Key Points

Question听 Among cognitively normal adults in the community, what is the risk from subjective cognitive decline (SCD), using SCD-plus (SCD+) criteria and assessed longitudinally, associated with mild cognitive impairment (MCI), Alzheimer disease (AD), and all-cause dementia?

Findings听 In this cohort study using longitudinal data from the Framingham Heart Study including 3585 participants, SCD+ was significantly associated with survival time to MCI, AD, and all-cause dementia. Associations were independent of APOE status and an AD polygenic risk score.

Meaning听 In a community setting, SCD+ was associated with an increased risk of future MCI, AD, and all-cause dementia with similar hazards estimated in clinic-based settings.

Importance听 Subjective cognitive decline (SCD) is recognized to be in the Alzheimer disease (AD) cognitive continuum. The SCD Initiative International Working Group recently proposed SCD-plus (SCD+) features that increase risk for future objective cognitive decline but that have not been assessed in a large community-based setting.

Objective听 To assess SCD risk for mild cognitive impairment (MCI), AD, and all-cause dementia, using SCD+ criteria among cognitively normal adults.

Design, Setting, and Participants听 The Framingham Heart Study, a community-based prospective cohort study, assessed SCD between 2005 and 2019, with up to 12 years of follow-up. Participants 60 years and older with normal cognition at analytic baseline were included. Cox proportional hazards (CPH) models were adjusted for baseline age, sex, education, APOE 蔚4 status, and tertiles of AD polygenic risk score (PRS), excluding the APOE region. Data were analyzed from May 2021 to November 2023.

Exposure听 SCD was assessed longitudinally using a single question and considered present if endorsed at the last cognitively normal visit. It was treated as a time-varying variable, beginning at the first of consecutive, cognitively normal visits, including the last, at which it was endorsed.

Main Outcomes and Measures听 Consensus-diagnosed MCI, AD, and all-cause dementia.

Results听 This study included 3585 participants (mean [SD] baseline age, 68.0 [7.7] years; 1975 female [55.1%]). A total of 1596 participants (44.5%) had SCD, and 770 (21.5%) were carriers of APOE 蔚4. APOE 蔚4 and tertiles of AD PRS status did not significantly differ between the SCD and non-SCD groups. MCI, AD, and all-cause dementia were diagnosed in 236 participants (6.6%), 73 participants (2.0%), and 89 participants (2.5%), respectively, during follow-up. On average, SCD preceded MCI by 4.4 years, AD by 6.8 years, and all-cause dementia by 6.9 years. SCD was significantly associated with survival time to MCI (hazard ratio [HR],鈥�1.57; 95% CI,鈥�1.22-2.03; P鈥�<.001), AD (HR,鈥�2.98; 95% CI,鈥�1.89-4.70; P鈥�<.001), and all-cause dementia (HR,鈥�2.14; 95% CI,鈥�1.44-3.18; P鈥�<.001). After adjustment for APOE and AD PRS, the hazards of SCD were largely unchanged.

Conclusions and Relevance听 Results of this cohort study suggest that in a community setting, SCD reflecting SCD+ features was associated with an increased risk of future MCI, AD, and all-cause dementia with similar hazards estimated in clinic-based settings. SCD may be an independent risk factor for AD and other dementias beyond the risk incurred by APOE 蔚4 and AD PRS.

Early detection of Alzheimer disease (AD) in its preclinical stages is critical for effective intervention and prevention. Subjective cognitive decline (SCD) has been recognized as a feature of preclinical AD and other dementias.^1-3 SCD refers to a self-perception of cognitive decline despite normal performance on traditional neuropsychological testing.⁴ Numerous studies demonstrate that SCD may be an early marker for cognitive decline or the presence of AD pathology, even for those individuals with no objective cognitive deficits.^3,5-8 The updated AD research framework of the National Institute on Aging and Alzheimer Association now recognizes SCD in the cognitive continuum within the cognitively unimpaired stage.^4,9,10

