Sex differences in modifiable dementia risk factors: Findings from the Rush Memory and Aging Project

Abstract

INTRODUCTION

We aimed to calculate population attributable fractions (PAFs) for incident dementia and examine sex differences in individuals with no cognitive impairment (NCI) or mild cognitive impairment (MCI).

METHODS

Longitudinal data from the Rush University Memory and Aging Project (MAP) were analyzed. Cox proportional‐hazards models were used to estimate covariate‐adjusted hazard ratios for incident dementia and calculate weighted PAFs within each cognitive status/sex subgroup.

RESULTS

The analytical sample comprised 1481 NCI (76.7% female) and 515 MCI (69.7% female) participants. Overall PAFs were similar for NCI (18.2%) and MCI (18.6%) subgroups, however, sex differences were evident. Males had higher PAFs than females in both NCI (42.5% vs. 25.1%) and MCI (51.5% vs. 12.4%), with differing risk factor profiles.

DISCUSSION

These findings support the notion that dementia risk is modifiable after the onset of MCI and that the number of potentially preventable dementia cases may be higher in males than in females.

Highlights

• The proportion of potentially preventable dementia cases was similar for no cognitive impairment (NCI) and mild cognitive impairment (MCI) individuals.

• For both cognitive states, a higher proportion of potentially preventable dementia cases was observed in males compared to females.

• The profiles of modifiable risk factors contributing to dementia differed between males and females.

• Lifestyle related risk factors were more prominent contributors to preventable dementia in males.

• Psychosocial risk factors, such as depression and social isolation, were important contributors in females.

1. BACKGROUND

Knowledge of dementia risk factors and treatments is advancing to meet the growing global public health challenge that dementia presents. ¹ , ² , ³ , ⁴ , ⁵ A key concept in understanding dementia prevention potential is population attributable fractions (PAFs), which estimate the proportion of dementia cases in a population that could be prevented if specific risk factors were eliminated. ⁶ , ⁷ According to the 2024 Lancet Commission into dementia prevention, intervention and care, approximately 45% of worldwide dementia cases can potentially be prevented by avoiding 14 modifiable risk factors. ⁵ These factors encompass a range of lifestyle and health‐related elements, including physical inactivity, obesity, hypertension, diabetes, smoking, excessive alcohol use, depression, social isolation, air pollution, traumatic brain injury (TBI), hearing loss, low levels of education, and, added in 2024, vision loss, and high cholesterol. ⁵ Although previous research has established their associations with variability in dementia incidence across broad population groups, a detailed knowledge of how their importance may vary across population subgroups is lacking. ⁸ , ⁹ , ¹⁰ , ¹¹ Precision medicine builds on this foundational knowledge and offers clinicians the ability to tailor early preventative interventions more effectively to each individual's unique risk profile. ¹²

Cognitive status is one characteristic that potentially moderates the importance of dementia risk factors. Individuals with mild cognitive impairment (MCI) represent a population vulnerable to further cognitive decline, with a markedly elevated risk of progressing to dementia compared to those with intact cognition. ¹³ , ¹⁴ Few epidemiological studies into dementia apply thorough clinical evaluations at study entry to determine participant inclusion, likely resulting in participant samples of both individuals with no cognitive impairment (NCI) and MCI; fewer still have determined differences between these subgroups with respect to the impact of dementia risk factors. This gap was addressed in a recent study by Wezeman and colleagues that estimated PAFs for dementia risk factors in NCI and MCI subgroups of two cohort studies in the United States (US) and Greece. ¹⁴ Results were mixed: overall PAFs were similar for both subgroups in the Greek cohort but they varied in the US cohort, with a higher PAFs among NCI than MCI. This variability highlights the importance of considering regional differences and sample characteristics when estimating PAFs, underscoring that results cannot necessarily be generalized across populations.

