Key Points

Question听 When does the degree of frailty accelerate before dementia, and how is it associated with dementia risk?

Findings听 In this analysis of 29鈥�849 participants from 4 cohort studies in the US and UK, frailty trajectories accelerated 4 to 9 years before the onset of dementia. Even among participants whose baseline frailty measurement occurred before that acceleration period, frailty was positively associated with dementia risk.

Meaning听 Frailty may have value in identifying at-risk populations for clinical trials and as a target for dementia prevention strategies.

Importance听 An accessible marker of both biological age and dementia risk is crucial to advancing dementia prevention and treatment strategies. Although frailty is a candidate for that role, the nature of the relationship between frailty and dementia is not well understood.

Objective听 To clarify the temporal relationship between frailty and incident dementia by investigating frailty trajectories in the years preceding dementia onset.

Design, Setting, and Participants听 Participant data came from 4 prospective cohort studies: the English Longitudinal Study of Ageing, the Health and Retirement Study, the Rush Memory and Aging Project, and the National Alzheimer Coordinating Center. Data were collected between 1997 and 2024 and were analyzed from July 2023 to August 2024. The settings were retirement communities, national-level surveys, and a multiclinic-based cohort. Included individuals were 60 years or older and without cognitive impairment at baseline. Included individuals also had data on age, sex, education level, and ethnicity and a frailty index score calculated at baseline.

Exposure听 Frailty was the main exposure, with participants鈥� degrees of frailty quantified using retrospectively calculated frailty index scores.

Main Outcomes and Measures听 Incident all-cause dementia ascertained through physician-derived diagnoses, self- and informant-report, and estimated classifications based on combinations of cognitive tests.

Results听 The participant number before exclusions was 87鈥�737. After exclusions, data from 29鈥�849 participants (mean [SD] age, 71.6 [7.7] years; 18鈥�369 female [62%]; 257鈥�963 person-years of follow-up; 3154 cases of incident dementia) were analyzed. Bayesian generalized linear mixed regression models revealed accelerations in frailty trajectories 4 to 9 years before incident dementia. Overall, frailty was positively associated with dementia risk (adjusted hazard ratios [aHRs] ranged from 1.18; 95% CI,鈥�1.13-1.24 to 1.73; 95% CI,鈥�1.57-1.92). This association held among participants whose time between frailty measurement and incident dementia exceeded the identified acceleration period (aHRs ranged from 1.18; 95% CI,鈥�1.12-1.23 to 1.43; 95% CI,鈥�1.14-1.80).

Conclusions and Relevance听 These findings suggest that frailty measurements may be used to identify high-risk population groups for preferential enrolment into clinical trials for dementia prevention and treatment. Frailty itself may represent a useful upstream target for behavioral and societal approaches to dementia prevention.

Dementia most commonly arises in older people due to mixed age-related neuropathologies,^1,2 suggesting that the aging process influences disease susceptibility.^3,4 A deeper understanding of the relationship between aging and late-life dementia could inform drug discovery and effective behavioral and societal strategies to dementia prevention. To optimize these approaches, it is important to have a readily measurable target that reflects biological aging and associates with dementia risk. Accumulating evidence indicates that frailty may be a viable candidate for that role.^5-7

Frailty can be understood as a gradable health state of increased vulnerability due to the accumulation of multiple age-related health deficits and reduced physiological reserve.⁸ At any age, frailty is positively associated with all-cause mortality and to a greater degree than are common laboratory-based estimates of biological age,^6,9 indicating that higher frailty reflects older biological age.^5,8 As observational data consistently show dementia occurring more frequently among individuals who have a higher degree of frailty,^7,10-13 frailty may represent a target for interventions aimed at reducing dementia risk.

Accumulating evidence suggests that frailty is modifiable. For example, physical activity, which has been associated with lower dementia risk and better cognition in both observational studies and randomized clinical trials,^14,15 is a strong determinant of frailty.¹⁶ A systematic review of 21 randomized clinical trials found that interventions involving both exercise delivered in group sessions and nutrition supplementation were the most effective in reducing frailty.¹⁷ Addressing the determinants of frailty, including physical activity and its other social, physical, and health-related modifiable factors,¹⁸ may therefore represent a strategy to reduce risk of developing clinical dementia symptoms. However, considering the long preclinical phase of Alzheimer disease (up to 15-20 years)^19,20 and the strong likelihood of reverse causality in the association of frailty with dementia risk, further investigation is required.

