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 001920851
007 cr mnu|||uuuuu
008 080110s2006 flu sbm 000 0 eng d
datafield ind1 8 ind2 024
subfield code a E14-SFE0001852
The association of head circumference with selected cognitive outcomes in older adults in Charlotte County, Florida
h [electronic resource] /
by Cathleen Copenhaver.
[Tampa, Fla] :
b University of South Florida,
ABSTRACT: OBJECTIVE: The brain reserve hypothesis was examined in a secondary analysis of cross-sectional data from a community-based sample of 468 older adults residing in Charlotte County, Florida. The objective of the analysis was to determine the association between head circumference and eight cognitive outcomes and to assess any potential effect modification of existing associations by Apolipoprotein E (APOE) genotype. METHODS: Cognitive outcomes include scores from the Modified Mini-Mental State Exam (3MS), the Hopkins Verbal Learning Test-Revised (HVLT-R), Stroop Color-Word Test, Trail-Making Test A and B, and a word-stem completion task measuring implicit memory. Descriptive statistics were calculated for each variable. Head circumference and dependent cognitive outcomes were modeled as dichotomous variables using logistic regression, adjusting for gender, age, education, income, height, and Spot The Word test score, a measure of pre-morbid IQ. For dichotomized test scores, poor outcomes (cases) were defined as having scores in the lowest quintile; the remaining top four quintiles were considered non-cases. RESULTS: small head circumference was significantly associated with low 3MS scores [OR(95%CI): 2.97 (1.12, 7.89), p=0.03], after adjustment for age, income and pre-morbid IQ. The association remained statistically significant after adjustment for gender and education as well. After adjustment, head circumference was not found to be statistically significantly associated with any other cognitive outcome. No effect modification was found by APOE genotype or years of education. CONCLUSION: This analysis confirms previous findings that exposure to low head circumference significantly impacts cognition in late life.
Thesis (M.S.P.H.)--University of South Florida, 2006.
Includes bibliographical references.
Text (Electronic thesis) in PDF format.
System requirements: World Wide Web browser and PDF reader.
Mode of access: World Wide Web.
Title from PDF of title page.
Document formatted into pages; contains 61 pages.
Adviser: Amy Borenstein, Ph.D.
x Public Health
t USF Electronic Theses and Dissertations.
The Association of Head Circumference W ith Selected Cognitive Outcomes in Older Adults in Charlotte County, Florida by Cathleen Copenhaver A thesis submitted in partial fulfillment of the requirements for the degree of Master of Science in Public Health Department of Epidemiology and Biostatistics College of Public Health University of South Florida Co-Major Professor: Amy R. Borenstein, Ph.D. Co-Major Professor: James A. Mortimer, Ph.D. Yougui Wu, Ph.D. Date of Approval: November 17, 2006 Keywords: brain reserve, alzheimer's disease, aging, 3ms, apolipoprotein e Copyright 2006 Cathleen Copenhaver
Acknowledgements I would like to extend special apprecia tion toward the members of my thesis committee. Their support, friendly advice a nd professional opinions inspired motivation, enthusiasm, and practicality when they were needed most. I would also like to thank members of the Department of Epidemiology a nd Biostatistics, professors and students, who provided moral sup port throughout all stag es of this project.
i Table of Contents List of Tables ii Abstract iii Chapter 1: Introduction 1 Chapter 2: Background 3 Chapter 3: Methods 15 Design and Population 15 Measures 16 Analysis 20 Chapter 4: Results 24 Descriptive Analysis 24 Bivariate Analysis 25 Multivariate Analysis 27 Interaction With APOE 36 Chapter 5: Discussion 38 References 46 Appendix A: Additional Tables 58
ii List of Tables 1. Descriptive Measures for Head Circum ference, Covariates and Outcomes 24 1a. Mean and Standard Deviation for Continuous Head Circumference and Covariates 24 1b. Mean and Standard Deviation for Continuous Outcomes 25 1c. Frequency and Relative Frequenc y for Gender and APOE Status 25 2. Frequency and Relative Frequency of Dichotomized Variables 26 3. Model Selection for Main Effect of H ead Circumference and APOE Interaction 28 3a. Model Selection for HC with 3MS 28 3b. Model Selection for HC with ImR 29 3c. Model Selection for HC with DeR 30 3d. Model Selection for HC with CuR 31 3e. Model Selection for HC with Recognition 32 3f. Model Selection for HC with Stroop 33 3g. Model Selection for HC with Trails 34 3h. Model Selection for HC with Implicit 35 A. Unadjusted Logistic Regression of Head Circumference, Outcomes and Covariates 58 A1. Dichotomized Outcomes Modeled with Dichotomous Head Circumference (Crude) 58
iii A2. Dichotomized Outcomes Modeled with Covariates (Crude) 59 A3. Dichotomized Head Circumference Modeled with Covariates (Crude) 61
iv The Association of Head Circumfe rence With Selected Cognitive Outcomes in Older Adults in Charlotte County, Florida Cathleen Copenhaver ABSTRACT OBJECTIVE: The brain reserve hypothesis was examined in a secondary analysis of cross-sectional data from a community-based sample of 468 older adults residing in Charlotte County, Florida. The objective of th e analysis was to determine the association between head circumference and eight cognitiv e outcomes and to assess any potential effect modification of existing associa tions by Apolipoprotein E (APOE) genotype. METHODS: Cognitive outcomes include scores from the Modified Mini-Mental State Exam (3MS), the Hopkins Verbal Learning Test-Revised (HVLT-R), Stroop Color-Word Test, Trail-Making Test A and B, and a wo rd-stem completion task measuring implicit memory. Descriptive statistics were calculat ed for each variable. Head circumference and dependent cognitive outcomes were modeled as dichotomous variables using logistic regression, adjusting for gender, age, educati on, income, height, and Spot The Word test score, a measure of pre-morbid IQ. For di chotomized test scores, poor outcomes (cases) were defined as having scores in the lowest quintile; th e remaining top four quintiles were considered non-cases. RESULTS: Sma ll head circumference was significantly associated with low 3MS scores [OR(95%CI): 2.97 (1.12, 7.89), p=0.03], after adjustment for age, income and pre-morbid IQ. The association remained statistically
v significant after adjustment for gender and e ducation as well. After adjustment, head circumference was not found to be statistical ly significantly associated with any other cognitive outcome. No effect modifica tion was found by APOE genotype or years of education. CONCLUSION: This analysis conf irms previous findings that exposure to low head circumference significantly impacts cognition in late life.
1 Chapter 1 Introduction This thesis offers an analysis of data from a community-based survey of older adults in Charlotte County, Florida. The purpose of the Charlotte County Healthy Aging Study (CCHAS) is to identify risk factors for cognitive function, and to understand life satisfaction and quality of lif e among an elder population. The objective of the current analysis is to study head circumference as a risk factor for cognitive function under the brain reserve hypothesis, and to examine a ny existing association for modification by genetic predisposition to Alzheimers disease. Head circumference has been studied previ ously as a risk factor for dementia and for neuropsychological outcomes. However, many of the cognitive outcomes examined in this analysis have not yet been adequately addressed within the literature on head size and cognition under the brain reserve hypothesis. This thesis seeks to address that gap by analyzing the risk due to smaller head ci rcumference of poor outcomes among eight wellknown neuropsychological tests. These tests are often used within a standard battery for identifying dementia and cognitive impairm ent among the elderly. While it is unlikely that head circumference will become part of a standard clinical risk profile for dementia, the etiological knowledge gained from discove ry of its influence on these tests will be invaluable. Through this knowledge, a better understanding of th e determinants of clinical presentation of dementia can be achieved, leading to potential new prevention
2 therapies that capitalize on m odifiable components of individual reserve capacity against cognitive decline. In the following pages, a detailed explan ation of the brain reserve hypothesis will be presented, along with thorough coverage of the existing literatu re on the subject of head size and cognition. This background will be found in chapter two, followed by a description of the population of CCHAS and the methods used in that study and in this analysis in chapter three. Description of the eight cognitive tests under study will also be provided in chapter three. In chapter four, the results of th is analysis will be presented, and a detailed discussion of the results and their interpretation will follow in chapter five.
