Midlife and late‐life population attributable fractions of risk factors for dementia in the United States: The Dementia Risk Prediction Project

Abstract

INTRODUCTION

Dementia prevalence is associated with modifiable factors. We quantified the contribution of dementia risk factors in midlife (45–64 years) and late life (≥ 65 years) in the United States.

METHODS

Data from six community‐based cohorts in the Dementia Risk Prediction Project (DRPP) were used. We estimated risk factor prevalence using nationally representative data. Cohort‐specific Cox regression models were used to estimate the association between modifiable risk factors and incident dementia in midlife and late life. Hazard ratios were pooled using meta‐analysis then used to calculate population attributable fractions (PAFs) and potential impact fractions.

RESULTS

Midlife and late‐life risk factors contributed to 22.7% and 16.5% of total dementia cases, respectively. Midlife obesity (PAF: 7.7%; 95% confidence interval [CI]: 4.9%–10.5%), lower education (PAF: 8.1%; 95% CI: 5.2%–11.1%), and late‐life physical inactivity (PAF: 10.4%; 95% CI: 6.2%–14.5%) were the greatest contributors.

DISCUSSION

Midlife and late‐life modifiable risk factors contribute to dementia risk, highlighting a need for interventions across the life course.

Highlights

• Our sample included 37,931 participants across six pooled, longitudinal US cohorts.

• We observed midlife and late‐life risk factors contributed to 22.7% and 16.5% of dementia cases, respectively.

• Midlife obesity, late‐life physical inactivity, and lower education appear to be the greatest contributors to dementia risk.

1. BACKGROUND

The number of individuals with dementia in the United States is estimated to nearly double from 5.3 million in 2019 to 10.5 million in 2050. ¹ When including a monetary value for unpaid caregiving, in addition to direct costs, the economic burden of dementia is projected to increase from $305 billion in 2020 to $1.5 trillion in 2050 within the United States. ² With a rapidly aging population, addressing modifiable risk factors is essential to prevent or delay the onset of dementia. Previous research has found that 40% of dementia cases may be prevented by eliminating 12 modifiable risk factors with the greatest impact: low education, hearing loss, hypertension, excessive alcohol consumption, obesity, smoking, depression, social isolation, physical inactivity, diabetes, air pollution, and traumatic brain injury (TBI). ³ In response to the significant burden of dementia, the US Department of Health and Human Services has set a goal of reducing dementia risk factors by 15% before 2030. ⁴

The relationships between modifiable risk factors and dementia risk vary with age at assessment of risk factors. Although midlife risk factors have been associated with a higher risk of dementia, non‐significant or even inverse associations have been found in late‐life cohorts for risk factors, including diabetes, hypertension, alcohol use, and obesity. ⁵ , ⁶ , ⁷ , ⁸ , ⁹ Additionally, the strength of the associations between low education, depression, smoking, physical inactivity, and incident dementia varies with age at the measurement of these risk factors. ¹⁰ Although the associations between modifiable risk factors and dementia risk have been characterized, few studies have calculated the associations between these risk factors across the lifespan and incident dementia. ¹¹ Previous analyses have also been limited by the sample size of existing cohorts and the lack of racial and ethnic diversity.

This study adopted a life‐course approach to estimate the population attributable fractions (PAFs) for dementia for significant mid‐ and late‐life modifiable risk factors in the United States, using data from the Dementia Risk Prediction Project (DRPP). The DRPP is a consortium of longitudinal cohort studies that includes several community‐based cohorts that enrolled minoritized and underrepresented racial and ethnic groups. We also calculated potential impact fractions (PIFs) to estimate the cases of dementia that could reasonably be prevented by reducing the prevalence of these risk factors. ⁴ These estimates may help inform clinical and public health interventions to reduce the prevalence of dementia.