Understanding the dynamics of frailty trajectories in the years before dementia can inform use of its measurement in dementia prevention programs and in the targeted recruitment of high-risk populations into clinical trials for dementia. We aimed to clarify the temporal relationship between frailty and incident dementia, examining how the timing of frailty measurement relative to dementia onset influences its association with dementia risk. We pursued 2 objectives: (1) to determine when an acceleration in the accumulation of frailty due to impending dementia is first observable and (2) to measure the association of frailty and dementia risk after controlling for any impact of that predementia frailty acceleration period.

We analyzed participant data from 4 large prospective cohort studies: the English Longitudinal Study of Ageing (ELSA), the Health and Retirement Study (HRS), the Rush Memory and Aging Project (MAP), and the National Alzheimer Coordinating Center (NACC). Study procedures were reviewed and approved by local institutional review boards, and written informed consent was obtained from study participants. Details of study methodology, ethics approval, and data access are included in the eMethods in Supplement 1. Participants were included if they were 60 years or older at baseline, were without cognitive impairment, and had at least 1 follow-up assessment. Participants self-identified race and ethnicity, which were categorized here into the following groups: Black, White, and other/unknown, which included American Indian or Alaska native, Asian, Native Hawaiian or Pacific Islander, and multiethnic group (for the ELSA study, other ethnic group also included Black). Race and ethnicity information was included in this study due to being a known determinant of frailty.⁸ To remove the influence of early-onset dementia cases that often occur exclusively due to genetic causes,²¹ participants were excluded if they developed dementia before age 65 years. At baseline, participants were required to have data available on age, sex, education level, and ethnicity and sufficient data to calculate a frailty index score. Frailty index scores were only calculated where participants had information available on at least 30 deficits used in that study鈥檚 frailty index.²² This study followed the Strengthening the Reporting of Observational Studies in Epidemiology () reporting guidelines.

Given that mixed dementia is what occurs chiefly in late life,^1,2 the study outcome was all-cause dementia. The method of determining this outcome differed between studies. In the ELSA cohort, classifications were derived through either a self-report of physician diagnosis of dementia or a mean score of 3.4 or greater on the 16-item Informant Questionnaire on Cognitive Decline in the Elderly completed by family members or caregivers, which represents a decline in the ability of daily function compared with 2 years prior of a magnitude indicating dementia.²³ In the HRS cohort, classifications of dementia were obtained using the Langa-Weir Classification of Cognitive Function method, which applies validated cut points to summary scores obtained from a range of cognitive tests (scores ranged from 0-27; scores of 0-6 indicated the presence of dementia).²⁴ In the MAP cohort, presumptive diagnoses of dementia and Alzheimer disease were calculated via an algorithmic decision tree using accepted clinical criteria and confirmed by a clinician.²⁵ In the NACC cohort, either a consensus team or a single physician used standard diagnostic criteria to classify participants as having all-cause dementia.^25,26

Frailty was the main exposure in this study, with each participant鈥檚 degree of frailty quantified using retrospectively calculated frailty index scores. The frailty index is a measure of health state, combining information from multiple physiological systems and closely reflecting an individual鈥檚 risk for adverse health events and mortality independently of chronological age.⁸ The variables included in a frailty index are routinely collected clinical data such as symptoms, signs, disabilities, and diseases that meet standard criteria.²² Frailty index scores had been developed and validated previously in each cohort.^12,27-30 Each frailty index was adapted for our investigation by ensuring that deficits closely reflecting cognition were removed from their composition (Table 1). As frailty index scores represent the proportion of total health deficits of an individual, higher scores indicate the accumulation of more age-related health deficits and worse health. The eMethods in Supplement 1 contains more information on the frailty index approach.