3 Chapter 2 Background Dementia and cognitive impairment among the elderly cause considerable public health concern, particularly in populations that have growing numbers of older adults. In North America, the prevalence of dementia has been estimated between 6 and 10 percent of adults aged 65 years and older, with nearly two-thirds of all dementia cases diagnosed as Alzheimers disease (AD). 1-5 The prevalence of both all-cause dementia and AD rise dramatically with age, increasing from belo w an estimated 3 percent for those aged 6574, to approximately 11 percent between ages 75-84, to over an estimated 30 percent in persons aged 85 years and older. 1-3 Mild cognitive impairment, considered a state of risk for dementia and Alzheimers disease, has also been measured in the community using different diagnostic crit eria at a prevalence between 3 and 19 percent, with an estimated conversion rate to dementia of 11-33 percent over two years. 6, 7 Etiologic factors that could be targeted for primary prevention, or that could aid in an understanding of the etiology of clinical presentation, are bei ng widely studied. Among them, the study of individual capacity to avoid the symptoms of dementia or cognitive impairment during life has gained significant attention in the lite rature, largely in the field of Alzheimers disease. The concept of individual brain reserve originated from autopsy studies of Alzheimers disease, where it was observed that individuals may m eet the criteria for
4 neuropathologic AD at autopsy, even at very old ages, without having manifested symptoms of cognitive decline. 8-12 These observations led to hypotheses regarding the individual differences in apparent tolerance for the pathology resulting in the maintenance of relatively normal functioning. The specific idea of brain reserve was posed following one such autopsy study, in which the observed differences between the demented and cognitively normal included a higher brain weight and greater number of large neurons at autopsy of the non-demented individuals. 8 A hypothesis of brain reserve was formalized, 13 stating that individuals may possess more neural substrate (larger brains ) and redundant neural networks that allow for normal cognition in the presence of neuropathology. Conversely, those with small head sizes and therefore smaller brains may be at greater risk for cognitive decline given the same pathology. The brain reserve hypothesis a nd its role in a threshold model of dementia were further specified by Mortimer. 14, 15 In this model, clinical expression of dementia in the individual is dependent both upon a propensit y to accumulate pathological lesions and upon the attainment of a critical threshold of neural reserve below which normal cognition can no longer be maintained. 14 Risk factors for neuropathology and for clinical expression may therefore be cons idered separately. In Alzhei mers disease, for example, risk factors for neuropathology include gene tic predisposition, Downs syndrome, head injury, diabetes, and cardiovascular and cereb rovascular conditions. Clinical expression, on the other hand, may be dependent upon br ain development, body growth, early-life socioeconomic conditions, income, education and IQ. 16, 17
5 The risk factors for clinical expressi on, as opposed to those for neuropathology, are contributors to the unifie d concept of brain reserve. In clarifying this concept, Mortimer stipulated that reserve could assume any of three forms: the number or density of neurons attained in adolescence, the coll ection of cognitive strategies and test-taking abilities (akin to cognitive re serve, discussed below), and the amount of functional brain tissue at any age. 15 Individual differences in these three subtypes of brain reserve, combined with individually-determined rates of pathological accumulation, influence the trajectory of descent toward the threshold, resul ting in earlier clinical presentation of abnormal cognition for those with smaller maxi mal attained brain size, fewer cognitive strategies, and faster rates of accumulation. The current study aims to identify whether smaller attained brain size increases the risk of poor cognitive perfor mance in an analysis that controls for both individual cognitive ability and a predictor of the rate of accumulation of Alzheimer pathology in a community sample. This analysis will use head circumference, pre-morbid IQ and gene tic predisposition to Alzheimers disease as proxy measures, respectively, for the three conditions under which cognitive decline is hypothesized to clinically manifest. There is also an interest as to whether the predisposition toward faster rates of accumula tion interacts with smaller attained brain size to modify the association between h ead size and cogniti ve performance. Prior to reviewing the literature investig ating the associations between head size and cognition, it may be informative to discuss the determinants of head size measurements. The growth of the cranium is driven by brain growth achieved during childhood. 16 It is known that both brain weight and intracranial volume achieve at least 75% of their maximum size by the age of three and achieve adult size by age 15. 18, 19
6 Brain growth is largely dependent upon genetic and environmental factors, although the extent of separate influence from these fact ors is unclear. For example, it has been observed that poor nutrition leads to dela yed or abnormal brain development not amenable to catch-up growth 17, 20 It has also been found in animal studies that an enriched environment increases brai n weight and dendritic branching. 21 Additionally, secular increases in brain size have been noted in de veloped countries, 14, 22 furthering the case for the role of environment in brain gr owth. On the other hand, twin and family studies indicate a potentially large role for genetic, as compared to environmental, contribution. Estimates of the genetic influe nce on head size have placed heritability of intracranial volume at roughly 80% 23 and of head circumference at close to 60%. 24 Nevertheless, the authors caution against the pos sibility that these heritability estimates may be inflated, citing the difficulty in th e separation of environmental and genetic factors of influence. It is possible, theref ore, that both environmen tal and genetic factors play substantial roles in the growth of the brain, and thus attained head size. Most of the research on estimates of ma ximal attained brain size with cognition has generated consistent support for the brain reserve hypothesis. Several studies have found a consistent association between head circumference and measures of cognition among those diagnosed with dementia. 25-27 In one of the earli est studies examining a large (n=1985) population-based sample of Japanese-Americans in King County, Washington for prevalent dementia ( Kame Project), a significant association (p=0.006) between head circumference and Cognitive Abilities Screening Instrument (CASI) scores was found among those diagnosed with AD. 25 It was found that scores among those patients with lower HC were worse than those among patients with higher HC, an
7 association that was not seen among nondemented individuals. Another study examining a hospital-based sample of AD patients in Japan found that intracranial volume (ICV), a measure of maximal brai n size, was correlated between 0.289 and 0.396 (p<0.05) with several measures of intelligen ce, including the Wechsler Adult Intelligence Scale-Revised (WAIS-R) and Ravens Colored Progressive Matrices (RPM). 26 A third and more recent study identified a statistica lly significant correlation of 0.33 (p=0.01) between ICV and MMSE scores among those with AD and VaD. 27 The same study, using linear regressi on, found that a 66.7 cm 3 increase in ICV was associated with a 1point change in MMSE, adjusted for age and gender. Compelling evidence in support of the brain reserve hypothesis has come from studies that compare cognitively impaired subjec ts with normal controls. Schofield et al. found, for example, that head circumferen ce in the lowest gender-specific quintile, compared with the other four quintiles, wa s associated with increased prevalence of probable and possible AD with an OR ( 95%CI) of 2.9 (1.4, 6.1) in women and 2.3 (0.6, 9.8) in men. 28 Wolf et al. examined ICV across normal controls and patients with prevalent mild cognitive impairment (M CI), AD and vascular dementia (VaD). 29 The authors found a decreasing prevalence of cognitive impairment among increasing quartiles of ICV (chi-square statistic for linear trend = 8.5, df=1, p=0.003). Later work by Wolf et al. on a different population demonstrated an in creased risk of cognitive impairment [OR: 2.8 (1.3, 6.0)], including MC I, and of dementia [OR: 2.5 (1.2, 5.1)] for ICV in the lowest quartile compared to the top three quartiles. 27 Additionally, Mortimer et al. found in a study of 294 Catholic sisters that head circumference in the lower two
8 tertiles and low education (<16 years) interacted to substantially increase the risk of prevalent dementia [OR: 4.3 (1.9, 9.6)]. 30 A third line of research on the role of brain reserve in the timing of clinical presentation has generated slightly less cons istent results. Seve ral studies have found strong inverse associations between time to onset of cognitive impairment and head size. Schofield et al., in a study of 28 female AD patients from a New York clinic, found that age at first symptom onset was significan tly correlated (r=0.64, p=0.01) with a crosssectional measure of intracranial area. 31 In the follow-up phase of the Kame Project, mentioned previously, two subsequent analys es revealed head circumference as an important risk factor for incident AD among those at genetic risk for the disease, defined as possession of one or more APOE 4 alleles. 32, 33 The first analys is found that head circumference in the lowest tertile (<21.4 inches) combined with positive 4 status predicted earlier onset of incident AD at a hazard ratio (HR) of 14.1 (3.0-65.0), adjusted for gender and education. 32 The second analysis found that developmental variables including small head circumference ( 54.4 cm, or 21.4 inches) and having four or more children in the household at age 2-3 years we re independently pred ictive of earlier onset of AD among 4-positive subjects, while vascular risk factors were more important among 4-negative individuals. 33 These findings illustrates the importance of reserve factors (e.g. HC) in delaying clinical presenta tion of disease among those individuals who may be at genetic risk for fa ster accumulation of pathology. There have been a few studi es that have not found evid ence that pre-morbid brain size protects against cognitive decline. In one case-control examination of familial (early onset) and sporadic (late ons et) AD, no difference was found in mean ICV across groups,
9 controlling for age, education and gender. 34 The author also found that ICV, the dependent variable in a lin ear regression, did not differ by age or age at onset of symptoms, although the approach of adding an indicator of dementia type (familial or sporadic) to the model may have offered less than ideal control for this factor without stratified analysis. Fu rthermore, it is possible that findings may have varied had the authors examined the range of risk (lowes t quantile) instead of continuous ICV. A second analysis by Edland et al. with more thorough control of potential confounders and examination of both tertiles and continuous ICV in a case-control study of 166 subjects with and 184 subjects without AD found no significant difference in mean ICV between patients and controls accounting for age, education, APOE 4 status, and birth year. 35 ICV in the lowest tertile did not predict AD for men or women, and age at onset was not correlated with ICV. Although the study appears well-controlled, the population in which this study was conducte d tended to be a well-educated sample 36 and the author cited the limitation that adverse developmental conditions (leading to reduced adult brain size) probably were not common among the sample. It is possible, therefore, that there were insufficient numbers of indi viduals with small ICV. Additionally, it was unclear whether the authors specified the lowe st tertile in a gender-specific manner, as did Wolf et al. and Schofield et al. mentioned above. 27-29 Analyses from the Kame Project, which similarly found no mean diffe rence of HC between cases and non-cases, also did not specify HC in a gender-specific fashion. 25, 32, 33 Given likely differences in brain structure and function in men and women, 37 specification of the gender-specific range of risk in head size may be subject to less bias fr om misclassification and may result in more valid measures of risk. Ge nder-specific quantiles also help to overcome
10 problems of collinearity between gender a nd head size in modeling, such as those encountered in the Kame analysis. 33 It is unknown as to the ex tent to which these issues may have contributed to the null finding in the results presented in Edland et al. 35 The most recent study to produce null fi ndings provided both a large-scale casecontrol analysis and a longitudinal followup of 450 non-demented controls examining head circumference and time to conversion to AD. 38 This study found no mean difference in HC across case-status in th e case-control analysis and no association between either continuous HC or the lower gender-specific quartile of HC and time to conversion to AD, although the sample size and relatively few conversion events may have limited the prospective analysis. The analyses were adjusted for age, APOE 4 status, family history of AD, gender and edu cation. While there are many strengths to this study in addressing some of the analytic concerns, findings in previous analyses of time-to-AD with head size as a predictor 33 indicate the importance of stratifying results by APOE 4 status to examine reserve while c ontrolling for genetic predisposition toward faster decline. Examining the reserve hypothesis more generally, several studies have found associations between head size and cognitive outcomes, particularly cognitive decline, among non-demented and community-living individuals. Among the first to do so, Reynolds et al. examined head size and MMS E scores in a community-based survey of the MoVIES cohort (n=852), which followed initially non-demented subjects of mostly Caucasian background with relatively low socioeconomic status from 1993 until 1996. 39 In a baseline analysis, and us ing gender-stratifie d logistic regression with low MMSE ( 10 th percentile) as the dependent variable, each 1-centimeter increase in HC was
11 inversely associated with low MMSE scor es [OR (95%CI) women: 0.83 (0.69, 0.99); OR men: 0.79 (0.63, 0.98)]. A similar finding was observed by Tisserand et al. in a baseline examination of multiple cognitive outcom es with HC among initially non-demented adults. 40 In that study, increasing continuous HC adjusted for age and gender was significantly and positively associated w ith continuous MMSE scores and several measures of intelligence using the Groningen Intelligence Test, adjusted for gender, sex, and either height, socioeconomic background or educational level. The authors also found significantly increased time to completion of the Stroop Color-Word task, scored so that longer times indicate slower processing speeds, with lower HC. 40 A third study also confirmed the association between adult head size and cognition, finding that mean intelligence scores as measured by the AH4 test at two test periods 3.5 years apart in a sample with a m ean age of 70 years increased with increasing quartile of HC (p for trend <0.01 at both tria ls, adjusted for age, sex, education, social class at birth, history of cerebrovascular di sease, and health profile emotional scores). 41 Conversely, no such trend was observed for cr oss-sectional measures of immediate and delayed memory measured by the Logical Me mory Test, an observation replicated by Tisserand et al., though there wa s a trend toward decline in immediate recall (p=0.03) and delayed recall (p=0.07) over time for t hose with smaller head circumferences. 40, 41 Finally, two studies examined the associ ation between ICV and cognition with inconsistent results. One study comprehens ively measured both ICV and regional brain volumes in 97 healthy, non-medicated men with a mean of 68 years of age. 42 ICV correlated with measures of pre-morbid intelligence (National Adult Reading Test, r=0.304, p<0.005; Ravens Standard Progressive Matrices (RPM), r=0.39, p<0.000) and
12 visual memory, but not immedi ate or delayed verbal memory (from the Auditory Verbal Learning Test, or AVLT, among others). The second study measured ICV and cognitive functioning also using RPM and AVLT in a sample of healthy men born in 1921, aged 79 years, finding no association between th e passive reserve represented by ICV. 43 The measure was adjusted for intelligence scor es measured at age 11, measures of brain burden by number of white-matter hyperinten sities, age and sex. A comparison between the studies is made very difficult by the di fferences in adjustment and subject age: whereas the former study did not make adjust ments for other factors, the latter study more rigorously explored the reserve hypot hesis by controlling for pathological burden and childhood intellectual function, which may provide a cognitive re serve in terms of more flexible or efficient networks that ma y protect against declin e in lieu of brain reserve. 43, 44 However, the statistical adjustment and inclusion of childhood IQ in the study by Staff et al. may have weakened th e association to the extent that it was undetectable in that sample. Additionally, th e much older subjects in that study may have been selected by survival to have slower rates of accumulated pathology and less effect of head circumference, as was suggested in the Kame Project analyses above. Unfortunately, no data on APOE 4 status was available for either study, rendering further speculation about the effect of ge netic risk for AD pathological accumulation impractical. The consideration of APOE 4 status as a proxy for accumulated pathology and of pre-morbid intelligence as a measure of cognitive reserve is important in the study of the reserve hypothesis. Cognitive reserve has been proposed as a complementary means of explaining the disparity between sufficien t pathological accumulation and clinical
13 presentation of dementia under the reserve hypothesis. 44 It states that those individuals possessing more efficient and flexible cognitive networks may compensate for pathology more successfully during life than those who do not have such networks. Staff et al. and Mortimer et al., among others mentioned previo usly, have used measures of pre-morbid (i.e. pre-decline, whether age-related or pathological) intelligence and/or education level as a proxy for such reserve. 30, 43 Most commonly, the measure of adjustment is education, given a known asso ciation with head size 29, 35, 40 and with cognitive measures, 45, 46 although pre-morbid intelligence measures may be a more sensitive individual measure of cognitive reserve. 47, 48 Diverse measures of increasing physical reserve, such as physical activity in youth, height and limb length, which may indicate a type of robustness against declin e in age, have also been dem onstrated to have an inverse association with cognitive decline. 49-53 None of the studies of physical reserve, however, have accounted for the correlation betw een body measurements and head size. Consistent adjustment for these measures of burden and reserve have been absent in studies of head size with cognitive impairment or decline. While measures of cognitive reserve are often controlled for in analyses with demented subjects, it has produced inconsistent results among the cognitiv ely normal. Adjustment for height and APOE 4 status has also been sporadic in th e literature, even though the substantial findings involving these measures indicate a need to consider them as standard potential confounders in the study of cognitive decline. In the current analysis, head size as measured by HC is compared with eight c ognitive outcomes in a sample of communityliving older adults. Both APOE 4 status and height, along with education and a measure of pre-morbid IQ, 54 are considered as potential conf ounders, and data are examined for
14 potential effect modification by APOE 4 status. BMI is also included among the covariates as a convenient, though imprecise proxy for potential risk from vascular, diabetic or metabolic syndrome factors. 55, 56 The main hypothesis being tested is th at small head circumference, determined in a manner that includes the ge nder-specific range of risk, is associated with a greater odds of poor cognitive outcome among subjects in the community. The secondary goal of the analysis is to test fo r effect-modification by APOE 4 status in any association found between head circumference and cognitive outcome. Literature support and applicability of the hypothesis to the specific cognitive outcomes analyzed in this study will be discussed in further detail in the Discussion chapter.
