A comparison of confidence distribution approaches for rare event meta-analysis.

Meta-analysis of rare event data has recently received increasing attention due to the challenging issues rare events pose to traditional meta-analytic methods. One specific way to combine information and analyze rare event meta-analysis data utilizes confidence distributions (CDs). While several CD methods exist, no comparisons have been made to determine which method is best suited for homogeneous or heterogeneous meta-analyses with rare events. In this article, we review several CD methods: Fisher's classic P-value combination method, one that combines P-value functions, another that combines confidence intervals, and one that combines confidence log-likelihood functions. We compare these CD approaches, and we propose and compare variations of these methods to determine which method produces reliable results for homogeneous or heterogeneous rare event meta-analyses. We find that for homogeneous rare event data, most CD methods perform very well. On the other hand, for heterogeneous rare event data, there is a clear split in performance between some CD methods, with some performing very poorly and others performing reasonably well.

A permutation-based approach for heterogeneous meta-analyses of rare events.

The increasingly widespread use of meta-analysis has led to growing interest in meta-analytic methods for rare events and sparse data. Conventional approaches tend to perform very poorly in such settings. Recent work in this area has provided options for sparse data, but these are still often hampered when heterogeneity across the available studies differs based on treatment group. We propose a permutation-based approach based on conditional logistic regression that accommodates this common contingency, providing more reliable statistical tests when such patterns of heterogeneity are observed. We find that commonly used methods can yield highly inflated Type I error rates, low confidence interval coverage, and bias when events are rare and non-negligible heterogeneity is present. Our method often produces much lower Type I error rates and higher confidence interval coverage than traditional methods in these circumstances. We illustrate the utility of our method by comparing it to several other methods via a simulation study and analyzing an example data set, which assess the use of antibiotics to prevent acute rheumatic fever.

Relative risk for Alzheimer disease based on complete family history.

OBJECTIVE: The inherited component for Alzheimer disease (AD) risk has focused on close relatives; consideration of the full family history may improve accuracy and utility of risk estimates. METHODS: A population resource including a genealogy of Utah pioneers from the 1800s linked to Utah death certificates was used to estimate relative risk for AD based on specific family history constellations, including from first- to third-degree relatives. RESULTS: Any affected first-degree relatives (FDR) significantly increased risk of AD (≥1 FDRs: relative risk [RR] 1.73, 95% confidence interval [CI] [1.59-1.87]; ≥2 FDRs: RR 3.98 [3.26-4.82]; ≥3 FDRs: RR 2.48 [1.07-4.89]; ≥4 FDRs: RR 14.77 [5.42-32.15]). Affected second-degree relatives (SDR) increased risk even in the presence of affected FDRs (FDR = 1 with SDR = 2: RR 21.29 [5.80-54.52]). AD only in third-degree relatives (TDR) also increased risk (FDR = 0, SDR = 0, TDR ≥3: RR 1.43 [1.21-1.68]). Mixed evidence was observed for differences in risk based on maternal compared to paternal inheritance; higher risks for men than women with equivalent family history, and higher risk for individuals with at least one affected FDR regardless of the relative's age at death, were observed. CONCLUSIONS: This population-based estimation of RRs for AD based on family history ascertained from extended genealogy data indicates that inherited genetic factors have a broad influence, extending beyond immediate relatives. Providers should consider the full constellation of family history when counseling patients and families about their risk of AD.

Nutritional Status is Associated With Severe Dementia and Mortality: The Cache County Dementia Progression Study.

PURPOSE: Studies have reported faster cognitive/functional decline in persons with dementia (PWD) with malnutrition. We investigated whether baseline nutritional status predicted severe dementia and mortality in a population-based sample. PATIENTS: A maximum of 300 PWD were assessed annually for up to 8.6 years. METHODS: Nutritional status was assessed using a modified Mini-Nutritional Assessment (mMNA). Severe dementia was defined as: "severe" rating on the Clinical Dementia Rating or Mini-Mental State Examination score ≤10. Using Cox proportional hazards models, we examined the association between baseline mMNA score (or its subcomponents) with each outcome. Covariates included demographics; dementia onset age, type, and duration; APOE genotype; and residency with caregiver. RESULTS: Compared with "well-nourished," "malnourished" PWD had 3-4 times the hazard of severe dementia [hazard ratio (HR), 4.31; P=0.014] and death (HR, 3.04; P<0.001). Those "at risk for malnutrition" had twice the hazard of severe dementia (HR, 1.98; P=0.064) and 1.5 times the hazard of death (HR, 1.46; P=0.015). mMNA subcomponents of food group intake, weight loss, body mass index, mobility, health status, protein consumption, and mid-arm circumference predicted one or both outcomes. CONCLUSIONS: Nutritional status is an important predictor of clinical outcomes in dementia and may provide an avenue for intervention.