2. METHODS

2.1. Data

The DRPP is a consortium of 16 international longitudinal cohorts that have pooled and harmonized data at the individual level to study risk factors associated with Alzheimer's disease and related dementias. In this analysis we used data from the six US‐based studies: Atherosclerosis Risk in Communities (ARIC) study, ¹² Cardiovascular Health Study (CHS), ¹³ Framingham Heart Study (FHS), ¹⁴ Kuakini Honolulu‐Asia Aging Study (HAAS), ¹⁵ Multi‐Ethnic Study of Atherosclerosis (MESA), ¹⁶ and the Sacramento Area Latino Study on Aging (SALSA). ¹⁷ Briefly, ARIC is a community‐based prospective cohort study that recruited a total of 15,792 mostly White and Black participants aged 45 to 64 years to examine myocardial infarction and coronary heart disease (CHD) incidence. ¹² CHS aimed to identify factors associated with CHD and stroke in older men and women within four US communities. ¹³ The Original FHS cohort was established in 1948 to investigate risk factors for cardiovascular disease within a sample of Massachusetts residents. Since then, the study has enrolled multigenerational and unrelated diverse cohorts spanning a broad age range: Offspring (enrolled from 1971 to 1975), New Offspring Spouse (NOS; enrolled from 2003 to 2005), Generation 3 (enrolled from 2002 to 2005), Omni 1 (enrolled from 1994 to 1998), and Omni 2 (enrolled from 2003 to 2005). ¹⁴ For this analysis, only data from the FHS Original, Offspring, and NOS cohorts were used. HAAS is a continuation of the Honolulu Heart Program (HHP), which followed middle‐aged Japanese American men for incident cardiovascular disease, investigating rates and risk factors associated with cognitive decline and dementia. ¹⁵ MESA studied subclinical cardiovascular disease in a diverse cohort of middle‐aged and older White, Black, Hispanic, and Chinese adults from six US communities. ¹⁶ SALSA followed the health and cognition of older Hispanic and Latino adults recruited from Sacramento County, California. ¹⁷ Baseline enrollment dates of these studies ranged from 1948 to 2003 (Table S1 in supporting information). Individuals included in the DRPP were at least 18 years old and had information on dementia ascertainment. The average follow‐up after the first visit at any age across all six US cohorts was 16.25 years.

RESEARCH IN CONTEXT

1. Systematic review: The authors conducted a comprehensive review of existing literature using original research articles and reviews to examine the associations between key modifiable risk factors and dementia in midlife and late life.

2. Interpretation: We estimated population attributable fractions and potential impact fractions for four midlife and five late‐life modifiable dementia risk factors. Our analyses provide greater support for associations found in related literature by using a diverse, longitudinal sample with individual follow‐up across both midlife and late‐life risk factor exposure.

3. Future directions: This study highlights the importance of modifiable risk factors for dementia in both midlife and late life. Future studies investigating the effect of reducing physical inactivity, obesity, and lower education across the life course may be particularly important to dementia prevention.

We excluded DRPP participants from non–US‐based cohorts and participants < 45 years of age at the time of risk factor assessment. We also excluded participants with prevalent dementia at midlife or late‐life risk factor assessment as well as those who were censored at midlife or late‐life baseline due to a lack of subsequent visits (Figure 1). The institutional review board at Northwestern University approved all study procedures for the DRPP (STU00214895).

FIGURE 1.

Analytic sample derivation.

The outcome of interest was incident all‐cause dementia on or after age 45. The ARIC dementia diagnosis was ascertained three ways: (1) International Classification of Diseases (ICD) hospitalization discharge codes or death certificate codes during follow up; (2) in‐person cognitive testing after visit 5 (2011–2013), including a complete neuropsychological battery and informant interviews, when appropriate; and (3) telephone screening using the Telephone Instrument of Cognitive Status‐modified (TICS‐m) among participants who did not attend visit 5 in person. Possible dementia cases were then adjudicated by a team of experts, using all available information (Table S2 in supporting information). ¹² CHS defined dementia status by adjudicated review of neurological, neuropsychological, and magnetic resonance imaging exam results according to Diagnostic and Statistical Manual of Mental Disorders Fourth Edition criteria. ¹⁸ A consensus dementia diagnosis was reached for HAAS participants with a low score on the Cognitive Abilities Screening Instrument (CASI) using neurologic examination, neuropsychological testing, informant interview, and brain imaging. ¹⁵ MESA used relevant ICD‐10 codes upon hospitalization or death to identify possible dementia cases. ¹⁹ SALSA participants with a low score on the Modified Mini Mental State Exam (3MS) or the Spanish and English Verbal Learning Test (SEVLT) were flagged for a dementia evaluation by a team of neurologists. ¹⁷ Participants in FHS cohorts scoring below a threshold on a Mini‐Mental State Examination (MMSE) based upon education attainment were evaluated by a review panel consisting of a neurologist and neuropsychologist for clinical diagnosis. ¹⁴ Participants with death events and an indication of dementia at the same time point were categorized as having a dementia outcome.