15 Chapter 3 Methods Design and Population This study analyzes secondary data from the Charlotte County Healthy Aging Study (CCHAS), a cross-sectional study of pr imarily Caucasian community-living older adults in Charlotte County, Flor ida. Two census tracts from this region were selected for CCHAS based on information from the 1990 U.S. Census that indicated it contained the highest proportion of resident s aged 65 and older in Florid a. One tract contained 7,093 inhabitants in 1990 (45.2% aged over 65 years) and the other contai ned 6,233 inhabitants in 1990 (37.4% over 65 years). Census bloc ks were sampled consecutively from a randomly ordered list obtained for each of the two larger census tracts. Prior to recruitment, a large-scale publicity campaign generated support for the study through newspaper articles, presentations to community organizations local television and radio appearances, and the formation of a Comm unity Advisory Council of 12 community leaders. 57 Eligible subjects were identified by canvassing all households, documenting all addresses within each block, and then revisiti ng the blocks at differe nt times on different days to request household information on name, age and sex of all persons in residence. Unreachable households were those that did no t answer the door after two attempts. A stratified sampling procedure was used within the selected census blocks until the desired
16 sample size was reached. The final samp le was to contain approximately equal proportions of male and female participants, a total of 126 persons within each of two age strata: age 60-74 and 75-84. 57, 58 The sampling method was also modified slightly to facilitate obtaining as representative a samp le as possible of the independently living elderly population of Charlotte County (for example, by including all members on each side of a street when a census block divided them). Of 4017 households surveyed in the selected tracts, census data were obtained for 2164 (54%) in the first phase of the study. From these, 1394 subjects were identified as eligible and invited to part icipate via postal mail and follo w-up telephone calls, including multiple eligible subjects per household. Of those eligible, 584 (42%) were unreachable, defined as no answer to up to nine telephone attempts at contact. Of those eligible and reachable, 468 participated in the study, 306 refused, and 36 accepted and later declined. From age, gender and education data on a portion of those who refused or did not complete all phases, it was determined that those who completed all phases were comparable to those who did not. Refusers we re similar in age to completers, were more likely to be women and to have less education (p<0.05). Sampled participants were primarily Caucasian (>98%). 57, 58 Measures CCHAS Procedures Exposure, outcomes, and potentially c onfounding and/or modifying variables for each subject were assessed during two persona l, structured interviews conducted over approximately one week, both in the homes of the subjects or at a neutral location
17 according to subject preference. 58 Income and years of education were assessed during the first interview, in which trained personnel administered a risk factor questionnaire. Interviewers recorded physic al measures, including head circumference (HC), weight and height, and assessed cognitive outcomes during the second interview one week later. Blood samples were collected by a phlebot omist during the week between the two interviews. The phlebotomis t visited the subjects home, where blood samples were taken to measure cholesterol levels, A1C le vels for diabetes, folate, and to obtain DNA for genotyping of APOE. All data were collected between November 1, 1997 and June 30, 1998. Procedures for CCHAS were approve d by the University of South Florida Institutional Review Board and written inform ed consent collected for each participating subject. Variables Head Circumference Head circumference (HC) is the primary exposure of interest in the curren t analysis. HC was measured by pl acing a non-distensible, flexible measuring tape at the line of the eyebro ws and passing it snugl y around the outermost occipital protuberance, returning to the eyebrows. HC was measured in inches for each subject and rounded to the nearest inch. 32 Cognitive Outcomes The following defines the eight cognitive outcomes of interest. All outcomes were dichotomized as described in the section on Bivariate Statistics in the Methods. a. The Modified Mini-Mental State Examin ation: The 3MS is a standardized measure of general cognitive ability in which the subject is asked a series of
18 questions. It samples a range of cognitive functions and the total score, used for this analysis, is scaled from 0 to 100. 59 b. Memory subscales of the Hopkins Verb al Learning Test-Revised (HVLT), 60 including Immediate Recall (ImR), Dela yed Recall (DeR), Cued Recall (CuR) and Recognition (Cite Small here): For each of the four measures, a word list of 12 words was used. For ImR, ther e are three learning trials, and the number of words remembered immediatel y after each trial is recorded and the three totals summed. For DeR, the subject is asked to remember the list in the three trials of free recall after an interference test; the total number of words recalled for each trial are summed. In the CuR subsection, administered directly after the DeR subsection, the subject is asked to recall the words according to the category they belong t o. There are three categories, and the numbers of words recalled in each categor y are summed for a total cued score. The score used in this analysis is the gain from cues, measured as the DeR score subtracted from the summed total cued score. Finally, the subject is read a list of 24 words in cluding the 12 they were asked to remember, and is asked to respond as to whether or not th e word read was on th e original list of 12. The correct and incorrect posit ive responses are summed, and the incorrect positive responses are subtracted from the correct positive responses to yield a discrimination index score representing Recognition. c. Stroop Color-Word Test 61 (Stroop): In this test, the subject is asked to read from three panels containing, respecti vely, words referring to colors (e.g. Blue), only colors with no words, and mismatched color-word combinations.
19 The score used in this analysis is th e number of words completed on the colorword task within a time limit. d. Trail-Making Test Parts A and B 62 (Trails): In these tests, subjects are asked to draw a line trail through randomly scattered bubbles containing numbers only (Part A) or numbers and letters (Part B) in sequential and sequential/alphabetical order, respectively. The score used for this analysis is the difference between the amount of time it took for th e subject to correctly finish Parts A and B. e. Implicit Memory 63 (Implicit): For this test, the subject was asked to read a word list containing nonsense and real wo rds and to identify the real words. After an interference test (Trails), the subject was asked to complete word stems. Half of the stems were from r eal words they had seen on the previous task. The score used for this analysis, the priming score, is the total number of stems completed with words subjects ha d seen before minus the total number of stems completed with wo rds not seen before. Other Variables of Interest Age was measured as the age at examination and calculated using SAS variables for the unique dates of examination and birth. Education was measured as the number of formal years of regular school completed, as reported by the subject. Income was scored as an ordina l variable representing nine levels of annual household income from below $10,000 per year (Level 1) to above $150,000 per year (Level 9). Scales to measure height were calibrated to one another and taken into the field. Height was measured against a wa ll and was in inches, rounded to the nearest inch. BMI was calculated as the weight in kilograms of the subject divided by the
20 squared height in meters. Pre-morbid intelligence was assessed using a standard measure, Spot The Word Test 1. 54 In this test, participants are asked to identify the real words in a list of 60 word pairs containing one real and one nonsense word. Each correctly identified word earns one point, with a maximum score of 60. Higher scores indicate higher crystallized intelligence, an indication of higher pre-morbid intelligence. APOE Genotype Blood samples were prepared by separa ting leukocytes from whole blood and lysing the cells using prepared stock solutions. The DNA was extracted by centrifugation and washed with ethanol before resuspensi on in tris-EDTA acid buffer solution. APOE genotyping was performed using standard digestion and PCR amplification. 57, 58 From the dual-allele genotype, APOE st atus was dichotomized as an 4 allele being present or absent, defined in the methods of analysis below. Analysis All analyses were performed using the SA S Statistical Software package, version 9.3. Univariate Statistics Histograms were generated to visually a ssess the distribution of each continuous variable, and the mean and standard deviation of each was calculated to describe central tendency and dispersion. Means and standard deviation for the exposure measure, HC, were also stratified by gender, and tested for difference in means using a t-test for independent means. 64 For categorical variables, fre quency and relative frequency were calculated to describe di stribution across levels.