Genome-wide association study for variants that modulate relationships between cerebrospinal fluid amyloid-beta 42, tau, and p-tau levels.

BACKGROUND: A relationship quantitative trait locus exists when the correlation between multiple traits varies by genotype for that locus. Relationship quantitative trait loci (rQTL) are often involved in gene-by-gene (G×G) interactions or gene-by-environmental interactions, making them a powerful tool for detecting G×G. METHODS: We performed genome-wide association studies to identify rQTL between tau and Aß42 and ptau and Aß42 with over 3000 individuals using age, gender, series, APOE ε2, APOE ε4, and two principal components for population structure as covariates. Each significant rQTL was separately screened for interactions with other loci for each trait in the rQTL model. Parametric bootstrapping was used to assess significance. RESULTS: We found four significant tau/Aß42 rQTL from three unique locations and six ptau/Aß42 rQTL from five unique locations. G×G screens with these rQTL produced four significant G×G interactions (one Aß42, two ptau, and one tau) with four rQTL where each second locus was from a unique location. On follow-up, rs1036819 and rs74025622 were associated with Alzheimer's disease (AD) case/control status; rs15205 and rs79099429 were associated with rate of decline. CONCLUSIONS: The two most significant rQTL (rs8027714 and rs1036819) for ptau/Aß42 are on different chromosomes and both are strong hits for pelvic organ prolapse. While diseases of the nervous system can cause pelvic organ prolapse, it is unlikely related to the ptau/Aß42 relationship but may suggest that these two loci share a pathway. In addition to a ptau/Aß42 rQTL and association with AD case/control status, rs1036819 is a strong rQTL for case/control status/Aß42 and for tau/Aß42. It resides in the ZFAT gene, which is related to autoimmune thyroid disease. For tau, rs9817620 interacts with the tau/Aß42 rQTL rs74025622. It is in the CHL1 gene, which is a neural cell adhesion molecule and may be involved in signal transduction pathways. CHL1 is related to BACE1, which is a ß-secretase enzyme that initiates production of the ß-amyloid peptide involved in AD and is a primary drug target. Overall, there are numerous loci that affect the relationship between these important AD endophenotypes and some are due to interactions with other loci. Some affect the risk of AD and/or rate of progression.

Interpreting Mini-Mental State Examination Performance in Highly Proficient Bilingual Spanish-English and Asian Indian-English Speakers: Demographic Adjustments, Item Analyses, and Supplemental Measures.

Purpose: Performance on the Mini-Mental State Examination (MMSE), among the most widely used global screens of adult cognitive status, is affected by demographic variables including age, education, and ethnicity. This study extends prior research by examining the specific effects of bilingualism on MMSE performance. Method: Sixty independent community-dwelling monolingual and bilingual adults were recruited from eastern and western regions of the United States in this cross-sectional group study. Independent sample t tests were used to compare 2 bilingual groups (Spanish-English and Asian Indian-English) with matched monolingual speakers on the MMSE, demographically adjusted MMSE scores, MMSE item scores, and a nonverbal cognitive measure. Regression analyses were also performed to determine whether language proficiency predicted MMSE performance in both groups of bilingual speakers. Results: Group differences were evident on the MMSE, on demographically adjusted MMSE scores, and on a small subset of individual MMSE items. Scores on a standardized screen of language proficiency predicted a significant proportion of the variance in the MMSE scores of both bilingual groups. Conclusions: Bilingual speakers demonstrated distinct performance profiles on the MMSE. Results suggest that supplementing the MMSE with a language screen, administering a nonverbal measure, and/or evaluating item-based patterns of performance may assist with test interpretation for this population.

Family-based tests for associating haplotypes with general phenotype data: Improving the FBAT-haplotype algorithm.