Many studies on SCD have been conducted in clinical or volunteer convenience samples who are easier to recruit and have higher rates of concerns about cognition than community samples. Clinical samples tend to have higher rates of conversion from normal cognition to mild cognitive impairment (MCI) and from MCI to dementia than community samples. This is likely due to clinical samples having higher risk profiles for conversion (eg, higher rates of APOE 蔚4, family history of dementia, educational level, and income).^11-13 Additionally, individuals in clinical settings seek help for their cognitive difficulties, which may be distinct and/or more severe than difficulties endorsed in community studies. As SCD may, at times, be a feature of normal aging rather than an intermediary between normal cognition and MCI, SCD may pose a smaller risk for conversion to MCI in community than in clinical samples because the baseline risk of conversion is also smaller in community samples.^14,15 With caveats related to study design, the evidence bears this out, with SCD often showing larger effects for conversion in clinic samples and smaller or absent effects in community samples.^1,15-27 These differences are consequential because community studies are less susceptible to selection bias and are more representative of the general population where early screening would usually take place.

To date, community studies of SCD have been limited by 1 or more of the following: small sample size, a single SCD assessment, lack of careful cognitive assessment in all or most participants, or lack of a criterion-standard consensus diagnosis for AD and dementia.^28,29 Assessing SCD over multiple time points with repeated expressions of concern could be crucial for improved SCD measurement. Careful cognitive assessment in all or most participants is also crucial to ensure that SCD is occurring in the absence of objective impairment. These improvements in study design could lead to a more accurate assessment of the true association between SCD and dementia in the general population.²⁹

The APOE 蔚4 allele is a well-established genetic risk factor for developing AD.^30-34 The relationships between APOE 蔚4, SCD, and objective impairment remain unclear,³⁵ despite several studies on this topic. A systematic review³⁵ found that among individuals with normal cognition, APOE 蔚4 frequency was comparable between those with and without SCD. As SCD is considered part of the cognitive continuum, the lack of association with APOE 蔚4 is surprising. Narrative reviews suggested that both APOE 蔚4 and SCD conferred an individual and multiplicative risk of conversion to objective cognitive impairment. This interaction between SCD and APOE 蔚4 status may be sex specific, with a significant association in women, but not men.^8,36 Lastly, it seems plausible that genetic risk for AD beyond APOE 蔚4 may confer risk for SCD. A polygenic risk score (PRS) measures an individual鈥檚 genetic likelihood of having a particular trait over multiple genomic variants鈥攊n the case of AD PRS, it can demonstrate an individual鈥檚 propensity for developing AD.³⁷ To our knowledge, the association between an AD PRS and SCD has not been evaluated.

Given these knowledge gaps, we investigated the association between SCD and risk of conversion to MCI, AD, and all-cause dementia, using data collected in the Framingham Heart Study (FHS), a large community study with comprehensively characterized cognition and criterion-standard consensus dementia diagnoses. We leveraged data collected on SCD across multiple visits at which participants were objectively cognitively normal and estimated the hazard ratio (HR) of conversion in the Cox proportional hazards (CPH) model. Based on recommendations from the international working group known as the SCD Initiative, we incorporated into the study design several characteristics of SCD (termed SCD-plus features, hereafter referred to as SCD+) suggested to increase risk for future objective cognitive decline.^38,39 We aimed to (1) examine the association between SCD and conversion to MCI, AD, and all-cause dementia, (2) examine how the size of associations may change with introduction of SCD+ features, (3) compare HRs estimated with and without adjustment for AD genetic risk factors (APOE 蔚4, AD PRS), and (4) test whether the interactions of SCD with these genetic risk factors and with sex, education, and depression (factors known to be associated with both SCD and AD) are associated with conversion.

The FHS is a prospective community longitudinal study, including multigenerational cohorts with initial recruitment in Framingham, Massachusetts.^40-43 Participants were initially invited to a regular clinic examination, ie, core visit; neuropsychological testing was offered to all participants as part of a large ancillary study.^44,45 The institutional review board at Boston University approved the study protocol. Written informed consent was obtained from all participants. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology () reporting guidelines.