Sex is another patient characteristic readily attainable in clinics that may be useful in targeted dementia risk reduction efforts. Emerging evidence suggests that sex may play a crucial role in shaping dementia risk and progression. ¹² , ¹⁵ , ¹⁶ , ¹⁷ However, findings on the effect of biological sex on dementia risk are inconsistent, with some studies reporting higher age‐adjusted dementia incidence rates in women than men in certain countries. ¹⁸ , ¹⁹ Sex may also impact the risk and rate of transitions between cognitive states, such as from NCI to MCI, and from MCI to dementia. ²⁰ , ²¹ Women are often found to have a higher prevalence of known dementia risk factors, such as depression and cardiovascular conditions, which may further influence these transitions. ²² , ²³ , ²⁴

Although previous work has examined how PAFs for dementia vary by cognitive status or sex, none have estimated them while concurrently considering variability in both characteristics. Statistical adjustment for sex fails to acknowledge the heterogeneity between health factors, exposures, and outcomes between the sexes, ²⁵ , ²⁶ and assumes that the relationship between risk factors and outcomes is equal for both. Moreover, most studies on PAFs have analyzed cross‐sectional data, applying weights derived from other populations. Regional PAFs play a crucial role in informing local policy, and the accuracy of these estimates depends on the representativeness of the population‐based sample. Here, we aimed to assess sex differences in the relationship between modifiable risk factors and potentially preventable cases of dementia across NCI and MCI subgroups, utilizing a community‐based cohort study in Northeastern Illinois, United States.

2. METHODS

2.1. Data source

This study used data from the Rush University Memory and Aging Project (MAP). MAP is a longitudinal, clinical‐pathological cohort study focusing on dementia and other chronic diseases of ageing. ²⁷ Participants were recruited from residential facilities, senior and subsidized housing, church groups, and social service agencies in Northeastern Illinois, Unites States. Since its inception in 1997, clinical assessments, including cognitive performance tests, motor function evaluations, and collection of experiential and other risk factors have been conducted annually. Enrolment was open to individuals who were able and willing to sign informed consent and an Anatomical Gift Act, agreeing to donate their brain, spinal cord, and other biospecimens after death. Participants also signed a repository consent, allowing their data to be repurposed for other studies. The MAP study was approved by the Institutional Review Board of Rush University Medical Center, Chicago, Illinois, United States. Data access can be requested at www.radc.rush.edu.

2.2. Participants

Participants were included if they were aged 50 years and older, were evaluated as NCI or MCI at baseline visit, and had at least one follow‐up visit. 1996 eligible participants formed the analytical sample (Figure 1).

FIGURE 1.

Flowchart of study participant selection. MAP, Memory and Aging Project.

2.3. Sex

MAP collects information on sex as male or female. The terms “gender” and“‘sex” are, however, used interchangeably in the participant interview and clinical evaluation forms. Sex relates to biological and physiological characteristics of males and females while gender relates to socially constructed characteristics of men and women. ¹⁵ , ¹⁶ Although they are used inconsistently in existing literature, for clarity, the terms ‘male’ and ‘female’ will be used exclusively when discussing sex differences in this study.

2.4. Cognitive status

A clinical diagnosis of cognitive status is determined at each assessment through a three‐stage process: computer scoring of cognitive tests, clinical judgment by a neuropsychologist, and final diagnostic classification by a clinician. All participants undergo a standardized clinical evaluation, including 19 cognitive tests scored by computer to assess impairment across five cognitive domains. A neuropsychologist, blinded to participant demographics, reviews these scores and makes a preliminary diagnosis, which is then confirmed or revised by a clinician after comprehensive review of all available data. ²⁸ Dementia and Alzheimer's disease (AD) are diagnosed based on National Institute of Neurologic and Communicative Disorders and Stroke/AD and Related Disorders Association criteria, requiring evidence of cognitive decline and memory impairment. ²⁹ , ³⁰ MCI is diagnosed when cognitive impairment is present but does not meet the criteria for dementia. ³¹ Participants without dementia and MCI are categorized as NCI. ³² All‐cause dementia was the main outcome in this study.

RESEARCH IN CONTEXT

1. Systematic review: This study is built upon our prior systematic review examining sex differences in risk factors in the progression of mild cognitive impairment (MCI) to dementia.