21 Defining the Variables for Analysis The exposure variable, HC, was first st ratified by gender and then frequency, relative and cumulative relative frequencies for each -inch interval were examined across the entire range. This was done to identify an acceptable dichotomization point that would accurately reflect the range of risk by gende r in this sample. Once an appropriate point was identified, male and fe male subjects were classified by their exposure status separately then combined within a single dich otomous variable. Outcomes were also dichotomized. Those s ubjects with scores falling at or below the bottom 20 th percentile were classified as havi ng a poor outcome, as compared with normal outcomes above the bottom 20 th percentile. (Note: For Trails, the absolute value of the difference was scored, where larger di fferences indicated a poor outcome, therefore scores falling at or above the top 20 th percentile were classified as a poor outcome.) Two exceptions, CuR and Implicit memory, were dic hotomized at the inte rval directly below that in which the 20 th percentile resided due to a sma ll range and tight clustering about the mean. Body Mass Index was dichotomized as obese and non-obese, using a BMI of 30 to indicate obesity. 65 All other covariates, except gender and APOE, were kept as continuous variables. APOE status was de scribed as either possessing one or more APOE 4 alleles ( 4-positive) or possessing no 4 alleles ( 4-negative). Bivariate Analysis Logistic regression was performed a nd crude odds ratios, 95% confidence intervals, and p-values were generated for each outcome with exposure, and for each outcome with each covariate. For each outco me with the exposure, the crude association was also analyzed stratified by gender. Logistic regression was then performed for the
22 exposure with each covariate. Those variables having associations reaching a significance of p 0.10 with both the exposure and outcome were considered as potential confounders. Concomitant predictors of outco me also were identified. All odds ratios are reported for the odds of low head circ umference (exposed), poor outcome, increasing value (continuous covariates), being female (Gender), obese (BMI) and 4-positive (APOE). Multivariate Analysis and Modeling The main objective of the multivariate analysis was to identify the most parsimonious model that described the odds rati os for the eight outco mes as they related to head circumference. The approach taken was to illustrate the crude, standard-adjusted, fully-adjusted, and trimmed models. The standard-adjusted model contained those variables most consistently identified as potential confounders and used as model variables in the literature: age, gender and ye ars of education. The fully-adjusted model contained all confounding variables and conc omitant predictors identified for each outcome at the specified significance within this sample. The trimmed model, accepted as the final model for each outcome, was id entified by a manual backwards selection procedure for those HC-Outcome associati ons reaching significance at p<0.05 in the full model. In this procedure, variables were removed singly from the full model and the effect of their removal was ranked in terms of percent change in the point estimate of the exposure. The least change earned a value of one, the next largest change two, and so on. Variables were then removed sequentially from the full model in order of rank until a noticeable (10%) difference in the point estim ate from the full model was observed. The model analyzed just prior to the noted change in estimate was selected for analysis of
23 potential joint confounders, which were added if found, completing the final model. 66 In the instance that no statistically significan t association was observed between head circumference and the outcome, the fully adjusted model was accepted over the crude as the final model. Although no inference can be drawn from either model, the fully adjusted model was chosen as final over th e crude in these non-statistically-significant instances primarily so that the final model could reflect the best-fi tting model as indicated by the Likelihood Ratio statistic. All model summaries list odds ratios, 95% confidence intervals, and p-values for the models as calculated both with the tota l sample and stratified by gender. Final models were analyzed fo r fit using the Hosmer-Lem eshow Goodness of Fit test. 67 Interactions All final models were analyzed for in teraction using logi stic regression by reintroducing the main effect term for APOE where necessary and adding a term for the interaction between the expos ure variable and APOE status to each final model. Interactions were accepted as potentia lly inferential at a significance of p 0.1. A post hoc analysis of model fit was completed separately for the addition of APOE and the interaction term, and the data were also ex amined further for potential interaction by analyzing the two-by-two tables for outco me and HC stratified by APOE status.
Chapter 4 Results Descriptive Analysis A histogram (not shown) for each continuous variable of interest was generated in order to characterize their dist ributions in this sample. A ll continuous variables displayed a roughly normal distribution with little skew, implicating the mean and standard deviation as acceptable measures of central tendency and dispersion. Results of the univariate analysis of conti nuous variables are found in Table 1a and 1b. The frequencies and relative frequencies of ge nder and APOE status are found in Table 1c. There was a significant difference in the mean head circumference by gender (t-test p<0.0001), indicating the need to examine the range of risk stratified by gender. M22.7(0.7)73.5(6.2)14.6(3.2)4.4(1.5) F21.6(0.6)72.6(6.3)13.2(2.7)3.9(1.6) All22.2(0.8)73.0(6.2)13.9(3.0)4.2(1.6) M49.7(6.2)68.9(3.4)28.4(4.6) F49.3(5.9)62.9(2.8)26.9(5.8) All49.5(6.0)65.8(4.3)27.6(5.3) Mean (SD) Mean (SD) IQHeightBMI Table 1. Descriptive Measures fo r HC, Covariates and Outcomes Table 1a. Mean and Standard Devia tion for Continous HC and Covariates HC AgeEducationIncome 24
91.6(7.6)20.1(5.5)7.5(2.8)1.0(1.6) 9.5(2.0)28.1(9.6)74.8(57.6)2.1(2.7) TrailsImplicit Mean (SD) Mean (SD) Recog.Stroop Table 1b. Mean and Standard Deviation for Continous Outcomes 3MSImRDeRCuR 228(49)240(51)360(78)102(22) N (%) Table 1c. Frequency a nd Relative Frequency for Gender and APOE Status Gender APOE MaleFemale 4 Neg 4 Pos Bivariate Analysis After stratifying by gender and examining the frequencies across the range of HC for each stratum, it was determined that usi ng the interval in which fell the exact bottom tertile, quartile or quintile (s tandard levels of comparison in the literature) would not allow sufficient diversity between the dichot omized levels to ensure that the low classification was capturing risk. Therefore, the interval directly below that which contained the bottom 20 th percentile was chosen as the appropriate point at which to dichotomize the variable. As Table 2 illu strates, this created an exposed group containing 5.8 percent of the sample. Tabl e 2 also shows the relative frequency across levels for outcomes and BMI after dichotom izing in the interval of the bottom 20 th percentile for outcome and at or above 30 ( obese) for BMI. CuR and Implicit outcomes show that slightly less of the sample, 12.6 a nd 19.2 percent respectively, was classified as a poor outcome for those variables. 25
LHLHLHLHLH N 274411013679936910336559409 % 5.894.221.678.421.278.922.078.012.687.4 LHLHLHLHONO N 1283409837010136790378121347 % 27.472.720.979.121.678.419.280.825.974.2 L: Low H: High/Normal O: Obese NO: Non-obese Recog.StroopTrailsImplicitBMI Table 2. Frequencies a nd Relative Frequencies of Dichotomized Variables HC3MSImRDeRCuR Appendix A shows the results of all crude logistic analysis for the exposure with outcomes (Table A1), and for covariates with both exposure and outcome (Tables A2 and A3). In crude analysis, HC wa s significantly associated at p 0.001 with 3MS [OR(95%CI): 4.38 (1.99, 9.66)], an associ ation that persisted for both sexes when stratified by gender [Male OR: 3.09 (1.07, 8.96) ; Female OR: 6.58 (1.99, 21.8)]. HC did not approach statistical si gnificance for any other outcome. Larger HC also was statistically significantly associated with more years of education [OR for low HC with increasing number of years of education: 0.86 (0.76, 0.98); p=0.02], increasing height [OR: 0.91 (0.83, 1.00); p=0.05] and most si gnificantly increasing IQ [OR: 0.92 (0.87, 0.98); p=0.004]. There were also non-significan t associations (p=0.09) of HC with the possession of an APOE 4 allele [OR: 0.28 (0.07, 1. 21)], which may suggest that individuals with higher HC were more likely to possess an APOE-e4 allele in this sample, and with increasing age [OR: 1.06 (0.99, 1.13)], suggesting th at individuals with smaller HC tended to be somewhat older than those with larger HC. Confounders and concomitant predictors were identified for each outcome using Tables A2 and A3. Table A2 shows the confounders and their associations for each 26
27 outcome, associated with both outcome and HC at p 0.1. Concomitant predictors (T) and confounders (C) for each outcome are summarized in Table 3 in the Full Model (F). No confounders or predictors were identified for CuR. Multivariate Analysis The resulting odds ratios and corres ponding confidence intervals for poor outcome with exposure to low HC can be seen in Table 3 for the crude, standardadjusted, full and trimmed models. All fi nal models demonstrated good fit with the Hosmer-Lemeshow test for goodness of fit. Th ese results, as well as the results of the final model selection process, are presented below by outcome: 3MS (Table 3a) After adjustment for standard and full-model variables, the estimate of the OR was reduced slightly from 4.38 to 2.96, but remained significant (p=0.03). From the stratified results of the full model we see that while the OR for men remained consistently above 2, the OR for women was higher [4.35 (1.00, 18.9)], reaching significance at p=0.05. Mode l selection procedures (MSP) revealed education and income as variables with little impact on the exposure-outcome associa tion in the model. Removal of both resulted in a sl ightly higher odds ratio for al l three estimates, and greater significance in the total sample and male subsample (Model 5). The change in estimate for the total sample was less than 10% of the unstratified full model, the previously identified criterion for allowing removal of the variable, but cons iderably changed the estimate in males. Income was identified as the variable responsible for the large change, a predictor variable that s howed more potential as a confounder in men than women