For family-based association studies, Horvath et al. proposed an algorithm for the association analysis between haplotypes and arbitrary phenotypes when the phase of the haplotypes is unknown, that is, genotype data is given. Their approach to haplotype analysis maintains the original features of the TDT/FBAT-approach, that is, complete robustness against genetic confounding and misspecification of the phenotype. The algorithm has been implemented in the FBAT and PBAT software package and has been used in numerous substantive manuscripts. Here, we propose a simplification of the original algorithm that maintains the original approach but reduces the computational burden of the approach substantially and gives valuable insights regarding the conditional distribution. With the modified algorithm, the application to whole-genome sequencing (WGS) studies becomes feasible; for example, in sliding window approaches or spatial-clustering approaches. The reduction of the computational burden that our modification provides is especially dramatic when both parental genotypes are missing. For example, for eight variants and 441 nuclear families with mostly offspring-only families, in a WGS study at the APOE locus, the running time decreased from approximately 21 hr for the original algorithm to 0.11 sec after our modification.

Sex Differences in Risk for Alzheimer's Disease Related to Neurotrophin Gene Polymorphisms: The Cache County Memory Study.

Neurotrophins, including nerve-growth factor and brain-derived neurotrophic factor, have been implicated in Alzheimer's disease (AD). Associations between AD and neurotrophin signaling genes have been inconsistent, with few studies examining sex differences in risk. We examined four single-nucleotide polymorphisms (SNPs) involved in neurotrophin signaling (rs6265, rs56164415, rs2289656, rs2072446) and risk for AD by sex in a population-based sample of older adults. Three thousand four hundred and ninety-nine individuals without dementia at baseline [mean (standard deviation) age = 74.64 (6.84), 58% female] underwent dementia screening and assessment over four triennial waves. Cox regression was used to examine time to AD or right censoring for each SNP. Female carriers of the minor T allele for rs2072446 and rs56164415 had a 60% (hazard ratio [HR] = 1.60, 95% confidence interval [CI] = 1.02-2.51) and 93% (HR = 1.93, 95% CI = 1.30-2.84) higher hazard for AD, respectively, than male noncarriers of the T allele. Furthermore, male carriers of the T allele of rs2072446 had a 61% lower hazard (HR = 0.39, 95% CI = 0.14-1.06) than male noncarriers at trend-level significance (p = .07). The association between certain neurotrophin gene polymorphisms and AD differs by sex and may explain inconsistent findings in the literature.

Closer caregiver and care-recipient relationships predict lower informal costs of dementia care: The Cache County Dementia Progression Study.

INTRODUCTION: Identifying factors associated with lower dementia care costs is essential. We examined whether two caregiver factors were associated with lower costs of informal care. METHODS: A total of 271 care dyads of the Cache County Dementia Study were included. Estimates of informal costs were based on caregiver reports of time spent in care-related activities and inflation-adjusted 2012 Utah median hourly wages. Caregiver coping and emotional closeness with the care-recipient were assessed using the Ways of Coping Checklist-Revised and Relationship Closeness Scale, respectively. RESULTS: Higher closeness was associated with 24% lower costs (expß = 0.763 [95% confidence interval: 0.583-0.999]) in linear mixed models controlling for demographics and baseline dementia severity and duration. Problem-focused coping was not associated with informal costs (P = .354). DISCUSSION: Caregiver closeness, a potentially modifiable factor, predicted lower dementia informal care costs over time. Future studies examining the care environment in closer dyads may identify specific care-related behaviors or strategies that are associated with lower costs.

Nutritional Status is Associated with Faster Cognitive Decline and Worse Functional Impairment in the Progression of Dementia: The Cache County Dementia Progression Study1.

Nutritional status may be a modifiable factor in the progression of dementia. We examined the association of nutritional status and rate of cognitive and functional decline in a U.S. population-based sample. Study design was an observational longitudinal study with annual follow-ups up to 6 years of 292 persons with dementia (72% Alzheimer's disease, 56% female) in Cache County, UT using the Mini-Mental State Exam (MMSE), Clinical Dementia Rating Sum of Boxes (CDR-sb), and modified Mini Nutritional Assessment (mMNA). mMNA scores declined by approximately 0.50 points/year, suggesting increasing risk for malnutrition. Lower mMNA score predicted faster rate of decline on the MMSE at earlier follow-up times, but slower decline at later follow-up times, whereas higher mMNA scores had the opposite pattern (mMNA by time ß= 0.22, p = 0.017; mMNA by time2 ß= -0.04, p = 0.04). Lower mMNA score was associated with greater impairment on the CDR-sb over the course of dementia (ß= 0.35, p <  0.001). Assessment of malnutrition may be useful in predicting rates of progression in dementia and may provide a target for clinical intervention.