2. Interpretation: Our findings reveal similar overall population attributable fractions (PAFs) for no cognitive impairment (NCI) and MCI, but males showed higher PAFs than females in both cognitive states, indicating greater potential preventability in men. Risk factor profiles differed by sex, with lifestyle factors prominent in males and psychosocial factors in females. This underscores the importance of considering sex and cognitive status in dementia prevention.

3. Future directions: Future research should investigate mechanisms behind these sex differences and examine younger and more diverse populations to refine and generalize prevention strategies.

2.5. Modifiable risk factors

Fourteen modifiable risk factors for dementia were selected for analysis, spanning sociodemographics, health conditions, and psychological factors. Their inclusion was based on the Lancet Commission update on dementia prevention in 2024 and our previous systematic review into sex differences in the associations of risk factors with progression of MCI to dementia. ³³ ,  

They were low education, smoking, excessive alcohol use, underweight, physical inactivity, diabetes, hypertension, stroke, heart attack, TBI, depression, social isolation, unmarried marital status, and vision impairment. Medical conditions such as history of hypertension, diabetes, TBI, stroke, and heart attack were obtained at each visit based on self‐report questionnaires. Diagnosis of depression was made by clinician based on clinical review at each visit. TBI was assessed using the Brain Injury Screening Questionnaire. Education, marital status, smoking, and alcohol use were self‐reported at baseline. Frequency of social activity was assessed at each visit using a six‐item questionnaire on how often the participants engaged in common types of activities involving social interaction. Physical activity was assessed using questions adapted from the 1985 National Health Survey which measures the sum of hours per week that the participants engaged in five categories of activities. Near‐vision acuity was determined with the Rosenbaum Pocket Vision Screener. Body mass index (BMI) was calculated using weight and height measurements at each visit recorded by a trained technician. A systematic review and metaanalysis found that mid‐life obesity and late‐life underweight increase the risk of cognitive impairment and dementia, while late‐life overweight and obesity are associated with a reduced risk of dementia ³⁴ ; as our study sample was aged almost 80 at baseline (Table 1), we included low BMI/underweight as a risk factor. Other factors identified through the Lancet Commission update such as diet and air pollution were not included as the data were not available in MAP. All variables were dichotomized to represent the presence (1) of absence (0) of the risk factor; cut‐points were chosen based on risk‐factor definitions used in Livingston et al. and other previous literature (Supporting Information Table S1). ³³ , ³⁵ , ³⁶

TABLE 1.

Descriptive statistics of the analyzed sample overall and stratified by sex.

Characteristic	Total sample	Male	Female
N (%)	1996	501 (25.1)	1495 (74.9)
Age, years, and mean (SD)	79.7 (7.43)	80.2 (6.8)	79.5 (7.6)
Race, N (%)
White	1875 (93.9)	479 (95.6)	1396 (93.4)
Black	100 (5.0)	18 (3.6)	82 (5.5)
Asian	9 (0.5)	1 (0.2)	8 (0.5)
Other	12 (0.6)	3 (0.6)	9 (0.6)
APOE ε4 status N (%)
Noncarrier	1540 (77.2)	392 (78.2)	1130 (75.6)
Carrier	456 (22.8)	109 (21.8)	365 (24.4)
Cognitive status at baseline, N (%)
NCI	1481 (74.2)	345 (68.9)	1136 (76.0)
MCI	515 (25.8)	156 (31.1)	359 (24.0)

Note: Data include imputed values. Missing values were present in APOE ε4 status (N = 111) and Race (N = 5).

Abbreviations: APOE, apolipoprotein E; MCI, mild cognitive impairment; NCI, no cognitive impairment; and SD, standard deviation.