The current study included participants with normal cognition at age 60 years or older. Visits before age 60 years were not considered, and participants with MCI or dementia at or before age 60 years were excluded. Participants were followed up from 2005 to 2019 for up to 12 years. Participants self-identified with the following race and ethnicity groups: Asian, non-Hispanic Black, Hispanic, non-Hispanic White, and other, which included Indian, Native American, and Pacific Islander. Race and ethnicity are included because they are well established risk factors for dementia.

To be included, participants must have completed items assessing SCD during at least 1 study visit at which time they were cognitively normal. We included all core and neuropsychological visits at which SCD was assessed and arranged visits longitudinally. We termed the combined set of core and/or neuropsychological visits as the hybrid set. The baseline visit was considered the first visit at or after age 60 years, at which SCD was assessed. More information on the methods used in this study are available in the eMethods in Supplement 1.

SCD items were presented in slightly different ways at core and neuropsychological visits. At core visits, participants were queried 鈥淒o you feel your memory is becoming worse?鈥� At neuropsychological visits, participants were queried 鈥淒o you have any concerns about your memory or thinking?鈥�

Diagnoses and dates of onset of MCI, AD, and all-cause dementia were assigned by a dementia review panel, including neurologists and neuropsychologists. The panel used criteria from the DSM-IV to diagnose dementia,⁴⁶ and criteria from the National Institute of Neurological and Communicative Disorders and Stroke and the Alzheimer Disease and Related Disorders Association were used to diagnose AD.^47,48 MCI was determined based on evidence of subjective and objective decline in one or more cognitive domains,⁴⁹ meeting widely accepted criteria for MCI.^50,51

APOE genotypes were determined from blood samples.^52,53 Individuals were coded as 蔚4 carriers (蔚2/蔚4, 蔚3/蔚4, and 蔚4/蔚4) and 蔚2 carriers (蔚2/蔚2, 蔚2/蔚3, and 蔚2/蔚4).⁵⁴ We derived 2 different PRSs (with and without the APOE region) using summary statistics from the AD genome-wide association study by Kunkle and colleagues.⁵⁵ PRS-CS, a Bayesian PRS method using continuous shrinkage priors,⁵⁶ was applied. Risk scores were evenly divided into low-, moderate-, and high-risk groups, and the categorized tertile values were considered in further analyses.

A visit was considered valid if the participant was cognitively normal at the time of the visit and the question about SCD was administered. To compare basic characteristics between participants with and without SCD, participants were grouped into 1 of 3 categories that were consistent with the CPH models described subsequently. Participants in the only-SCD group endorsed SCD at each valid visit. Participants in the no-SCD group did not endorse SCD at their last valid visit. Participants in the mixed-SCD group endorsed SCD at their last valid visit but not at all valid visits. Comparisons used Fisher exact test for categorical variables and analysis of variance for continuous variables.

We used CPH models to compare time with MCI, AD, and all-cause dementia between those with and without SCD, which was treated as time varying. Figure 1 describes and shows exemplar participants demonstrating the different ways of SCD assignment and measuring survival time in years. Base models were adjusted for age, sex, and educational level. Age was defined as the age at the baseline visit in the non-SCD group and as the age at which SCD was considered to begin in the SCD group. Subsequent CPH models were adjusted for potential confounders, including genetic factors (APOE 蔚4 status, APOE 蔚2 status, and categorized PRS), depression, cardiovascular disease (CVD), current smoking, hypertension, and type 2 diabetes. Models accounted for the relatedness among individuals by using a mixed-model fitting familial structure as a random intercept. We constructed Kaplan-Meier curves for MCI, AD, and all-cause dementia, stratified by SCD and non-SCD groups, based on CPH models with age, sex, educational level, APOE 蔚4 status, and APOE 蔚2 status adjustments.

