided by its standard error such

PAGE 78

64 that W = / SE Confidence Intervals are based on the natural log of the odds ratio (model coefficient), its standard error and an alpha level, providing both a lower limit and an upper limit around the odds ratio. Confidence intervals provide more explanatory value regarding the coefficient. A wider interval indi cates a less precise measure of the variable (Agresti, 2002; Kahn & Sempos, 1989; Tabachnick & Fidell, 2001) Variables were entered into the logistic regression model using a block entry method. This type of regression is based on theory or prior research in which the researcher determines the order of entry of variables into the model. This compares to stepwise regression where variables are entere d and retained in the model based only on statistical significance (Tab achnick & Fidell, 2001). Model goodness of fit was tested using the Hosmer-Lemeshow statistic. This statistic tests overall fit of the model but does not provide specific information on problems with model fitting. This statistic compares the observed with the expected observations in different partitions of th e data using the Pearson statistic. A nonsignificant results indicates a good fit of the model since the observed values are not significantly different from the expected values (Tabachnick & Fidell, 2001). Covariates which are in the OOIS (Type of Delivery, Induction, Apgar Score, Fetal Anomaly and Multiple Birth) were una vailable for infants who were transferred from another hospital and admitted to the NICU. The contribution of these covariates to the models for Research Question 1 and Research Question 4 were assessed through the following methods to determine if the inclusi on of transfers in mu ltivariate models, but the exclusions of these covariates, significan tly influenced the final model selected. A

PAGE 79

65 logistic regression model was developed fo r Research Question 1 with only in-born infants using all of the covariates. The mo st parsimonious model was selected. Then, a second model was developed with the same sample of in-born infants but with the exclusion of the following covariates: Type of Delivery, Induction, Apgar Score, Fetal Anomaly and Multiple Birth. The two models were compared to determine the change in the log likelihood ratio statistic between the sa turated model (full-risk adjustment model) and reduced model (reduced-ris k adjustment model) and th e Hosmer-Lemeshow statistic for overall fit of each model. Second, a R eceiver Operating Charac teristic curve was developed for each model to determine how th e sensitivity and specificity of the model changed when the five covariates are excluded. Based on the results of both of these analyses, a decision was made to exclude transfers from the analysis. A descriptive analysis of transfers and the analysis comp aring the full-risk adjustment model to the reduced-risk adjustment model are provided in Chapter 4. Research Questions Models tested for each research question are provided below. Research question 1. Is there an association between the da y of admission (weekday versus weekend) to the NICU and the infants outcome? DEATH=WDA + BW + SGA + APGAR + VENT + RACE + GENDER + MULTIPLE + ANOMALY + TD + error

PAGE 80

66 Research question 2. Is there an association betw een the time of admission (d ay versus night) to the NICU and the infants outcome? DEATH=NA + BW + SGA + APGAR + VENT + RACE + GENDER + MULTIPLE + ANOMALY + TD + IND + error Research question 3. Is there effect modification between day of admission and time of admission? DEATH=WDA + NA + WDA*NA + BW + SGA + APGAR + VENT + RACE + GENDER + MULITPLE + ANOMALY + TD + IND + error Research question 4. Do staffing patterns of nurses in the NICU mediate the association between day or time of admission to the NICU and the infants outcome? 1) DEATH=WDA + BW + SGA + APGAR + VENT + RACE + GENDER + MULTIPLE + ANOMALY + TD + IND + IBS + CAP + RNHR + NIRDAY + error 2) DEATH=NA + BW + SGA + APGAR + VENT + RACE + GENDER + TD + IND + MULTIPLE + ANOMALY + IBS + CAP + RNSHIFT+ NIRSHIFT + error

PAGE 81

67 3) DEATH=WDA + NA + WDA*NA + BW + SGA + APGAR + VENT + RACE + GENDER + + MULTIPLE + ANOMALY + TD + IBS + CAP + RNHR + NIRDAY + error Sample Size Calculation The total number of admissions consis ted of 3,511 neonatal admissions to the NICU from October 2001-December 2006. If all admissions were included in the analyses, there was 80% power to detect the lowest odds ratio of 1.3 with a 95% confidence interval and a tolerance value of .8. There was 80% power to detect an odds ratio of 1.4 with a 95% confidence interval but a lower tolerance value of .5 (Kromrey, 2007). Tolerance assess the multicolinearity amo ng variables in a regression equation. It is computed as 1 SMC (squared multiple correlation of a variable) where that variable is the dependent variable in a model with al l the other independent variables. At a value of 1, there is singularity between that vari able and the other i ndependent variables (Tabachnick & Fidell, 2001).T olerance has its value between 1 and .1 with 1 indicating no multicolinarity among the independent variables.

PAGE 82

68 Chapter Four: Results The purpose of this study was to investi gate the time of admission to a Neonatal Intensive Care Unit (NICU) and its associati on with in-hospital mortality among a cohort of neonates at a regi onal perinatal center. Two different time point s were considered. The first time point was admission on the weekend versus the weekday. The second time point was admission during the nighttime shift versus the day shift. The secondary purpose of the study was to investigate if registered nurse staffing affected this association between NICU admission day or admission time and in-hospital death. Study Population This section summarizes the data linking procedures described in detail in the next section and provides an overview of the study population. After linking the NICU and OOIS databases and excluding cases based on a priori definitions, the total admissions included in the study were 1,846 admissions. The total admissions to the TGH NICU during the study period of October 1, 2001 December 31, 2006 were 3,511. Of those admissions, 43 were readmissions of infants and these were excluded from the study. These infants had two admission dates in the NICU with the first admission showing a discharge destination of Acute Care Facility. There was a subsequent admission date after the first discharge. Th ere were 59 TGH births which could not be

PAGE 83

69 linked to the OOIS and these were excl uded from the study. Table 3 shows the comparison of TGH births which were linked and those which were not linked. Table 3 Comparison of NICU Births Linked to OOIS with NICU Births Not Linked to OOISa (N=1905) NICU Births Linked NICU Births Unlinked Number Percent Number Percent ChiSquare (df) p-value Exposure Variables Weekend Admissionb 434 23.5% 11 18.6% .76(1) .385 Covariates (non-referent category) Male Infantc 971 52.6% 34 57.6% .61(2) .739 Non-Black Race 1291 69.9% 46 79.7% 2.59(1) .108 Infant on ECMO 10 0.5% 2 3.4% 7.41(1) .006** Infant on Ventilation 457 24.8% 15 25.4% .01(1) .907 Small for Gestational Age 392 21.2% 11 18.6% .23(1) .631 Outcome Variable Death Prior to Discharge 125 6.8% 3 5.1% .26(1) .610 Total 1846 96.9% 59 3.1% aThe following covariates are found in the OOIS and are therefore excluded from this table: multiple birth, APGAR score, delivery type, induction, fetal anomaly. bWeekend Admission includes Saturday and Sunday only since admission time is in the OOIS. cGender unknown for 1 Linked case. **p < .01 The unlinked TGH births had a significantly higher proportion of infants on ECMO (3.4%, p=.006) than those births which were linked. However, this represents only two infants and is not expected to bias the study results. Re sults of the t-test (Appendix C) on linear variab les between these two groups showed a significant difference between birth weight and gestation with unlinked births being heavier (mean difference=349.216, p=.006) and having older gestati onal age (mean difference=1.543, p=.008). The effect size (difference in means/SD of linked births) is .37 for birth weight

PAGE 84

70 and .35 for gestation. This is a medium eff ect size and given the small proportion of admissions unlinked (3.1%) in the sample used for this study the exclusion of these cases is not expected to bias the results. There was no significant difference on weekend admission or death prior to discharge. Of the remaining 3,409 admissions, 2,683 were born at Tampa General Hospital and 726 were transferred from another hospital. For TGH births, 1,842 were admitted to the NICU directly from Labor & Delivery (L&D), 837 were admitted to the Newborn Nursery prior to their admission in the NICU, two were classified as being admitted to the Morgue (even though they were in the NICUD) and two were unknown as to discharge after L&D. These four infants were discharged alive from the NICUD and were included in the analysis. The 839 infants adm itted to the Newborn Nursery were excluded from the study. Table 4 shows the comparison of infants admitted to the Newborn Nursery first with those infants admitted to the NICU first. The infants admitted to the Newborn Nursery prior to their NICU admission were a healthier group of infants than those adm itted to the NICU directly from Labor & Delivery. They differed significantly from di rect NICU admissions in terms of less ventilation, less small for gestational age infants, fewer multiple births, a lower proportion with Apgar scores less than se ven at five minutes, a lower proportion delivered by cesarean section and lower proportion of infants with the presence of a fetal anomaly. More Newborn Nursery infants were induced compared to direct NICU admissions.

PAGE 85

71 Table 4 Comparison of Infants Admitted to NICU First with Infants Admitted to Newborn Nursery First (N=2683) NICU Newborn Nursery Number Percent Number Percent ChiSquare (df) p-value Exposure Variables Weekend Admission 547 29.6% 239 28.6% .32(1) .570 Nighttime Admission 798 43.2% 381 45.5% 1.23(1) .268 Covariates (non-referent category) Male Infanta 971 52.6% 476 56.9% 4.62(2) .099 Non-Black Race 1291 69.9% 597 71.3% .54(1) .465 Infant on ECMO 10 0.5% 1 0.1% 2.52(1) .113 Infant on Ventilation 457 24.8% 23 2.7% 185.18(1) p<.001*** SGA 392 21.2% 137 16.4% 8.62(1) .003** Multiple Birth 310 16.8% 53 6.3% 53.87(1) p<.001*** APGAR < 7b 261 14.2% 8 1.0% 111.24(1) p<.001*** Cesarean Section 1070 58.0% 331 39.5% 81.44(1) p<.001*** Induction 207 11.2% 141 16.8% 16.19(1) p<.001*** Fetal Anomaly 170 9.2% 28 3.3% 28.97(1) p<.001*** Outcome Variable Death Prior to Discharge 125 6.8% 1 0.1% 56.93(1) p<.001*** Total 1846 68.8% 837 31.2% aGender unknown for 1 NICU case. bAPGAR score missing values for 8 NICU and 2 NBN cases. **p<.01; ***p<.001 The results of t-tests on birth weight, gestation, Apgar score and ventilation days were also significantly diffe rent. Newborn Nursery infant s were heavier (3,144 versus 2,090.55, p<.001), had an older gestational age (38.2 versus 33.32, p<.001), had a higher Apgar score at less than seven minutes ( 8.79 versus 7.91, p<.001) and had fewer days on ventilation (.12 versus 3.45, p<.001) (Appendix E). There wa s no significant difference

PAGE 86

72 regarding the exposures, and infants admitted to the Newborn Nursery first were significantly less likely to die prio r to discharge (0.1% versus 6.8%, p<.001). Data Linking Procedures The NICUD was the primary databases to which all other databases were linked. The first link was the OOIS linked to the NICUD. After all mother-infant pairs were confirmed, the nurse staffing data from the ANSOS Database JCAHO report were linked with the NICUD-OOIS database. There were a total of 3,511 admissions to the NICU during the study time period of Oc tober 1, 2001 December 31, 2006. There were a total of 726 admissions which were transfers. These could not be linked to the NICUD. The remaining 2,785 admissions were av ailable for linking with the OOIS. Deterministic Linking The first pass of data matched NICU admissions (NICUD) with mothers in the OOIS database based on mothers first and last name, address and date of birth/delivery. This pass yielded 2,343 matches (84% of cases available for linking). Probabilistic Linking The second pass of data matched NICU admission with mothers in the OOIS database based on mothers first and last name and date of bi rth/delivery. This pass yielded 187 matches (6.7% of cases availabl e for linking). These matches were then verified comparing infants gender, mothers ag e, infants birth weight and race of infant

PAGE 87

73 with race of mother between the two databases. Discrepancies within these variables were further investigated to determine if the correct mother-infant pair was linked. For birth weight, differences of more than one ounce between the two databases were investigated. Where there were differences of birth weight greater than one ounce, gender, race or age, the OOIS was used to determine if the same woman gave birth to multiples and the wrong twin was matched to the NICU database or to determine if another woman with the same name gave birth on the same day and the wrong motherinfant pair was matched. In no instance of these 187 matches was another woman of the same name found to have given birth at T GH on the same day. Therefore, these were assumed to be the correct mother-infant pairs. Manual Linking The third pass of data matched NICU admission with mothers in the OOIS database manually. This pass yi elded 196 matches (7% of cases available for linking). Both databases were sorted by mothers la st and first name. These pairs were linked based on mothers first and last name, address and/or date of birth. In all but 17 of these cases, the mothers first and last name and addressed matched but were in a different format between the two databases. In 17 of th ese cases, the mothers first and last name and address matched, but the dates of birth were different. One of these was a readmission. Four of these were at shift change and could account for the one day difference in dates. The remaining 12 were verified by determining if another woman with the same name gave birth on the same day and the wrong mother-infant pair was

PAGE 88

74 matched. In no instance of these 12 matches was another woman of the same name found to have given birth at TGH on the date of bi rth listed in either the NICUD or the OOIS. Therefore, these were assumed to be the correct mother-infant pair. Measures between the two databases whic h contained the same information were compared using correlation coefficients and crosstabulations to determine the level of agreement between the two databases and furt her test the integrity of the linkage. The birth weight of the infant between the NIC UD and the OOIS had a correlation coefficient of .985 ( p<.001) and the correlation coefficient for mothers age was .988 ( p<.001). Crosstabulations of infants gender showed an agreement of 99% on female gender and 98.8% on male gender; for infants race with mothers race, the agreement for race Black was 90.2% and for race Other was 98.6%. Since an infant classified as Black could have a mother who is White, it is not expected th at the agreement on the race variable would be 100%. The level of agreement between these two databases on variables which contained the same information provides fu rther support for correct linkage between mother and infant pairs. Unlinked Tampa General Hospital Births There were 59 TGH births that were admitted to the NICU which could not be linked to the OOIS database using mothers na me (2.1% of cases available for linking). Table 3, already discussed, provides the comp arison of TGH births which were linked with those births which could not be linke d and were excluded from the analysis.

PAGE 89

75 Data Preparation Once the databases were linked, the data were prepared for analyses. All identifying information was deleted from the fi nal merged database. This included patient names and addresses; dates of birth, admission and discharge; and dates of nursing shifts. NICU and OB Databases Both the NICUD and OOIS databases were in Excel format as text variables. The Excel files were exported into SPSS 15.0 for Windows. All variable s retained in the study were recoded to numeric variables. The outcome variables, exposure variables and covariates were all recoded according to the definitions in Chapter 3. Admission day was recoded into day of week using the SPSS compute statement for date function. Admission time was recorded into a nu meric variable using the SPSS compute statement for time function. Census was computed using the number of admissions and discharges in the NICUD. Frequency distributions for each date were run a nd exported into Excel. The census on September 30, 2001 was obtained from TGH. The census was then calculated in the excel database ((beginning dail y census + daily admissions) daily discharges=ending daily census) and imported into SPSS. Once in SPSS, it was merged into the main database using the Keyed Table command to merge the same record into multiple records in the main data base. Capacity was defined as the census at midnight of each day/42 available beds.

PAGE 90

76 RN Database The ANSOS database was in 91 individual text files for each three-week period of the study. These files were copied into an Excel spreadsheet and formatted. Dates were reviewed to ensure that each date in th e study period was represented. This spreadsheet was then exported in SPSS. Registered nur sing hours by day and shift were then recoded according to the defin ition in Chapter 3. In order to include registered nursing hours for infants admitted between midnight a nd 7:00 am, a lag function was used on the nursing shift variables to compute the number of nurses on the evening shift and night shift from the previous day for each admission in the database. Descriptive Analyses The difference between the two cohorts used in the study, weekend versus weekday admissions and nighttime versus day admissions, were compared using chisquare and t-tests. For crossta bulations with nominal level va riables a chi-square test was used and for continuous variab les a t-test was used. Alpha was set at .05 for significance. For continuous variables, the distribution be tween the two groups was investigated using mean, standard deviation, minimum a nd maximum value, kurtosis and skew. Weekend Admissions Compared to Weekday Admissions Tables 5 and 6 provide the results of weekend admissions compared to weekday admissions. Table 5 provides the results of crosstabulations on ke y covariates and the outcome variable with the chi-square statistic and Table 6 provides the results of t-test on

PAGE 91

77 key covariates. The results of descriptive analyses providing the standard deviation, minimum and maximum values, kurtosis and skew are given in Appendix E. Table 5 Weekend Admissions Compared to Weekday Admissions (N=1846) Weekend Weekday Number Percent Number Percent ChiSquare(df) p-value Covariates (non-referent category) Male Infanta 291 53.2% 680 52.3% .52(2) .770 Non-Black Race 363 66.4% 928 71.4% 4.72(1) .030* Birth Weight < 2500 grams 384 70.2% 902 69.4% 1.06 .745 Gestation < 37 weeks 420 76.8% 962 74.1% 1.51 .218 Small for Gestational Age 114 20.8% 278 21.4% .07(1) .788 APGAR < 7 at 5 Minutesb 83 15.3% 178 13.7% .75(1) .388 Infant on Ventilation 137 25.0% 320 25.0% .04(1) .852 Infant on ECMO 3 0.5% 7 0.5% .001(1) .980 Multiple Birth 88 16.1% 222 17.1% .28(1) .599 Fetal Anomaly 47 8.6% 123 9.5% .35(1) .552 Cesarean Section 310 57.0% 760 59.0% .62(2) .732 Induction 55 10.1% 152 11.7% 1.05(1) .306 Outcome Variable Death Prior to Discharge 38 6.9% 87 6.7% .04(1) .846 Total 547 29.6% 1299 70.4% aGender unknown on 1 Weekday case. bApgar score missing on 4 Weekend and 4 Weekday cases. *p<.05 The only significant difference between weekend and weekday admissions was weekend admissions were more likely to be Black (33.6% versus 28.6%, p=.030; note: table frequency is for non-Black race). Wh ile weekend admissions had a slightly higher proportion of deaths, it was not statistic ally significant (6.9% versus 6.7%, p=.846).