Participant age, sex, education level, and ethnicity were included as covariates due to possibly confounding the relationship between frailty and incident dementia. In all datasets, age was measured in years at baseline; sex was a self-reported binary variable (male or female); education was reported at baseline, and for consistency between studies, was recoded into a 3-category variable (ie, lower, intermediate, and higher education) (eMethods in Supplement 1); and race and ethnicity were collected via self-report. The eMethods in Supplement 1 contains information regarding mortality data, which were used in censoring.

Full details of the statistical analyses are included in the eMethods in Supplement 1. In brief, to determine when an acceleration in the accumulation of frailty associated with impending dementia is first observable (objective 1), we modeled trajectories in frailty index scores (the dependent variable) using a backward timescale with bayesian generalized linear mixed regression models.³¹ In each model, population-level effects of time were fitted using natural cubic splines, which allow for nonlinear trajectories in frailty index scores (eg, rate of increase in frailty may hasten with advancing age³²), and included both a random intercept and slope (linear fit) for participants. We estimated the start of the predementia frailty acceleration period as the year after which the size of differences in frailty index scores between the incident dementia group and the censored group were observed to be statistically significant and increase consistently. We next measured the association of frailty and incident dementia after controlling for any impact of the predementia frailty acceleration period (objective 2). To do this, we first used Cox proportional hazards models to examine the relationships between frailty index scores and dementia risk. This model was then estimated separately within 2 subgroups. The first subgroup included participants whose time between baseline frailty measurement and event (incident dementia, censor) was less than or equal to the predementia frailty acceleration period (as estimated in objective 1). The second subgroup included participants whose time between baseline frailty measurement and event was greater than the predementia frailty acceleration period. All statistical models included covariates of age, sex, education, and race and ethnicity. Multiple sensitivity analyses were conducted to assess the robustness of associations of frailty index scores and incident dementia. All P values were 2-sided, and P <.05 was considered statistically significant. All statistical analyses were conducted using R, version 4.2.1 (R Foundation for Statistical Computing).

The participant number before exclusions was 87鈥�737. After exclusions, data from 29鈥�849 participants (mean [SD] age, 71.6 [7.7] years; 18鈥�369 female [62%]; 11鈥�480 male [38%]) were included in this analysis (Table 2 and eFigure 1 in Supplement 1). Participants identified with the following race and ethnicities: 2645 Black (9%), 25鈥�288 White (85%), and 1916 other/unknown (6%). Most participants were contributed by NACC (12鈥�582 [42%]) and the least by MAP (1451 [5%]). In total, 257鈥�963 person-years of follow-up and 3154 cases of incident dementia were analyzed. Among the cohorts, participants in MAP were oldest (mean [SD] age, 79.0 [7.0] years) and had the highest degrees of frailty (mean [SD] frailty index score, 0.19 [0.09]), on average, corresponding to the highest observed rates of incident dementia.

To determine when an acceleration in the accumulation of frailty associated with impending dementia might be first observable (objective 1), we modeled frailty index scores using backward timescales and adjusted for potential confounders. In the years before incident dementia or censor, frailty index scores tended to increase (Figure 1). Among the censored groups, gradual increases in frailty index scores were observed in all datasets, although these were smallest in NACC. Among the incident dementia groups, we observed accelerations in the rates of increase in frailty index scores in the years proximal to dementia. These accelerations began 4 to 9 years before dementia, varied by dataset, and were particularly pronounced in ELSA and NACC and less so in MAP and HRS, although still present in those datasets. That divergence in frailty trajectories associated with incident dementia was supported by the model results, whereby, for all datasets but HRS, the inclusion of an event group (incident dementia or censored) by time interaction term resulted in improved model fit (eTable 1 in Supplement 1). The population-level effects from the interaction model (ie, that which included the event group by time interaction term) are presented in eTable 2 in Supplement 1. Among participants who developed dementia, covariate-adjusted expected frailty index scores were, on average, higher in females than in males by 18.5% in ELSA (95% CI,鈥�7.3%-32.3%), 20.9% in HRS (95% CI,鈥�12.7%-28.4%), and 16.2% in MAP (95% CI,鈥�5.1%-28.4%) but not different in NACC (1.5%; 95% CI,鈥夆垝3.9% to 7.2%) (Figure 2).