Additional sensitivity analyses investigated the SCD+ criteria. We limited models to participants from the set of neuropsychological visits only and the set of core visits only, as the SCD questions differed by whether concern was present (neuropsychological only). We shifted participants from the SCD to non-SCD group if SCD was consecutively endorsed for more than 5 years to better capture SCD onset within 5 years of objective impairment. We shifted participants from the SCD to the non-SCD group if SCD was endorsed only once at their last valid visit to better capture persistence of SCD. We made all 3 changes noted previously in the same model including only using neuropsychological visit (to capture concern), restricting consecutively endorsed SCD to 5 years, and requiring 2 consecutive visits with SCD, including the last. Data were analyzed from May 2021 to November 2023 using R software, version 4.3 (R Project for Statistical Computing). All P values were 2-sided, and a P value <.05 was considered statistically significant.

Overall, this study included 3585 individuals (mean [SD] baseline age, 68.0 [7.7] years; 1975 female [55.1%]; 1610 male [44.9%]). Participants identified with the following races and ethnicities: 90 Asian (2.5%), 113 non-Hispanic Black (3.2%), 97 Hispanic (2.7%), 3283 non-Hispanic White (91.6%), and 2 other (0.1%). There were 1804 college graduates (50.3%), and 770 APOE 蔚4 carriers (21.5%). A total of 236 participants (6.6%) developed MCI, 73 (2.0%) developed AD, and 89 (2.5%) developed all-cause dementia over the study period (Figure 2). On average, SCD preceded MCI by 4.4 years, AD by 6.8 years, and all-cause dementia by 6.9 years. Table 1⁵⁷ shows study sample characteristics overall and stratified into 3 groups: SCD at each valid visit (only SCD, n鈥�=鈥�1034), SCD present at the last valid visit but not all valid visits (mixed SCD, n鈥�=鈥�562; for a total of 1596 participants [44.5%] with SCD), and SCD not present at the last valid visit (no SCD, n鈥�=鈥�1989). The baseline age across groups was similar, and the mean (SD) age of SCD beginning was 69.8 [8.0] years. Compared with the no-SCD group, the only-SCD group had significantly more women (613 of 1034 [59.3%] vs 1043 of 1989 [52.4%]) and depression (128 of 1034 [21.0%] vs 114 of 1989 [9.9%]). The groups did not significantly differ by APOE status or by other factors鈥擟VD, current smoking, hypertension, and type 2 diabetes. Compared with the no-SCD group, future cognitive impairment was also significantly higher in the only-SCD group: MCI (89 of 1034 [8.6%] vs 115 of 1989 [5.8%]), AD (35 of 1034 [3.4%] vs 29 of 1989 [1.5%]), and all-cause dementia (40 of 1034 [3.9%] vs 39 of 1989 [2.0%]). Participants were followed up for 10鈥�783 person-years and had a mean (SD) of 2.1 (1.5) valid visits. We observed the same trends in the set of neuropsychological visits (eTable 1 in Supplement 1).

Survival analyses revealed that SCD was significantly associated with survival time to MCI, AD, and all-cause dementia (Table 2 and Figure 3). In the base model (adjusting for age, sex, and educational level but not APOE status), the HR was 1.60 (95% CI, 1.24-2.06; P鈥�<鈥�.001) for MCI, 4.33 (95% CI, 2.38-7.85; P鈥�<鈥�.001) for AD, and 2.17 (95% CI, 1.46-3.23; P鈥�<鈥�.001) for all-cause dementia. When we included APOE 蔚4 status as a covariate, the HR of SCD remained similar for MCI (HR,鈥�1.57; 95% CI, 1.22-2.03; P鈥�<鈥�.001) and all-cause dementia (HR, 2.14; 95% CI, 1.44-3.18; P鈥�<鈥�.001) but was reduced, although still significant, for AD (HR,鈥�2.98; 95% CI, 1.89-4.70; P鈥�<鈥�.001). We also observed that the HR of APOE 蔚4 status was significant for MCI (HR,鈥�1.72; 95% CI, 1.27-2.32; P鈥�<鈥�.001) and AD (HR,鈥�1.88; 95% CI, 1.09-3.26; P鈥�=鈥�.02). In the model with both APOE 蔚4 and 蔚2 status, we found no significant association due to 蔚2 status, and the SCD HR remained similar.