Using the dementia risk factors with the greatest impact from the 2020 report of the Lancet Commission, ³ we considered eight modifiable risk factors that had available data across the DRPP consortium. Risk factors were measured during midlife, defined as 45 to 64 years of age, and late life, defined as ≥ 65 years. ³ These risk factors included education level, obesity, hypertension, type 2 diabetes, smoking, physical inactivity, excessive alcohol use, and depression. Individual‐level data on each variable was harmonized by the DRPP according to national guidelines and available data (Table S3 in supporting information). ²⁰ , ²¹ Risk factors were dichotomized based on previous research. Lower education was defined as ≤ high school education versus any college or vocational‐level education. Obesity was defined as a body mass index (BMI) ≥ 30 kg/m². Hypertension was defined as systolic blood pressure ≥ 140 mmHg, diastolic blood pressure of ≥ 90 mmHg, or taking anti‐hypertensive medication, consistent with previous research. ²² Diabetes was defined as HbA1C ≥ 6.5%, fasting plasma glucose ≥ 126 mg/dL, or casual glucose ≥ 200 mg/dL and the use of anti‐diabetic medication. Smoking was categorized as current smoker versus former or never smoker. Physical inactivity was defined as zero metabolic equivalent of a task (MET) minutes/week. Excessive alcohol consumption was defined as alcohol consumption of > 112 g per week for women and > 210 g per week for men. ²³ Depression was defined as a Center for Epidemiological Studies Depression (CESD) score ≥ 16 for the 20‐item assessment, ≥ 10 for the 10‐item assessment, or ≥ 9 for the 11‐item assessment. ²⁴ , ²⁵ Data for hearing loss, social isolation, air pollution, and TBI have not been harmonized by the DRPP and, therefore, were not included in this analysis. Vision loss and low‐density lipoprotein cholesterol were new additions in the July 2024 report of the Lancet Commission and, therefore, not included in this analysis. ²⁶

We also included several covariates. At enrollment or a subsequent visit, each cohort collected self‐reported racial and/or ethnic groups and biological sex and performed genetic testing for the apolipoprotein E (APOE) genotype. Age was collected at each visit by every cohort.

2.2. Statistical analysis

We described the baseline characteristics of our sample using frequencies (percent) and means (standard deviation [SD]). We also examined baseline characteristics by cohort.

Missing data varied by variable and cohort (Table S4 in supporting information). To account for missing data, we used the R package jomo to impute missing variables and covariates 25‐fold. ²⁷ Imputed values outside of their expected range were set to the cohort‐specific variable mean.

We calculated cohort‐stratified hazard ratios (HRs) for both midlife (ARIC, FHS Original, combined FHS Offspring and NOS, and MESA) and late life (ARIC, CHS, FHS Original, combined FHS Offspring and NOS, HAAS, MESA, and SALSA) risk factors using cause‐specific Cox proportional hazard models. Models were adjusted for all modifiable risk factors as well as age, sex, race/ethnicity (White, Black, Hispanic, or Other), and APOE ε4 allele (ε2/ε2 or ε2/ε3, ε2/ε4, ε3/ε3, or ε3/ε4 or ε4/ε4). The follow‐up period was calculated beginning at each individual's first visit within the midlife or late‐life age windows until incident dementia, death, or censoring. Individuals could contribute to both the midlife and late‐life models. The validity of the proportional hazards assumption was evaluated through visual inspection of the log–log Kaplan–Meier curves. Cohort‐specific effect estimates were pooled across the 25 imputed datasets using Rubin rules. We then used the inverse variance method to pool the HRs for each risk factor across cohorts in midlife and late ‐life, separately, using random‐effects meta‐analyses via the R package meta. In sensitivity analyses, we excluded individuals who were underweight, defined as a BMI < 18.5, because very low weight may be associated with dementia. We also excluded individuals who experienced early‐onset dementia (< 65 years), as early‐onset dementia often involves different pathologies compared to late‐onset dementia. ²⁸

$$\mathrm{We}\mathrm{calculated}\mathrm{PAFs}\mathrm{using}\mathrm{Levin}:PA{F}_e=\frac{P_e\left(H{R}_e-1\right)}{\left(P_e\left[H{R}_e-1\right]+1\right)}$$

where $P_e$ is the prevalence of the risk factor in the population, and HR is the hazard ratio for each risk factor. ²⁹ We used 1000 Monte Carlo simulations to combine the uncertainty across prevalence and PAF estimates to calculate 95% confidence intervals (CIs) for PAFs. ³⁰ Each individual risk factor's PAF was weighted according to its communality to calculate combined PAF estimates for midlife and late life. ³ We used bootstrapping with 250 resamples to calculate 95% CIs for the combined PAF estimates. PIFs were calculated using the formula where $P_e^{\prime }$ represents a hypothetical proportional reduction in the prevalence of a risk factor. ²⁹ We calculated PIFs at several potential levels of risk factor reduction to examine the impact of increasingly aggressive intervention.

To obtain the midlife and late‐life population prevalence ($P_e$) of each risk factor e in the United States, we used data from the National Health and Nutrition Examination Survey (NHANES) from 2013 to 2020 (pre‐COVID). ³¹ To account for the survey design, we applied survey weights to estimate national prevalences. We used the Taylor series linearization method to estimate the standard errors of all prevalences.