PAGE 92

78 Table 6 T-Test Results for Weekend Admissions Compared to Weekday Admissions (N=1846) Weekend Weekday t SE p-value (n=547) (n=1299) Mean Birth Weight in Grams 2053.070 2106.330 1.090 48.480 .272 Mean Gestation in Weeks 33.090 33.420 1.470 .223 .141 Mean APGAR Score 7.860 7.930 .862 .082 .389 Mean Days on Ventilation 3.260 3.520 .339 .778 .734 Mean Number of RNs 54.870 56.280 2.780 .508 .006** Mean Number of RN Hours 331.950 341.990 3.240 3.100 p<.001*** Mean Census 46.140 46.320 .522 .346 .602 Mean RN to Infant Ratio 1.190 1.210 4.360 .006 p<.001*** Mean Capacity 1.098 1.102 .522 .008 .602 **p<.01; ***p<.001 Infants admitted on the weekend compar ed to the weekday, did not differ significantly in terms of acuity. However, nursing hours and RN to infant ratio was significantly different. There were less mean nursing hours (331.95 versus 341.99, p=<.001) and a lower mean RN to infant ratio (1.190 versus 1.210, p<.001) for infants admitted on the weekend. Capacity was not significantly different. Nighttime Admissions Compared to Day Admissions Tables 7 and 8 provide the results of nighttime admissions compared to day admissions. Table 7 provides the results of crosstabulations on ke y covariates and the outcome variable with the chi-square statistic and Table 8 provides the results of t-test on key covariates. The results of descriptive analyses providing the standard deviation, minimum and maximum values, kurtosis and skew are given in Appendix F.

PAGE 93

79 Table 7 Nighttime Admissions Compared to Day Admissions (N=1846) Nighttime Day Number Percent Number Percent ChiSquare (df) p-value Covariates (non-referent category) Male Infanta 416 52.1% 555 53.0% .90(2) .637 Non-Black Race 576 72.2% 715 68.2% 3.37(1) .066 Birth Weight < 2500 grams 576 77.2% 710 67.7% 4.21(1) .040* Gestation < 37 weeks 614 76.9% 768 73.3% 3.23(1) .073 Small for Gestational Age 162 20.3% 230 21.9% .73(1) .392 APGAR < 7 at 5 Minutesb 146 18.4% 115 11.0% 19.94(1) p<.001*** Infant on Ventilation 209 26.2% 248 23.7% 1.55(1) .213 Infant on ECMO 1 .1% 9 .9% 4.52(1) 033* Multiple Birth 140 17.5% 170 16.2% .57(1) .451 Fetal Anomaly 70 8.8% 100 9.5% .32(1) .571 Cesarean Section 406 51.0% 664 63.0% 31.02(2) p<.001*** Induction 96 12.0% 111 10.6% .94(1) .332 Outcome Variable Death Prior to Discharge 63 7.9% 62 5.9% 2.81(1) .094 Total 798 43.2% 1048 56.8% aGender unknown on 1 Day case. bApgar score missing on 2 Nighttime and 4 Day cases. *p<.05; ***p<.001 Infants admitted at night differed significantly from infants admitted during the day on birth weight, Apgar score, ECMO and cesarean section rates. Nighttime admissions had a higher proportion of low bi rth weight infants (77.2% versus 67.7%, p=.040) and subsequently a higher proportion of infants with Apgar scores less than seven at five minutes (18.4% versus 11.0%, p=<.001). Infants admitted during the day were more likely to be on ECMO (.9% versus .1%, p=.033), however these numbers are very small and should be in terpreted cautiously. Day admi ssions were delivered by c-

PAGE 94

80 section at a higher rate than ni ghttime admissions (63% versus 51%, p=<.001). Infants admitted at night were not significantly more likely to die before discharge than infants admitted during the day (7.9% versus 5.9%, p=.094) even though they had a higher proportion of deaths. Table 8 T-Test Results for Nighttime Admissions Compared to Day Admissions (N=1846) Nighttime Day t SE p-value (n=798) (n=1048) Mean Birth Weight /Grams 2028.910 2137.480 2.450 44.620 .014* Mean Gestation in Weeks 33.040 33.530 2.370 .206 .018* Mean APGAR Score 7.740 8.050 3.980 .075 p<.001*** Mean Days on Ventilation 3.700 3.250 -.633 .717 .527 Mean Number of RNs 27.540 29.160 6.261 .258 p<.001*** Mean Number of RN Hours 163.500 175.200 7.541 1.550 p<.001*** Mean Census 46.290 46.250 -.123 .319 .902 Mean RN:Infant Ratio .593 .628 11.646 .003 p<.001*** Mean Capacity 1.102 1.101 -.123 .008 .902 *p<.05; ***p<.001 Infants admitted at night had significa ntly lower mean birth weights (2028.91 versus 2137.48, p=.014), younger mean gestational age (33.04 versus 33.53, p=.018) and lower mean Apgar scores (7.74 versus 8.05, p<.001). Nighttime admissions also had significantly fewer mean nursing hours (mean difference of 11.70 nursing hours, p<.001) and lower mean RN to infant ratios (.593 versus .628, p<.001) compared to day. Bivariate Correlations Bivariate correlations betw een birth weight and hosp ital-level nursing and census variables are given in Table 9. Bivariate co rrelations between the hospital-level nursing and census variables are given in Table 10.

PAGE 95

81 Table 9 Bivariate Correlations with Birth Weig ht and Hospital-Level Variables (N=1846) Birth Weight of Infant Number of RNs by Day of Admission -.021 Number of RN Hours by Day of Admission -.018 Number of RNs by Shift at Admission -.017 Number of RN Hours by Shift at Admission -.014 Census/Day -.038 Capacity/Day -.038 RN to Infant Ratio by Day of Admission .024 RN to Infant Ratio by Shift at Admission .029 There was no significant correlation between birth weight and any of the nursing variables (number of registered nurses, hours of registering nur sing or RN to infant ratio) for day of admission or shift at admission. There was also no si gnificant correlation between birth weight and census or capacity. Table 10 Bivariate Correlations with Hospital-Level Variables (N=1846) RN:Infant RN:Infant Hrs/Day RN/Shift Hrs/Shift Day Shift Census Capacity RN Day .996*** .967*** .959*** .626*** .575*** .885*** .885*** RN Hours/Day .967*** .958*** .623*** .580*** .882*** .882*** RN Shift .997*** .597*** .667*** .861*** .861*** RN Hours/Shift .593*** .674*** .850*** .850*** RN:Infant Day .885*** .203*** .203*** RN:Infant Shift .206*** .206*** Census 1.00*** ***p<.001 The nursing variables and the census variable s were all significantly correlated as expected. Since RN hours is based on number of RNs per day or sh ift, these variables have a correlation coefficient close to one (RN number per day with RN hours per

PAGE 96

82 day=.996, p=<.001; (RN number per shift with RN hours per shift=.997, p=<.001).Capacity is defined through the cen sus numbers and these variables have a perfect correlation. The correlation of nurse to infant ratios is strongly correlated with nursing hours with a coefficient of .623 ( p=<.001) for day of admission and a coefficient of .674 ( p=<.001) for shift at admission. While nur sing hours are strongly correlated with capacity (.882, p=<.001 for day of admission; .850, p=<.001 for shift at admission), the nurse to infant ratio has a much lower correlation with capacity (.203, p=<.001 for day of admission and .206, p=<.001 for shift at admission). Multivariate Analysis Multivariate models using logistic regression were developed to measure the association between the expos ure variables and the outcome while simultaneously controlling for covariates which might be c onfounders. Table 11 shows the results for the bivariate relationships for both exposures a nd all covariates. Tables 12 26 show the results for multivariate logistic regression models for the four research questions. For multivariate models, Dfbetas were run to determine if there were any cases in the model with influence. Dfbeta measures the change in the model coefficient of an independent variable when a case considered to be an outlier is removed from the analysis. Cases with Dfbeta values greater than one may be outliers (Tabachnick & Fidell, 2001). To determine how each of those cases aff ected the outcome of model selection, each was removed from the logistic regression model and changes to the log likelihood ratio statistic and Hosemer-Lemeshow test for goodness of fit were evaluated

PAGE 97

83 to determine if the removal of the case changed the statistics for the selected model. Only one model, Model 4B, contained any cases w ith Dfbeta values exceeding 1. Removal of these cases did not change the model results and therefore those cases were retained. Bivariate Relationships for Exposur es and Covariates with Outcome Table 11 provides the results of bivariate relationships between the exposures, covariates and the outcome variable. In formation provided includes the regression coefficient, coefficient standard error, Wald statistic, unadjusted odds ratio with confidence intervals and significance test. Neither weekend nor nighttime exposures had significant unadjusted odds ratios for death before discharge. Weekend admissi on showed an increase of 2.7% in the odds of death and nighttime admission showed an increase of 36% in the odds of death. However, neither was significant as indicated by an insignificant Wald test ( p=.895 and p=.100 respectively) and confidence intervals which c ontained the value of one.

PAGE 98

84 Table 11 Bivariate Relationships for Exposures and Covariates with Death Before Discharge (n=1837) Confidence Interval B SE Wald (df) Exp (B) Lower Upper Sig. Weekend Admission .027 .204 .017(1) 1.027 .689 1.531 .895 Nighttime Admission .308 .187 2.712(1) 1.360 .943 1.962 .100 Birth Weight/100 Gram -.145 .015 88.701(1) .865 .839 .891 p<.001*** SGA .096 .223 .185(1) 1.101 .711 1.704 .667 APGAR < 7 2.161 .196 121.652(1) 8.680 5.912 12.743 p<.001*** Infant on Ventilation 3.352 .282 141.266(1) 28.570 16.437 49.660 p<.001*** Non-Black Race .353 .219 2.603(1) 1.423 .927 2.186 .107 Male Infant .112 .188 .354(1) 1.118 .774 1.615 .552 Fetal Anomaly 1.648 .217 57.915(1) 5.199 3.400 7.949 p<.001*** Multiple Birth .300 .230 1.697(1) 1.350 .860 2.119 .193 Cesarean Section .200 .193 1.076(1) 1.221 .837 1.782 .300 Induction -1.380 .514 7.212(1) .252 .092 .689 p<.001*** RN Hours by Day -.002 .001 1.955(1) .998 .995 1.001 .162 RN Hours by Shift -.004 .003 1.684(1) .996 .991 1.002 .194 Capacity/Day -.472 .569 .688(1) .624 .204 1.904 .407 RN:Infant Ratio Day -1.169 .825 2.009(1) .311 .062 1.564 .156 RN:Infant Ratio Shift -2.196 1.431 2.355(1) .111 .007 1.838 .125 **p<.01; ***p<.001 Model terms: weekend=1, nighttime=1, birth weight=100 gram increments, sga=1, APGAR<7=1, ventilation=1, non-Black race=1, male=1, anomaly=1, multiple birth=1, c-section=1, induction=1, RN hours=actual hours per 24 hour period, RN hours per shift = actual hours per 12 hour shift, capacity=census/42, RN:Infant ratio per day=#RN/cens us; RN:Infant ratio by shift=#RN shift/census. Covariates which were signifi cant were those related to the infants acuity. Birth weight showed an unadjusted odds ratio of .865 (95% CI=.839-.891) with a narrow confidence interval. For every 100 gram in crease in birth weight, there was a 15.6% increase in the odds of survival. An Apgar sc ore of less than seven at five minutes was associated with a significant increase in the odds of death (UOR=8.68, 95% CI=5.912.743). Being on a ventilator ha d the strongest a ssociation with dying. Infants on a ventilator were 28.5 times more likely to die before discharge (UOR=28.57, 95%

PAGE 99

85 CI=16.437-49.660). Infants with the presence of fetal anomaly had a five-fold increase in the odds of dying (UOR=5.20, 95% CI=3.4-7.95) Induction was the last covariate to show a significant association with dying. It showed a protective effect with an almost 75% reduced risk of dying (UOR=.252, 95% CI=.092-.689). Research Question 1 The research question under investigation was : Is there an association between the day of admission (weekday versus weekend) to the NICU and the infants outcome? The multivariate analysis found no significant association between weekend admission and death prior to discharge. The model which was tested is defined as Model 1. There were no cases which had Dfbeta values greater than one for this model. Table 12 displays the results of the model with the best fit, Table 13 displays the results of the log likelihood tests on each model and Appendix G provides each model in each block of entry for the logistic regression analysis. The blocks for entry for the first three models were defined as: Block 1: Exposure variable (s) (weekend, nighttime or weekend*nighttime) Block 2: Infant acuity (birth weight, ventilator, APGAR, SGA) Block 3: Infant characteristics (anomaly, race, gender, multiple) Block 4: OB interventions (induction, cesarean section). Model 1: DEATH=WDA + BW + VENT + APGAR+ SGA + ANOMALY + RACE + GENDER + MULTIPLE + IND + TD + error

PAGE 100

86 Table 12 Logistic Regression Model 1 for Weekend Exposure with Death Before Discharge (n=1837) Confidence Interval B S.E. Wald df Exp(B) Lower Upper Sig. Weekend Admission .058 .247 .055 1 1.060 .653 1.721 .814 Birth Weight -.120 .019 40.569 1 .887 .855 .920 p<.001*** Infant on Ventilation 2.316 .326 50.574 1 10.134 5.353 19.186 p<.001*** APGAR < 7 1.208 .236 26.110 1 3.347 2.106 5.319 p<.001*** SGA .109 .291 .141 1 1.115 .630 1.973 .708 Fetal Anomaly 2.672 .350 58.255 1 14.474 7.287 28.750 p<.001*** Non-Black Race .762 .265 8.236 1 2.142 1.273 3.604 .004** Male Infant .063 .229 .075 1 1.065 .680 1.668 .784 Multiple Birth .205 .283 .526 1 1.228 .705 2.139 .468 **p<.01; ***p<.001 Model terms: weekend=1, birth weight=100 gr am increments, ventilation=1, APGAR=1, sga=1, anomaly=1, non-Black race=1, male=1, multiple=1. In the best fitting model, the exposure variable remained non-significant even though the odds ratio increased from 1.027 to 1.083. Significant variables in this model were birth weight, Apgar score, ventilation and non-Black ra ce. Birth weight remained a consistent measure with its unadjusted odds ratio (UOR=.865, 95% CI=.839-.891). The odds ratios for both Apgar score and ventilat ion declined from th eir unadjusted rates. Non-Black race, which had a non-significant una djusted odds ratio, became significant in the multivariate model (AOR=2.127, 95% CI=1.261-3.69). Table 13 -2 Log Likelihood Tests for Model 1 with Weekend Exposure Difference Model in Log Critical Model -2LL df Likelihoods Sig. Chi-Square (df) Model 1.1: Exposure 902.668 1 .896 Model 1.2: Acuity 612.022 5 290.646 p<.001*** 9.4877(4) Model 1.3: Infant Characteristics 539.460 9 72.562 p<.001*** 9.4877(4) Model 1.4: Saturated 534.190 11 5.27 .072 5.9914(2) ***p<.001

PAGE 101

87 Model 1.3 was the best fitting model. The difference in log likelihoods between Model 1.3 and Model 1.2 was significantly different, indicating the Model 1.3 had more explanatory power ( 2 = 72.562, 4 df, p=<.001). The difference between Model 1.3 and the Saturated Model, Model 1.4, was not significantly different ( 2 = 5.27, p=.072) indicating that the addition of obstetrical inte rventions, inductions and cesarean sections, did not add any significant value to the explanatory power of the model. The HosemerLemeshow test for goodness of fit for M odel 1.3 had a chi-square value of 6.326 ( p=.611) indicating the observed probabi lities did not significantly differ from the expected probabilities. Comparison of full-risk adjustment mode l with reduced-risk adjustment model. A comparison was made of the model with all risk-adjustment variables, ModelFA (birth weight, APGAR score, ventilation, small for gestational age, gender, race, anomaly, multiple birth, induction, cesarean se ction) with a model that had a reduced number of variables for risk-adjustment, ModelRA (birth weight, ventilation, small for gestation age, gende r and race). ModelFA is given in Tables 12 and 13. ModelRA is provided in Tables 14 and 15. The complete analysis for ModelRA is provided in Appendix I. This comparison was used to dete rmine if transfers could be included in the analyses. There was no information on transfers in the OOIS; so, information on riskadjustment (APGAR, anomaly, multiple, induc tion, cesarean section) was not available on these infants.