When we included a categorized AD PRS as a covariate, we found no significant associations due to the PRS with and without the APOE region, and the SCD HR remained similar (eTables 2 and 3 in Supplement 1). In models including depression as a covariate, there were small reductions in the SCD HRs for all 3 outcomes (<5%), although the SCD hazards remained significant (eTable 3 in Supplement 1). When we included other CVD-related factors as covariates, the HR of SCD remained similar (eTable 3 in Supplement 1). We did not observe significant SCD interactions with sex, education, depression, or APOE 蔚4 for any outcomes (eTable 4 in Supplement 1). Three-way interactions among SCD, APOE 蔚4, and sex (or depression) were also not significant (eTable 4 in Supplement 1).

When we restricted participants to those from the set of neuropsychological visits (SCD with concern) (eTable 5 in Supplement 1), the HRs were modestly larger compared with models using the set of hybrid visits and with models using the set of core visits (eTable 6 in Supplement 1) for MCI and AD dementia but not for all-cause dementia. When we shifted participants from the SCD group to the non-SCD group if SCD was consecutively endorsed for more than 5 years, the APOE 蔚4鈥�adjusted HRs were increased by 30% to 50% compared with the primary models (eTable 7 in Supplement 1). When we shifted participants from the SCD group who endorsed SCD only once at their last valid visit (eTable 8 in Supplement 1), the APOE 蔚4鈥�adjusted HRs were not consistently different compared with the primary models. When we incorporated all 3 of these conditions, the APOE 蔚4鈥�adjusted HRs for MCI, AD dementia, and all-cause dementia were 7.62 (95% CI,鈥�5.44-10.66), 6.02 (95% CI,鈥�3.01-12.05), and 4.44 (95% CI,鈥�2.40-8.20), respectively, values that are 1.6- to 4.5-fold larger than the main analyses (eTable 9 in Supplement 1).

To our knowledge, this was the largest study of SCD in a community sample with a longitudinal assessment of SCD. We found strong associations of SCD with the risk of incident MCI, AD, and all-cause dementia, which are consistent with previous reports suggesting that SCD may be an early predictor of objective measures of cognitive impairment^1-3,28,58 and AD pathology.^5-7 On average, SCD preceded MCI by 4.4 years, AD by 6.8 years, and all-cause dementia by 6.9 years, supporting the premise that SCD could be an early manifestation of neurodegeneration, located on a continuum between normal cognition and MCI.⁵ Also, we showed that the hazard of SCD on future cognitive decline was independent of APOE 蔚4 status. We did not find differences by sex, education, APOE 蔚4 status, or an AD PRS in the association of SCD on our outcomes. Among potential confounding factors, we found that depression was significantly associated with both SCD and the onset of cognitive impairment and dementia, but adjustment for depression did not substantially alter hazards of SCD on time to outcomes.

The international working group, the SCD Initiative, has described several characteristics of SCD (SCD+ features) that may increase the risk for future objective cognitive decline.³⁸ We incorporated several of these characteristics into our primary analyses (SCD in memory, irrespective of other domains of cognition, SCD onset over the age of 60 years, consistent SCD over time), and/or sensitivity analyses (worries associated with SCD, onset of SCD within the past 5 years, and more rigid definition of consistent SCD over time). Overall, these analyses provided validation for the SCD+ features. The primary model demonstrated HRs that were similar to those of a recent meta-analysis⁵⁹ (that combined community and clinic samples). Sensitivity analyses demonstrated HRs that were larger for worries associated with SCD and onset of SCD within the past 5 years. A sensitivity analysis that incorporated all the previously listed features demonstrated markedly larger HRs (1.6- to 4.5-fold larger HRs). Our study findings suggest that these recommendations were effective for identifying those who are more likely to convert and may be used for designing effective screening tools.