2.6. Statistical analysis

Baseline characteristics were summarized using counts and percentages for categorical variables and means and standard deviations (SDs) for continuous variables. To calculate associations (hazard ratios) between risk factors and incident dementia, we used Cox proportional hazards models while adjusting for the potential confounders of age, race (White, Black or African American, Asian, and other individuals) and apolipoprotein E (APOE) ε4 status (carrier or noncarrier of an APOE ε4 allele). All dementia risk factors and confounders were included concurrently in Cox models, thereby evaluating the individual associations of each risk factor with dementia incidence while mitigating potential confounding effects. Follow‐up was until the onset of dementia or, for participants without any record of dementia, the last assessment at which they were known to be free from dementia.

PAFs for dementia were calculated for all risk factors by using the following formula, where P is the prevalence of the risk factor and HR is the hazard ratio derived from the Cox models:

$$\frac{P\left(HR-1\right)}{1+P\left(HR-1\right)}$$

Overall PAF was calculated using the following formula, where each PAF⁺ is in the set of strictly positive PAF values. Negative PAFs were omitted in the calculation of the overall PAF ensuring that the value reflects only the potential impact of risk factors with positive associations with dementia incidence:

$$1-\left[\left(1-PA{F_1}^{+}\right)\left(1-PA{F_2}^{+}\right)\text{…}\right]$$

Weighted PAF is the relative contribution of each risk factor to the overall PAF. Weighted PAF was calculated using the following formula, where $\sum PAF$ is the sum of all individual PAFs:

$$\frac{PAF}{\sum PAF}XOverallPAF$$

The overall weighted PAF was calculated using the same formula for calculating overall PAF but based on the weighted PAF for each individual risk factor:

$$1-\left[\left(1-WeightedPA{F}_1\right)\left(1-WeightedPA{F}_2\right)\text{…}\right].$$

Analyses were first conducted in the total sample, then within cognitive status subgroups, and finally by cognitive status and sex subgroups. Nonsignificant risk factors were considered in the calculation of the overall PAF as even if a factor has a weaker association with dementia, its prevalence in a population can still contribute meaningfully to overall disease burden. ¹⁴

Multiple imputation by chained equations (MICE) was used to handle the missing data (Supporting Information Table S2). This method generated fifty imputed datasets to account for the uncertainty in the imputed values. The pooled results from these datasets are presented. To assess the impact of the imputation, sensitivity analyses were conducted using only participants with complete data (Supporting Information Table S3). All data were analyzed using RStudio Statistical Software (v4.3.2; R Core Team 2023).

3. RESULTS

3.1. Sample characteristics

The analytical sample consisted of 1996 participants, of whom 74.9% were female and 93.9% were White. Males were marginally older, with an average age of 80.2 (standard deviation, SD = 6.8) years, while females had an average age of 79.5 (SD = 7.6) years (Table 1). The majority were NCI (74.2%), while 25.8% were classified as having MCI. Over 13,744 person‐years of follow‐up, 580 cases of dementia were observed, equating to an absolute risk of 29.1% and an incidence rate of 4.2 cases per 100 person‐years (Table 2). The dementia incidence rate was higher in participants with MCI compared to NCI (10.1 vs. 3.0 cases per 100 person‐years, respectively) and slightly higher in males than females (4.4 vs. 4.2 cases per 100‐person years, respectively), overall.

TABLE 2.

Descriptive dementia incidence statistics stratified by cognitive status and sex.

	Total sample			NCI			MCI
Statistic	Total	Male	Female	Total	Male	Female	Total	Male	Female
Sample size (N)	1996	501	1495	1481	345	1136	515	156	359
Incident dementia (N)	580	141	439	336	68	268	244	73	171
Absolute risk (%)	29.1	28.1	29.4	22.7	19.7	23.6	47.4	46.8	47.6
Person‐years of follow‐up (years)	13744.3	3174.0	10570.3	11317.6	2534.3	8783.3	2426.7	639.7	1787.0
Dementia incidence rate per 100 person‐years	4.2	4.4	4.2	3.0	2.7	3.1	10.1	11.4	9.6

Abbreviations: MCI, mild cognitive impairment; NCI, no cognitive impairment.