PAGE 102

88 Table 14 Logistic Regression ModelRA for Weekend Exposure with Death Before Discharge (n=1837) Confidence Interval B S.E. Wald df Exp(B) Lower Upper Sig. Weekend Admission .083 .227 .133 1 1.086 .696 1.694 .715 Birth Weight -.001 .000 13.765 1 .999 .999 1.000 p<.001*** Infant on Ventilation 3.047 .310 96.489 1 21.062 11.466 38.688 p<.001*** SGA .454 .266 2.917 1 1.575 .935 2.653 .088 Non-Black Race .645 .241 7.135 1 1.906 1.187 3.060 .008 Male Infant .022 .210 .011 1 1.023 .678 1.543 .915 ***p<.001 Model terms: weekend=1, birth weight=100 gram in crements, ventilation=1, sga=1, non-Black race=1, male=1. Table 15 -2 Log Likelihood Tests for ModelRA with Weekend Exposure Difference Model in Log Critical Model -2LL df Likelihoods Sig. Chi-Square (df) Model RA.1: Exposure 902.668 1 .896 Model RA.2: Acuity 642.581 4 260.087 p<.001*** 7.8147(3) Model RA.3: Infant Characteristics 634.943 6 7.638 .022* 5.9914(2) ***p<.001 ModelRA has a -2 log likelihood score of 634.943. The difference between ModelRA and ModelFA -2 log likelihood sta tistic is 95.483 with 5 de grees of freedom. The critical chi-square at 5(df) is 11.0705. Therefore, the differen ce is significant and the null hypothesis that both models are equal is reje cted. The Hosmer-Lemeshow test (HL) for ModelFA is non-significant ( 2 =6.326, df=8, p =.611), while the HL test for ModelRA is significant ( 2 =20.093, df=8 p=.022). Therefore, the ModelRA lacks explanatory power over the full risk adjustment model and is not a good fit. The complete models for ModelRA are provided in Appendix H.

PAGE 103

89 Receiver Operating Characteristic curv es were run for both models. The ROC curves are provided in Appendix I. The Area Under the Curve (AUC) for ModelFA was .932 (.916, .948; SE =.008) and .895 (873, .917; SE =.011) for ModelRA. SPSS does not currently provide a method for comparing tw o ROCs for significant differences (Stephan, Wesseling, Schink, & Jung, 2003). However, given the analysis conducted with the two models using the log likelihood ratio test and the HL test, and given the greater AUC for ModelFA, the model with the full-risk adjustment methodology had gr eater explanatory power and therefore transfers we re excluded from the analysis. TGH NICU admissions compared to transfers. Descriptive analyses compared in-born admissions with transfers to determine how the exclusion of transfers would bias the results. The tw o cohorts were compared to determine statistically significant differences Table 16 compares in-born admissions with transfers or out-born admissions.

PAGE 104

90 Table 16 Comparison of In-Born NICU Admissions with Transfer NICU Admissionsa (N=2572) In-Born Transfer Number Percent Number Percent ChiSquare (df) p-value Exposure Variables Weekend Admissionb 434 23.5% 183 25.2% .82(1) .365 Covariates (non-referent category) Male Infantc 971 52.6% 426 58.7% 8.35(2) .015* Non-Black Race 1291 69.9% 633 87.2% 83.32(1) p<.001*** Infant on ECMO 10 0.5% 24 3.3% 30.52(1) p<.001*** Infant on Ventilation 457 24.8% 314 43.3% 84.81(1) p<.001*** SGA 392 21.2% 104 14.3% 1892.10(1) p<.001*** Outcome Variable Death Prior to Discharge 125 6.8% 62 8.5% 2.42(1) .120 Total 1846 71.8% 726 28.2% aThe following covariates are found in the OOIS and are therefore excluded from this table: multiple birth, APGAR score, delivery type, induction, fetal anomaly. bWeekend Admission includes Saturday and Sunday only since admission time is in the OOIS. bGender unknown for 1 In-born and 1 Transfer cases. *p<.05; ***p<.001 Transfers differed significantly from in -born infants on gender, race, ECMO, ventilation and small for gestational age. They were more likely to be male ( p=.015) and be of a non-Black race ( p=<.001). They were also more likely to be on ECMO ( p<.001), be on ventilation ( p<.001) and be small for gestational age ( p=<.001). However, as a group, t-test results showed they had a higher mean birth weight (2,411.72 versus 2,090.55, p=<.001) and were of an older ge stational age ( 34.41 versus 33.32, p=<.001).

PAGE 105

91 Even though they had a higher proportion of infants on ventilation, they did not have significantly more days on ventilation (4.40 versus 3.45, p=.159). Results of t-test are given in Appendix J. Transfers were not significantly more likely to be admitted on the weekend and were not significantly more likely to die prior to discharge when compared to in-born infants. The unadjusted odds ratio for out-bor n infants was also not significant (UOR=1.286, 95% CI=.936, 1.766). When entered into ModelRA, the odds ratio was still non-significant (AOR=.837, 95% CI=.582, 1.205) Given that they did not differ significantly on the exposure or outcome variable, it is not exp ected that the exclusion of transfers will significantly bias the results. Re search Question 2 The research question under investigation was : Is there an association between the time of admission (evening or nighttime versus day) to the NICU and the infants outcome? The multivariate analysis found no si gnificant association between nighttime admission and death prior to discharge. The model which was tested is defined as Model 2. There were no cases with Dfbeta values greater than one for this model. Table 17 displays the results of the model with the best fit, Table 18 displays the results of the log likelihood tests on each model and Appendix K provides each model for each bl ock in the logistic regression analysis.

PAGE 106

92 Model 2: DEATH=NA + BW + VENT + APGAR+ SGA + ANOMALY + RACE + GENDER + MULTIPLE + +IND + TD + error Table 17 Logistic Regression Model 2 for Nighttime Exposure with Death Before Discharge (n=1837) Confidence Interval B S.E. Wald df Exp(B) Lower Upper Sig. Nighttime Admission .128 .232 .304 1 1.136 .722 1.789 .582 Birth Weight -.120 .019 40.152 1 .887 .855 .921 p<.001*** Infant on Ventilation 2.323 .326 50.622 1 10.203 5.381 19.347 p<.001*** APGAR < 7 1.188 .239 24.639 1 3.279 2.052 5.241 p<.001*** SGA .127 .293 .188 1 1.136 .639 2.018 .665 Fetal Anomaly 2.681 .351 58.474 1 14.603 7.345 29.035 p<.001*** Non-Black Race .751 .265 8.025 1 2.120 1.260 3.565 .005** Male Infant .064 .229 .078 1 1.066 .681 1.670 .780 Multiple Birth .214 .284 .570 1 1.239 .710 2.162 .450 **p<.01; ***p<.001 Model terms: nighttime=1, birth weight=100 gram increments, ventilation=1, APGAR=1, sga=1, anomaly=1, non-Black race=1, male=1, multiple=1. Nighttime admission was not significantly associated with risk of dying in the adjusted model. The odds ratio declined in the multivariate analysis (UOR=1.360 and AOR=1.136) when covariates related to in fants acuity were added. Birth weight, ventilation, Apgar score, anomaly and non-Black race were also significant in Model 2. Birth weight, Apgar score, ventilation, anom aly and non-Black race we re all consistent with their adjusted odds ratios in Model 1. Cesarean section was not significant.

PAGE 107

93 Table 18 -2 Log Likelihood Tests for Model 2 with Nighttime Exposure Difference Model in Log Critical Model -2LL df Likelihoods Sig. Chi-Square (df) Model 2.1: Exposure 899.980 1 .100 Model 2.2: Acuity 611.884 5 288.96 p<.001*** 9.4877(4) Model 2.3: Infant Characteristics 539.212 9 72.672 p<.001*** 9.4877(4) Model 2.4: Saturated 534.168 11 5.044 .080 5.9914 (2) ***p<.001 Model 2.3 was the best fitting model. The difference in log likelihoods between the Model 2.3, and Model 2.2 was significantl y different, indicating that Model 2.3 had more explanatory power ( 2 = 9.4877, 4 df, p<.001). The Saturated Model, 2.4, was not significantly different from Model 2.3 and therefore the null hypot hesis that the two models are equal is not rejected. The Ho semer-Lemeshow test for goodness of fit had a chi-square value of 6.326 (p=.615) indicating the obser ved probabilities did not significantly differ from the e xpected probabilities for Model 2.3 and that this model is the best fit for the data. These result s agree with the results of Model 1. Research Question 3 The research question under investigation was: Is there effect modification between day of admission and time of admission? The multivariate analysis found no significant association between the exposure variables and death prior to discharge. This model tested for effect modifica tion between the two exposure variables, weekend and nighttime. The model which was te sted is defined as Model 3. There were no cases with Dfbeta values greater than one for this model. Table 19 displays the results

PAGE 108

94 of the model with the best fit, Table 20 disp lays the results of the log likelihood tests on each model and Appendix L provides each model for each block of entry in the logistic regression analysis. Model 3: DEATH=WDA + NA + WDA*NA + BW + VENT + APGAR+ SGA + ANOMALY + RACE + GENDER + MULTIPLE + IND + TD + error Table 19 Complete Logistic Regression Model 3 for Effect Modification with Death Before Discharge (n=1837) Confidence Interval B S.E. Wald df Exp(B) Lower Upper Sig. Weekend Admission .205 .363 .321 1 1.228 .603 2.499 .571 Nighttime Admission .211 .275 .590 1 1.235 .721 2.115 .442 Weekend*Nighttime -.301 .498 .364 1 .740 .279 1.967 .546 Birth Weight -.120 .019 40.489 1 .887 .855 .920 p<.001*** Infant on Ventilation 2.323 .327 50.556 1 10.203 5.378 19.354 p<.001*** APGAR < 7 1.187 .239 24.584 1 3.277 2.050 5.238 p<.001*** SGA .130 .294 .197 1 1.139 .641 2.026 .657 Fetal Anomaly 2.668 .350 58.078 1 14.409 7.255 28.617 p<.001*** Non-Black Race .766 .267 8.238 1 2.150 1.275 3.627 .004* Male Infant .070 .229 .093 1 1.072 .684 1.681 .761 Multiple Birth .217 .285 .582 1 1.242 .711 2.170 .445 **p<.01; ***p<.001 Model terms: weekend=1, nighttime=1, birth weight=100 gram increments, ventilation=1, APGAR=1, sga=1, anomaly=1, non-Black race=1, male=1, multiple=1. Significant results followed the same pa ttern observed in Model 1 and Model 2. Significant covariates were bi rth weight, ventilation, Apgar sc ore, fetal anomaly and nonBlack race. The value of these covariates in terms of their coefficients, odds ratios and

PAGE 109

95 confidence intervals were similar with the other two models. Neither exposure variable was significant nor was the eff ect modification significant. Table 20 -2 Log Likelihood Tests for Model 3 with Effect Modification Difference Model in Log Critical Model -2LL df Likelihoods Sig. Chi-Square (df) Model 3.1: Exposures 899.248 3 .329 Model 3.2: Acuity 611.568 7 287.680 p<.001*** 9.4877(4) Model 3.3: Infant Characteristics 538.819 11 72.749 p<.001*** 9.4877(4) Model 3.4: Saturated 533.743 13 5.076 0.079 5.9914 (2) ***p<.001 Model 3.3 was the best fitting model. The difference in log likelihoods between Model 3.3 and Model 2.3 with exposures a nd acuity was significantly different, indicating Model 3.3 had mo re explanatory power ( 2 = 9.4877, 4 df, p =<.001). The Hosemer-Lemeshow test for goodness of f it had a chi-square value of 5.644 ( p=.687) indicating the observed probabi lities did not significantly differ from the expected probabilities. The difference between the Model 3.3 and the Satura ted Model 3.4 was not significantly different so the null hypothesis that these tw o models are equal is not rejected. These results are almost the same as those achieved with Model 1 and Model 2. Research Question 4 The research question under investigation was: Do staffing patterns of nurses in the NICU mediate the associat ion between day or time of admission to the NICU and the

PAGE 110

96 infants outcome? The multivariate analyses found no significant association between the exposure variables and the death prior to discharge. Three models were tested: one for ea ch exposure and one for the effect modification term. Model 4B had 10 cases w ith Dfbetas greater than one for the covariate nurse to infant rati o by shift and 1 case with Dfbetas greater than one for the covariate capacity. Each of these cases was removed from the model and the statistical findings of the best fitting model did not ch ange. These models are defined as Model 4A, Model 4B and Model 4C. Tables 21, 23 and 25 display the results of the model with the best fit. Tables 22, 24 and 26 display the resu lts of the log likelihood tests on each model. Appendices M, N and O provide the results each model for each block of entry in the three logistic regression analyses for this se ction. The blocks for entry were defined as: Block 1: Exposure variable (s) (weekend, nighttime, weekend*nighttime) Block 2: Infant acuity (birth we ight, ventilation, Apgar score, SGA) Block 3: Infant characteristics (anomaly, gender, race, multiple) Block 4: OB interventions (induction, cesarean section) Block 5: Hospital-level c ovariates (RN hours, capacity, nurse to infant ratio). Model 4A contained the exposure variab le for weekend admission with all the covariates including registered nursing hours per day, capacity per day and RN to infant ratio per day. This model is defined below.

PAGE 111

97 Model 4A: DEATH=WDA + BW + VENT + APGAR + SGA + ANOMALY + IND + RACE + GENDER + MULTIPLE + TD + RNHR + CAP + NIRDAY+ error Table 21 presents the results for research question 4A. The mode l selected, 4A.3 (see Table 22) is the same as the model selected for research question 1, Is there an association between the day of admission (w eekday versus weekend) to the NICU and the infants outcome? Model 1.3. Table 21 Logistic Regression Model 4A for Weekend Exposur e with Hospital-Level Covariates Included with Death Before Discharge (n=1837) Confidence Interval B S.E. Wald df Exp(B) Lower Upper Sig. Weekend Admission .058 .247 .055 1 1.060 .653 1.721 .814 Birth Weight -.120 .019 40.569 1 .887 .855 .920 p<.001*** Infant on Ventilation 2.316 .326 50.574 1 10.134 5.353 19.186 p<.001*** APGAR < 7 1.208 .236 26.110 1 3.347 2.106 5.319 p<.001*** SGA .109 .291 .141 1 1.115 .630 1.973 .708 Fetal Anomaly 2.672 .350 58.255 1 14.474 7.287 28.750 p<.001*** Non-Black Race .762 .265 8.236 1 2.142 1.273 3.604 .004** Male Infant .063 .229 .075 1 1.065 .680 1.668 .784 Multiple Birth .205 .283 .526 1 1.228 .705 2.139 .468 **p<.01; ***p<.001 Model terms: weekend=1, birth weight=100 gr am increments, ventilation=1, APGAR=1, sga=1, anomaly=1, non-Black race=1, male=1, multiple=1. The results presented in Table 22 show th e results for each step of model entry with the Saturated Model, 4A.5, being the bl ock where registered nursing hours for the day, capacity for the day and nurse to infant ratio for the day were entered. The addition

PAGE 112

98 of nursing and capacity covari ates was not significant ( 2=2.342, p=.504; critical 2 =7.8147, df=3). Each of these covariates was also non-significant (Appendix M). Table 22 -2 Log Likelihood Tests for Model 4A with Week end Exposure with Hospital -Level Covariates Difference Model in Log Critical ChiModel -2LL df Likelihoods Sig. Square (df) Model 4A.1: Exposure 902.668 1 .896 Model 4A.2: Acuity 612.022 5 290.646 p<.001*** 9.4877(4) Model 4A.3: Infant Characteristics 539.460 9 72.562 p<.001*** 9.4877(4) Model 4A.4: OB Interventions 534.190 11 5.270 .072 5.9914(2) Model 4A.5: Saturated 531.848 14 2.342 .504 7.8147(3) ***p<.001 Model 4B contained the exposure variab le nighttime admission with all the covariates including registered nursing hours by 12-hour shift, capacity for the entire day and the nurse to infant ratio for the shift. This model is defined below. Model 4B: DEATH=NA + BW + VENT + APGAR + SGA + ANOMALY + RACE + GENDER + MULTIPLE + IND + TD + SHIFTHR + CAP + NIRSHIFT+ error Table 23 presents the results for research question 4B. The mode l selected, 4B.3 (see Table 24) is the same as the model selected for research question 2, Is there an association between the time of admission (nighttime versus daytime) to the NICU and the infants outcome? Model 2.3.