Dementia severity and the longitudinal costs of informal care in the Cache County population.

BACKGROUND: Dementia costs are critical for influencing healthcare policy, but limited longitudinal information exists. We examined longitudinal informal care costs of dementia in a population-based sample. METHODS: Data from the Cache County Study included dementia onset, duration, and severity assessed by the Mini-Mental State Examination (MMSE), Clinical Dementia Rating Scale (CDR), and Neuropsychiatric Inventory (NPI). Informal costs of daily care (COC) was estimated based on median Utah wages. Mixed models estimated the relationship between severity and longitudinal COC in separate models for MMSE and CDR. RESULTS: Two hundred and eighty-seven subjects (53% female, mean (standard deviation) age was 82.3 (5.9) years) participated. Overall COC increased by 18% per year. COC was 6% lower per MMSE-point increase and compared with very mild dementia, COC increased over twofold for mild, fivefold for moderate, and sixfold for severe dementia on the CDR. CONCLUSIONS: Greater dementia severity predicted higher costs. Disease management strategies addressing dementia progression may curb costs.

Dementia caregivers' coping strategies and their relationship to health and well-being: the Cache County Study.

OBJECTIVES: Prior research identifies that psychological outcomes among dementia caregivers are associated with their use of coping strategies. Few studies have tested the association of coping and health longitudinally. METHOD: This study examined factors associated with the use of coping strategies over time and their associations with physical and mental health outcomes in a population-based sample of 226 dementia caregivers in Cache County, Utah, USA. Caregivers annually completed the Ways of Coping Checklist-Revised, the Beck Anxiety Inventory, and a health interview. Care-recipient cognitive and functional abilities were obtained using the Mini-Mental State Exam and the Clinical Dementia Rating. Neuropsychiatric symptoms were assessed using the Neuropsychiatric Inventory. RESULTS: Caregivers most frequently identified providing care as a problem (37.6%). Linear mixed models of caregiver coping strategies found that the use of most strategies were stable except for increasing Avoidance among adult child caregivers (ß = 0.14, p = 0.048). On average, increased Wishful Thinking (ß = 2.48, p < 0.001) or Blames Self (ß = 1.06, p = 0.002) was associated with higher anxiety scores. Increased use of Blames Others among males (interaction, ß = 0.28, p = 0.02) and greater use of Wishful Thinking among younger caregivers (interaction, ß = -0.01, p = 0.01) were associated with more caregiver health conditions. Coping strategies were not associated with change in anxiety or health conditions over time. CONCLUSION: Our results emphasize the importance of caregiver coping strategies on caregiver health and well-being and may identify subgroups of persons at risk for worse outcomes.

Use of diuretics is associated with reduced risk of Alzheimer's disease: the Cache County Study.

Although the use of antihypertensive medications has been associated with reduced risk of Alzheimer's disease (AD), it remains unclear which class provides the most benefit. The Cache County Study of Memory Health and Aging is a prospective longitudinal cohort study of dementing illnesses among the elderly population of Cache County, Utah. Using waves I to IV data of the Cache County Study, 3417 participants had a mean of 7.1 years of follow-up. Time-varying use of antihypertensive medications including different class of diuretics, angiotensin converting enzyme inhibitors, ß-blockers, and calcium channel blockers was used to predict the incidence of AD using Cox proportional hazards analyses. During follow-up, 325 AD cases were ascertained with a total of 23,590 person-years. Use of any antihypertensive medication was associated with lower incidence of AD (adjusted hazard ratio [aHR], 0.77; 95% confidence interval [CI], 0.61-0.97). Among different classes of antihypertensive medications, thiazide (aHR, 0.7; 95% CI, 0.53-0.93), and potassium-sparing diuretics (aHR, 0.69; 95% CI, 0.48-0.99) were associated with the greatest reduction of AD risk. Thiazide and potassium-sparing diuretics were associated with decreased risk of AD. The inverse association of potassium-sparing diuretics confirms an earlier finding in this cohort, now with longer follow-up, and merits further investigation.

The association of traumatic brain injury with rate of progression of cognitive and functional impairment in a population-based cohort of Alzheimer's disease: the Cache County Dementia Progression Study.