A recent meta-analysis⁵⁹ of studies evaluating the association between SCD and time to MCI and dementia found the HRs for dementia and MCI to be 1.90 (95% CI,鈥�1.52-2.36) and 1.72 (95% CI,鈥�1.18-2.52), respectively. This meta-analysis combined studies conducted in clinic, volunteer, and community cohorts.⁵⁹ Heterogeneity would be expected across these studies based on differences in recruitment sources. SCD may pose a smaller risk for conversion to MCI and dementia in community samples than in clinical samples because the baseline risk of conversion is smaller in community samples.¹⁴ In the meta-analysis,⁵⁹ moderator analyses on recruitment sources demonstrated that SCD posed significantly greater risk of developing MCI in clinical cohorts compared with community cohorts. Although there were not statistically significant differences for similar studies of dementia, results were in the same direction, but there was lack of power to detect a significant difference. That the association sizes from our community study were similar to (and actually larger in sensitivity analyses than) the meta-analysis that combined studies of varying recruitment sources speaks to the value of the SCD+ criteria to improve prediction.

In survival models adjusted for genetic risk factors of AD, SCD remained significantly associated with our outcomes. Specifically, compared with models that were not adjusted for APOE 蔚4 status, in adjusted models, the magnitude of SCD HR was slightly reduced but still significant for MCI, AD, and all-cause dementia. Further, we did not find that SCD was significantly more common among APOE 蔚4 carriers than noncarriers. The heritability of AD is 56% to鈥�79%, and approximately one-half of heritability is explained by APOE 蔚4.^30-32 Interestingly, our findings suggest that the effects of SCD and APOE 蔚4 are largely independent and that SCD does not mediate the effect of APOE 蔚4 on the incidence of AD. Further, we observed no SCD鈥壝椻��APOE 蔚4 or related 3-way interactions associated with time to MCI, AD, and all-cause dementia. Our findings are largely consistent with a meta-analysis indicating that APOE 蔚4 does not predispose individuals to developing SCD.³⁵ It provides further evidence that the 础笔翱蔚4 allele confers a similar risk for cognitive impairment regardless of whether SCD is present.⁸ Additionally, even after the APOE 蔚4 adjustment, the hazards of SCD were substantially larger than the hazards of APOE 蔚4 itself.

Given that SCD is thought to be an intermediary between normal cognition and MCI, but did not mediate the relationship between APOE 蔚4 and cognitive impairment, it seemed plausible that it may instead mediate the relationship between other (non鈥�APOE 蔚4) AD genetic risk factors and AD. For this reason, we also tested whether the AD PRS group with the APOE region excluded differed by SCD status and tested whether adding it to the CPH models would impact the hazard of SCD on time to cognitive impairment. We found that the PRS groups did not differ by SCD status, and their inclusion in the CPH models had no significant impact on the association of SCD on time to cognitive impairment. These findings should be interpreted with caution because, unlike APOE 蔚4, the AD PRS was not associated with time to any of our outcomes, probably due to the relatively small number of AD cases in the study sample.

Of the 1596 participants with SCD, only 121 progressed to cognitive impairment during the study period, and accordingly, the sensitivity (51%) and specificity (56%) of SCD were fair. This is not unique to the FHS and has been observed elsewhere.⁶⁰ Importantly, clinicians make judgments based on positive and negative predictive values of a test, not their sensitivity and specificity. By contrast, the negative predictive value of SCD was excellent (94%). For a patient who does not meet criteria for SCD, the probability of developing cognitive impairment in the future is small, and additional screening may not be warranted. For a population with a similar rate of dementia as that of the FHS, this may reduce the need for further screening of greater than 50% of the population.