PAGE 113

99 Table 23 Logistic Regression Model 4B for Nighttime Exposu re with Hospital-Level Covariates Included with Death Before Discharge (n=1837) Confidence Interval B S.E. Wald df Exp(B) Lower Upper Sig. Nighttime Admission .128 .232 .304 1 1.136 .722 1.789 .582 Birth Weight -.120 .019 40.152 1 .887 .855 .921 p<.001*** Infant on Ventilation 2.323 .326 50.622 1 10.203 5.381 19.347 p<.001*** APGAR < 7 1.188 .239 24.639 1 3.279 2.052 5.241 p<.001*** SG .127 .293 .188 1 1.136 .639 2.018 .665 Fetal Anomaly 2.681 .351 58.474 1 14.603 7.345 29.035 p<.001*** Non-Black Race .751 .265 8.025 1 2.120 1.260 3.565 .005** Male Infant .064 .229 .078 1 1.066 .681 1.670 .780 Multiple Birth .214 .284 .570 1 1.239 .710 2.162 .450 **p<.01; ***p<.001 Model terms: nighttime=1, birth weight=100 gram increments, ventilation=1, APGAR=1, sga=1, anomaly=1, non-Black race=1, male=1, multiple=1. The results presented in Table 24 show th e results for each step of model entry with the Saturated Model, 4B.5, being the bl ock where registered nursing hours for the shift, capacity for the day and nurse to infant ratio for the shift were entered. The addition of nursing and capacity covari ates was not significant ( 2=5.143, df=3, p=.162). Each of these covariates was also non-significant (Appendix N). Table 24 -2 Log Likelihood Tests for Model 4B with Nigh ttime Exposure with Hospital-Level Covariates Difference Model in Log Critical Chi Model -2LL df Likelihoods Sig. Square (df) Model 4B.1: Exposure 899.980 1 .100 Model 4B.2: Acuity 611.884 5 288.96 p<.001*** 9.4877(4) Model 4B.3: Infant Characteristics 539.212 9 72.672 p<.001*** 9.4877(4) Model 4B.4: OB Interventions 534.168 11 5.044 .080 5.9914 (2) Model 4B.5: Saturated 528.328 14 5.84 .120 7.8147 (3) ***p<.001

PAGE 114

100 Model 4C contained the effect modifi cation term weekend*nighttime admission with all the covariat es including registered nursing hours by 24-hour shift, capacity for the entire day and the nurse to infant rati o for the day. This model is defined below. Model 4C: DEATH=WDA + NA + WDA*NA + BW + VENT + APGAR + SGA + ANOMALY + RACE + GENDER + MULTIPLE + IND + TD + RNHR + CAP + NIRDAY+ error Table 25 presents the results for research question 4C. The model selected, 4C.3 (see Table 26) is the same as the model selected for research question 3, Is there effect modification between day of admission and time of admission?, Model 3.3. Table 25 Logistic Regression Model 4C for Effect Modifica tion with Hospital-Level Covariates Included with with Death Before Discharge (n=1,837) Confidence Interval B S.E. Wald df Exp(B) Lower Upper Sig. Weekend Admission .205 .363 .321 1 1.228 .603 2.499 .571 Nighttime Admission .211 .275 .590 1 1.235 .721 2.115 .442 Weekend*Nighttime -.301 .498 .364 1 .740 .279 1.967 .546 Birth Weight -.120 .019 40.489 1 .887 .855 .920 p<.001*** Infant on Ventilation 2.323 .327 50.556 1 10.203 5.378 19.354 p<.001*** APGAR < 7 1.187 .239 24.584 1 3.277 2.050 5.238 p<.001*** SGA .130 .294 .197 1 1.139 .641 2.026 .657 Fetal Anomaly 2.668 .350 58.078 1 14.409 7.255 28.617 p<.001*** Non-Black Race .766 .267 8.238 1 2.150 1.275 3.627 .004* Male Infant .070 .229 .093 1 1.072 .684 1.681 .761 Multiple Birth .217 .285 .582 1 1.242 .711 2.170 .445 **p<.01; ***p<.001 Model terms: weekend=1, nighttime=1,birth weight=100 gram increments, ventilation=1, APGAR=1, sga=1, anomaly=1, non-Black race=1, male=1, multiple=1.

PAGE 115

101 The results presented in Table 26 show th e results for each step of model entry with the Saturated Model, 4C.5, being the bl ock where registered nursing hours for the day, capacity for the day and nurse to infant ratio for the day were entered. The addition of nursing and capacity covari ates was not significant ( 2=2.169, p=.538; critical 2 =7.8147, df=3). Each of these covariates was also non-significant (Appendix O). Table 26 -2 Log Likelihood Tests for Model 4C with Effect Modification with Hospital Level Covariates Difference Model in Log Critical Model -2LL df Likelihoods Sig. Chi-Square (df) Model 4C.1: Exposures 899.248 3 .329 Model 4C.2: Acuity 611.568 7 287.680 p<.001*** 9.4877(4) Model 4C.3: Infant Characteristics 538.819 11 72.749 p<.001*** 9.4877(4) Model 4C.4: OB Interventions 533.743 13 5.076 .079 5.9914 (2) Model 4C.5: Saturated 531.573 16 2.169 .538 7.8147 (3) ***p<.001 Summary of the Research The results presented in this section fa iled to support the association between weekend admission and nighttime admission to a NICU and subsequent higher risk of death prior to discharge. The multivariate anal ysis showed that the higher risk of death was associated with infant acuity and other characteristics such as birth weight, being on a ventilator, an Apgar score of less than 7 at 5 minutes, having an anomaly detected during the fetal period and race of non-Black. The additional of covariates related to nursing hours or NICU capacity did add explain the higher odds of death observed in the sample for the study.

PAGE 116

102 Chapter Five: Discussion The purpose of this study was to investi gate the time of admission to a Neonatal Intensive Care Unit (NICU) and its associati on with in-hospital mortality among a cohort of neonates at a regi onal perinatal center. Two different time point s were considered. The first time point was admission on the weekend versus the weekday. The second time point was admission during the nighttime shift versus the day shift. The secondary purpose of the study was to investigate if registered nurse staffing affected this association between NICU admission day or admission time and in-hospital death. Main Study Findings This section presents the main study findings as they relate to the research questions under investigation. Discussion begi ns with the exposure variables, weekend admission and nighttime admission, and their relationship with the outcome variable, death prior to discharge. Th e significant findings related to infant acuity and other characteristics are given. Then, the findi ngs related to nurse staffing are provided. Association of Day or Time of NICU Admission with Mortality Neither admission to the NICU during the weekend nor during th e nighttime shift was associated with an increased risk of dyi ng prior to discharge from the hospital. The

PAGE 117

103 descriptive analyses showed that infants who are admitted on the weekend or during the nighttime have a higher proportion of deaths (6.9% weekend versus 6.7% weekday; 7.9% nighttime versus 5.9% day). When weekend admissions were defined as Saturday or Sunday only, there was still no significant difference in mortality. Saturday and Sunday admissions had a lower proportion of deaths compared to weekday admissions (6.5% versus 6.9%; p=.762). This higher proportion of deaths for weekday admissions may indicate the influence of nighttime admissions since in this study Friday night after 7:00 pm and Monday morning prior to 7:00 am were defined as nighttime admissions. However, multivariate analysis found that the odds ratio of dying if admitted on the weekend was non-significant (AOR =1.060, 95% CI=.653, 1.721) and if admitted during the nighttime was also non-sign ificant (AOR=1.136, 95% CI=.772, 1.789). The addition of an effect modificat ion term did not provide a si gnificant associa tion regarding admission by day or time and infant death (AOR=.740, 95% CI=.721, 2.115). Nighttime admissions had a significantly higher proportion of low birth weight infants (less than 2,500 grams) compared to day admissions (77.2% versus 68.2%, p=.040). The case mix of infants admitted during the nighttime could explain the difference in mortality. Multivariate models confirm that case mix is an underlying cause for the increased mortality during both the weekend and nighttime. Covariates related to infant acuity and infant characteristics were the only covariates to consistently remain significant regardless of the exposure being te sted or the addition of other covariates. When variables related to infants acuity and other characteristics were entered into the model, the adjusted odds ratio for nigh ttime admissions decreased from 1.36 (95%

PAGE 118

104 CI=943, 1.962) to 1.14 (95% CI=.722, 1.789). Th e adjusted odds ratio for weekend admissions remained relatively unchanged. Lee et. al. (2003) and Abdel-Latif et. al. (2006) used NICU data in their investigation of nighttime admissions and a higher odds of mortality. However, they limited the outcome to death within the first seven days after admission instead of the entire length of stay. They also included infants less than 32 weeks gestation since these infants were considered those most sensitive to the effects of patient care after admission. Lee found an increased odds of death of 60% (AOR=1.6, 95% CI=1.1, 2.3) if admitted at night. However, Abdel-Latif found no significa nt association with early neonatal death and nighttime admission to the NICU (AOR=1.07, 95% CI=.881, 1.30). The strongest association was found with the presence of a fetal anomaly. The AOR for a fetal anomaly ranged from 14.5 (95% CI=7.29, 28.75) to 14.6 (95% CI=7.35, 29.04). An infant being on ventilation showed a ten-fold increased odds of dying that remained consistent in each model tested a nd with each exposure. An Apgar score of less than 7 at five minutes remained consistently strong in each model with an adjusted odds ratio of 3.4 (95% CI=2.11, 5.3) for weekend admissions and 3.3 (95% CI=2.05, 5.24). The other significant covariate was a non-Black race which had a consistent risk of dying two times higher than Black infants. Birth weight was a strong protective factor since it was measured as a continuous variable. For every 100 gram increase in birth weight, the odds of dying declined by 11%. This is in agreement with the Clinical Risk Index for Babies (CRIB) and the SNAP (score for Neonatal Acute Physiology) risk-adjustment scoring systems. Both use

PAGE 119

105 birth weight and Apgar score. Birth weight is used as a ca tegorical variable and Apgar uses the same categories used in this study. The CRIB uses birt h defects which are defined as none, not acutely life threatening and acutely life threatening. Congenital anomalies which are considered potential lif e threatening such as anencephaly, Trisomy 18 and renal agenesis are excluded. SNAP does no t use birth defects in its scoring system (Cockburn et al., 1993; Richards on et al., 2001). In this st udy anomalies were those detected during the fetal period and they were not identified regarding the type. The ROC comparison of Model FA and Model RA showed an AUC similar to that found with the CRIB and SNAP with an Area Under the Curve of .93. The CRIB had an AUC of .90-.92 in the study by Cockburn and co lleagues (1993) and the SNAP had an AUC of .91 in the study by Richardson and coll eagues (2001). Furthe r analysis of the Model FA found when the covariate fetal anomaly was removed, the AUC was reduced to .908 (95% CI=.887, .929, SE =.011), when ventilation was removed the AUC was reduced to .896 (95% CI=.869, .924, SE =.014) and when both were removed the AUC was reduced to .850 (95% CI=.810, .890, SE =.020) (ROC curves found in Appendix P). The AUC for this study was higher than exp ected. However, both the presence of an anomaly and the use of ventilation proved to be strong predicators of the outcome. Association of Nurse Staffing Nurse staffing was not found in this study to have a significant relationship with the odds of dying prior to discharge. The adjusted odds ratio for nursing hours per day was .998 but was non-significant ( p=.911). Since this study inve stigated the outcome of

PAGE 120

106 death prior to discharge, nursing during the first 24 hours of admission may not have been as sensitive a measure. Of the 125 infants who died in this study, only 36 died within the first 24 hours of admission. The average time these infants remained in the NICU prior to death was 18.6 days. Measur ement of nursing hours during the entire length of stay may have provided more information. The use of registered nurs ing hours by shift produced cas es with leverage in the model and also a non-significant result for nur sing variables in Model 4B. The beta coefficient for registered nursing hours by shift produced a positive coefficient which was unexpected. It is possible that the use of daily census, as opposed to census by shift, may have influenced these results. To the extent that census doe s not fluctuate during the day, there should be no bias. To the extent that there are higher than normal admissions or discharges which occur throughout the day, daily census may not be a good measure when used with registered nursing hours by shif t. Discharge time is not in the NICUD so census by shift could not be calculated. Additiona lly, the beginning census for the study was obtained by TGH and census is de fined as bed count at midnight. The average number of admission per da y in the NICU is 3.03 and the average number of discharges per day is 1.94. One-thir d of the admissions were above this mean and one-third of the discharges were above this mean. There were 36 infants who died within one day of admission. For these infants, 14 were admitted on days with above average number of admissions.

PAGE 121

107 Limitations of the Study Sample Size The power analysis in Chapter 3 show ed that with 3,500 admissions there was 80% power to detect an odds ratio of 1.3 (95% confidence interval, to lerance of .8) given a rate of 3.8% for infant deaths. The sa mple size for the study was actually 1,846. The adjusted odds ratios in this study were 1.06 for weekend exposures and 1.14 for nighttime exposures. When the effect modification te rm was introduced, the adjusted odds ratios increased to 1.23 for weekend exposures and 1.24 for nighttime exposures. To re-evaluate the power analysis, a new analysis was computed for an infant death rate of 6.8% which was the observe d rate in the study. Tolerance was then computed for both exposures. Tolerance is measured as 1-R2. Each exposure variable was entered into a least squares regression equati on as the dependent variable with the other covariates as the independent variables (T abachnick & Fidell, 2001). Tolerance for the weekend exposure variable was .951 and for the nighttime exposure variable was .637. Given the results of the revised power analys is, with the tolerance levels computed, a sample size of 12,631 would have been needed to detect an odds ratio of 1.10 with .951 tolerance (weekend exposure ) and 18,947 to detect an odds ratio of 1.10 with .637 tolerance (nighttime exposure). For the highe r odds ratios observed with the effect modification term, a sample size of 3,451 would have been needed for the odds ratio of 1.228 with for weekend exposure and a sample size of 5,176 would have been needed to detect an odds ratio of 1.235. Therefore, statistical power to detect a significant

PAGE 122

108 relationship between weekend and nighttime ad missions and death prior to discharge at the observed odds ratios wa s lacking in this study. However, this study did have the statisti cal power to detect an odds ratio of 1.3 for the weekend exposure variable and 1.4 for the nighttime exposure variable. Lee et. al. (2003) found an increased risk of 60% ( AOR=1.6, 95% CI=1.2, 2.4) in their study of nighttime admissions to 17 NICUs in Canada. Other studies which investigated this association using linked vital statistics data also found odds ratios in the range of 1.282.09 for nighttime births (Heller et al., 2000; Luo & Karlberg, 2001; Stephansson et al., 2003), and 1.2 for weekend births (Hong et al., 2006). Study Setting Therefore, the lack of a st atistically significant impact ma y be due to the type of NICU studied. TGH is one of eleven Regional Perinatal Intensive Care Centers in Florida and one of six which uses ECMO. It is a Level III NICU. Only University of South Florida board certified neonatologists are allowe d to provide patient ca re in the unit. Lee and colleagues (2003) found that the presence of an in-house attending or fellow reduced the odds of dying at night in the NICU by 40% (AOR=0.6, 95% CI=0.4,0.9). Abdel-Latif et. al. (2006) did not find any significant association with nighttime admission to the NICU and neonatal death in their study after regionalization had occurred in Australia. TGH also received designation as a Ma gnet Hospital by the American Academy of Nursing. Aiken (2000) found that hospitals with this designation have a higher proportion of registered nurses with a Bachelors degree in Nursing and higher ratios of

PAGE 123

109 nurses to patients. Given the study setting, th e TGH NICU is staffed with very skilled nurses trained in providing neona tal care to high-risk infants. Numerous studies cited in the Literature Review showed the value of skilled nursing and appropriate nurse to patient ratios in reducing adverse events in th e hospital. The lack of a significant finding related to nursing may be due to a lack of statistical power or the nursing definition used. Missing Data Infants who were transferred in did not have data available for the following covariates: Apgar score, multiple birth, i nduction, c-section and fetal anomaly. The exclusion of transfers affected sample size which affected statistical power, but the inclusion of 726 additional cases would not ha ve sufficiently increased the power of this study. However, the unadjusted odds ratio did no t show a statistically significant increase in mortality for transfers and transfers did not die at a significantly higher rate if admitted during the weekend, even though they were a higher risk cohort of infants. Therefore, it is not expected that their exclusi on affected the study results. The data on fetal anomalies only indicated the presence or absence of an anomaly detected during the fetal period. The type of anomaly was unknown. Therefore, infants with anomalies with a high case fatality rate (anencephaly, renal agencies, Truism 13) could not be excluded from the study. CRIB which uses birth defects in its scoring system excludes infants with these anomalies. However, the multivariate models in the study should have controlled for the presence of life threatening anomalies.