BACKGROUND: There is limited research on factors that influence the rate of progression in Alzheimer's disease (AD). A history of traumatic brain injury (TBI) is associated with an increased risk for AD, but its role on the rate of dementia progression after the onset of AD has not been examined. METHODS: A population-based cohort of 325 persons with incident AD was followed for up to 11 years. The sample was 65% female with a mean (SD) age of dementia onset = 84.4 (6.4) years. History of TBI was categorized as number, severity (with or without loss of consciousness), and timing in relation to dementia onset (within ten years or more than ten years). Cognition was assessed by the Consortium to Establish a Registry of AD battery, and functional ability was assessed by the Clinical Dementia Rating Sum of Boxes. RESULTS: In linear mixed models, a history of TBI within ten years of onset showed faster progression of functional impairment (LR x2 = 10.27, p = 0.006), while those with TBI more than ten years before dementia onset had higher scores on a measure of list learning (ß = 1.61, p = 0.003) and semantic memory (ß = 0.75, p = 0.0035). CONCLUSIONS: History of TBI and its recency may be a useful factor to predict functional progression in the course of AD.

Vascular risk factors and neuropsychiatric symptoms in Alzheimer's disease: the Cache County Study.

OBJECTIVE: Knowledge of potentially modifiable risk factors for neuropsychiatric symptoms (NPS) in Alzheimer's disease (AD) is important. This study longitudinally explores modifiable vascular risk factors for NPS in AD. METHODS: Participants enrolled in the Cache County Study on Memory in Aging with no dementia at baseline were subsequently assessed over three additional waves, and those with incident (new onset) dementia were invited to join the Dementia Progression Study for longitudinal follow-up. A total of 327 participants with incident AD were identified and assessed for the following vascular factors: atrial fibrillation, hypertension, diabetes mellitus, angina, coronary artery bypass surgery, myocardial infarction, cerebrovascular accident, and use of antihypertensive or diabetes medicines. A vascular index (VI) was also calculated. NPS were assessed over time using the Neuropsychiatric Inventory (NPI). Affective and Psychotic symptom clusters were assessed separately. The association between vascular factors and change in NPI total score was analyzed using linear mixed model and in symptom clusters using a random effects model. RESULTS: No individual vascular risk factors or the VI significantly predicted change in any individual NPS. The use of antihypertensive medications more than four times per week was associated with higher total NPI and Affective cluster scores. CONCLUSIONS: Use of antihypertensive medication was associated with higher total NPI and Affective cluster scores. The results of this study do not otherwise support vascular risk factors as modifiers of longitudinal change in NPS in AD.

Impact of offspring death on cognitive health in late life: the Cache County study.

OBJECTIVE: Experiencing the death of a child is associated with negative short-term mental health consequences, but less is known about cognitive outcomes and whether such associations extend to late life. We tested the hypothesis that experiencing an offspring death (OD) is associated with an increased rate of cognitive decline in late life. METHODS: This population-based longitudinal study observed four cognitive statuses spaced 3-4 years apart, linked to an extensive database containing objective genealogic and vital statistics data. Home visits were conducted with 3,174 residents of a rural county in northern Utah, initially without dementia, aged 65-105. Cognitive status was measured with the Modified Mini-Mental State Exam at baseline and at 3-, 7-, and 10-year follow-ups. OD was obtained from the Utah Population Database, which contains statewide birth and death records. RESULTS: In linear mixed models, controlling for age, gender, education, and apolipoprotein E status, subjects who experienced OD while younger than age 31 years experienced a significantly faster rate of cognitive decline in late life, but only if they had an ε4 allele. Reclassifying all OD (regardless of age) according to subsequent birth of another child, OD was only related to faster cognitive decline when there were no subsequent births. CONCLUSION: Experiencing OD in early adulthood has a long-term association with cognitive functioning in late life, with a gene-environment interaction at the apolipoprotein E locus. Subsequent birth of another child attenuates this association.

Variants in PPP3R1 and MAPT are associated with more rapid functional decline in Alzheimer's disease: the Cache County Dementia Progression Study.