All analyses were conducted in R version 4.3.0 using the jomo, dplyr, survival, and meta packages. ²⁷ , ³² , ³³ , ³⁴ Statistical significance was set a priori at α = 0.05 throughout.

3. RESULTS

Our sample included 37,931 unique participants who experienced a total of 5610 cases of incident dementia. The analysis of modifiable risk factors in midlife included 4171 incident dementia cases (from 26,336 participants, 56.9% female), and the analysis of modifiable risk factors in late life included 4690 cases (from 29,398 participants, 49.2% female; Tables 1 and S5 in supporting information). Participants in the midlife and late‐life cohorts had a median (25th–75th percentile) follow‐up of 25.5 (17.5, 30.2) and 10.5 (6.5, 17.7) years, respectively. The median age (25th–75th percentile) of dementia onset in the midlife and late‐life risk factor cohorts was 81.9 (76.7, 86.3) and 83.2 (78.7, 87.3) years, respectively. Individuals with incident dementia had higher prevalences of lower education among both the midlife (61.5% vs. 49.2%) and late‐life risk factor cohorts (61.2% vs. 54.3%) compared to individuals without dementia (Table 1).

TABLE 1.

Demographic characteristics of DRPP participants by age period of risk factor measurement and dementia status.

	Midlife		Late ‐life
	Dementia N = 4171	No dementia N = 22167	Dementia N = 4690	No dementia N = 24708
Sex
Female	2524.0 (60.5%)	11,974.0 (54.0%)	2630.0 (56.1%)	11,839.0 (47.9%)
Male	1647.0 (39.5%)	10,193.0 (46.0%)	2060.0 (43.9%)	12,869.0 (52.1%)
Age, years a	54.1 (6.1)	52.2 (5.8)	70.1 (5.9)	70.4 (5.2)
Race/ethnicity
Hispanic	18.0 (0.4%)	887.0 (4.0%)	236.0 (5.0%)	2491.0 (10.1%)
NH Black	927.0 (22.2%)	4306.0 (19.4%)	803.0 (17.1%)	3472.0 (14.1%)
NH White	3212.0 (77.0%)	16,491.0 (74.4%)	3307.0 (70.5%)	14,887.0 (60.3%)
Other	14.0 (0.3%)	483.0 (2.2%)	344.0 (7.3%)	3858.0 (15.6%)
Lower education b	2567.0 (61.5%)	10,911.0 (49.2%)	2869.0 (61.2%)	13,425.0 (54.3%)
Obesity	1054.0 (25.3%)	5761.0 (26.0%)	1180.0 (25.2%)	6583.0 (26.6%)
Hypertension c	1682.0 (40.3%)	8068.0 (36.4%)	2985.0 (63.6%)	16,242.0 (65.7%)
Diabetes	404.0 (9.7%)	2085.0 (9.4%)	805.0 (17.2%)	4281.0 (17.3%)
Current smoking	1055.0 (25.3%)	6335.0 (28.6%)	561.0 (12.0%)	2832.0 (11.5%)
Excessive alcohol d	394.0 (9.4%)	2356.0 (10.6%)	374.0 (8.0%)	2235.0 (9.0%)
Physical inactivity e	1200.0 (28.8%)	4843.0 (21.8%)	708.0 (15.1%)	3201.0 (13.0%)
Depression f	451.0 (10.8%)	2417.0 (10.9%)	455.0 (9.7%)	2014.0 (8.2%)
APOE ε4 allele
ε2/ε2 or ε2/ε3	389.0 (9.3%)	2759.0 (12.4%)	444.0 (9.5%)	2984.0 (12.1%)
ε2/ε24	107.0 (2.6%)	535.0 (2.4%)	112.0 (2.4%)	524.0 (2.1%)
ε3/ε3	2304.0 (55.2%)	14048.0 (63.4%)	2721.0 (58.0%)	16576.0 (67.1%)
ε3/ε4 or ε4/ε4	1371.0 (32.9%)	4825.0 (21.8%)	1413.0 (30.1%)	4624.0 (18.7%)

Note: Midlife defined as ages 45 to 64 years. Late life defined as ages ≥ 65 years. Percentage reported as column percentage.

Abbreviations: APOE, apolipoprotein E; CESD, Center for Epidemiological Studies Depression; DRPP, Dementia Risk Prediction Project; MET, metabolic equivalent of a task; NH‐non‐Hispanic; SD, standard deviation.

^a Mean (SD); measured at first visit within respective age window.

^b Lower education defined as ≤ high school education.