PAGE 124

110 Both CRIB and SNAP use physiological measures in their risk-adjustment scoring systems. These types of measure were not available for this study. However, the ROC analysis showed comparable results to both CRIB and SNAP. Nurse Staffing The staffing database contained informati on on scheduling only. Therefore, it was assumed that if a nurse was scheduled to wor k, that nurse came to work on that day and shift and worked the entire shift. To the exte nt that this did not happen, the coefficient in the models would overestimate the effects of nursing and potentially bi as the results away from the null. If this were more likely to happen during the weekend or nighttime shifts, the misclassification bias would be non-ra ndom. Since the study design did not include review of personnel records, there is not way to know if this happened. However, the study did not find any significant association wi th nurse staffing patterns and therefore it is unlikely that any overestima tion of the effect occurred. This study did average the effects of nursi ng over an entire shift or day and was unable to study actual nurse to patient ratios or actual wo rkload per nurse. This could have biased the result s toward the null and therefore faile d to show an effect. Since the outcome variable under study was death prior to discharge, limiting the nursing variables to the first 12 to 24 hours under study would have not measured the impact of nursing over the entire course of stay. Nursing experience and years of education were important covariates in studies discussed in the Literature Review related to adverse events and nurse staffing. That data

PAGE 125

111 was not collected for this study. The use of pool nurses or nurses from a non-critical care unit can also adversely affect patient outcomes. This inform ation was also not available for this study. Physician Staffing Physician staffing could not be used for this study. The database which contained physician scheduling was missing 37% of th e study time period. Given that Tampa General Hospital is a regional center, it is not expected that the exclusion of this information seriously biased the results of the study. TGH NICU is a closed unit with board certified neonatologist s from USF College of Medi cine providing physician care. As a teaching hospital, TGH also uses residents in the NICU. The TGH NICU is staffed with neonatologists 24 hours per day, se ven days per week. Lee (2003) found in a multiple site study conducted in Canada of regional centers that the presence of an attending or fellow neonatologist in the NICU at night reduced mortality by 40%. Research which uses nurse or physician st affing is faced with many issues. First, it is often unknown, especially if secondary data are used, if those scheduled were actually in the unit for the time they were sc heduled to be there. Second, the actual caseload of a nurse is often unknown and caseloa d in this study was an estimate based on the census and nursing hours in the unit. Third, the level of experience and education of nurses is often unknown and difficult to extract from secondary databases used for scheduling. For physicians staffing information, the year of residency or fellowship may

PAGE 126

112 not be known. Physicians scheduled for the un it may be on-call and not in the unit during the time for which they are scheduled. Definition of Exposure Variables Studies in this area of in vestigation varied in how we ekend or nighttime births or admissions were defined. Some studies include d holidays with weekends, evaluated them separately or kept them on the day of the week they occurred. Studies defined weekend either as Saturday or Sunday or incl uded Friday night and Monday morning. The two studies that looked specifically at NICU admissions and data, defined nighttime admissions differently. Day admissions were defined as 8:00 am to 6:00 pm and nighttime admissions were defined as 6:01 pm to 7:59 am in one study (Abdel-Latif et al., 2006) and daytime as 8:00 am to 5:00 pm and nighttime as 5:00 pm to 8:00 am in the other (Lee et al., 2003). The definition s eems to depend on the hospital(s) under study and how they define the shifts. Additionally, much of the re search used linked birth a nd infant death records and investigated the day or time of birth, wh ile later research focused on the time of admission to the neonatal intensive care unit. A lack of consistency in definitions of weekend or nighttime admissions hinders the synthesis of this research into specific conclusions that can aid hospita ls in staffing patterns and other interventions to reduce infant deaths.

PAGE 127

113 Public Health Implications The infant mortality rate increased for th e first time in four decades in 2002. The advances seen during the nineties with respect to reductions in infant mortality have stalled. These reductions have been associated with improve ments in neonatal care (e.g. surfactants, mechanical ventilation) and reductions in Sudden Infant Death Syndrome (Sappenfield, 2007). Public health should not lose focus on the importance of the birth process, access to quality prenatal care and the hospital of delivery. This study agreed with other research which shows the concentrations of cesarean sections and inductions during the weekday are increasing. These deliveries peak on Friday and peak between the hours of 7 am to 8 am (Goodman, Sappenfield, & Thompson, 2007), thereby potentially changing the case mix for NICU admissions on th e weekend and at night. This study showed that delivery in an a ppropriate facility for high-risk infants enhances the chance for survival. It is impor tant that public health programs monitor the system of high-risk obstetrical care in co mmunities to ensure that women deliver at hospitals equipped with appropriately staffed neonatal intensive care units. Research in Florida has found that there is no significant racial disparity of infants who die within the first seven days of life and weigh at least 1,500 gr ams. This is attributable, in part, to the regional system of NICUs in Florida (L. Stanley, personnel community with J. Murphy, 2003). During the past ten years, Florida has seen an increase of 45,000 in the number of live births. Surveillance of regional centers and Level III NICUs at private hospitals

PAGE 128

114 should be maintained to evaluate if current capacity can absorb and can continue to absorb the subsequent increase in NICU admissions. The regionali zation of NICU care has contributed to a reduction in infant deaths in the st ate. Policy makers and public health advocates need to ensure that funding fo r this system is adequately maintained and meets the critical need for NICU beds. Further, in Florida the rate of women delivering without prenatal care has been increasing since 2002. This jeopardizes the ab ility to ensure that high-risk pregnant women are seen by appropriate maternal fetal medicine providers and deliver in Level III facilities. In 2006, there were 4,100 (1.9%) wo men in Florida who delivered without prenatal care compared to 1,859 (.9%) in 2002. Black women comprised 21% of all live births in Florida but repres ented over one-third of thos e women who delivered without prenatal care (Florida Department of Health, 2006). Many public he alth clinics are at capacity as more pregnant women find themselves without health insurance either due to employer-based coverage being eliminated or reduced, reductions in Medicaid or difficulty in obtaining Medicaid due to ci tizenship documentati on required under the Deficit Reduction Act of 2005 (Stanley, 2006). P ublic health needs to monitor this issue and how lack of access to prenatal care will affect where women with high-risk pregnancies deliver and how t hose pregnancies are managed. Public health research and programming often focuses on individual level risk factors. In order to more fully understand infa nt mortality, research needs to address not only individual-level risk fact ors but also contextual-level variables and hospital-level variables and the interaction between the th ese three. Often, public health programs are

PAGE 129

115 focused on changing individual level be haviors or risks without acknowledging, addressing or understanding how the context in which a woman lives or the health care system she is exposed to influences her re productive choices, outco mes and behaviors. Future Research Implications This section discusses future research implications as they relate to this studys findings. To the extent that the Tampa Ge neral Hospitals NICUD would be used for other research projects, the mothers medical record number in the NICUD would enable easier database linkage with the mothers obste trical record. Obstetri cal data on transfers should also be collected to allow for rese arch involving all neonates in the unit. Additional studies should evalua te if mortality by day or ti me of admission is different for infants at less than 32 week s gestation or different for deat hs within the first week of life. This study also found that for Black wome n, they were more likely to deliver during the weekend. The underlying reasons for this finding are important to investigate. The use of cesarean sections and inductions during the week and how t hose influence case mix of NICU admissions during the weekend and night should also be monitored. In order to adequately an swer the questions regarding admission day or time and it association with infant mortal ity, studies need to include more than one site and type of hospital. A multi-site study would allow the re searcher to investigate different staffing patterns of both physician and nurses and diffe rent types of hospitals while controlling for the influence of case mix. For instance, t eaching hospitals could be compared to nonteaching hospitals, thos e with 24 hour physician coverage could be compared with those

PAGE 130

116 without that type of covera ge, hospitals with magnet designation could be compared to those hospitals without that status. A sufficiently large sample size would be needed to detect a small odds ratio while controlling for acuity and the effects of hospital type and staffing. The number needed to detect an odds ratio in this study would have been costly. With approximately 200,000 births in Florida, approximately 15% needing NICU care and the exclusionary criteria in this study, an es timated one to two years of NICU admissions for the state would be needed to detect a significa nt association of the level observed. Both nurse and physician staffing should acc ount for factors other than scheduled time in the unit. Years of experience and tr aining are important f actors in provision of patient care. Actual workload for nurses is a more sensitive measure than average census for the day. Since the research question concerned admission day or time and subsequent death, future studies, if there is a large e nough sample size, should limit death to the first 24 hours to one week of admission and limit nursing and physician staffing to the time period related to that admission. If death is exte nded to time before discharge, then nurse and physician staffing should cover the time in the unit. Cost-effectiveness analysis is needed to determine how an increase of one registered nurse per shift affects a decreas e of neonatal deaths and by what amount. The cost of adding one nurse per shift over the numbe r of shifts in the course of a year could add into the hundred of thousands of dollars. Hospitals need to understand the benefit in order to make an informed decision regarding registered nurse staffing.

PAGE 131

117 In conclusions, public health research needs to continue multiple avenues of investigation in order to better understand the underlying is sues associated with infant mortality. These investigations should i nvolve the study of individual-level risks, contextual level factors and hospital-leve l factors and the interaction among each of them.

PAGE 132

118 List of References Abdel-Latif, M., Bajuk, B., & Lui, K. ( 2006). Mortality and morbidities among very premature infants admitted after hours in an Australian Neonatal Intensive Care Unit Network. Pediatrics, 117 (5), 1632-1639. Acolet, D., Elbourne, D., McIntosh, N., Weindling, M., Korkodilos, M., Haviland, J., et al. (2005). Project 27/28: inquiry into quality of neonatal care and its effect on the survival of infants who were born at 27 and 28 weeks in England, Wales, and Northern Ireland. Pediatrics, 116 (6), 1457-1465. Adams, M., Read, J., & Rawlings, J. ( 1993). Preterm delivery among black and white enlisted women in the United States army. Obstetrics and Gynecology, 81 65-71. Agresti, A. (2002). Categorical Data Analysis (2nd ed.). Hoboken: John Wiley & Sons, Inc. Aiken, L., Clarke, S., Cheung, R. B., Sloane, D., & Silber, J. (2003). Educational levels of hospital nurses and surgical patient mortality. Journal of the American Medical Association, 290(12), 1617-1623. Aiken, L., Clarke, S., Sloane, D., Sochalsk i, J., & Silber, J. (2002). Hospital nurse staffing and patient mortality, nur se burnout and job dissatisfaction. Journal of the American Medical Association, 288 (16), 1987-1993. Aiken, L., Havens, D., & Sloane, D. (2000) The Magnet nursing services recognition program: a comparison of two groups of magnet hospitals. Amercian Journal of Nursing, 100(3), 26-36. Alexander, G., Himes, J., Kaufman, R., Mo r, J., & Kogan, M. (1996). A United States national reference for fetal growth. Obstetrics & Gynecology, 87 (2), 163-168. Alexander, G., Kogan, M., Bade r, D., Carlo, W., Allen, M., & Mor, J. (2003). US birth weight/gestational age-specific neonatal mortality: 1995-1997 rates for Whites, Hispanics and Blacks. Pediatrics, 111(1), 61-66. Alexander, G., Kogan, M., Himes, J., & Gold enberg, R. (1999). R acial differences in birthweight for gestational age and infant mortality in extremely-low-risk US populations. Paedatric and Perinatal Epidemiolgoy, 13 205-217.

PAGE 133

119 Alexander, G., Tompkins, M., Allen, M., & Hulsey, T. (1999). Trends and racial differences in birth weight and related survival. Maternal Child Health Journal, 3(1), 71-79. Allen, M., Alexander, G., Tompkins, M., & Hulsey, T. (2000). Raci al differences in temporal changes in newborn viability and survival by gestational age. Paediatric and Perinatal Epidemiology, 14 152-158. American Academy of Pediatrics, & Am erican College of Obstetricians and Gynecologists. (2006). The Apgar score. Pediatrics, 117 (4), 1444-1447. Baskett, T. (2000). Virginia Apgar and the newborn Apgar Score. Resuscitation, 47 215217. Bell, C., & Redelmeier, D. (2001). Mortality among patients admitted to hospitals on weekends as compared with weekends. New England Journal of Medicine, 345(9), 663-668. Buka, S., Brennan, R., Rich-Edwards, J., Raudenbush, S., & Earis, F. (2003). Neighborhood support and the birth weight of urban infants. Amercian Journal of Epidemiology, 157 (1), 1-8. Callaghan, L., Cartwright, D., O' Rourke, P., & Davies, M. (2003) Infant to staff ratios and risk of mortality in ve ry low birthweight infants. Archives of Disease in Childhood Fetal and Neonatal Edition, 88 94-97. Carmel, S., & Rowan, K. (2001). Variation in intensive care unit outcomes: a search for the evidence of organizational factors. Current Opinion in Critical Care, 7(4), 284-296. Casey, B., McIntire, D., & Leveno, K. (2001). The continuing value of the Apgar score for the assessment of newborn infants. New England Journal of Medicine, 344 (7), 467-471. Childrens Defense Fund. (1992). The State of America's Children 1992 Washington, DC: By Author. Children's Medical Services. (2006). Regional Perinatal Intens ive Care Centers Annual Report, Fiscal Year 2005-2006 (annual report). Tallaha ssee: Florida Department of Health. Cho, S.-H., Ketefian, S., Barkauskas, V. H., & Smith, D. G. (2003). The effects of nurse staffing on adverse events, morbidit y, mortality, and medical costs. Nursing Research, 52 (2), 71-79.

PAGE 134

120 Cockburn, F., Cooke, R., Gamsu, H., Greenough, A., Hopkins, A., McIntosh, N., et al. (1993). The CRIB (clinical risk index for babi es) score: a tool for assessing initial neonatal risk and comparing performa nce of neonatal intensive care units. Lancet, 342, 193-198. Darr, D. (2005). Evaluation of the Florida Me dicaid Healthy Start Waiver (program evaluation). Tampa: Univer sity of South Florida. Dowding, V., Duignan, N., Henry, G., & M acDonald, D. (1987). Induction of labour, birthweight and perinatal mortality by day of the week. British Journal of Obstetrics and Gynaecology, 94 413-419. Elsmen, E., Pupp, I., & Hellstrom-Westas, L. (2004). Preterm male infants need more initial respiratory and circulatr oy support than femal infants. Acta Paediatrica, 93 529-533. Finster, M., & Wood, M. (2005). The Apgar score has survived the test of time. Anesthesiology, 102 (4), 855-857. Florida Department of Health, Offi ce of Vital Sta tistics. (2006). Prental Care Entry Available from www.Floridacharts.com. Goh, A., Lum, L., & Abdel-Latif, M. (2001). Impact of 24 hour critical care physician staffing on case-mix adjusted mortlaity in paediatic intensive care. Lancet, 357 445-446. Goldenberg, R., Cliver, S., & Mulvihill, F. (1 996). Medical, psychosocial and behavioral risk factors do not explain the increased risk for low birth weight among Black women. American Journal of Obstetrics and Gynecology, 175 1317-1324. Goodman, D., Sappenfield, W., & Thompson, D. (2007). Preterm Birth and Intrpartum Intervention in Florida: What is the Connection? Paper presented at the March of Dimes Florida Chapter Preterm Bi rth Summit, Tampa, Florida. Gould, J., Qin, C., Marks, A., & Chavez, G. (2003). Neonatal mortality in weekend vs weekday births. Journal of the American Medical Association, 289 (22), 29582962. Hall, L. M., Doran, D., & Pink, G. (2004). Nurse staffing models, nursing hours, and patient safety outcomes. Journal of Nursing Administration, 34 (1), 41-45. Hamilton, P., & Restrepo, E. (2003). Weekend birth and higher neonatal mortality: A problem of patient acuit y or quality of care? Journal of Obstetric, Gynecologic, and Neonatal Nursing, 32 (6), 724-733.

PAGE 135

121 Healy, A., Malone, F., & LM, S. (2006). Early access to prenatal care: Implications for racial disparity in perinatal mortality. Obstetrics and Gynecology, 107 625-631. Heinz, D. (2004). Hospital nurse staffing a nd patient outcomes: A review of current literature. Dimensions of Critical Care Nursing, 23(1), 44-50. Heller, G., Misselwitz, B., & Schmidt, S. (2000). Early neonatal mortality, asphyxia related deaths, and timing of low ri sk births in Hesse, Germany, 1990-8: observational study. British Medical Journal, 321 274-275. Hendry, R. (1981). The weekend A dangerous time to be born? British Journal of Obstetrics and Gynaecology, 88 1200-1203. Hodnett, E., & Fredericks, S. (2003). Support during pregnancy for women at increased risk of low birthweight babies (systematic review No. Art. No.:CD000198. DOI:10.1002/14651858.CD000198). Hong, J., Kang, H., Yi, S.-W., Han, Y., Nam, C., Gombojav, B., et al. (2006). A comparison of perinatal mortality in Korea on holidays and working days. British Journal of Obstetrics and Gynaecology, 113 1235-1238. Horbar, J. D., Badger, G. J., Lewit, E. M., Rogowski, J., & Shiono, P. H. (1997). Hospital and patient characteristics a ssociated with variation in 28-day mortality rates for very low birth weight infants. Pediatrics, 99 (2), 149-156. Ingemarsson, I. (2003). Gender aspects of preterm birth. British Journal of Obstetrics and Gynaecology, 110 (Suppl 20), 34-38. Kaaresen, P., Dohlen, G., Fundingsrud, H., & Dahl, L. (1998). The use of the CRIB (clinical risk index for babies) score in auditing the performance of one neonatal intensive care unit. Acta Paediatrica, 87 195-200. Kahn, H., & Sempos, C. (1989). Statistical Methods in Epidemiology (Vol. 12). New York: Oxford University Press. Khoshnood, B., Wall, S., & Lee, K. (2005). Risk of low birth weight associated with advanced maternal age among four et hnic groups in the United States. Maternal Child Health Journal, 9 (1), 3-9. Kolas, T., Saugstad, O., Daltveit, A., Nils en, S., & Oian, P. (2006). Planned cesarean versus planned vaginal delivery at term: comparison of newborn infant outcomes. American Journal of Obst etrics and Gynecology, 195 1538-1543. Kromrey, J. (2007). SAS Program to Determine Sample Size by Odds Ratio and Tolerance Values. In L. Stanley (Ed.). Tampa.