This study has limitations. The rates of MCI (6.6%), AD (2.0%), and all-cause dementia (2.5%) were low. Because FHS participants have access to more health care opportunities, their burden of chronic diseases tends to be lower. The SCD questions were not administered at all visits, which could have led to missed cases or altered identification of when SCD occurred. More comprehensive scales have been developed that measure SCD with more granularity. We did not use these scales to minimize burden on participants who undergo extensive assessment beyond SCD. Two additional SCD+ features (seeking medical help, observer confirmation of cognitive decline) have been proposed, but unfortunately, they were not systematically collected in our study. Both are more readily available in clinical samples. AD biomarkers were not available to help elucidate underlying pathology, and only a small subset came to autopsy. We did conduct separate analyses for AD dementia and all-cause dementia. Although an AD clinical syndrome (ie, amnestic multidomain dementia) may be explained by AD pathology, other pathologies are commonly comorbid and may contribute to an AD clinical syndrome and all-cause dementia. Plausibly, the anatomic brain region where pathology accumulates, more than the type of pathology, may influence SCD. Lastly, the study sample was composed primarily of White, non-Hispanic participants, half of whom were college graduates, and most of whom resided in the northeast US. Generalizability to other groups may be limited.

In this cohort study including a large community sample with a longitudinal assessment of SCD and an analytic approach that embraced SCD+ features, we found significant associations between SCD and time to MCI, AD, and all-cause dementia, with large magnitudes, similar to those observed in clinical settings. Our findings provide added evidence that even in the community, SCD, particularly with the SCD+ features, is an important harbinger of future objective cognitive decline and dementia and argues for its use for early screening in the community. Further, SCD appears to be an independent risk factor for AD, beyond the risk incurred by genetics.

Accepted for Publication: April 17, 2024.

Published Online: July 3, 2024. doi:10.1001/jamapsychiatry.2024.1678

Corresponding Author: Jesse Mez, MD, MS, Department of Neurology, Boston University Chobanian & Avedisian School of Medicine, 72 E Concord St, L-522 Instructional Bldg, Boston, MA 02118 (jessemez@bu.edu).

Author Contributions: Drs Mez and Kang had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Concept and design: Kang, Mahajan, Gurnani, Lunetta, Farrer, Mez.

Acquisition, analysis, or interpretation of data: Kang, Li, Spat-Lemus, Durape, Chen, Devine, Auerbach, Ang, Sherva, Qiu, Lunetta, Au, Farrer, Mez.

Drafting of the manuscript: Kang, Mahajan, Mez.

Critical review of the manuscript for important intellectual content: Kang, Li, Spat-Lemus, Durape, Chen, Gurnani, Devine, Auerbach, Ang, Sherva, Qiu, Lunetta, Au, Farrer, Mez.

Statistical analysis: Kang, Durape, Chen, Sherva, Mez.

Obtained funding: Au, Farrer, Mez.

Administrative, technical, or material support: Gurnani, Devine, Auerbach, Ang, Mez.

Supervision: Li, Spat-Lemus, Devine, Lunetta, Farrer, Mez.

Conflict of Interest Disclosures: Dr Devine reported receiving grants from the National Institute on Aging during the conduct of the study. Dr Ang reported receiving grants from the National Institutes of Health (NIH) during the conduct of the study. Dr Au reported receiving grants from National Institute on Aging; scientific advisory board fees from Signant Health and Novo Nordisk; and consulting fees from Davos Alzheimer鈥檚 Collaborative nonprofit organization outside the submitted work. Dr Farrer reported receiving grants from the NIH during the conduct of the study. Dr Mez reported receiving grants from the NIH during the conduct of the study. No other disclosures were reported.

Funding/Support: This work was supported by the Framingham Heart Study鈥檚 National Heart, Lung, and Blood Institute contract (N01-HC-25195; HHSN268201500001I), and National Institutes of Health grants from the National Institute on Aging (R01AG008122, R01AG016495, R01AG033040, RF1AG054156, RF1AG062109, R01AG061028, U19AG068753, P30AG072978).

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.

Data Sharing Statement: See Supplement 2.

Additional Contributions: We thank the Framingham Heart Study (FHS) participants for their decades of dedication and the FHS staff for their hard work in collecting and preparing the data. No financial compensation was received for these contributions.