Expected frailty index scores, calculated from the interaction model while holding the covariates of age, sex, education, and ethnicity constant, were then compared between the incident dementia and censored groups at each year (Figure 1). Compared with the censored groups, these frailty scores were consistently higher in the incident dementia groups, 20, 13, 12, and 8 years before dementia in HRS, MAP, ELSA and NACC, respectively (Figure 1 and eTable 3 in Supplement 1). At the point of dementia detection, frailty index scores were most elevated in ELSA (0.21 points higher than censored participants), elevated to a similar degree in both MAP and NACC (0.13 and 0.12 points higher, respectively) and to a lesser extent in HRS (0.04 points higher). The start of the predementia frailty acceleration period, ie, the year after which the size of differences in frailty index scores between the incident dementia group and the censored group were observed to be statistically significant and increase consistently, was estimated at 9, 6, 4 and 4 years before dementia for NACC, MAP, ELSA and HRS (Figure 1 and eTable 3 in Supplement 1), which was similar in both males and females (eFigures 2-5 in Supplement 1). Mean differences in expected frailty index scores and P values are presented in eTable 3 in Supplement 1.

We next measured the association of frailty index scores and incident dementia after controlling for the predementia frailty acceleration period (objective 2). We did this by using Cox proportional-hazards models to determine the associations of frailty with incident dementia for participants whose time between baseline frailty measurement and event (incident dementia or censored) was greater than the cohort-specific predementia frailty acceleration period (as estimated under objective 1). The size of analyzed samples, the predementia frailty acceleration periods, and the number of deficits included in frailty indices varied in the main and sensitivity analyses (eTable 4 in Supplement 1).

In the main analyses, in each dataset, each 0.1 increase in frailty index scores (equivalent to 4-5 additional health deficits) was associated with higher dementia risk (Figure 3 and eTable 4 in Supplement 1). This association was strongest in NACC, weakest in HRS, and similar in ELSA and MAP, with hazard ratios ranging from 1.18 (95% CI,鈥�1.13-1.24) to 1.73 (95% CI,鈥�1.57-1.92). When the time between frailty measurement and incident dementia or censor was considered, in most datasets, associations remained similar in both groups (ie, in participants whose time between frailty measurement and incident dementia or censor was less than or equal to the predementia frailty acceleration period, and in participants whose time between measurement and outcome exceeded that period). Here, event timing 脳 frailty index score interaction terms were not statistically significant in the ELSA (estimate = 0.93; P鈥�=鈥�.51), MAP (estimate = 1.10; P鈥�=鈥�.48), or NACC (estimate = 1.08; P鈥�=鈥�.57) cohorts but were in the HRS (estimate = 0.77; P鈥�<鈥�.001) cohort. Across datasets and in participants whose baseline frailty measurement was conducted before the predementia acceleration period had begun, the associations of frailty index scores with dementia risk were consistently positive and statistically significant. There, each 0.1 increase in frailty index scores was associated with hazard ratios ranging from 1.18 (95% CI,鈥�1.12-1.23) to 1.43 (95% CI,鈥�1.14-1.80) and in the absence of meaningful differences in this association between males and females (eFigures 2-5 in Supplement 1). The results from both sensitivity analyses demonstrated a robustness in these findings, whereby frailty index scores calculated before the predementia frailty acceleration period remained associated with incident dementia at a statistically significant level even when that period was extended by 2 years (sensitivity analysis 1). Likewise, our results were robust to removing health deficits that were independently associated with incident dementia from the calculation of frailty index scores (sensitivity analysis 2).