(Appendix A, Table A3). Income was th erefore reintroduced to produce a more conservative final model (Model 4). The final estimate, adjusted for age, education and income, was nearly identical to the full m odel and had a slightly smaller confidence interval [OR: 2.97 (1.12, 7.89)]. Given that only education was removed during the MSP, assessment for joint confounding was not necessary for this outcome. NOR() p 1 4684.38(1.99,9.66) <0.0001 2 4623.37(1.40,8.11) 0.01 XXX 3F 4212.90(1.09,7.71) 0.03 CCTC 4* 4212.97(1.12,7.89) 0.03 CTC 5 4633.11(1.18,8.25) 0.02 CC I 417 CTCRIIQ Height APOE Age Education Gender IncomeTable 3. Model Selection for Main Effect of Head Circumference and APOE Interaction No interaction observed***Final Model F: Full Model X: Standard Adjustment Variable C: Confounder T: Concomitant I: Interaction R: Required **Term for interaction not statistically significant (p=0.99)APOE*HCOVERALL Model 95% CI Table 3a. Model Selection for HC with 3MS HC NOR() p OR( ) p 1 4683.09(1.07,8.96) 0.04 6.58(1.99,21.8) 0.002 2 4622.36(0.69,8.07) 0.17 5.25(1.40,19.6) 0.01 3F 4212.00(0.51,7.84) 0.39 4.29(1.00,18.4) 0.05 4* 4212.04(0.52,7.94) 0.31 4.51(1.03,19.7) 0.05 5 4632.46(0.64,9.01) 0.17 4.60(1.02,20.8) 0.05*Final Model F: Full Model HC ( Strati f ied b y Gender ) MF Model 95% CI 95% CI 28
ImR (Table 3b) The crude association between low HC and poor outcome showed an initially positive direction [OR (95%CI): 1.62 (0.69, 3.82); p=0.27; Model 1], but after both standard and full adjustment the OR move d toward the null and showed no significant association between exposure a nd poor outcome (Model 3). NOR() p 1 4681.62(0.69,3.82) 0.27 2 4621.05(0.40,2.76) 0.92 XXX 3*F 4191.09(0.37,2.94) 0.94 CCTTCC I 456 CCTTCCRI*Final Model F: Full Model X: Standard Adjustment Variable C: Confounder T: Concomitant I: Interaction R: Required **Term for interaction not statistically significant (p=0.50)OVERALL Model 95% CI No interaction observed** Table 3b. Model Selection for HC with ImR HCAge Education Gender Income IQ Height APOE APOE*HC NOR() p OR() p 1 4681.37(0.45,4.18) 0.58 1.90(0.49,7.39) 0.35 2 4620.83(0.22,3.06) 0.78 1.33(0.33,5.44) 0.69 3*F 4190.73(0.17,3.13) 0.67 1.51(0.33,6.88) 0.60*Final Model F: Full Model HC ( Strati f ied b y Gender ) MF Model 95% CI 95% CI DeR (Table 3c) Though the crude OR for delayed recall s howed a positive di rection initially, standard and full adjustment changed the direct ion of the association for the total sample (Models 2 and 3). No associa tions were found to be statistic ally significant, therefore no 29
selection procedures were performed. Th e male subsample showed a consistently positive direction in all adjusted models, but also highly insignificant (p=0.68). In women, the crude direction of association was negative and continued to drop and gain significance with adjustment, resulting in an OR (95%CI) = 0.37 (0.06, 2.19) at p=0.28 in the final model. NOR() p 1 4681.53(0.65,3.61) 0.33 2 4620.96(0.36,2.52) 0.93 XXX 3*F 4160.80(0.28,2.29) 0.67 CCTTCC I 416 CCTTCRI*Final Model F: Full Model X: Standard Adjustment Variable C: Confounder T: Concomitant I: Interaction R: Required **Term for interaction not statistically significant (p=0.64)OVERALL Model 95% CI No interaction observed** Table 3c. Model Selection for HC with DeR HCAge Education Gender Income IQ Height APOE APOE*HC NOR() p OR() p 1 4682.01(0.68,5.92) 0.20 0.89(0.19,4.19) 0.88 2 4621.32(0.39,4.54) 0.66 0.55(0.11,2.77) 0.47 3*F 4161.34(0.34,5.32) 0.68 0.37(0.06,2.19) 0.28*Final Model F: Full Model HC ( Strati f ied b y Gender ) MF Model 95% CI 95% CI CuR (Table 3d) Since there were no confounders identified for CuR, the full model used standard adjustment variables (Model 2) and resulted in no substa ntial change. Therefore the standard-adjusted model was chosen as th e final one (Model 1). In both models, no significant association was obser ved. The direction of asso ciation for the total sample 30
was positive (p=0.60), was positive in men [OR: 2.12 (0.53, 8.46); p=0.29], and inverse in women [OR: 0.61 (0.07, 4.97); p=0.64]. NOR() p 1 4681.22(0.41,3.66) 0.72 2* 4621.35(0.44,4.11) 0.60 XXX I 462 RI Table 3d. Model Selection for HC with CuR HCAge Education Gender Income IQ Height APOE APOE*HC*Final Model F: Full Model X: Standard Adjustment Variable C: Confounder T: Concomitant I: Interaction R: Required **Term for interaction not statistically significant (p=0.99)OVERALL Model 95% CI No interaction observed** NOR() p OR( ) p 1 4682.17(0.57,8.29) 0.26 0.54(0.07,4.30) 0.56 2* 4622.12(0.53,8.46) 0.29 0.61(0.07,4.97) 0.64*Final Model F: Full Model HC ( Strati f ied b y Gender ) MF Model 95% CI 95% CI Recognition (Table 3e) Crude and adjusted models agreed in the total sample and in both gender strata for Recognition, showing an inverse association that strengthened and gained significance with adjustment. However, no significance was observed in any model, therefore no model selection procedures were performed. In the final model, the OR (95%CI) for the total sample was 0.38 (0.12, 1.15) (p=0.09)(Mod el 3). In males, the point estimate for OR was 0.51 (0.13, 2.02), non-significant p=0. 34, and in females it dropped substantially to 0.18 (0.02, 1.58), also non-significant at p=0.12. 31
NOR() p 1 4680.75(0.30,1.90) 0.54 2 4620.45(0.16,1.30) 0.14 XXX 3*F 4190.38(0.12,1.15) 0.09 CCTTCC I 415 CCTTCCRI*Final Model F: Full Model X: Standard Adjustment Variable C: Confounder T: Concomitant I: Interaction R: Required **Term for interaction not statistically significant (p=0.98)OVERALL Model 95% CI No interaction observed** Table 3e. Model Selection for HC with Recognition HCAge Education Gender Income IQ Height APOE APOE*HC NOR() p OR( ) p 1 4680.98(0.32,2.97) 0.97 0.32(0.04,2.57) 0.29 2 4620.58(0.16,2.11) 0.41 0.25(0.03,2.01) 0.19 3*F 4190.51(0.13,2.02) 0.34 0.18(0.02,1.58) 0.12*Final Model F: Full Model HC ( Strati f ied b y Gender ) MF Model 95% CI 95% CI Stroop (Table 3f) The crude OR for low HC with poor outcome was initially positive at 1.64 (0.70, 3.87) (p=0.26) in the total sample (Model 1), but after adjustment fo r standard variables dropped to unity (p=0.98) (Model 2). After fu ll adjustment (Model 3), the point estimate dropped somewhat further, leaving an OR of 0.81 (0.29, 2.25) p=0.68. The male subsample exhibited positive associations in all models [final OR: 1.73 (0.43, 6.95); p=0.44], and the females exhibited inverse point estimates [final OR: 0.42 (0.08, 2.18); p=0.30]. No significant associations were observed. 32
NOR() p 1 4681.64(0.70,3.87) 0.26 2 4620.99(0.37,2.64) 0.98 XXX 3*F 4190.81(0.29,2.25) 0.68 CCTCC I 415 CCTCCRI*Final Model F: Full Model X: Standard Adjustment Variable C: Confounder T: Concomitant I: Interaction R: Required **Term for interaction not statistically significant (p=0.30)OVERALL Model 95% CI No interaction observed** Table 3f. Model Selection for HC with Stroop HCAge Education Gender Income IQ Height APOE APOE*HC NOR() p OR( ) p 1 4682.88(0.97,8.57) 0.06 0.71(0.15,3.36) 0.67 2 4622.06(0.56,7.59) 0.28 0.45(0.09,2.23) 0.33 3*F 4191.73(0.43,6.95) 0.44 0.42(0.08,2.18) 0.30*Final Model F: Full Model HC ( Strati f ied b y Gender ) MF Model 95% CI 95% CI Trails (Table 3g) This analysis also showed a reversed direction from a positive crude estimate toward an increasingly inve rse estimate throughout adjustment. The total sample estimate dropped from a crude OR (95%CI) of 1.58 (0.67, 3.71) (p=0.30) to an OR of 0.78 (0.28-2.19) (p=0.63) in the final model (M odel 3). The trend was the same in males and females, though the drop was more precipit ous in males throughout adjustment [final Male OR: 0.40 (0.06, 2.52); p=0.33] whereas the female estimates approached unity [final Female OR: 1.12 (0.30, 4.21); p=0.87] No associations were found to be significant, and no selection procedures were performed. 33
NOR() p 1 4681.58(0.67,3.71) 0.30 2 4620.91(0.34,2.47) 0.85 XXX 3*F 4200.78(0.28,2.19) 0.63 CCTC I 416 CCTCRI*Final Model F: Full Model X: Standard Adjustment Variable C: Confounder T: Concomitant I: Interaction R: Required **Term for interaction not statistically significant (p=0.47)OVERALL Model 95% CI No interaction observed** Table 3g. Model Selection for HC with Trails HCAge Education Gender Income IQ Height APOE APOE*HC NOR() p OR( ) p 1 4681.44(0.44,4.74) 0.55 1.78(0.52,6.15) 0.36 2 4620.62(0.12,3.32) 0.57 1.28(0.35,4.60) 0.71 3*F 4200.40(0.06,2.52) 0.33 1.12(0.30,4.21) 0.87*Final Model F: Full Model HC ( Strati f ied b y Gender ) MF Model 95% CI 95% CI Implicit (Table 3h) Only one potential confounder, IQ, was identified for this outcome. Though IQ has consistently remained a substantial c onfounder for most outcomes, it was found that its addition did not produce estimates that we re substantially diffe rent from the crude estimates. An additional model was run (not shown), including the standard adjustment variables and IQ, but the results were not substantially different from the standard adjusted model (Model 2). The variable of impact in that model and Model 2 was determined to be education. Since the model with education (Model 2) offered a noticeably more conservative estimate, and sinc e it is in line with standard adjustment practices, Model 2 was adopted as final. Neve rtheless, no associations were significant. 34
The direction of the estimate in all mode ls was positive for both the total and gender stratified samples. NOR() p 1 4681.51(0.62,3.69) 0.37 2* 4621.22(0.47,3.18) 0.69 XXX 3F 4641.43(0.58,3.55) 0.44 C I 458 XXX RI*Final Model F: Full Model X: Standard Adjustment Variable C: Confounder T: Concomitant I: Interaction R: Required **Term for interaction not statistically significant (p=0.99)OVERALL Model 95% CI No interaction observed** Table 3h. Model Selection for HC with Implicit HCAge Education Gender Income IQ Height APOE APOE*HC NOR() p OR( ) p 1 4681.79(0.54,5.93) 0.34 1.28(0.33,4.93) 0.72 2* 4621.30(0.33,5.09) 0.71 1.28(0.33,5.01) 0.73 3F 4641.65(0.48,5.66) 0.39 1.27(0.33,4.91) 0.73*Final Model F: Full Model HC ( Strati f ied b y Gender ) MF Model 95% CI 95% CI Interaction With APOE No significant interactions were f ound by adding an interaction term for APOE and HC (and APOE itself in models that lacked this term). P-values for the interaction term can be found in Model I in Tables 3a-h. All p-values were a bove p=0.3 and did not meet the criteria for potential inference. Post hoc analysis of model fit with the inclus ion of APOE revealed that model fit was significantly improved for ImR, CuR, Recognition, and Trails though APOE did not 35
36 reach significance as a predictor for either of these models [ImR OR: 1.71 (0.93, 3.13), p=0.08; CuR OR: 0.92 (0.47, 1.82) p=0.81; Re cogn. OR: 1.44 (0.83, 2.47), p=0.19; Trails OR: 0.92 (0.47, 1.82), p=0.39]. Neither term changed the measure of the association between HC and outcome. The interaction term was found to improve the fit of the model beyond that observed for APOE singly for Recognition, however the parameter estimate and standard error were prohibitively large ( =15.56, SE=692.2) and indicated against interpretation. Given the lack of significance or model improvement with the addition of APOE or the interaction term, a tw o-by-two analysis was completed for HC with each outcome, stratified by APOE. Examination of the tables showed that only two subjects possessed both low HC and at least one APOE 4 allele, making impossible any meaningful interpretation of effect modification by APOE status.