3.2. PAFs of modifiable dementia risk factors in the total sample, NCI, and MCI subgroups

Although the prevalence of dementia risk factors varied between the total sample and the NCI and MCI subgroups, in each case the least common was low education (prevalences ranging 6.2%–6.6%) and the most common was unmarried marital status (prevalences ranging 57.7%–68.2%) (Table 3). Of the 14 risk factors, in the total sample and the NCI subsample, three were positively associated with dementia risk at a statistically significant level, five had hazard ratios for dementia above 1.00 although were not statistically significant, and seven had hazard ratios at or below 1.00. In the MCI subsample, no risk factors were positively associated with dementia risk at a statistically significant level, although 11 had hazard ratios for dementia above 1.00. The overall weighted PAF for all positively associated risk factors was 22.0% in the total sample, 18.2% in the NCI subsample, and 18.6% in the MCI subsample. Although overall PAFs were similar in NCI and MCI subsamples, individual weighted PAFs varied. In the NCI subsample, weighted PAFs ranged from 0.5% (diabetes) to 4.8% (depression). In the MCI subsample, weighted PAFs ranged from 0.1% (heart attack) to 4.9% (smoking).

TABLE 3.

Weighted PAF of modifiable risk factors in the total sample, NCI, and MCI subgroups.

	OVERALL (N = 1,996)			NCI (N = 1,481)			MCI (N = 515)
Variables	HR (95% CI)	Prevalence (%)	Weighted PAF (%)	HR (95% CI)	Prevalence (%)	Weighted PAF (%)	HR (95% CI)	Prevalence (%)	Weighted PAF (%)
Excessive alcohol	0.99 (0.75 – 1.31)	11.8	–	0.92 (0.63‐ 1.35)	11.7	–	1.14 (0.74 – 1.76)	11.8	1.4
Underweight	1.07 (0.86 – 1.32)	19.0	1.2	1.04 (0.77 ‐ 1.39)	18.0	0.7	1.04 (0.76 – 1.43)	21.4	0.7
Depression	1.73 (1.28 ‐2.35) ‡	7.0	4.3	1.81 (1.22 ‐ 2.70) ‡	6.8	4.8	1.43 (0.88 – 2.33)	7.4	2.8
Diabetes	0.95 (0.71 – 1.27)	12.7	–	1.04 (0.72 ‐ 1.50)	13.1	0.5	1.00 (0.61 – 1.63)	11.7	–
Low education	1.20 (0.86 – 1.67)	6.3	1.1	1.30 (0.87 ‐ 1.96)	6.2	1.7	1.27 (0.70 – 2.28)	6.6	1.6
Heart attack	1.02 (0.75 – 1.38)	8.7	0.2	1.00 (0.66 ‐ 1.53)	8.1	–	1.01 (0.64 – 1.60)	10.3	0.1
Hypertension	0.87 (0.73 – 1.03)	53.3	–	0.96 (0.76 – 1.21)	53.1	–	0.76 (0.58 – 1.00) *	53.8	–
Social isolation	1.54 (1.19 – 1.99) ‡	11.3	5.1	1.52 (1.00 – 2.30) *	9.0	4.1	1.09 (0.78 – 1.54)	17.9	1.4
Unmarried	1.05 (0.85 – 1.29)	60.1	2.6	1.00 (0.76 ‐ 1.31)	57.7	–	0.93 (0.65 – 1.30)	68.2	–
Physical inactivity	1.08 (0.91 – 1.28)	39.9	2.8	1.11 (0.88 ‐ 1.39)	39.2	4.1	1.04 (0.79 – 1.37)	42.1	1.5
Smoking	1.04 (0.87 – 1.24)	41.9	1.5	0.98 (0.77 ‐ 1.24)	41.5	–	1.13 (0.86 – 1.49)	43.3	4.9
Stroke	1.37 (1.03 – 1.83) †	9.5	2.8	1.56 (1.07 ‐ 2.28) †	7.4	3.6	1.02 (0.65 – 1.60)	13.2	0.2
TBI	0.83 (0.58 – 1.21)	5.6	–	0.85 (0.54 ‐ 1.33)	6.4	–	1.12 (0.61 – 2.05)	6.6	0.7
Vision impairment	1.21 (0.97 – 1.51)	16.1	2.9	1.05 (0.76 ‐ 1.45)	14.2	0.6	1.27 (0.92 – 1.74)	21.6	4.8
Total weighted PAF (%)	22.0			18.2			18.6

Note: Associations were derived from Cox proportional hazard models adjusted for age, race, and APOE ε4 status. Weighted PAF values that are negative or zero were excluded when calculating the total weighted PAF.