With the purpose of clarifying the temporal relationship between frailty and dementia, we modeled frailty trajectories in the years preceding dementia and assessed how the timing of frailty measurement relative to dementia onset influences its association with dementia risk. From this analysis of almost 30鈥�000 individuals participating in 4 cohort studies in the UK and US, we reported 4 main findings: (1) an elevated degree of frailty was observed 8 to 20 years before dementia onset; (2) the rate of decline in health and function in prodromal dementia, as reflected in a higher degree of frailty, accelerated from 4 to 9 years before dementia onset; (3) frailty was higher in females compared with males among those who developed dementia, with the greatest differences observed further from dementia onset; and (4) frailty was a robust risk factor for incident dementia even when its measurement occurred before the predementia frailty acceleration period. These results offer insight into the natural course of declining health in the subclinical stages of neurodegenerative diseases, position frailty index scores as a measure effective in identifying high-risk individuals for inclusion into treatment and prevention trials for dementia, and support the notion that frailty may serve as an upstream dementia risk factor.

Previous reports have suggested a preclinical phase of Alzheimer disease up to 15 to 20 years in length,^19,20 with changes in health and function first detectable at a population level from 10 years before dementia onset. Examples of these include higher health care usage and lower social engagement (2 years before diagnosis),^33,34 accelerated cognitive decline (6-10 years prior),^20,35 and more depressive symptoms (10 years prior).³⁶ Even though we observed a degree of heterogeneity in frailty trajectories between the datasets, in each case, the predementia frailty acceleration period was estimated to lie within that 10-year prodromal period, supporting those earlier studies. Consequently, 1 explanation for elevated frailty in the years proximal to dementia relates to the adverse impacts of neurodegenerative changes.

Aside from neurodegenerative processes hastening frailty accumulation, another explanation for our findings is that accelerated biological aging is a dementia cause rather than a consequence. In support, strong links have been established between changes in the hallmarks of aging and the development of neurodegenerative diseases,^3,4 and chronological age itself has long been understood as a key risk factor. Rapidly increasing frailty index scores, observed here up to 9 years before dementia onset, may therefore signal an exhaustion of systemic reserves leaving affected individuals vulnerable to diseases that might otherwise have remained subclinical.⁸ This loss of reserve associated with higher frailty has been demonstrated previously in dementia, where frailty was associated with weaker relationships between dementia and neuropathological burden and polygenic risk despite persistently high dementia rates.^7,37,38 Another mechanism through which frailty may precipitate dementia is by accelerating the accumulation of age-related brain changes. In support, cross-sectional analyses of cognitively intact individuals have linked frailty with smaller brain volumes and more vascular neuropathology.^39,40 This interpretation of our findings highlights the potential of frailty as a target of modifiable risk for dementia.^16,18

Regardless of the nature of the relationship between the predementia frailty acceleration period and subsequent dementia, the findings from our analyses align with the position that frailty could be a strong dementia risk factor. In individuals whose measurement of frailty occurred before the predementia frailty acceleration period had begun, and in both males and females, we observed positive associations between frailty index scores and incident dementia. Our findings join previous reports of a robust association between frailty and incident dementia, even when adjusting for a polygenic dementia risk score and a marker of area-level deprivation,⁷ adjusting for the competing risk of death,¹¹ including only nontraditional risk factors in the composition of the frailty index,¹³ or when conceptualizing frailty as a phenotype.¹⁰

A considerable strength of our investigation was the use of 4 different cohort studies across 2 continents, which varied in participant characteristics and in study methodologies (eg, settings, methods of dementia detection, time between repeat assessments). These differences contributed to variability in our statistical findings, both in terms of the frailty trajectories and in the strength of associations between frailty index scores and incident dementia. In datasets with annual monitoring of functioning and physician-derived dementia classifications (ie, NACC, MAP), the onset of the predementia frailty acceleration period was detected earlier compared with in those with biennial assessments involving self-reported or estimated dementia classifications (ie, ELSA, HRS), likely due to higher accuracy and timeliness in diagnoses. Despite these differences, by applying a consistent analytical approach to each dataset and reviewing results independently, we observed an encouraging consistency in findings supportive of strong external validity.