PAGE 136

122 Lang, T. A., Hodge, M., Olson, V., Romano, P. S. & Kravitz, R. L. (2004). Nurse-patient ratios: A systematic review on the eff ects of nurse staffing on patient, nurse employees, and hospital outcomes. Journal of Nursing Administration, 34 (7-8), 326-337. Lee, S. K., Lee, D. S. C., Andrews, W. L., Baboolal, R., Pendray, M., Stewart, S., et al. (2003). Higher mortality rates among inborn infants admitted to neonatal intensive care units at night. Journal of Pediatrics, 143, 592-597. Lundsberg, L., Bracken, M., & Saftlas, A. (1997). Low-to-moderate gestational alcohol use and intrauterine growth retardatio n, low birth weight and preterm delivery. Annuals of Epidemiology, 7 498-508. Luo, Z., & Karlberg, J. (2001). Timing of birth and infant and early ne onatal mortality in Sweden 1973-1995: longitudinal birth register study. British Medical Journal, 323, 1327-1330. Luo, Z., Liu, S., Wilkins, R., & Kramer, M. ( 2004). Risks of stillbirth and early neonatal death by day of week. Canadian Medical Association Journal, 170(3), 337-341. MacDorman, M., Declercq, E., Menacker, F ., & Malloy, M. (2006). Infant and neonatal mortality for primary cesarean and vagina l births to women with "no indicated risk," United States, 1998-2001 birth cohorts. Birth, 33(3), 175-182. MacFarlane, A. (1978). Variations in number of births and perinatal mortality by day of week in England and Wales. British Medical Journal, 2 1670-1673. Mangold, W. (1981). Neonatal mortality by the day of the week in the 1974-1975 Arkansas live birth cohort. American Journal of Public Health, 71(6), 601-605. Mathers, C. (1983). Births and perinatal deaths in Australia: varia tions by day of week. Journal of Epidemiology and Community Health, 37 57-62. Mathews, T., Menacker, F., & MacDorman, M. (2004). Infant mortality statistics from the 2002 period linked birth/infant death data set. National Vital Statistics Reports, 53 (10), 1-30. McGrady, G., Sun, J., Rowley, D., & Hogue, C. (1992). Preterm delivery and low birth weight among first-born infants of black and white college graduates. American Journal of Epidemiology, 136, 266-276. Merlo, J., Gerdtham, U.-G., Eckerlund, I ., Hakansson, S., Otterblad-Olausson, P., Pakkanen, M., et al. (2005). Hospital level of care and neonatal mortality in lowand high-risk deliveries: reassessing the question in Sweden using multilevel analysis. Medical Care, 43 (11), 1092-1100.

PAGE 137

123 Morse, S., Wu, S., Ma, C., Ariet, M., Resnick, M., & Roth, J. (2006). Racial and gender differences in the viability of extremel y low birth weight infants: A populationbased study. Pediatrics, 117, 106-112. Mugford, M., Szczepura, A., Lodwick, A., & S tilwell, J. (1988). Factors affecting the outcome of maternity care II: Neona tal outcomes and resources beyond the hospital of birth. Journal of Epidemiology and Community Health, 42 170-176. Needleman, J., Buerhaus, P., Mattke, S., St ewart, M., & Zelevinsky, K. (2002). Nursestaffing levels and the qual ity of care in hospitals. New England Journal of Medicine, 346 (22), 1715-1722. O'Campo, P., Xue, X., Wang, M.-C., & Ca ughy, M. (1997). Neighborhood risk factors for low birthweight in Baltimore: a multilevel analysis. American Journal of Public Health, 87 (7), 1113-1118. Pearl, M., Braverman, P., & Abrams, B. (2001). The relationship of neighborhood socioeconomic characteristics to birthw eight among 5 ethnic groups in California. American Journal of Public Health, 91 (11), 1808-1814. Phibbs, C., Bronstein, J., Buxton, E., & Phibbs R. (1996). The effects of patient volume and level of care at the hospital of birth on neonatal mortality. JAMA, 276 (13), 1054-1059. Phipps, M., & Sowers, M. (2002). De fining early adolescent childbearing. American Journal of Public Health, 92 125-128. Pickett, K., & Pearl, M. (2001). Multilevel analyses of neighbourhood socioeconomic context and health outcomes: a critical review. Journal of Epidemilogy Community Health, 55(55), 111-122. Pollack, M., Cuerdon, T., Patel, K., Ruttimann, U., Getson, P., & Levetwon, M. (1994). Impact of quality-of-care factors on pediatric intens ive care unit mortality. Journal of the American Medical Association, 272 (12), 941-946. Pronovost, P., Angus, D., Dorman, T., Robinson, K., Dremsizov, T., & Young, T. (2002). Physician staffing patterns and clinical outcomes in critically ill patients: a systematic review. Journal of the American Medical Association, 288 (17), 21512162. Rajaratnam, J., Burke, J., & O'Campo, P. (2006). Maternal and child health and neighborhood context: The selection and c onstruction of area-level variables. Health & Place, 12 547-556.

PAGE 138

124 Richardson, D., Corcoran, J., Escobar, G ., & Shoo, K. L. (2001). SNAP-II and SNAPPEII: simplified newborn illness sevreity and mortality risk scores. The Journal of Pediatrics, 138 (1), 92-100. Richardson, D., Gray, J., Gortmaker, S., Go ldmann, D., Pursley, D., & McCormick, M. (1998). Declining severity adjusted mo rtality: evidence of improving neonatal intensive care. Pediatrics, 102 (4), 893-899. Richardson, D., Shah, B., Frantz, I., Bednarek, F., Rubin, L., & McCormick, M. (1999). Perinatal risk and severity of illness in newborns at 6 neonatal intensive care units. American Journal of Public Health, 89 (4), 511-516. Richardson, D., Tarnow-Mordi, W., & Shoo, K. L. (1999). Risk adjustment for quality improvement. Pediatrics, 103 (1), 255-265. Roberts, E. (1997). Neighborhood social enviro nments and the distribution of low birth weight in Chicago. Amercian Journal of Public Health, 87(4), 597-603. Rogowski, J., Horbar, J. D., Staiger, D., Ke nny, M., Carpenter, J., & Geppert, J. (2004). Indirect vs direct hospital quality indicat ors for very low-birt h-weight infants. Journal of the American Medical Association, 291 (2), 202-209. Rogowski, J., Staiger, D., & Horbar, J. D. (2004). Variations in the quality of care for very-low-birthweight infant s: implications for policy. Health Affairs, 23 (5), 8897. Rothberg, M. B., Abraham, I., Lindenauer, P. K., & Rose, D. N. (2005). Improving nurse-to-patient staffing ratios as a cost-effective safe ty intervention. Medical Care, 43(8), 785-791. Sanderson, M., Sappenfield, W., Jespersen, K., Liu, Q., & Baker, S. (2000). Association between level of delivery hospital and neonatal outcomes among South Carolina Medicaid recipients. American Journal of Obstetrics and Gynecology, 183 15041511. Sappenfield, W. (2007). Florida's Lack of Declin e in Infant Mortality: Potential Roles for Healthy Start Coalitions. Paper presented at the FAHSC Summer Meeting, Sanabel Island, Florida. Sasichay-Akkadeshanunt, T., Scalzi, C., & Ja wad, A. (2003). The relationship of nurse staffing and patient outcomes. Journal of Nursing Administration, 33(9), 478-485. Schoendorf, K., Hogue, C., Kleinman, J., & Ro wley, D. (1992). Mortality among infants of black as compared with white college-educated parents. New England Journal of Medicine, 326 1522-1526.

PAGE 139

125 Stanley, L. (2006). Understanding the Issue of Uninsu red Pregnant Women and Ways in Which Communities Can Begin to Address the Issue. Paper presented at the 2006 Partners in Perinatal Health Sharing Solutions Conference, Clearwater Beach, Florida. Stephan, C., Wesseling, S., Schink, T., & Jung, K. (2003). Comparison of Eight Computer Programs for Receiver-Op erating Characteristic Analysis. Clinical Chemistry, 49(3), 433-439. Stephansson, O., Dickman, P., Johansson, A., Kieler, H., & Cnattingius, S. (2003). Time of birth and risk of intrap artum and early neonatal death. Epidemiology, 14 (2), 218-222. Stilwell, J., Szczepura, A., & Mugford, M. (1988). Factors affecting the outcome of materinity care I: Relationship between staffing and perina tal deaths at the hospital of birth. Journal of Epidemiology and Community Health, 42 157-169. Tabachnick, B., & Fidell, L. (2001). Using Multivariate Statistics (4th ed.). Boston: Allyn and Bacon. Tamow-Mordi, W., Hau, C., Warden, A., & Shearer, A. (2000). Hospital mortality in relation to staff workload: A 4-year st udy in an adult inte nsive care unit. Lancet, 356, 185-189. Tampa General Hospital. (2001). 2001 Annual Report Retrieved February 2, 2007 from www.tgh.org/sections/01_about/about_tgh.htm Tampa General Hospital. (2007). Facts About Tampa General Hospital (TGH) Retrieved January 18, 2007 from http://www.tgh.org/sections/01_about/fact_sheet/fact_sheet.htm Tenner, P., Dibrell, H., & Taylor, R. (2003). Improved survival w ith hospitalists in a pediatric intensive care unit. Critical Care Medicine, 31 (3), 847-852. Thorngern-Jerneck, K., & Herbst, A. (2001) Low 5-minute Apgar score: A populationbased register study of 1 million term births. Obstetrics and Gynecology, 98 (1), 65-70. Tibby, S. M., Correa-West, J., Durward, A ., Ferguson, L., & Murdoch, I. A. (2004). Adverse events in a paediatric intensive care unit: Relationship to workload, skill mix and staff supervision. Intensive Care Medicine, 30 1160-1166. Tourangeau, A., Cranley, L., & Jeffs, L. (2006). Impact of nursing on hospital patient mortality: a focused review a nd related policy implications. Quality and Safety in Health Care, 15 4-8.

PAGE 140

126 Tourangeau, A., Giovannetti, P., Tu, J ., & Wood, M. (2002). Nursing-related determinants of 30-day mortality for hospitalized patients. Canadian Journal of Nursing Research, 33 (4), 71-88. Tucker, J., Parry, G., McCabe, C., Nicols on, P., & Tarnow-Mordi, W. (2002). Patient volume, staffing, and workload in relati on to risk-adjusted outcomes in a random stratified sample of UK neonatal intensive care units: a prospective evaluation. Lancet, 359, 99-107. Urato, A., Craigo, S., Chelmow, D., & O'Brie n, W. (2006). The association between time of birth and fetal inju ry resulting in death. American Journal of Obstetrics and Gynecology, 195 1521-1526. Visscher, W., Feder, M., Burns, A., Brady, T., & Bray, R. (2003). The impact of smoking and other substance use by urban women on the birth weight of their infants. Substance Use and Misuse, 38 (8), 1063-1093. Windham, G., Hopkins, B., Fenster, L., & Swan, S. (2000). Prenatal active or passive tobacco smoke exposure and risk of pr eterm delivery or low birth weight. Epidemiology, 11 (4), 427-433. Zhu, B., & Le, T. (2003). Effect of interpregna ncy interval on infant low birth weight: A retrospective cohort study using the Mich igan maternally linked birth database. Maternal Child Health Journal, 7 169-178. Zhu, B., Rolfs, R., Nangle, B., & Horan, J. (1999). Effect of th e interval between pregnancies on perinatal outcomes. New England Journal of Medicine, 340 589594.

PAGE 141

127 Appendices

PAGE 145

131 Appendix C: T-Test Results for Link ed Births with Unlinked Births Table C1 T-Test Results for NICU Births Linked to OOIS with NICU Births Not Linked to OOISa (N=1905) NICU Births Linked NICU Births Unlinked t SE p (n=1846) (n=59) Mean Birth Weight in Grams 2090.55 2439.76 -2.77 126.02 .006** Mean Gestation in Weeks 33.32 34.86 -2.65 .58 .008** Mean Ventilator Days 3.45 2.51 .47 1.99 .639 aThe following covariates are found in the OOIS and are therefore excluded from this table: multiple birth, APGAR score, delivery type, induction, fetal anomaly. **p<.01

PAGE 146

132 Appendix D: T-Test Results for NICU Ad missions with Newborn Nursery Admissions Table D1 T-Test Results of Infants Admitted to NICU First with Infants Admitted to Newborn Nursery First (N=2683) NICU Newborn t SE p (n=1846) (n=837) Mean Birth Weight in Grams 2090.550 3144.000 -33.62 31.33 p<.001*** Mean Gestation in Weeks 33.320 38.200 -39.80 .120 p<.001*** Mean APGAR Score 7.910 8.790 -20.50 .040 p<.001*** Mean Ventilator Days 3.450 .120 -15.60 .060 p<.001*** ***p<.001

PAGE 147

133 Appendix E: Descriptive Statistics Weekend to Weekday Admissions Table E1 Descriptive Statistics for Weekend Admissions Co mpared to Weekday Admissions (n=1846) Mean SD Min Max Kurtosis Skew Weekend Admissions (n=547) Birth Weight in Grams 2053.07 928.76 385.00 4970.00 -.297 .476 Gestation in Weeks 33.09 4.46 22.00 42.00 -.425 -.299 APGAR Score 7.86 1.59 1.00 10.00 3.430 -1.81 Days on Ventilation 3.26 12.37 0.00 162.00 68.400 7.16 Number of RNs/Day 54.87 9.63 17.50 75.00 -.246 -.353 Number of RN Hrs/Day 331.95 58.66 23.50 469.00 -.225 -.336 Census/Day 46.14 6.65 22.00 61.00 .401 -.317 RN:Infant Ratio/Day 1.19 .104 .84 1.56 -.195 -.085 Capacity/Day 1.098 .158 .524 1.45 .401 -.317 Weekday Admissions (n=1299) Birth Weight 2106.33 960.28 378.00 5100.00 -.240 .526 Gestation 33.42 4.35 22.00 43.00 -.447 -.319 APGAR Score 7.93 1.61 0.00 10.00 5.990 -2.26 Days on Ventilation 3.52 16.33 0.00 267.00 114.020 9.26 Number of RNs/Day 56.25 10.72 24.00 84.00 -.231 -.363 Number of RN Hrs/Day 341.99 65.83 145.00 519.00 -.219 -.345 Census/Day 46.32 6.85 16.00 63.00 .879 -.549 RN:Infant Ratio/Day 1.21 .117 .89 1.66 .200 .123 Capacity/Day 1.102 .163 .381 1.50 .879 -.549

PAGE 148

134 Appendix F: Descriptive Statistics, Nighttime to Day Admissions Table F1 Descriptive Statistics for Nighttime Admissions Compared to Day Admissions (n=1846) Mean SD Min Max Kurtosis Skew Nighttime Admissions (n=798) Birth Weight in Grams 2028.910 916.720 380.000 5100.000 -.250 .507 Gestation in Weeks 33.040 4.520 23.000 42.000 -.516 -.272 APGAR Score 7.740 1.750 0.000 10.000 3.700 -1.860 Days on Ventilation 3.700 14.620 0.000 165.000 57.400 6.850 Number of RNs 27.540 5.250 10.500 43.500 -.259 -.272 Number of RN Hours 163.500 31.220 62.000 258.000 -.238 -.252 Census 46.290 6.680 16.000 61.000 .686 -.439 RN:Infant Ratio .593 .064 .390 .850 .420 .130 Capacity 1.102 .158 .381 1.450 .686 -.44 Day Admissions (n=1048) Birth Weight 2137.480 974.300 378.000 4980.000 -.280 .505 Gestation 33.530 4.270 22.000 43.000 -.381 -.3438 APGAR Score 8.050 1.470 0.000 10.000 6.700 -2.370 Days on Ventilation 3.250 15.730 0.000 267.000 145.900 10.480 Number of RNs 29.160 5.680 11.500 45.000 -.119 -.32 Number of RN Hours 175.200 34.330 70.000 270.000 -.118 -.298 Census 46.250 6.880 16.000 63.000 .781 .513 RN:Infant Ratio .628 .060 .410 .840 .341 .064 Capacity 1.101 .16 .381 1.500 .781 -.51

PAGE 149

135 Appendix G: Logistic Regression for Model 1 Table G1 Complete Logistic Regression Models for Weekend Exposure with Death Before Discharge (n=1837) Block 1.1: Weekend Exposure Confidence Interval B S.E. Wald df Exp(B) Lower Upper Sig. Weekend Admission .027 .204 .017 1 1.027 .689 1.531 .895 -2 Log Likelihood=902.668; 2=.017, df=1, p=.896 Block 1.2: Weekend Exposure + Acuity Confidence Interval B S.E. Wald df Exp(B) Lower Upper Sig. Weekend Admission .000 .231 .000 1 1.000 .635 1.573 .999 Birth Weight -.052 .014 14.268 1 .949 .924 .975 p<.001*** Infant on Ventilation 2.629 .321 66.904 1 13.859 7.382 26.021 p<.001*** APGAR < 7 1.231 .223 30.347 1 3.424 2.210 5.305 p<.001*** SGA .576 .272 4.501 1 1.779 1.045 3.030 .034* -2 Log Likelihood=612.022; 2=290.646, df=4, p=<.001 Block 1.3: Weekend Exposure + Acuity + Infant Characteristics Confidence Interval B S.E. Wald df Exp(B) Lower Upper Sig. Weekend Admission .058 .247 .055 1 1.060 .653 1.721 .814 Birth Weight -.120 .019 40.569 1 .887 .855 .920 p<.001*** Infant on Ventilation 2.316 .326 50.574 1 10.134 5.353 19.186 p<.001*** APGAR < 7 1.208 .236 26.110 1 3.347 2.106 5.319 p<.001*** SGA .109 .291 .141 1 1.115 .630 1.973 .708 Fetal Anomaly 2.672 .350 58.255 1 14.474 7.287 28.750 p<.001*** Non-Black Race .762 .265 8.236 1 2.142 1.273 3.604 .004** Male Infant .063 .229 .075 1 1.065 .680 1.668 .784 Multiple Birth .205 .283 .526 1 1.228 .705 2.139 .468 -2 Log Likelihood=539.460; 2=72.562, df=4, p=<.001

PAGE 150

136 Appendix G: Logistic Regression for Model 1 (continued) Block 1.4: Weekend Exposure + Acuity + Infant Characteristics+ OB Interventions Confidence Interval B S.E. Wald df Exp(B) Lower Upper Sig. Weekend Admission .079 .248 .102 1 1.083 .665 1.762 .749 Birth Weight -.117 .019 38.737 1 .889 .857 .923 p<.001*** Infant on Ventilation 2.403 .331 52.677 1 11.054 5.777 21.151 p<.001*** APGAR < 7 1.213 .238 26.007 1 3.364 2.111 5.363 p<.001*** SGA .286 .302 .896 1 1.331 .736 2.407 .344 Fetal Anomaly 2.669 .350 58.151 1 14.428 7.265 28.652 p<.001*** Non-Black Race .755 .267 7.994 1 2.127 1.261 3.590 p<.001*** Male Infant .047 .231 .041 1 1.048 .667 1.647 .839 Multiple Birth .281 .290 .939 1 1.325 .750 2.339 .332 Induction -.911 .623 2.141 1 .402 .119 1.363 .143 Cesarean Section -.484 .251 3.721 1 .617 .377 1.008 .054 -2 Log Likelihood=534.190; 2=5.270, df=2, p=.072 **p<.01; ***p<.001 Model terms: weekend=1, birth weight=100 gr am increments, ventilation=1, APGAR=1, sga=1, anomaly=1, non-Black race=1, male=1, multiple=1, induction=1, c-section=1.