^c Hypertension was defined as systolic blood pressure ≥ 140 mmHg, diastolic blood pressure of ≥ 90 mmHg, or taking anti‐hypertensive medication.

^d Excessive alcohol defined as alcohol consumption of > 112 g per week for women and > 210 g per week for men.

^e Physical inactivity defined as zero MET minutes/week.

^f Depression defined as a CESD score ≥ 16 for the 20‐item assessment, ≥ 10 for the 11‐item assessment, or ≥ 9 for the 10‐item assessment.

Several midlife risk factors were more prevalent among participants who developed dementia compared to participants who did not, including hypertension (40.3% vs. 36.4%). Mean BMI in both mid‐ and late‐life risk factor groups was largely consistent among those who did and did not develop dementia. Physical inactivity in both mid‐ and late‐life was more prevalent in individuals who developed dementia (28.8% and 15.1%) compared to individuals who did not (21.8% and 13.0%). Depression in late life was more prevalent in individuals who developed dementia (9.7%) compared to individuals who did not develop dementia (8.2%). Participant characteristics by cohort are presented in Table S5. Continuous risk factors, such as BMI, are presented in Table S6 in supporting information.

Lower education (HR 1.2, 95% CI: 1.1–1.4), obesity (HR 1.2, 95% CI: 1.1–1.3), diabetes (HR 1.3, 95% CI: 1.0–1.7), and depression (HR 1.1, 95% CI: 1.0–1.2) measured in midlife were associated with higher risk of dementia after adjustment for age, sex, race, ethnicity, and APOE ε4 allele (Table 2). High school education or lower (HR 1.2, 95% CI: 1.2–1.4), smoking (HR 1.3, 95% CI: 1.2–1.4), diabetes (1.2, 95% CI: 1.1–1.3), physical inactivity (HR 1.2, 95% CI: 1.1–1.3), and depression (HR 1.3, 95% CI: 1.2–1.5) measured in late life were significantly associated with a higher risk of dementia after adjustment. In sensitivity analyses, restricting the analysis to those who were not underweight did not significantly alter the results (Table S7 in supporting information). Restricting the analysis to those who received dementia diagnoses after age 65 years (N = 95 with early‐onset dementia excluded) also did not meaningfully alter results (Table S8 in supporting information).

TABLE 2.

Midlife and late‐life hazard ratios, population attributable fractions, and potential impact fractions for modifiable risk factors of dementia in US DRPP cohorts based on population prevalences of risk factors in NHANES.

Dementia risk factor	P e (95% CI)	HR (95% CI)	PAF (95% CI)	Communality	PIF (95% CI)
Midlife
Lower education a	36.4 (34.7, 38.0)	1.24 (1.13, 1.37)	8.13 (5.15, 11.11)	0.59	1.2 (0.6, 1.8)
Obesity	42.6 (40.9, 44.4)	1.19 (1.11, 1.29)	7.67 (4.88, 10.47)	0.71	1.2 (0.6, 1.7)
Hypertension b	42.1 (40.3, 43.8)	1.09 (0.90, 1.31)	3.50 (−3.30, 10.30)	0.85	0.5 (−0.6, 1.7)
Diabetes	19.6 (17.7, 21.5)	1.30 (1.00, 1.68)	5.51 (1.44, 9.58)	0.65	0.8 (−0.1, 1.7)
Smoking	20.0 (18.7, 21.4)	1.15 (0.97, 1.37)	2.94 (−0.12, 6.01)	0.65	0.4 (−0.1, 1.0)
Physical inactivity c	49.1 (47.3, 50.8)	2.13 (0.84, 5.37)	35.60 (9.92, 61.27)	0.68	5.3 (−0.5, 11.2)
Excessive alcohol d	9.5 (8.4, 10.7)	1.03 (0.92, 1.15)	0.29 (−0.72, 1.30)	0.52	0.0 (−0.1, 0.2)
Depression e	8.9 (7.9, 9.9)	1.11 (1.01, 1.23)	0.98 (0.15, 1.82)	0.55	0.2 (0.0, 0.3)
Combined PAF	‐	‐	22.7 (13.4, 29.6)		‐
Late‐life
Lower education a	40.7 (38.8, 42.6)	1.24 (1.15, 1.35)	8.95 (6.34, 11.56)	0.46	1.3 (0.8, 1.9)
Obesity	38.8 (36.8, 40.8)	1.02 (0.95, 1.09)	0.61 (−2.10, 3.33)	0.48	0.1 (−0.3, 0.5)
Hypertension b	68.4 (66.4, 70.4)	1.12 (0.95, 1.32)	7.57 (−2.41, 17.55)	0.62	1.1 (−0.4, 2.7)
Diabetes	28.0 (25.5, 30.5)	1.19 (1.10, 1.29)	5.02 (3.02, 7.02)	0.64	0.8 (0.4, 1.1)
Smoking	8.3 (7.2, 9.3)	1.26 (1.15, 1.38)	2.11 (1.46, 2.76)	0.60	0.3 (0.2, 0.5)
Physical inactivity c	56.4 (54.3, 58.4)	1.21 (1.10, 1.32)	10.38 (6.24, 14.52)	0.47	0.6 (0.9, 2.3)
Excessive alcohol d	4.4 (3.5, 5.3)	1.01 (0.90, 1.12)	0.02 (−0.46, 0.51)	0.80	0.0 (−0.1, 0.1)
Depression e	7.6 (6.6, 8.7)	1.32 (1.15, 1.51)	2.38 (1.57, 3.19)	0.48	0.4 (0.1, 0.6)
Combined PAF	‐	‐	16.5 (12.5, 20.4)		‐