Our results should be interpreted with respect to limitations. First, a degree of reverse causality remains likely in the absence of a randomized design, particularly considering the up to 20-year preclinical phase of dementia.^19,20 Second, to enhance consistency and comparability in analyses between cohorts, we did not include potentially relevant covariates in statistical models unless they were universally available. Although we included education level, which is an important marker of socioeconomic status, we did not include other markers of social deprivation that may be causally associated with dementia.⁴¹ Third, genetic risk for dementia, often approximated using APOE 蔚4 status, was not adjusted for. Nonetheless, previous reports of strong associations between frailty and incident dementia even after adjusting for social deprivation,⁷ and within both carriers and noncarriers of the gene apolipoprotein E (APOE) 蔚4,¹² lead us to maintain confidence in our findings.

This cohort study offers novel insights into the temporal relationship between frailty and dementia and provides robust observational evidence that frailty may serve as a risk factor for dementia even when measured distally to dementia onset. These findings suggest that frailty measurements can be used to identify high-risk population groups for preferential enrolment into clinical trials for dementia prevention and treatment, and that frailty itself may represent a useful upstream target for behavioral and societal approaches to dementia prevention.

Accepted for Publication: September 12, 2024.

Published Online: November 11, 2024. doi:10.1001/jamaneurol.2024.3774

Correction: This article was corrected on November 15, 2024, to fix the short title. The word family was corrected to frailty.

Corresponding Author: David D. Ward, PhD, Centre for Health Services Research, Faculty of Medicine, The University of Queensland, Princess Alexandra Hospital, 34 Cornwall St, Bldg 33, Woolloongabba, QLD 4102, Australia (david.ward@uq.edu.au).

Author Contributions: Dr Ward had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Concept and design: Ward, Flint, Littlejohns, Foote, Wallace, Llewellyn, Ranson, Hubbard, Rockwood.

Acquisition, analysis, or interpretation of data: Ward, Flint, Littlejohns, Canevelli, Wallace, Gordon, Llewellyn, Rockwood, Stolz.

Drafting of the manuscript: Ward, Flint, Wallace.

Critical review of the manuscript for important intellectual content: Flint, Littlejohns, Foote, Canevelli, Wallace, Gordon, Llewellyn, Ranson, Hubbard, Rockwood, Stolz.

Statistical analysis: Ward, Flint, Wallace, Stolz.

Obtained funding: Llewellyn, Ranson.

Administrative, technical, or material support: Ranson.

Supervision: Ward, Canevelli, Llewellyn, Hubbard.

Conflict of Interest Disclosures: Dr Foote reported receiving grants from National Institute on Aging (RF1AG073593) during the conduct of the study. Dr Rockwood reported holding copyright over the Clinical Frailty Scale and Pictorial Fit-Frail Scale, which are made freely available for noncommercial education and research, as well as nonprofit health care with completion of a permission agreement stipulating that users will not change or charge for or commercialize the scales; in addition, for-profit entities (including pharma) pay a licensing fee, 15% of which is retained by the Dalhousie University Office of Commercialization and Innovation Engagement, and all remaining license fees are donated to the Dalhousie Faculty of Medicine Advancement Fund. In the past 3 years, licenses have been negotiated with Rebibus Therapeutics Inc, Cook Research Incorporated, W.L. Gore Associates Inc, Pfizer Inc, Cellcolabs AB, AstraZeneca UK Limited, Qu Biologics Inc, Biotest AG, BioAge Labs Inc, Congenica, Icosavax Inc outside the submitted work; in addition, Dr Rockwood reported having a patent for Electronic Goal Attainment Scaling pending Application made and in the past 3 years Dr Rockwood reported receiving honoraria for invited lectures, rounds and academic symposia on frailty from: Burnaby Family Practice, Chinese Medical Association, University of Nebraska-Omaha, the Australia New Zealand Society of Geriatric Medicine, the Atria Institute, University of British Columbia, McMaster University, and the Fraser Health Authority. Dr Rockwood reported currently serving on a data safety monitoring board for EIP Pharma Inc and being a member for the past 3 years of the ADMET-2 advisory board (Johns Hopkins), and the Wake Forest University Medical School Centre advisory board. No other disclosures were reported.