PAGE 151

137 Appendix H: Logistic Regression for Reduced-Risk Model Table H1 Complete Logistic Regression Model for Weekend E xposure with Death Before Discharge Death Before Discharge, Reduced Risk Model (n=1837) Block RA.1: Weekend Exposure Confidence Interval B S.E. Wald df Exp(B) Lower Upper Sig. Weekend Admission .027 .204 .017 1 1.027 .689 1.531 .895 -2 Log Likelihood=902.668; 2=.017, df=1, p=.896 Block RA.2: Weekend Exposure + Acuity Confidence Interval B S.E. Wald df Exp(B) Lower Upper Sig. Weekend Admission .038 .225 .029 1 1.039 .669 1.614 .865 Birth Weight .000 .000 11.613 1 1.000 .999 1.000 .001** Infant on Ventilation 3.060 .310 97.461 1 21.330 11.618 39.160 p<.001*** SGA .509 .263 3.729 1 1.663 .992 2.788 .053 -2 Log Likelihood=612.022; 2=290.646, df=4, p=<.001 Block RA.3: Weekend Exposure + Acuity + Infant Characteristics Confidence Interval B S.E. Wald df Exp(B) Lower Upper Sig. Weekend Admission .083 .227 .133 1 1.086 .696 1.694 .715 Birth Weight -.001 .000 13.765 1 .999 .999 1.000 p<.001*** Infant on Ventilation 3.047 .310 96.489 1 21.062 11.466 38.688 p<.001*** SGA .454 .266 2.917 1 1.575 .935 2.653 .088 Non-Black Race .645 .241 7.135 1 1.906 1.187 3.060 .008 Male Infant .022 .210 .011 1 1.023 .678 1.543 .915 -2 Log Likelihood=539.460; 2=72.562, df=4, p=<.001 **p<.01; ***p<.001 Model terms: weekend=1, birth weight=100 gram in crements, ventilation=1, sga=1, non-Black race=1, male=1.

PAGE 152

Appendix I: Receiver Operating Characteristic Curves for ModelFA and ModelRA 1 Specificity1.0 0.8 0.6 0.4 0.2 0.0 Sensitivity1.0 0.8 0.6 0.4 0.2 0.0 ROC CurveDiagonal segments are produced by ties. Figure 1: ROC for Full Risk Adjustment Model 138

PAGE 153

Appendix I: Receiver Operating Characteristic Curves for ModelFA and ModelRA (continued) 1 Specificity1.0 0.8 0.6 0.4 0.2 0.0 Sensitivity1.0 0.8 0.6 0.4 0.2 0.0 ROC CurveDiagonal segments are produced by ties. Figure 2: ROC for Reduced Risk Adjustment Model 139

PAGE 154

140 Appendix J: T-Test Results for In-Born and Transfers Table J1 T-Test Results of In-Born NICU Admissions with Transfer NICU Admissionsa (N=2572) In-Born Transfers t SE p (n=1846) (n=726) Mean Birth Weight in Grams 2090.55 2411.72 -7.04 45.63 p<.001*** Mean Gestation in Weeks 33.32 34.41 -4.99 .220 p<.001*** Mean Ventilator Days 3.45 4.40 -1.41 .680 .159 aThe following covariates are found in the OOIS and are therefore excluded from this table: multiple birth, APGAR score, delivery type, induction, fetal anomaly. ***p<.001

PAGE 155

141 Appendix K: Logistic Regression for Model 2 Table K1 Complete Logistic Regression Models for Nighttime Exposure with Death Before Discharge (n=1837) Block 2.1: Nighttime Exposure Confidence Interval B S.E. Wald df Exp(B) Lower Upper Sig. Nighttime Admission .308 .187 2.712 1 1.360 .943 1.962 .100 -2 Log Likelihood=899.980; 2=2.706, df=1, p=.100 Block 2. 2: Nighttime Exposure + Acuity Confidence Interval B S.E. Wald df Exp(B) Lower Upper Sig. Nighttime Admission .081 .218 .138 1 1.084 .708 1.661 .710 Birth Weight -.052 .014 13.919 1 .949 .924 .976 p<.001*** Infant on Ventilation 2.637 .322 66.863 1 13.970 7.425 26.285 p<.001*** APGAR < 7 1.217 .226 28.870 1 3.376 2.166 5.262 p<.001*** SGA .589 .274 4.631 1 1.802 1.054 3.080 .031 -2 Log Likelihood=611.884; 2=288.096, df=4, p=<.001 Block 2.3: Nighttime Exposure + Acuity + Infant Characteristics Confidence Interval B S.E. Wald df Exp(B) Lower Upper Sig. Nighttime Admission .128 .232 .304 1 1.136 .722 1.789 .582 Birth Weight -.120 .019 40.152 1 .887 .855 .921 p<.001*** Infant on Ventilation 2.323 .326 50.622 1 10.203 5.381 19.347 p<.001*** APGAR < 7 1.188 .239 24.639 1 3.279 2.052 5.241 p<.001*** SGA .127 .293 .188 1 1.136 .639 2.018 .665 Fetal Anomaly 2.681 .351 58.474 1 14.603 7.345 29.035 p<.001*** Non-Black Race .751 .265 8.025 1 2.120 1.260 3.565 .005** Male Infant .064 .229 .078 1 1.066 .681 1.670 .780 Multiple Birth .214 .284 .570 1 1.239 .710 2.162 .450 -2 Log Likelihood=539.212; 2=72.672, df=4, p=<.001

PAGE 156

142 Appendix K: Logistic Regression for Model 2 (continued) Block 2.4: Nighttime Exposure + Acuity + Infant Characteristics+ OB Interventions Confidence Interval B S.E. Wald df Exp(B) Lower Upper Sig. Nighttime Admission .083 .234 .124 1 1.086 .686 1.718 .725 Birth Weight -.117 .019 38.487 1 .889 .857 .923 p<.001*** Infant on Ventilation 2.401 .331 52.629 1 11.038 5.769 21.117 p<.001*** APGAR < 7 1.200 .241 24.862 1 3.319 2.071 5.320 p<.001*** SGA .291 .303 .919 1 1.337 .738 2.423 .338 Fetal Anomaly 2.676 .351 58.233 1 14.533 7.308 28.900 p<.001*** Non-Black Race .748 .267 7.856 1 2.112 1.252 3.563 .005** Male Infant .049 .230 .045 1 1.050 .669 1.650 .831 Multiple Birth .283 .290 .948 1 1.327 .751 2.343 .330 Induction -.906 .621 2.124 1 .404 .120 1.367 .145 Cesarean Section -.470 .252 3.468 1 .625 .381 1.025 .063 -2 Log Likelihood=534.168; 2=5.044, df=2, p=.080 **p<.01; ***p<.001 Model terms: nighttime=1, birth weight=100 gram increments, ventilation=1, APGAR=1, sga=1, anomaly=1, non-Black race=1, male=1, multiple=1, induction=1, c-section=1.

PAGE 157

143 Appendix L: Logistic Regression for Model 3 Table L1 Complete Logistic Regression Model 3 for Effect Modification with Death Before Discharge (n=1837) Block 3.1: Weekend + Nighttime + Weekend*Nighttime Confidence Interval B S.E. Wald df Exp(B) Lower Upper Sig. Weekend Admission .166 .301 .306 1 .580 1.181 .655 2.129 Nighttime Admission .414 .224 3.422 1 .064 1.513 .976 2.347 Weekend*Nighttime -.349 .410 .722 1 .395 .706 .316 1.577 -2 Log Likelihood=899.248; ; 2=3.438, df=3, p=.329 Block 3.2: Weekend + Nighttime + Weekend*Nighttime + Acuity Confidence Interval B S.E. Wald df Exp(B) Lower Upper Sig. Weekend Admission .129 .337 .146 1 1.137 .588 2.202 .702 Nighttime Admission .161 .260 .384 1 1.175 .705 1.958 .535 Weekend*Nighttime -.262 .466 .315 1 .770 .309 1.918 .574 Birth Weight -.053 .014 14.161 1 .949 .923 .975 p<.001*** Infant on Ventilation 2.633 .322 66.755 1 13.916 7.399 26.171 p<.001*** APGAR < 7 1.212 .227 28.619 1 3.361 2.155 5.239 p<.001*** SGA .587 .274 4.609 1 1.799 1.052 3.075 .032* -2 Log Likelihood=611.568; ; 2=287.680, df=4, p=<.001 Block 3.3: Weekend + Nighttime + Weekend*Nighttime + Acuity + Infant Characteristics Confidence Interval B S.E. Wald df Exp(B) Lower Upper Sig. Weekend Admission .205 .363 .321 1 1.228 .603 2.499 .571 Nighttime Admission .211 .275 .590 1 1.235 .721 2.115 .442 Weekend*Nighttime -.301 .498 .364 1 .740 .279 1.967 .546 Birth Weight -.120 .019 40.489 1 .887 .855 .920 p<.001*** Infant on Ventilation 2.323 .327 50.556 1 10.203 5.378 19.354 p<.001*** APGAR < 7 1.187 .239 24.584 1 3.277 2.050 5.238 p<.001*** SGA .130 .294 .197 1 1.139 .641 2.026 .657 Fetal Anomaly 2.668 .350 58.078 1 14.409 7.255 28.617 p<.001*** Non-Black Race .766 .267 8.238 1 2.150 1.275 3.627 .004** Male Infant .070 .229 .093 1 1.072 .684 1.681 .761 Multiple Birth .217 .285 .582 1 1.242 .711 2.170 .445 -2 Log Likelihood=538.819; ; 2=72.749, df=4, p=<.001

PAGE 158

144 Appendix L: Logistic Regre ssion for Model 3 (continued) Block 3.4: Weekend + Nighttime + Weekend*Nighttime + Acuity + Infant Characteristics.+ OB Interventions Confidence Interval B S.E. Wald df Exp(B) Lower Upper Sig. Weekend Admission .228 .363 .395 1 1.256 .617 2.558 .530 Nighttime Admission .163 .278 .342 1 1.177 .682 2.030 .559 Weekend*Nighttime -.295 .501 .347 1 .744 .279 1.988 .556 Birth Weight -.117 .019 38.765 1 .889 .857 .923 p<.001*** Infant on Ventilation 2.404 .332 52.582 1 11.071 5.780 21.204 p<.001*** APGAR < 7 1.200 .241 24.841 1 3.319 2.071 5.321 p<.001*** SGA .297 .304 .958 1 1.346 .742 2.442 .328 Fetal Anomaly 2.662 .350 57.841 1 14.320 7.212 28.435 p<.001*** Non-Black Race .764 .268 8.099 1 2.146 1.268 3.632 .004* Male Infant .054 .231 .054 1 1.055 .671 1.660 .815 Multiple Birth .284 .291 .952 1 1.328 .751 2.347 .329 Induction -.923 .623 2.190 1 .397 .117 1.349 .139 Cesarean Section -.468 .253 3.436 1 .626 .381 1.027 .064 -2 Log Likelihood=533.743; ; 2=5.076, df=2, p=.079 **p<.01; ***p<.001 Model terms: weekend=1, nighttime=1, birth weight=100 gram increments, ventilation=1, APGAR=1, sga=1, anomaly=1, non-Black race=1, male =1, multiple=1, induction=1, c-section=1.

PAGE 159

145 Appendix M: Logistic Regression for Model 4A Table M1 Complete Logistic Regression Models 4A for We ekend Exposure with Hospital-Level Covariates Included with Death Before Discharge (n=1837) Block 4A.1: Weekend Exposure Confidence Interval B S.E. Wald df Exp(B) Lower Upper Sig. Weekend Admission .027 .204 .017 1 1.027 .689 1.531 .895 -2 Log Likelihood=902.668; 2 =.017, df=1, p=.896 Block 4A2: Weekend Exposure + Acuity Confidence Interval B S.E. Wald df Exp(B) Lower Upper Sig. Weekend Admission .000 .231 .000 1 1.000 .635 1.573 .999 Birth Weight -.052 .014 14.268 1 .949 .924 .975 p<.001*** Infant on Ventilation 2.629 .321 66.904 1 13.859 7.382 26.021 p<.001*** APGAR < 7 1.231 .223 30.347 1 3.424 2.210 5.305 p<.001*** SGA .576 .272 4.501 1 1.779 1.045 3.030 .034* -2 Log Likelihood=612.022; 2=290.646, df=4, p=<.001 Block 4A.3: Weekend Exposure + Acuity + Infant Characteristics Confidence Interval B S.E. Wald df Exp(B) Lower Upper Sig. Weekend Admission .058 .247 .055 1 1.060 .653 1.721 .814 Birth Weight -.120 .019 40.569 1 .887 .855 .920 p<.001*** Infant on Ventilation 2.316 .326 50.574 1 10.134 5.353 19.186 p<.001*** APGAR < 7 1.208 .236 26.110 1 3.347 2.106 5.319 p<.001*** SGA .109 .291 .141 1 1.115 .630 1.973 .708 Fetal Anomaly 2.672 .350 58.255 1 14.474 7.287 28.750 p<.001*** Non-Black Race .762 .265 8.236 1 2.142 1.273 3.604 .004** Male Infant .063 .229 .075 1 1.065 .680 1.668 .784 Multiple Birth .205 .283 .526 1 1.228 .705 2.139 .468 -2 Log Likelihood=539.460; 2=72.562, df=4, p=<.001

PAGE 160

146 Appendix M: Logistic Regressi on for Model 4A (continued) Block 4A.4: Weekend Exposure + Acuity + Infant Characteristics+ OB Interventions Confidence Interval B S.E. Wald df Exp(B) Lower Upper Sig. Weekend Admission .079 .248 .102 1 1.083 .665 1.762 .749 Birth Weight -.117 .019 38.737 1 .889 .857 .923 p<.001*** Infant on Ventilation 2.403 .331 52.677 1 11.054 5.777 21.151 p<.001*** APGAR < 7 1.213 .238 26.007 1 3.364 2.111 5.363 p<.001*** SGA .286 .302 .896 1 1.331 .736 2.407 .344 Fetal Anomaly 2.669 .350 58.151 1 14.428 7.265 28.652 p<.001*** Non-Black Race .755 .267 7.994 1 2.127 1.261 3.590 .005** Male Infant .047 .231 .041 1 1.048 .667 1.647 .839 Multiple Birth .281 .290 .939 1 1.325 .750 2.339 .332 Induction -.911 .623 2.141 1 .402 .119 1.363 .143 Cesarean Section -.484 .251 3.721 1 .617 .377 1.008 .054 -2 Log Likelihood=534.190; 2 =5.270, df=2, p=.072 Block 4A.5: Weekend Exposure + Acuity + Infant Characteristics+ OB Interventions + Hospital Variables Confidence Interval B S.E. Wald df Exp(B) Lower Upper Sig. Weekend Admission .066 .251 .068 1 1.068 .652 1.748 .794 Birth Weight -.116 .019 38.067 1 .890 .858 .924 p<.001*** Infant on Ventilation 2.412 .333 52.634 1 11.160 5.816 21.415 p<.001*** APGAR < 7 1.208 .239 25.611 1 3.346 2.096 5.342 p<.001*** SGA .276 .304 .827 1 1.318 .727 2.390 .363 Fetal Anomaly 2.671 .351 57.997 1 14.453 7.268 28.740 p<.001*** Non-Black Race .760 .268 8.055 1 2.139 1.265 3.615 .005 Male Infant .064 .233 .075 1 1.066 .675 1.682 .784 Multiple Birth .288 .292 .970 1 1.334 .752 2.366 .325 Induction -.903 .627 2.075 1 .405 .119 1.385 .150 Cesarean Section -.445 .254 3.076 1 .641 .389 1.054 .079 RN Hours/Day -.002 .014 .013 1 .998 .971 1.027 .911 Capacity .243 4.414 .003 1 1.276 .000 7287.434 .956 Nurse:Infant Ratio -1.000 3.775 .070 1 .368 .000 600.899 .791 -2 Log Likelihood=531.848; 2 =2.342, df=3, p=.504 **p<.01; ***p<.001 Model terms: weekend=1, birth weight=100 gr am increments, ventilation=1, APGAR=1, sga=1, anomaly=1, non-Black race=1, male=1, multiple=1, induc tion=1, c-section=1, RN hours=hours per 24 hour day, capacity=census/42, N:I ration=#RN per 24 hour day/census per day.