37 Chapter 5 Discussion For the majority of cognitive outcomes a ssessed in this study, head circumference was not associated with poor cognitive outco mes. Of eight cognitive outcomes, one clear result, consistent with the brain reserve hypothesis and existing literature, emerged along with one suggestive but not statistically significant result that contradicted the a priori hypothesis. The statistically signifi cant finding for the association between low head circumference and poor 3MS outcome is consis tent with the published literature. While the 3MS has not previously been directly studi ed in relation to head size, results for the MMSE from which the 3MS was created have consistently shown a significant positive association with increasing head ci rcumference in non-demented subjects 39, 40 and with increasing intracranial volume in cognitively impaired subjects. 29 The association between head size and outcome has also been observed for other measures of global cognition, such as a positive correlation between mean CASI scores, a measure very similar to the 3MS, and HC in a community sample of Japanese-Americans in the Kame Project, 25 for several measures of general and specific intelligence scores with HC, 40, 41 intracranial area, 42 and intracranial volume, 26, 43 and for a modified Clinical Dementia Rating with intracranial area in a sample of female dementia patients. 31 In all of the studies, these measures of association reached statistical significance.
38 It is worth noting that the association between head circumference and 3MS in the current study remained statistically signifi cant after adjustment for a measure of premorbid IQ. Measures of IQ provide an impor tant control in the investigation of brain reserve, since it is itself hypothesized to be a form of reserve against cognitive decline independent of brain size. 44 In our analysis, it was also one of the strongest confounders of association, not surprisingly given the we ll-known correlation between head size and measures of intelligence. 26, 40-43, 68 Nevertheless, after adjustment for IQ, head circumference remained a strong independent predictor of 3MS scores in this sample. The non-statistically signif icant trend of an inverse association of HC with recognition memory, which is contrary to our hypothesis, has no support from the existing literature. Though AD patients are know n to answer more pos itively in yes/no recognition trials, studies have shown that the discrimination scores on recognition subtests generally do not differ across groups with varying cognitive pathologies, 69 and that the verbal recognition component of the HVLT, similar to recognition components on other verbal tests, does not improve the se nsitivity and specificity of the HVLT scores in predicting later clinical decline into dementia. 70, 71 Given its poor discrimination ability, one would not expect any signi ficant deviation away from unity. In our sample, poor recognition outcome showed the expected associations with all confounders except height, in which ther e was a significant a nd slightly elevated association of poor outcome w ith increasing height that lik ely was a reflection of the association between height and HC. It wa s positively correlated with intelligence, and remained so after adjustment in the full model, suggesting its similarity to other outcomes in terms of potential confounders. Given this, it appears unlik ely that some unknown
39 confounder could have generated the finding. Nevertheless, the association indicated in this analysis merits further investigation in samples with perhaps a larger cognitivelyimpaired subpopulation to more accurately stud y the effects of low head circumference on poor outcome. The findings for HC with the remaini ng outcomes did not reach any inferential level of statistical significance in our sample. Upon close inspection of the literature, particularly for immediate and de layed verbal recall, we find th at this is not at odds with previously published investig ations. In several of the studies mentioned above with significant findings for measures of head si ze with global cognition and intelligence outcomes there has consisten tly been no association found between these measures and immediate or delayed recall tests sim ilar in nature to that on the HVLT, 40-43 despite the excellent early predictive va lue these tests have demons trated for cognitive decline. 70-74 To date, there has been no research to support an associati on or lack thereof between HC and cued recall, the gain from cues that is particularly useful in predicting AD, other than generalized associations of this outcome with clinical disease. 75, 76 Logically, one could argue that accurate discriminators for AD or amnestic mild cognitive impairment would be poor candidates for associativ e analysis with a measure, such as HC, that is hypothesized to be protective of clinical manifestation for many types of progressive pathological in sult, especially in a population that is likely demonstrating mixed pathologies through low cognitive outcome s. Therefore, it is unsurprising that no association was observed for cued recall in this commun ity sample with undefined pathologies.
40 Among the Stroop interference test, Trails difference score (see Methods), and the implicit memory test, only the Stroop test has been reported as having an association with head size. Tisserand et al. found an inverse association with between time to complete the interference task and head circum ference in least-s quares regression ( from .18 to .48, depending on adjustment for age a nd sex alone, or with SES, height or education). 40 In the current study, no significan t association was observed between HC and number of words completed in a set time period (a direct association would be expected based on the reserve hypothesis and lite rature). It is possi ble that the sample size, nearly twice as large in Tisserand et al. as compared to the current study, contributed to the difference in findings. By contrast, no findings have been repor ted for an association between HC and the Trails difference score or implicit memory, although Trails was among the cognitive tests performed by Wolf et al. in her co mparison of HC and cognition among normal, MCI and demented subjects. 27 As a measure of executive function, 62, 77 poor outcome on the difference score is hypothesized to be associated with lower HC under the reserve hypothesis. However, it is agai n possible that magnitude of th e measure of association is too small to detect among a population of this size. It is worth discussing in mo re detail the difference in findings for an association between HC and the outcomes 3MS, Stroop and Trails, given that all three tests are measures of executive function and that thei r scores strongly correlate with each other (Pearson correlation coefficien ts in this sample range be tween 0.40 and 0.46, p<0.0001, data not shown). The most lik ely explanation given the data in this sample is that the tests, measuring executive function by different means, demonstrate disparate true
41 associations with HC. Crude analysis of the association between HC and these outcomes, given in Appendix A (Table A1), shows the init ial difference in the effect sizes for HC on 3MS and both Stroop and Trails, where for 3MS the OR>4.0 and for the latter two crude OR 1.6. In a retrospective power analysis under the assumption of an unmatched case control study with the given pr oportion of cases to controls and proportion of exposure to low HC among the controls, it was found that the power to detect an association at =0.05 in this sample would fall dramatically for the three outcomes as the hypothetical association drops below an OR of about 3. Gi ven this and the initial differences in effect size, it seems entirely possible that the a ssociation between HC and both Stroop and Trails scores is low to moderate in this sample, and that there is likely insufficient power to detect a statistica lly significantly elevated risk of poor outcome. The eighth outcome in this analysis, implicit memory scores from a word-stem completion task, also failed to demonstrate a statistically significant association with dichotomous HC. While there is no known literature on this association, existing literature on the association be tween Alzheimers disease cas e status and performance on word-stem completion tasks has generated mixed results. 78-81 A meta-analysis attempting to overcome the limitations of smaller studies suggested a weak but statistically significant correlation between case stat us and word-stem completion scores. 82 However, the authors cautioned that null findings should be interpreted carefully given the apparent need for large sample sizes and adequate power to detect the small association. Therefore, while the current analysis detect ed no statistically significant association between head size and cognitive outcome, th e conclusion that there is no biological impact of head size on implicit memory scor es can likely not be drawn from our data.
42 Apart from outcome-specific interpretations of these results, there are several points to consider in the analysis and interp retation of the overall findings. First, the sample in the Charlotte County Healthy Ag ing Study was drawn from a population in a popular retirement location in Florida, subject to substantial migration. It is likely that those migrating to the region might on averag e have higher income and better education, which would correlate with hi gher IQ scores, than those wh o do not. This self-selection may explain the relatively high mean years of education and level of income among our sample. These characteristics impart two possible implications for interpretation of the results. First, higher educa tion is known to be inversely associated with poor cognitive performance and dementia, 83, 84 indicating that poor cogniti on may be less prevalent in this community sample than in other populati ons studied in North America or Europe. Since poor cognition was defined relatively (as a quantile) for analysis, this may have contributed to fewer findings among the eight outcomes. Second, if the relative classification of poor cognition in this highly ed ucated sample led to misclassification of truly poor cognition, it would indi cate that the statistically significant finding that small head size predicted low 3MS scores might underestimate the true association with the same bottom quantile of scores among the general population. Another point of consideration is the use of multiple comparisons in this analysis. No corrections for multiple comparisons were used, which generally raises the possibility of finding a statistically significan t association by chance. This is of less concern in this analysis given that the eight endpoints we re dependent (the correlation between them, data not shown, was observed to be between 0.3 and 0.7, excepting cued recall and implicit memory scores). Nevertheless, sinc e correction accounting for these correlations
43 may still place the required per-comparison allowable error below the observed p-value of 0.03 for a familywise error rate of =0.05, 85 the possibility should be considered that a Type I error may have occurred. There are tw o reasons why it is un likely that the single significant finding in this study is the result of Type I erro r. First, the finding is substantially supported by similar results in previously published studies. Second, the observations, consistent in male and female subjects, are supported by a biologically plausible hypothesis. Therefore, it is conc luded that in this sample the statistically significant association identifie d between head size and 3MS scores is likely valid. Finally, it was found that the in teractive effects of APOE 4 status could not be examined in this sample due to a very sm all number of subjects who had both low HC and possessed one or more APOE 4 alleles (n=2). Crude anal ysis indicated that APOE 4-positive subjects performed less well on cognitive measures. Post hoc analyses of model fit after failure to detect interaction also showed that inclusion of APOE statistically significantly improved the fit of the final models with immediate recall, recognition and trails. The estimates of effect were generally too unstable to infer much from this information, although it was suggested that possession of an 4 allele independently increased risk of poor out come in the Trails test (p=0.08). In examining effect modification by APOE 4 status, this analysis attempted to replicate the results found in the Kame Project, in which it was seen that the effect of head circumference on incident AD was restricted to those with one or more 4 alleles. 32, 33 Had the data permitted, it was hypothesized that possession of an 4 allele may have modified the association between HC and poor outcome such that the effect would have been stronger in 4-positive subjects. Given the si gnificance of the findings from the
44 Kame Project, however, it is important that other large and welldesigned studies with sufficient exposed individuals analyze these associations further so that risk of impairment can be accurately described with respect to head size and genetic risk of AD.
45 References 1. Ebly EM, Parhad IM, Hogan DB, Fung TS. Prevalence and types of dementia in the very old: Results from the Cana dian study of health and aging. Neurology. 1994;44:15931600. 2. Graves AB, Larson EB, Edland SD, et al. Prevalence of dementia and its subtypes in the Japanese American population of king county, Washington state. The Kame Project. Am J Epidemiol 1996;144:760-771. 3. White L, Petrovitch H, Ross GW, et al. Prevalence of dementia in older Japanese American men in Hawaii: The Honolulu-Asia Aging Study. JAMA 1996;276:955-960. 4. Fillenbaum GG, Heyman A, Huber MS, et al. The prevalence and 3-year incidence of dementia in older black and white community residents. J Clin Epidemiol. 1998;51:587595. 5. Hendrie HC. Epidemiology of dementia and Alzheimers disease. Am J Geriatr Psychiatry. 1998;6:S3-18. 6. Gauthier S, Reisberg B, Zaudig M, et al. Mild cognitive impairment. Lancet 2006;367:1262-1270.