BACKGROUND: Single-nucleotide polymorphisms (SNPs) located in the gene encoding the regulatory subunit of the protein phosphatase 2B (PPP3R1, rs1868402) and the microtubule-associated protein tau (MAPT, rs3785883) gene were recently associated with higher cerebrospinal fluid (CSF) tau levels in samples from the Knight Alzheimer's Disease Research Center at Washington University (WU) and Alzheimer's Disease Neuroimaging Initiative (ADNI). In these same samples, these SNPs were also associated with faster functional decline, or progression of Alzheimer's disease (AD) as measured by the Clinical Dementia Rating sum of boxes scores (CDR-sb). We attempted to validate the latter association in an independent, population-based sample of incident AD cases from the Cache County Dementia Progression Study (DPS). METHODS: All 92 AD cases from the DPS with a global CDR-sb ≤1 (mild) at initial clinical assessment who were later assessed on CDR-sb data on at least two other time points were genotyped at the two SNPs of interest (rs1868402 and rs3785883). We used linear mixed models to estimate associations between these SNPs and CDR-sb trajectory. All analyses were performed using Proc Mixed in SAS. RESULTS: Although we observed no association between rs3785883 or rs1868402 alone and change in CDR-sb (P > .10), there was a significant association between a combined genotype model and change in CDR-sb: carriers of the high-risk genotypes at both loci progressed >2.9 times faster than noncarriers (P = .015). When data from DPS were combined with previously published data from WU and ADNI, change in CDR-sb was 30% faster for each copy of the high-risk allele at rs3785883 (P = .0082) and carriers of both high-risk genotypes at both loci progressed 6 times faster (P < .0001) than all others combined. CONCLUSIONS: We replicate a previous report by Cruchaga et al that specific variations in rs3785883 and rs1868402 are associated with accelerated progression of AD. Further characterization of this association will provide a better understanding of how genetic factors influence the rate of progression of AD and could provide novel insights into preventative and therapeutic strategies.

Caregiver coping strategies predict cognitive and functional decline in dementia: the Cache County Dementia Progression Study.

OBJECTIVES: Few longitudinal studies have studied the influence of the care environment on the clinical progression of dementia. We examined whether caregiver coping strategies predict dementia progression in a population-based sample. DESIGN: Longitudinal, prospective cohort study. SETTING: Cache County (Utah) population. PARTICIPANTS: A total of 226 persons with dementia, and their caregivers, were assessed semiannually for up to 6 years. MEASUREMENTS: Ways of Coping Checklist-Revised, Mini-Mental State Examination (MMSE), and Clinical Dementia Rating (CDR). RESULTS: Mean (SD) age of dementia onset in persons with dementia was 82.11 (5.84) years and mean caregiver age was 67.41 (13.95) years. Mean (SD) follow-up was 1.65 (1.63) years from baseline. In univariate linear mixed-effects models, increasing use of problem-focused and counting blessings by caregivers was associated with slower patient worsening on the MMSE. Problem-focused coping, seeking social support, and wishful thinking were associated with slower Clinical Dementia Rating Scale sum of boxes (CDR-sb) worsening. Considering covariates, increasing use of problem-focused coping was associated with 0.70 points per year less worsening on the MMSE and 0.55 points per year less worsening on the CDR-sb. Compared with no use, the "regular" use of this strategy was associated with 2 points per year slower worsening on the MMSE and 1.65 points per year slower worsening on the CDR-sb. CONCLUSIONS: Caregiver coping strategies are associated with slower dementia progression. Developing interventions that target these strategies may benefit dementia patients.

Stressful life events and cognitive decline in late life: moderation by education and age. The Cache County Study.

OBJECTIVE: Stressful life events (SLE) have been associated with increased dementia risk, but their association with cognitive decline has been inconsistent. In a longitudinal population-based study of older individuals, we examined the association between SLE and cognitive decline, and the role of potential effect modifiers. METHODS: A total of 2665 non-demented participants of the Cache County Memory Study completed an SLE questionnaire at Wave 2 and were revisited 4 and 7 years later. The events were represented via several scores: total number, subjective rating (negative, positive, and unexpected), and a weighted summary based on their impact. Cognition was assessed at each visit with the modified Mini-Mental State Exam. General linear models were used to examine the association between SLE scores and cognition. Effect modification by age, education, and APOE genotype was tested. RESULTS: Years of formal education (p = 0.006) modified the effect of number of SLE, and age (p = 0.009) modified the effect of negative SLE on the rate of cognitive decline. Faster decline was observed among those with fewer years of education experiencing more SLE and also among younger participants experiencing more negative SLE. There was no association between other indicators of SLE and cognitive decline. APOE genotype did not modify any of the aforementioned associations. CONCLUSIONS: The effects of SLE on cognition in late life are complex and vary by individual factors such as age and education. These results may explain some of the contradictory findings in the literature.