PAGE 161

147 Appendix N: Logistic Regression for Model 4B Table N1 Complete Logistic Regression Models 4B for Nighttime Exposure with Hospital-Level Covariates included with Death Before Discharge (n=1837) Block 4B.1: Nighttime Exposure Confidence Interval B S.E. Wald df Exp(B) Lower Upper Sig. Nighttime Admission .308 .187 2.712 1 1.360 .943 1.962 .100 -2 Log Likelihood=899.980; 2 =2.706, df=1, p=.100 Block 4B.2: Nighttime Exposure + Acuity Confidence Interval B S.E. Wald df Exp(B) Lower Upper Sig. Nighttime Admission .081 .218 .138 1 1.084 .708 1.661 .710 Birth Weight -.052 .014 13.919 1 .949 .924 .976 p<.001*** Infant on Ventilation 2.637 .322 66.863 1 13.970 7.425 26.285 p<.001*** APGAR < 7 1.217 .226 28.870 1 3.376 2.166 5.262 p<.001*** SGA .589 .274 4.631 1 1.802 1.054 3.080 .031* -2 Log Likelihood=611.884; 2 =288.096, df=4, p=<.001 Block 4B. 3: Nighttime Exposure + Acuity + Infant Characteristics Confidence Interval B S.E. Wald df Exp(B) Lower Upper Sig. Nighttime Admission .128 .232 .304 1 1.136 .722 1.789 .582 Birth Weight -.120 .019 40.152 1 .887 .855 .921 p<.001*** Infant on Ventilation 2.323 .326 50.622 1 10.203 5.381 19.347 p<.001*** APGAR < 7 1.188 .239 24.639 1 3.279 2.052 5.241 p<.001*** SGA .127 .293 .188 1 1.136 .639 2.018 .665 Fetal Anomaly 2.681 .351 58.474 1 14.603 7.345 29.035 p<.001*** Non-Black Race .751 .265 8.025 1 2.120 1.260 3.565 .005** Male Infant .064 .229 .078 1 1.066 .681 1.670 .780 Multiple Birth .214 .284 .570 1 1.239 .710 2.162 .450 -2 Log Likelihood=539.212; ; 2 =72.672, df=4, p=<.001

PAGE 162

148 Appendix N: Logistic Regressi on for Model 4B (continued) Block 4B.4: Nighttime Exposure + Acuity + Infant Characteristics+ OB Interventions Confidence Interval B S.E. Wald df Exp(B) Lower Upper Sig. Nighttime Admission .083 .234 .124 1 1.086 .686 1.718 .725 Birth Weight -.117 .019 38.487 1 .889 .857 .923 p<.001*** Infant on Ventilation 2.401 .331 52.629 1 11.038 5.769 21.117 p<.001*** APGAR < 7 1.200 .241 24.862 1 3.319 2.071 5.320 p<.001*** SGA .291 .303 .919 1 1.337 .738 2.423 .338 Fetal Anomaly 2.676 .351 58.233 1 14.533 7.308 28.900 p<.001*** Non-Black Race .748 .267 7.856 1 2.112 1.252 3.563 .005** Male Infant .049 .230 .045 1 1.050 .669 1.650 .831 Multiple Birth .283 .290 .948 1 1.327 .751 2.343 .330 Induction -.906 .621 2.124 1 .404 .120 1.367 .145 Cesarian Section -.470 .252 3.468 1 .625 .381 1.025 .063 -2 Log Likelihood=534.168; 2=5.044, df=2, p=.080 Block 4B.5: Weekend Exposure + Acuity + Infant Characteristics+ OB Interventions + Hospital Variables Confidence Interval B S.E. Wald df Exp(B) Lower Upper Sig. Nighttime Admission .064 .252 .064 1 1.066 .650 1.748 .800 Birth Weight -.119 .019 38.799 1 .888 .855 .922 p<.001*** Infant on Ventilation 2.431 .332 53.661 1 11.365 5.931 21.777 p<.001*** APGAR < 7 1.181 .243 23.679 1 3.257 2.024 5.241 p<.001*** SGA .280 .304 .850 1 1.323 .729 2.401 .357 Fetal Anomaly 2.734 .354 59.538 1 15.392 7.686 30.823 p<.001*** Non-Black Race .767 .268 8.165 1 2.153 1.272 3.644 .004** Male Infant .051 .233 .048 1 1.052 .667 1.661 .827 Multiple Birth .300 .292 1.054 1 1.349 .762 2.391 .305 Induction -.951 .634 2.250 1 .386 .111 1.338 .134 Cesarian Section -.478 .257 3.444 1 .620 .375 1.027 .063 RN Hours/Shift .048 .029 2.693 1 1.049 .991 1.111 .101 Capacity -7.457 4.453 2.805 1 .001 .000 3.561 .094 Nurse:Infant Ratio -15.264 8.043 3.602 1 .000 .000 1.647 .058 -2 Log Likelihood=529.025; 2=5.143, df=3, p=.162 **p<.01; ***p<.001 Model terms: weekend=1, birth weight=100 gr am increments, ventilation=1, APGAR=1, sga=1, anomaly=1, non-Black race=1, male=1, multiple=1, induc tion=1, c-section=1, RN hours=hours per 12 hour shift, capacity=census/42, N:I ration=#RN per 12 hour shift/census per day.

PAGE 163

149 Appendix O: Logistic Regression Model for 4C Table O1 Complete Logistic Regression Model 4C for Effect Modification with Hospital-Level Covariates with Death Before Discharge (n=1837) Block 4C.1: Weekend + Nighttime + Weekend*Nighttime Confidence Interval B S.E. Wald df Exp(B) Lower Upper Sig. Weekend Admission .166 .301 .306 1 .580 1.181 .655 2.129 Nighttime Admission .414 .224 3.422 1 .064 1.513 .976 2.347 Weekend*Nighttime -.349 .410 .722 1 .395 .706 .316 1.577 -2 Log Likelihood=899.248; 2=3.438, df=3, p=.329 Block 4C. 2: Weekend + Nighttime + Weekend*Nighttime + Acuity Confidence Interval B S.E. Wald df Exp(B) Lower Upper Sig. Weekend Admission .129 .337 .146 1 1.137 .588 2.202 .702 Nighttime Admission .161 .260 .384 1 1.175 .705 1.958 .535 Weekend*Nighttime -.262 .466 .315 1 .770 .309 1.918 .574 Birth Weight -.053 .014 14.161 1 .949 .923 .975 p<.001*** Infant on Ventilation 2.633 .322 66.755 1 13.916 7.399 26.171 p<.001*** APGAR < 7 1.212 .227 28.619 1 3.361 2.155 5.239 p<.001*** SGA .587 .274 4.609 1 1.799 1.052 3.075 .032* -2 Log Likelihood=611.568; 2=287.680, df=4, p=<.001

PAGE 164

150 Appendix O: Logistic Regression Model for 4C (continued) Block 4C.3: Weekend + Nighttime + Weekend*Nig httime + Acuity + Infant Characteristics Confidence Interval B S.E. Wald df Exp(B) Lower Upper Sig. Weekend Admission .205 .363 .321 1 1.228 .603 2.499 .571 Nighttime Admission .211 .275 .590 1 1.235 .721 2.115 .442 Weekend*Nighttime -.301 .498 .364 1 .740 .279 1.967 .546 Birth Weight -.120 .019 40.489 1 .887 .855 .920 p<.001*** Infant on Ventilation 2.323 .327 50.556 1 10.203 5.378 19.354 p<.001*** APGAR < 7 1.187 .239 24.584 1 3.277 2.050 5.238 p<.001*** SGA .130 .294 .197 1 1.139 .641 2.026 .657 Fetal Anomaly 2.668 .350 58.078 1 14.409 7.255 28.617 p<.001*** Non-Black Race .766 .267 8.238 1 2.150 1.275 3.627 .004** Male Infant .070 .229 .093 1 1.072 .684 1.681 .761 Multiple Birth .217 .285 .582 1 1.242 .711 2.170 .445 -2 Log Likelihood=538.819; 2=72.749, df=4, p=<.001 Block 4C.4: Weekend + Nighttime + Weekend*Nighttime + Acuity + Infant Characteristics + OB Interventions Confidence Interval B S.E. Wald df Exp(B) Lower Upper Sig. Weekend Admission .228 .363 .395 1 1.256 .617 2.558 .530 Nighttime Admission .163 .278 .342 1 1.177 .682 2.030 .559 Weekend*Nighttime -.295 .501 .347 1 .744 .279 1.988 .556 Birth Weight -.117 .019 38.765 1 .889 .857 .923 p<.001*** Infant on Ventilation 2.404 .332 52.582 1 11.071 5.780 21.204 p<.001*** APGAR < 7 1.200 .241 24.841 1 3.319 2.071 5.321 p<.001*** SGA .297 .304 .958 1 1.346 .742 2.442 .328 Fetal Anomaly 2.662 .350 57.841 1 14.320 7.212 28.435 p<.001*** Non-Black Race .764 .268 8.099 1 2.146 1.268 3.632 .004* Male Infant .054 .231 .054 1 1.055 .671 1.660 .815 Multiple Birth .284 .291 .952 1 1.328 .751 2.347 .329 Induction -.923 .623 2.190 1 .397 .117 1.349 .139 Cesarean Section -.468 .253 3.436 1 .626 .381 1.027 .064 -2 Log Likelihood=533.743; 2=5.076, df=2, p=.079

PAGE 165

151 Appendix O: Logistic Regression Model for 4C (continued) Block 4C.5: Weekend + Nighttime + Weekend*Nighttime + Acuity + Infant Characteristics.+ OB Interventions.+ Hospital-Level Covariates Confidence Interval B S.E. Wald df Exp(B) Lower Upper Sig. Weekend Admission .191 .367 .272 1 1.211 .590 2.486 .602 Nighttime Admission .121 .281 .184 1 1.128 .651 1.955 .668 Weekend*Nighttime -.245 .504 .236 1 .783 .292 2.102 .627 Birth Weight -.117 .019 38.208 1 .890 .858 .924 p<.001*** Ventilation 2.412 .333 52.523 1 11.157 5.811 21.422 p<.001*** APGAR < 7 1.199 .242 24.623 1 3.317 2.066 5.326 p<.001*** SGA .284 .305 .866 1 1.329 .730 2.416 .352 Fetal Anomaly 2.664 .351 57.708 1 14.359 7.221 28.555 p<.001*** Non-Black Race .769 .269 8.151 1 2.157 1.273 3.657 .004** Male Infant .070 .233 .090 1 1.072 .679 1.694 .764 Multiple Birth .289 .293 .971 1 1.335 .752 2.370 .324 Induction -.915 .628 2.124 1 .400 .117 1.371 .145 Cesarean Section -.438 .255 2.944 1 .646 .392 1.064 .086 RN Hours/Day -.002 .014 .012 1 .998 .971 1.027 .911 Capacity .268 4.429 .004 1 1.307 .000 7703.305 .952 Nurse:Infant Ratio -.971 3.793 .066 1 .379 .000 641.180 .798 -2 Log Likelihood=531.573; 2=2.169, df=3, p=.538 **p<.01; ***p<.001 Model terms: weekend=1, birth weight=100 gr am increments, ventilation=1, APGAR=1, sga=1, anomaly=1, non-Black race=1, male=1, multiple=1, induc tion=1, c-section=1, RN hours=hours per 24 hour day, capacity=census/42, N:I ration=#RN per 24 hour day/census per day.

PAGE 166

Appendix P: Receiver Operating Ch aracteristic Curves, Model 1 1 Specificity1.0 0.8 0.6 0.4 0.2 0.0 Sensitivity1.0 0.8 0.6 0.4 0.2 0.0 ROC CurveDiagonal segments are produced by ties. Figure 3: ROC for Anomaly Removed, Model 1 152

PAGE 167

Appendix P: Receiver Operating Characte ristic Curves, Model 1 (continued) 1 Specificity1.0 0.8 0.6 0.4 0.2 0.0 Sensitivity1.0 0.8 0.6 0.4 0.2 0.0 ROC CurveDiagonal segments are produced by ties. Figure 4: ROC with Ventilation Removed, Model 1 153

PAGE 168

Appendix P Receiver Operating Characte ristic Curves Model 1 (continued) 1 Specificity1.0 0.8 0.6 0.4 0.2 0.0 Sensitivity1.0 0.8 0.6 0.4 0.2 0.0 ROC CurveDiagonal segments are produced by ties. Figure 5: ROC for Anomaly and Ventilator Removed, Model 1 154

PAGE 169

About the Author Leisa Stanley received her Bachelor s Degree in Political Science and Communication in 1985 and her Ma sters Degree in Commun ication in 1987, both from Florida State University. She entered the Ph.D. program in Public Health at the University of South Florida with a joint c oncentration in Maternal Child Health and Epidemiology. She was inducted into Phi Ka ppa Phi Honor Society and Delta Omega Public Health Honor Society. She received a Maternal Child Health Bureau MCH Training Fellowship Grant. She is the Associate Executive Direct or of the Healthy Start Coalition of Hillsborough County. She directs th e planning and research activ ities of the Coalition in addition to the Fetal Infant Mortality Review Project. She has served on several statewide committees including the Birth Defects Registry Advisory Board, the Perinatal Periods of Risk Practice Collaborative and Black Infant Health Practice Collaborative. She has presented at both national and state conferences.


xml version 1.0 encoding UTF-8 standalone no
record xmlns http:www.loc.govMARC21slim xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.loc.govstandardsmarcxmlschemaMARC21slim.xsd
leader nam Ka
controlfield tag 001 001970181
003 fts
005 20081121110037.0
006 m||||e|||d||||||||
007 cr mnu|||uuuuu
008 081121s2008 flu sbm 000 0 eng d
datafield ind1 8 ind2 024
subfield code a E14-SFE0002421
035
(OCoLC)276123180
040
FHM
c FHM
049
FHMM
090
RA425 (ONLINE)
1 100
Stanley, Leisa J.
0 245
Association among neonatal mortality, weekend or nighttime admissions and staffing in a Neonatal Intensive Care Unit
h [electronic resource] /
by Leisa J. Stanley.
260
[Tampa, Fla.] :
b University of South Florida,
2008
3 520
ABSTRACT: The purpose of this study was to investigate the time of admission to a Neonatal Intensive Care Unit (NICU) and its association with in-hospital mortality among a cohort of neonates at a regional perinatal center. Two different time points were considered: admissions on the weekend versus the weekday and admissions during the nighttime shift versus the day shift. The secondary purpose of the study was to investigate if registered nurse staffing affected this association between NICU admission day or admission time and in-hospital death. Three separate databases were used which contained information on NICU admissions, hospital deliveries and nurse staffing. These databases were linked resulting in data for each individual mother-infant pair for each separate admission to the NICU. Readmissions to the NICU, NICU admissions which could not be linked with the delivery data, admissions from the Newborn Nursery and transfers from other hospitals were excluded from the study.^ The final study population consisted of 1,846 admissions from October 1, 2001 through December 31, 2006. Weekend admissions were lower than weekday admissions (29.6% versus 70.4%) and nighttime admissions were lower than day admissions (43.2% versus 56.8%). Infants admitted at nighttime were more likely to be low birth weight, have lower Apgar scores and less likely to be delivered by cesarean section. Weekend admissions did not differ significantly from weekday admissions, except weekend admissions were more likely to be Black (33.6% versus 28.6%, p=.30). After adjusting for infant's acuity and other covariates using multivariate logistic regression, the odds of dying on the weekend was not significantly different than weekday admissions (AOR=1.06, 95% CI=.653-1.721) and were not significantly different for nighttime admissions (AOR=1.14, 95% CI=.722-1.79). Nurse staffing was not a significant covariate.^ Covariates which were significant risk factors for death prior to discharge were non-Black race of the infant, Apgar score of less than 7 at five minutes, presence of a fetal anomaly, and use of ventilation during the stay. Infant's birth weight was a significant protective factor.
502
Dissertation (Ph.D.)--University of South Florida, 2008.
504
Includes bibliographical references.
516
Text (Electronic dissertation) in PDF format.
538
System requirements: World Wide Web browser and PDF reader.
Mode of access: World Wide Web.
500
Title from PDF of title page.
Document formatted into pages; contains 154 pages.
Includes vita.
590
Adviser: Karen (Kay) Perrin, Ph.D.
653
Infant death.
Acuity.
Nurse staffing.
Adverse events.
Obstetrical interventions.
690
Dissertations, Academic
z USF
x Public Health
Doctoral.
773
t USF Electronic Theses and Dissertations.
4 856
u http://digital.lib.usf.edu/?e14.2421