46 7. Ritchie K. Mild cognitive impairment: An epidemiological perspective. Dialogues Clin Neurosci 2004;6:401-408. 8. Katzman R, Terry R, DeTeresa R, et al. Clinical, pathological, and neurochemical changes in dementia: A subgroup with preserve d mental status and numerous neocortical plaques. Ann Neurol 1988;23:138-144. 9. Snowdon DA, Greiner LH, Mortimer JA, Riley KP, Greiner PA, Markesbery WR. Brain infarction and the clin ical expression of Alzheime r disease. The Nun Study. JAMA 1997;277:813-817. 10. Gosche KM, Mortimer JA, Smith CD Markesbery WR, Snowdon DA. Hippocampal volume as an index of Alzheimer neuropathology: Findings from the Nun Study. Neurology. 2002;58:1476-1482. 11. Neuropathology Group. Medical Research Council Cognitive Function and Aging Study. Pathological correlates of late-onset dementia in a multicentre, community-based population in England and Wales. neuropathol ogy group of the medical research council cognitive function and ageing study (MRC CFAS). Lancet 2001;357:169-175. 12. Riley KP, Snowdon DA, Markesbery WR. Alzheimer's neurofibrillary pathology and the spectrum of cognitive function: Findings from the nun study. Ann Neurol 2002;51:567-577. 13. Satz P. Brain reserve capacity on sympto m onset after brain inju ry: A formulation and review of evidence for threshold theory. Neuropsychology. 1993;7:273-295.
47 14. Mortimer JA. The continuum hypothesis of Alzheimers disease and normal aging: The role of brain reserve. Alzheimer's Research 1995;1:67-70. 15. Mortimer JA. Brain reserve and the clin ical expression of Alzheimers disease. Geriatrics 1997;52 Suppl 2:S50-3. 16. Borenstein Graves A. Alzheimer's dis ease and vascular dementia. In: Nelson LM, Tanner CM, Van Den Eeden, S.K., McGuire VM, eds. Neuroepidemiology: From Principles to Practice. New York: Oxford University Press; 2004:102-130. 17. Borenstein AR, Copenhaver CI, Mortimer JA. Early-life risk factors for Alzheimer disease. Alzheimer Dis Assoc Disord 2006;20:63-72. 18. Dekaban AS. Changes in brain weights du ring the span of human life: Relation of brain weights to body heights and body weights. Ann Neurol 1978;4:345-356. 19. Sgouros S, Goldin JH, Hockley AD, Wake MJ, Natarajan K. Intracranial volume change in childhood. J Neurosurg 1999;91:610-616. 20. Dobbing J. Infant nutrition and later achievement. Am J Clin Nutr 1985;41:477-484. 21. Schrott LM. Effect of training and environment on br ain morphology and behavior. Acta Paediatr Suppl 1997;422:45-47. 22. Miller AK, Corsellis JA. Evidence for a secular increase in human brain weight during the past century. Ann Hum Biol 1977;4:253-257.
48 23. Pennington BF, Filipek PA, Lefly D, et al A twin MRI study of size variations in human brain. J Cogn Neurosci 2000;12:223-232. 24. Ermakov S, Kobyliansky E, Livshits G. Quantitative genetic study of head size related phenotypes in ethnically homogeneous Chuvasha pedigrees. Ann Hum Biol 2005;32:585-598. 25. Graves AB, Mortimer JA, Larson EB Wenzlow A, Bowen JD, McCormick WC. Head circumference as a measure of cogniti ve reserve. associa tion with severity of impairment in Alzheimers disease. Br J Psychiatry 1996;169:86-92. 26. Mori E, Hirono N, Yamashita H, et al Premorbid brain size as a determinant of reserve capacity against intellectual decline in Alzheimers disease. Am J Psychiatry 1997;154:18-24. 27. Wolf H, Julin P, Gertz HJ, Winblad B, Wahlund LO. Intracranial volume in mild cognitive impairment, Alzheimers disease and vascular dementia: Evidence for brain reserve? Int J Geriatr Psychiatry 2004;19:995-1007. 28. Schofield PW, Logroscino G, Andrews HF Albert S, Stern Y. An association between head circumference and Alzheimers disease in a population-based study of aging and dementia. Neurology. 1997;49:30-37. 29. Wolf H, Kruggel F, Hensel A, Wahlund LO, Arendt T, Gertz HJ. The relationship between head size and intracranial volume in elderly subjects. Brain Res. 2003;973:7480.
49 30. Mortimer JA, Snowdon DA, Markesbery WR. Head circumference, education and risk of dementia: Findings from the nun study. J Clin Exp Neuropsychol 2003;25:671679. 31. Schofield PW, Mosesson RE, Stern Y, Mayeux R. The age at onset of Alzheimers disease and an intracranial area measurement. A relationship. Arch Neurol 1995;52:9598. 32. Borenstein Graves A, Mortimer JA, Bowen JD, et al. Head circumference and incident Alzheimers disease: Modification by ap olipoprotein E. Neurology. 2001;57:1453-1460. 33. Borenstein AR, Wu Y, Mortimer JA, et al. Developmental and vascular risk factors for Alzheimers disease. Neurobiol Aging 2005;26:325-334. 34. Jenkins R, Fox NC, Rossor AM, Harvey RJ, Rossor MN. Intracranial volume and Alzheimer disease: Evidence against the cerebral reserve hypothesis. Arch Neurol 2000;57:220-224. 35. Edland SD, Xu Y, Plevak M, et al. Tota l intracranial volume: Normative values and lack of association with Alzheimers disease. Neurology 2002;59:272-274. 36. Jack CR,Jr, Petersen RC, Xu YC, et al Medial temporal atrophy on MRI in normal aging and very mild Alzheimers disease. Neurology 1997;49:786-794.
50 37. Gur RC, Turetsky BI, Matsui M, et al. Sex differences in brain gray and white matter in healthy young adults: Correlatio ns with cognitive performance. J Neurosci. 1999;19:4065-4072. 38. Espinosa PS, Kryscio RJ, Mendiondo MS, et al. Alzheimer's disease and head circumference. J Alzheimers Dis 2006;9:77-80. 39. Reynolds MD, Johnston JM, Dodge HH, DeKosky ST, Ganguli M. Small head size is related to low mini-mental state examination scores in a community sample of nondemented older adults. Neurology 1999;53:228-229. 40. Tisserand DJ, Bosma H, Van Boxtel MP, Jolles J. Head size and cognitive ability in nondemented older adults are related. Neurology. 2001;56:969-971. 41. Gale CR, Walton S, Martyn CN. Foetal and postnatal head growth and risk of cognitive decline in old age. Brain 2003;126:2273-2278. 42. MacLullich AM, Ferguson KJ, Deary IJ, Seckl JR, Starr JM, Wardlaw JM. Intracranial capacity and brain volumes are a ssociated with cogniti on in healthy elderly men. Neurology. 2002;59:169-174. 43. Staff RT, Murray AD, Deary IJ, Whalle y LJ. What provides cerebral reserve? Brain. 2004;127:1191-1199. 44. Stern Y. The concept of cognitive reserve: A catalyst for research. J Clin Exp Neuropsychol 2003;25:589-593.
51 45. Katzman R. Education and the prevalence of dementia and Alzheimers disease. Neurology. 1993;43:13-20. 46. Plassman BL, Welsh KA, Helms M, Brandt J, Page WF, Breitner JC. Intelligence and education as predictors of cognitive state in late life: A 50-year follow-up. Neurology. 1995;45:1446-1450. 47. Stern Y. Cognitive reserve and Alzheimer disease. Alzheimer Dis Assoc Disord 2006;20:112-117. 48. Alexander GE, Furey ML, Grady CL, et al. Association of pr emorbid intellectual function with cerebral metabolism in Alzheime rs disease: Implicat ions for the cognitive reserve hypothesis. Am J Psychiatry 1997;154:165-172. 49. Fritsch T, Smyth KA, McClendon MJ, et al. Associations between dementia/mild cognitive impairment and cognitive perf ormance and activity levels in youth. J Am Geriatr Soc 2005;53:1191-1196. 50. Abbott RD, White LR, Ross GW, et al. Hei ght as a marker of childhood development and late-life cognitive function: The Honolulu-Asia aging study. Pediatrics 1998;102:602-609. 51. Beeri MS, Davidson M, Silverman JM, Noy S, Schmeidler J, Goldbourt U. Relationship between body height and dementia. Am J Geriatr Psychiatry 2005;13:116123.
52 52. Jeong SK, Kim JM, Kweon SS, Shin MH Seo MW, Kim YH. Does arm length indicate cognitive and functional reserve? Int J Geriatr Psychiatry. 2005;20:406-412. 53. Kim JM, Stewart R, Shin IS, Yoon JS. Limb length and dementia in an older Korean population. J Neurol Neurosurg Psychiatry 2003;74:427-432. 54. Yuspeh RL, Vanderploeg RD. Spot-the-w ord: A measure for estimating premorbid intellectual functioning. Arch Clin Neuropsychol 2000;15:319-326. 55. Decarli C. Vascular factor s in dementia: An overview. J Neurol Sci 2004;226:19-23. 56. Luchsinger JA, Tang MX, Shea S, Mayeux R. Hyperinsulinemia and risk of Alzheimer disease. Neurology 2004;63:1187-1192. 57. Small BJ, Graves AB, McEvoy CL, Crawford FC, Mullan M, Mortimer JA. Is APOE--epsilon4 a risk factor for cognitive impairment in normal aging? Neurology. 2000;54:2082-2088. 58. Borenstein AR, Mortimer JA, Wu Y, et al. Apolipoprotein E and cognition in community-based samples of African Americans and Caucasians. Ethn Dis 2006;16:915. 59. McDowell I, Kristjansson B, Hill GB, Hebe rt R. Community screening for dementia: The mini mental state exam (MMSE) and modified mini-mental state exam (3MS) compared. J Clin Epidemiol 1997;50:377-383. 60. Shapiro AM, Benedict RH, Schretlen D, Br andt J. Construct and concurrent validity of the Hopkins Verbal Learning Test-Revised. Clin Neuropsychol 1999;13:348-358.
53 61. Spieler DH, Balota DA, Faust ME. Str oop performance in healthy younger and older adults and in individu als with dementia of the Alzheimers type. J Exp Psychol Hum Percept Perform 1996;22:461-479. 62. Corrigan JD, Hinkeldey NS. Relationshi ps between parts A and B of the Trail Making test. J Clin Psychol. 1987;43:402-409. 63. Nyberg L, Winocur G, Moscovitch M. Corre lation between frontal lobe functions and explicit and implicit stem comp letion in healthy elderly. Neuropsychology 1997;11:7076. 64. Blair RC, Taylor RA. Biostatistics for the Health Sciences. Draft. Biostatistics I Course Text, Fall 2004. ed. University of S outh Florida: Department of Epidemiology and Biostatistics; 1999. 65. Ogden CL, Carroll MD, Curtin LR, McDowell MA, Tabak CJ, Flegal KM. Prevalence of overweight and obesity in the united states, 1999-2004. JAMA 2006;295:1549-1555. 66. Schwartz S. Coursepack for advan ced research methods (epidemiology 7703). University of South Florida:2006;1. 67. Agresti A. An Introduction to Categorical Data Analysis. New York: John Wiley & Sons, Inc.; 1996. 68. Andreasen NC, Flaum M, Swayze V,2nd, et al. Intelligence and brain structure in normal individuals. Am J Psychiatry 1993;150:130-134.