Note: Age‐stratified models adjusted for age, sex, race, ethnicity, and frequency of the APOE ε4 allele. Midlife is defined as ages 45 to 64 years. Late life is defined as ages ≥ 65 years.

Abbreviations: APOE, apolipoprotein E; CESD, Center for Epidemiological Studies Depression; CI, confidence interval; DRPP, Dementia Risk Prediction Project; HR, hazard ratio; MET, metabolic equivalent of a task; NHANES, National Health and Nutrition Examination Survey; PAF, population attributable fraction; PIF, potential impact fraction.

^a Lower education defined as ≤ high school education.

^b Hypertension was defined as systolic blood pressure ≥ 140 mmHg, diastolic blood pressure of ≥ 90 mmHg, or taking anti‐hypertensive medication.

^c Physical inactivity defined as zero MET minutes/week.

^d Excessive alcohol defined as alcohol consumption of > 112 g per week for women and > 210 g per week for men.

^e Depression defined as a CESD score ≥ 16 for the 20‐item assessment, ≥ 10 for the 11‐item assessment, or ≥ 9 for the 10‐item assessment.

The eight combined midlife risk factors contributed to 22.7% (95% CI: 13.4%–29.6%) of dementia cases, and the combined late‐life risk factors contributed to 16.5% (95% CI: 12.5%–20.4%) of dementia cases (Figure 2). Lower education in midlife was the greatest contributor to dementia (PAF 8.1%, 95% CI: 5.2%–11.1%), followed by obesity (PAF 7.7%, 95% CI: 4.9%–10.5%), diabetes (PAF 5.5%, 95% CI: 1.4%–9.6%), and depression (PAF 1.0%, 95% CI: 0.2%–1.8%). Physical inactivity (PAF 10.4%, 95% CI: 6.2%–14.5%), lower education (PAF 9.0%, 95% CI: 6.3%–11.6%), diabetes (PAF 5.0, 95% CI: 3.0%–7.0%), depression (PAF 2.4%, 95% CI: 1.6%–3.2%), and smoking (PAF 2.1, 95% CI: 1.5%–2.8%) were the greatest late‐life contributors to dementia.

FIGURE 2.

PAF for midlife (aged 45–64) and late‐life (aged 65+) risk factors for dementia in US cohorts of DRPP. CI, confidence interval; DRPP, Dementia Risk Prediction Project; PAF, population attributable fraction.

We also calculated the potential reduction in dementia cases, assuming a range of reductions in the prevalence of risk factors in midlife and late life (Figures 3 and 4). Using data from 2013 to 2020 on risk factor prevalence in the United States, we found that a 15% reduction in all midlife and late‐life risk factors could eliminate 176,230 (3.3%) and 227,939 (4.3%) US dementia cases, respectively. During midlife, 15% reductions in lower education (64,253, 1.2%), obesity (60,652, 1.2%), diabetes (43,544, 0.8%), and depression (7780, 0.1%) would reduce dementia cases. During late life, 15% reductions in physical inactivity (82,052, 1.6%), lower education (70,718, 1.3%), diabetes (39,673, 0.8%), depression (18,809, 0.4%), and smoking (16,687, 0.3%) would result in the greatest reductions in dementia cases (Table 2).

FIGURE 3.

Reduction in dementia cases given reduction in midlife risk factor prevalence for significantly associated risk factors.

FIGURE 4.

Reduction in dementia cases given reduction in late‐life risk factor prevalence for significantly associated risk factors.

4. DISCUSSION

We used individual‐level data pooled from six US longitudinal cohorts to estimate the contribution of mid‐ and late‐life exposure to key modifiable risk factors for dementia. We found that midlife risk factors contributed to 22.6% of dementia cases, and five late‐life risk factors contributed to 16.5% of dementia cases.