Funding/Support: This work was supported in part by the Deep Dementia Phenotyping (DEMON) Network, through the Frailty and Dementia Special Interest Group, and is an outcome of a workshop entitled 鈥淔railty and Precision Dementia Medicine鈥� funded by the University of Exeter Global Partnerships Fund (Dr Ranson); the Legal and General PLC as part of their corporate social responsibility (CSR) program (Mr Flint); the National Institute on Aging (NIA) grant RF1AG073593 (Dr Foote); a Banting Postdoctoral Fellowship awarded by the Canadian Institutes of Health Research (Dr Wallace); Alzheimer鈥檚 Research UK (Drs Llewellyn and Ranson); the National Institute for Health and Care Research Applied Research Collaboration South West Peninsula (Dr Llewellyn); the Dalhousie University Advancement as Clinical Research Professor of Frailty and Aging (Dr Rockwood). In addition, Dr Rockwood is a member of Team 14 (Frailty and Dementia), part of the Canadian Consortium on Neurodegeneration in Aging, which is supported by a grant from the Canadian Institutes of Health Research (grant CAN-163902) with funding from several partners. This work uses several datasets which have been funded by a number of agencies and grants. English Longitudinal Study of Ageing is funded by the NIA (grant R01AG017644), and by UK government departments coordinated by the National Institute for Health and Care Research. The Health and Retirement Study was produced and distributed by the University of Michigan with funding from the NIA (grant U01AG009740). The Rush Memory and Aging Project is supported by the National Institutes of Health (NIH; grant R01AG1791714). The National Alzheimer Coordinating Center (NACC) database is funded by NIA/NIH grant U24 AG072122. NACC data are contributed by the NIA-funded Alzheimer鈥檚 Disease Research Centers: P30 AG062429 (principal investigator [PI] James Brewer, MD, PhD), P30 AG066468 (PI Oscar Lopez, MD), P30 AG062421 (PI Bradley Hyman, MD, PhD), P30 AG066509 (PI Thomas Grabowski, MD), P30 AG066514 (PI Mary Sano, PhD), P30 AG066530 (PI Helena Chui, MD), P30 AG066507 (PI Marilyn Albert, PhD), P30 AG066444 (PI John Morris, MD), P30 AG066518 (PI Jeffrey Kaye, MD), P30 AG066512 (PI Thomas Wisniewski, MD), P30 AG066462 (PI Scott Small, MD), P30 AG072979 (PI David Wolk, MD), P30 AG072972 (PI Charles DeCarli, MD), P30 AG072976 (PI Andrew Saykin, PsyD), P30 AG072975 (PI David Bennett, MD), P30 AG072978 (PI Neil Kowall, MD), P30 AG072977 (PI Robert Vassar, PhD), P30 AG066519 (PI Frank LaFerla, PhD), P30 AG062677 (PI Ronald Petersen, MD, PhD), P30 AG079280 (PI Eric Reiman, MD), P30 AG062422 (PI Gil Rabinovici, MD), P30 AG066511 (PI Allan Levey, MD, PhD), P30 AG072946 (PI Linda Van Eldik, PhD), P30 AG062715 (PI Sanjay Asthana, MD, FRCP), P30 AG072973 (PI Russell Swerdlow, MD), P30 AG066506 (PI Todd Golde, MD, PhD), P30 AG066508 (PI Stephen Strittmatter, MD, PhD), P30 AG066515 (PI Victor Henderson, MD, MS), P30 AG072947 (PI Suzanne Craft, PhD), P30 AG072931 (PI Henry Paulson, MD, PhD), P30 AG066546 (PI Sudha Seshadri, MD), P20 AG068024 (PI Erik Roberson, MD, PhD), P20 AG068053 (PI Justin Miller, PhD), P20 AG068077 (PI Gary Rosenberg, MD), P20 AG068082 (PI Angela Jefferson, PhD), P30 AG072958 (PI Heather Whitson, MD), P30 AG072959 (PI James Leverenz, MD).

Role of the Funder/Sponsor: The funders had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Disclaimer: The views expressed are those of the authors and not necessarily those of Legal and General Public Limited Company.

Data Sharing Statement: See Supplement 2.