54 69. Lacritz LH, Cullum CM. The Hopkins Verbal Learning Test and CVLT: A preliminary comparison. Arch Clin Neuropsychol 1998;13:623-628. 70. Backman L, Jones S, Berger AK, Laukka EJ, Small BJ. Multiple cognitive deficits during the transition to Alzheimers disease. J Intern Med 2004;256:195-204. 71. Kuslansky G, Katz M, Verghese J, et al Detecting dementia with the Hopkins Verbal Learning Test and the Mini-Mental State Examination. Arch Clin Neuropsychol 2004;19:89-104. 72. Ivanoiu A, Adam S, Van der Linden M, et al. Memory evaluation with a new cued recall test in patients with mild cognitiv e impairment and Alzheimers disease. J Neurol 2005;252:47-55. 73. Tierney MC, Yao C, Kiss A, McDowe ll I. Neuropsychological tests accurately predict incident Alzheimer disease after 5 and 10 years. Neurology 2005;64:1853-1859. 74. Schrijnemaekers AM, de Jager CA, Hoge rvorst E, Budge MM. Cases with mild cognitive impairment and Alzheimers disease fail to benefit from repeated exposure to episodic memory tests as compared with controls. J Clin Exp Neuropsychol 2006;28:438-455. 75. Traykov L, Baudic S, Raoux N, et al. Patterns of memory impairment and perseverative behavior discriminate early Alzheimers disease from subcortical vascular dementia. J Neurol Sci 2005;229-230:75-79.
55 76. Yuspeh RL, Vanderploeg RD, Kershaw DA. Validity of a semantically cued recall procedure for the Mini-Mental State Examination. Neuropsychiatry Neuropsychol Behav Neurol 1998;11:207-211. 77. Hashimoto R, Meguro K, Lee E, Kasai M, Ishii H, Yamaguchi S. Effect of age and education on the Trail Making test and dete rmination of normative data for Japanese elderly people: The Tajiri Project. Psychiatry Clin Neurosci. 2006;60:422-428. 78. Camus JF, Nicolas S, Wenisch E, Morrone I, Blanchard F, Bakchine S. Implicit memory for words presented in short text s is preserved in Alz heimers disease. Psychol Med 2003;33:169-174. 79. Backman L, Almkvist O, Nyberg L, Ande rsson J. Functional changes in brain activity during priming in Alzheimers disease. J Cogn Neurosci 2000;12:134-141. 80. Kuzis G, Sabe L, Tiberti C, Merello M, Leiguarda R, Starkstein SE. Explicit and implicit learning in patients with Alzheimer di sease and Parkinson disease with dementia. Neuropsychiatry Neuropsychol Behav Neurol 1999;12:265-269. 81. Randolph C, Tierney MC, Chase TN. Imp licit memory in Alzheimers disease. J Clin Exp Neuropsychol 1995;17:343-351. 82. Meiran N, Jelicic M. Implicit memory in Alzheimers disease: A meta-analysis. Neuropsychology. 1995;9:291-303. 83. Mortimer JA, Graves AB. Education and other socioeconomic determinants of dementia and Alzheimers disease. Neurology 1993;43:S39-S44.
56 84. Stern Y, Gurland B, Tatemichi TK, Ta ng MX, Wilder D, Mayeux R. Influence of education and occupation on the incidence of Alzheimers disease. JAMA 1994;271:1004-1010. 85. Pocock SJ, Geller NL, Tsiatis AA. The analysis of multiple endpoints in clinical trials. Biometrics 1987;43:487-498.
57 Appendix A: Additional Tables
OR( ) p OR( ) p OR( ) p 4.38 (1.99,9.66) <0.001 3.09 (1.07,8.96) 0.04 6.58 (1.99,21.8) 0.002 1.62(0.69,3.82) 0.27 1.37(0.45,4.18) 0.58 1.90(0.49,7.4) 0.35 1.54(0.65,3.61) 0.33 2.01(0.68,5.92) 0.21 0.89(0.19,4.2) 0.88 1.22(0.41,3.66) 0.72 2.17(0.57,8.29) 0.26 0.54(0.07,4.3) 0.56 0.75(0.30,1.90) 0.54 0.98(0.32,2.97) 0.97 0.32(0.04,2.6) 0.29 1.64(0.70,3.87) 0.26 2.88 (0.97,8.57) 0.06 0.71(0.15,3.4) 0.67 1.58(0.67,3.17) 0.30 1.44(0.44,4.74) 0.55 1.78(0.52,6.2) 0.36 1.51(0.62,3.69) 0.37 1.79(0.54,5.93) 0.34 1.28(0.33,4.9) 0.72Potential confounders assoc. with outcome and total-sample exposure at p<0.1Discrim. Stroop Trails Implicit*Continous covariates. Crude OR is for probability of event (Low HC; Poor Outcom e) with each increasing unit of the covariate: Age and Educ in yrs; Income in levels 1(<10k)-9(>150k); IQ in points; Height in inches. **Categorical covariates. Crude OR is for probability of event with: Gender (female); BMI (obese); APOE ( 4+).3MS ImR DeR CuR Table A. Unadjusted Logistic Regression of Head Circumference, Outcomes and Covariates Table A1. Dichotomized Outcomes Modeled W ith Dichotomous Head Circumference (Crude) Dichotomized Outcome HC OVERALL N=468 Male N=228 Female N=240 95% CI 95% CI 95% CI 58
OR( ) p OR( ) p OR( ) p 1.11 (1.07,1.16) <0.001 0.79 (0.73,0.87) <0.001 0.79(0.51,1.22) 0.28 1.09 (1.05,1.13) <0.001 0.88 (0.82,0.95) 0.001 0.49 (0.31,0.77) 0.002 1.09 (1.05,1.13) <0.001 0.85 (0.78,0.92) <0.001 0.64 (0.41,1.00) 0.05 0.99(0.95,1.04) 0.75 0.98(0.89,1.07) 0.62 1.34(0.77,2.33) 0.30 1.06 (1.02,1.09) 0.0012 0.91 (0.85,0.97) 0.0067 0.53 (0.35,0.80) 0.0026 1.11 (1.06,1.15) <0.001 0.82 (0.75,0.89) <0.001 1.09(0.70,1.71) 0.69 1.11 (1.07,1.16) <0.001 0.80 (0.73,0.87) <0.001 1.12(0.72,1.74) 0.62 1.02(0.99,1.06) 0.24 0.95(0.88,1.02) 0.16 1.24(0.78,1.96) 0.37 OR( ) p OR( ) p OR( ) p 0.69 (0.58,0.83) <0.001 0.84 (0.81,0.88) <0.001 0.96(0.91,1.01) 0.13 0.76 (0.64,0.91) 0.002 0.91 (0.88,0.95) <0.001 1.06 (1.00,1.11) 0.05 0.72 (0.60,0.85) <0.001 0.89 (0.86,0.92) <0.001 1.01(0.96,1.06) 0.80 1.01(0.84,1.21) 0.94 0.97(0.93,1.02) 0.25 0.99(0.93,1.06) 0.78 0.89 (0.77,1.02) 0.10 0.92 (0.89,0.95) <0.001 1.05 (1.00,1.10) 0.05 0.72 (0.60,0.85) <0.001 0.92 (0.89,0.95) <0.001 0.94 (0.89,0.99) 0.03 0.65 (0.55,0.78) <0.001 0.90 (0.87,0.94) <0.001 0.98(0.93,1.03) 0.36 0.88(0.75,1.04) 0.13 0.97 (0.93,1.01) 0.10 0.97(0.92,1.02) 0.22 *TABLE A2 CONTINED ON THE NEXT PAGE**Continous covariates. Crude OR is for probability of event (Low HC; Poor Outcome) with each increasing unit of the covariate: Age and Educ in yrs; Income in levels 1(<10k)-9(>150k); IQ in points; Height in inches. ***Categorical covariates. Crude OR is for probab ility of event with: Gender (female); BMI (obese); APOE ( 4+).Potential confounders assoc. with outc ome and total-sample exposure at p<0.1Discrim. Stroop Trails Implicit 3MS ImR DeR CuR Dichotomized Outcome Income** N=424 IQ** N=464 Height** N=465 95% CI 95% CI 95% CI Discrim. Stroop Trails Implicit 3MS ImR DeR CuR Table A2*. Dichotomized Outcomes Modeled With Covariates (Crude) Dichotomized Outcome Age** N=467Education** N=463Gender*** N=46895% CI 95% CI 95% CI 59
OR( ) p OR( ) p 0.81(0.48,1.36) 0.42 1.15(0.67,1.95) 0.62 0.67(0.39,1.15) 0.15 1.30(0.77,2.18) 0.32 1.02(0.62,1.69) 0.92 1.62 (0.98,2.67) 0.06 0.97(0.52,1.82) 0.94 0.91(0.46,1.79) 0.79 1.11(0.70,1.76) 0.65 1.39(0.86,2.25) 0.17 0.79(0.47,1.34) 0.39 0.94(0.54,1.64) 0.83 0.93(0.56,1.54) 0.78 1.65(1.00,2.72) 0.52 1.21(0.73,2.02) 0.46 1.13(0.66,1.96) 0.65*Continous covariates. Crude OR is for probability of event (Low HC; Poor Outcome) with each increasing unit of the covariate: Age and Educ in yrs; Income in levels 1(<10k)-9(>150k); IQ in points; Height in inches. **Categorical covariates. Crude OR is for probability of event with: Gender (female); BMI (obese); APOE ( 4+). Potential confounders assoc. with outcome and total-sample exposure at p<0.1Table A2. Dichotomized Outcomes Modele d With Covariates (Crude) Continued Discrim. Stroop Trails Implicit 3MS ImR DeR CuR Dichotomized Outcome BMI** N=468 APOE** N=468 95% CI 95% CI 60
OR( ) p OR( ) p OR( ) p 1.06 (0.99,1.13) 0.09 1.01(0.93,1.10) 0.85 1.13(1.01,1.26) 0.03 0.86 (0.76,0.98) 0.02 0.83(0.71,0.98) 0.02 0.90(0.73,1.10) 0.30 0.75(0.34,1.63) 0.47 0.92(0.70,1.20) 0.52 0.79(0.53,1.18) 0.25 1.03(0.71,1.49) 0.88 0.92 (0.87,0.98) 0.004 0.91(0.84,0.98) 0.01 0.94(0.86,1.02) 0.14 0.91 (0.83,1.00) 0.05 0.83(0.73,0.94) 0.0047 0.83(0.67,1.03) 0.09 0.67(0.24,1.72) 0.37 0.87(0.27,2.82) 0.8111 0.30(0.04,2.38) 0.25 0.28 (0.07,1.21) 0.09 0.32(0.04,2.54) 0.28 0.25(0.03,2.00) 0.19Potential confounders assoc. with outc ome and total-sample exposure at p<0.1BMI** APOE***Continous covariates. Crude OR is for probability of event (Low HC; Poor Outcome) with each increasing unit of the covariate: Age and Educ in yrs; Income in levels 1(<10k)-9(>150k); IQ in points; Height in inches. **Categorical covariates. Crude OR is for probability of event with: Gender (female); BMI (obese); APOE ( 4+). --Income* IQ* Height* Age* Education* Gender** --Table A3 Dichotomized H ead Circumference Modeled With Covariates (Crude) Continous Covariates HC OVERALL M F 95% CI 95% CI 95% CI 61