We found that lower education, obesity, diabetes, and depression in midlife contributed to approximately one quarter of dementia cases, which differed from previous studies. Overall, our findings were more modest due to not finding a significant association between physical inactivity and dementia risk in midlife. A nationally representative study found that 41.0% of US dementia cases could be eliminated with 12 modifiable risk factors, with the greatest individual contributors being late‐life physical inactivity (20.1%), midlife obesity (20.9%), and midlife hypertension (20.2%). ³⁵ Another study using data from the US Behavioral Risk Factor Surveillance Survey (BRFSS) found that 36.9% of dementia cases were associated with eight modifiable risk factors, the most prominent also being midlife physical inactivity (11.8%), low education (11.7%), and midlife obesity (17.7%). ³⁶

Although we ,  ,   did not observe a significant association between physical inactivity and dementia risk in midlife, we did find physical inactivity significantly contributed to dementia risk in late life. The strength of the evidence between physical activity and risk of dementia is generally considered to be strong. ³⁷ A recent analysis of the UK Biobank found that midlife physical activity, both leisure time and occupational, was inversely associated with the risk of dementia, while midlife sedentary behavior was associated with increased risk of dementia. ³⁸ Importantly, physical activity in late life has been shown to reduce the risk of dementia among individuals with mild cognitive impairment. ³⁹

Our analysis indicated that promoting higher education levels would reduce dementia cases, which is consistent with the existing literature. One meta‐analysis estimated that each additional year of education reduces the risk for dementia by 7%. ⁴⁰ Another US‐based longitudinal study found that lower education attainment in midlife was linked to higher incident dementia. ⁴¹

Obesity in midlife but not late life contributed to dementia risk, which is also consistent with previous studies. A meta‐analysis of longitudinal studies found obesity in midlife was associated with a 41% increased risk of dementia, whereas obesity in late life was associated with a 17% lower risk. ⁴² The differing association between obesity and dementia in midlife compared to late life may be due to reverse causation or survivor bias, as well as the association between lower weight and chronic conditions such as cancer or weight loss in the preclinical phase of dementia due to behavioral and metabolic changes. Further study of the association between BMI categories and dementia risk is also warranted.

We found diabetes in midlife and late life to be a contributor to dementia, which is consistent with prior literature. An analysis in Whitehall II found that earlier age of type 2 diabetes onset was associated with a higher risk for incident dementia. ⁶ One study by the Swedish Twin Registry found that midlife and late‐life diabetes were associated with a higher risk of Alzheimer's disease and vascular dementia, with greater risk due to midlife diabetes. ⁴³ These findings support the role of diabetes as an important risk factor through both midlife and late life.

We observed modest contributions to dementia cases from depression in midlife and late life. Our findings were opposite of the SHARE study, which found a stronger association between depression and risk of dementia in midlife than late life. ⁴⁴ One study using US data estimated that late‐life depression contributed to 6.4% of dementia cases, which was higher than our estimate. ³⁵ Another study using longitudinal data on 1.8 million Danish citizens found an HR of 2.6 for midlife depression and 2.8 for late‐life depression for dementia, respectively. ⁴⁵ Our results support the need for depression screening and treatment throughout adulthood.

We did not observe that hypertension in midlife or late life was significantly associated with dementia risk. This is contrary to most prior literature, which indicates that hypertension in midlife is associated with a higher risk of dementia, while the association between late‐life hypertension and dementia is less clear. ⁴⁶ , ⁴⁷ , ⁴⁸ It is consistent with the results from SPRINT MIND that showed intensive blood pressure lowering (< 130 mmHg) compared to standard (< 140 mmHg) did not significantly reduce the incidence of dementia (HR, 0.8; 95% CI: 0.7–1.0). However, treatment reduced the incidence of mild cognitive impairment (HR, 0.8; 95% CI: 0.7–0.9), which was not available for all cohorts in DRPP. ⁴⁹ As dementia incidence increases with age, ⁵⁰ our observed non‐significant association between midlife hypertension and dementia could be attributed to the relatively young average age of DRPP participants during midlife (52.4 years, SD 5.9). Additionally, research has indicated that a J‐ or U‐shaped association may exist between late‐life blood pressure, in which both hypotension and hypertension may increase dementia risk. ⁵¹ Therefore, the lack of association we observed between late‐life hypertension and dementia may be attributed to the use of a dichotomous definition of hypertension. Further investigation into the complex relationship between hypertension and dementia may be warranted.

Current  smoking in midlife and late life was not associated with dementia risk, which was contrary to existing literature. One study on 33,108 participants in Northern California found that midlife smoking of > 0.5 packs per day was associated with higher risk of Alzheimer's disease and vascular dementia. ⁵² Similarly, a study conducted on Japanese community dwellers found that midlife and late‐life smoking contributed to dementia risk. ⁵³

Excessive alcohol consumption in midlife or late life was not associated with dementia in this analysis. Literature on the association between alcohol consumption and dementia is mixed. A meta‐analysis found excessive alcohol consumption, defined as > 14 drinks per week, was associated with a higher risk of dementia; however, associations between low to moderate alcohol consumption appear to be protective for dementia in some studies. ⁵⁴ The authors noted that alcohol consumption is associated with several healthy lifestyle factors, including lower BMI, higher physical activity, and higher socioeconomic status. ⁵⁴

The results of our study support current public health efforts such as Healthy People 2030, which include central goals of reducing overweight and obesity, decreasing diabetes, and increasing treatment for depression. ⁵⁵ , ⁵⁶ Thus far, results of these efforts have been mixed; however, policies to expand insurance coverage and increase access to health‐care providers may promote steps toward achieving a reduction in depression. Randomized access to Medicaid coverage in Oregon reduced undiagnosed depression by nearly 50% and untreated depression by >r 60%. ⁵⁷ Expanding programs such as the National Health Service Corps may address the 37% of the US population that resides in areas of mental health practitioner shortages. ⁵⁸

Our study has several limitations. First, the application of the Levin approach using adjusted relative risks instead of the true causal relative risk has been found to result in bias due to confounding. However, the Levin formula remains in widespread use, as the bias generally has minor effects that do not appreciably change findings. ⁵⁹ The causal contribution of late‐life risk factors, in particular, may be underestimated by reverse causation. Our study suggests that causal inference methods may be useful to investigate pathways relating modifiable risk factors and dementia. Second, loss to follow‐up was greater among participants reporting Black or other race, leading to attrition bias. We used multiple imputation to account for missing data in the sample. Third, DRPP cohorts varied in enrollment years, follow‐up intervals, study designs, and in the way variables were collected; therefore, cohort effects cannot be excluded. To reduce variation between cohorts and account for variables that were collected differently, variable definitions were aligned during the extensive harmonization process. Although DRPP includes several cohorts of minoritized racial and ethnic groups, the sample is not fully representative of the US population and differs from the US general population in several important ways, including a lower prevalence of depression. Nevertheless, the DRPP cohort is more representative of the United States than any single cohort. Fourth, the estimated number of preventable cases does not account for changes in mortality due to intervention. Concurrent reductions in mortality may simultaneously increase the risk of developing dementia. Fifth, due to cohorts enrolling participants in mid and late life, survivorship bias may be present. Sixth, due to data limitations and the harmonization process, a very low threshold was used for physical activity, resulting in a conservative effect estimate, which may have influenced our findings. Seventh, important modifiable risk factors such as social isolation, TBI, air pollution, and hearing loss were not included in our analysis due to a lack of harmonized data in the DRPP. Eighth, while meta‐analyses incorporated inverse‐variance weighting, ensuring larger cohorts contributed proportionally more to the pooled estimates, the degree of between‐cohort heterogeneity varied across risk factors; some showed no evidence of heterogeneity (I ² = 0%), whereas others indicated moderate to substantial heterogeneity (I² up to 76.6%), reflecting differences in cohort size and design that may have led to imprecise estimates and, consequently, less stable PAF estimates (Figures S1 and S2 in supporting information). Finally, the communality‐weighted method we used to estimate the combined midlife and late‐life PAFs may underestimate the true combined PAFs. ⁶⁰ , ⁶¹ As such, the combined PAFs may be conservative estimates of the proportion of dementia cases attributable to modifiable risk factors.

Overall, this study examined eight modifiable risk factors to estimate the PAF of dementia associated with midlife and late‐life exposure, using longitudinal data from six US cohorts in the DRPP. Our results demonstrated that both midlife and late‐life risk factors contribute to the risk of dementia, emphasizing the importance of healthy lifestyle behaviors throughout the life course and highlighting midlife as an important target for future interventions.

CONFLICT OF INTEREST STATEMENT

Dr. Petito reports unrelated research funding from Omron Healthcare Co., Ltd.; otherwise, there are no conflicts of interest. Author disclosures are provided in the Supporting Information.

DISCLAIMER

The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

CONSENT STATEMENT

Written informed consent was received from all human subjects.

Supporting information

Supporting Information

Supporting Information

Li JM, Gauen AM, Zmora R, et al. Midlife and late‐life population attributable fractions of risk factors for dementia in the United States: The Dementia Risk Prediction Project. Alzheimer's Dement. 2026;22:e71065. 10.1002/alz.71065