Abbreviations: APOE, apolipoprotein E; CI, confidence interval; HR, hazard ratio; MCI, mild cognitive impairment; NCI, no cognitive impairment; PAF, population attributable fraction; and TBI, traumatic brain injury.

^* p ≤ 0.05.

^† p ≤ 0.01.

^‡ p ≤ 0.0.

3.3. PAFs of modifiable dementia risk factors in NCI and MCI subgroups for males

For males, the most common risk factors in both NCI and MCI subgroups were hypertension (prevalences ranging 47.4%–50.4%) and smoking (prevalences ranging 48.7%–50.0%), and the lowest was depression, which affected less than 4% of males (Table 4). In the NCI subgroup, stroke and depression had statistically significant positive associations with dementia risk, six risk factors were positively associated with dementia risk but not at a statistically significant level, and seven had hazard ratios at or below 1.00. In the MCI subgroup, heart attack and physical inactivity were significantly associated with dementia risk, nine risk factors had hazard ratios for dementia above 1.00 although were not statistically significant, and four had hazard ratios at or below 1.00. The overall weighted PAF for all positively associated risk factors was 42.5% in the NCI subsample, and 51.5% in the MCI subsample. In the NCI subsample, weighted PAFs ranged from 1.7% (heart attack) to 11.1% (smoking). In the MCI subsample, weighted PAFs ranged from 1.2% (unmarried) to 17.7% (smoking) (Figures 2 and 3).

TABLE 4.

Hazard ratio, prevalence, and weighted PAF of males in MAP cohort stratified by NCI and MCI subgroups.

	Overall (N = 501)			NCI (N = 345)			MCI (N = 156)
Variables	HR (95% CI)	Prevalence (%)	Weighted PAF (%)	HR (95% CI)	Prevalence (%)	Weighted PAF (%)	HR (95% CI)	Prevalence (%)	Weighted PAF (%)
Excessive alcohol	0.77 (0.50 – 1.20)	24.9	–	0.92 (0.48 – 1.78)	25.5	–	0.71 (0.36 – 1.41)	23.2	–
Underweight	1.13 (0.68 – 1.86)	14.4	1.4	0.63 (0.23 – 1.76)	12.1	–	1.45 (0.74 – 2.84)	20.0	5.4
Depression	3.31 (1.49 – 7.39) ‡	3.4	5.5	4.09 (1.05 – 15.95) *	3.2	5.4	2.91 (0.90 – 9.40)	3.9	4.5
Diabetes	1.14 (0.69 – 1.88)	2.3	1.7	1.32 (0.67 – 2.62)	19.1	3.5	1.34 (0.57 – 3.17)	12.2	2.6
Low education	1.04 (0.49 – 2.24)	5.2	0.2	0.83 (0.26 – 2.70)	4.9	–	1.38 (0.41 – 4.67)	5.8	1.4
Heart attack	1.26 (0.77 – 2.06)	14.6	2.7	1.19 (0.59 – 2.40)	15.4	1.7	2.49 (1.16 – 5.34) *	12.8	10.6
Hypertension	0.72 (0.50 – 1.05)	49.5	–	0.61 (0.35 – 1.06)	50.4	–	0.79 (0.44 – 1.41)	47.4	–
Social isolation	1.51 (0.92 – 2.50)	14.2	5.1	1.46 (0.61 – 3.51)	11.3	3.0	1.19 (0.61 – 2.29)	20.5	2.5
Unmarried	1.00 (0.67 – 1.49)	29.5	–	0.73 (0.37 – 1.43)	25.2	–	1.05 (0.57 – 1.92)	38.1	1.2
Physical inactivity	1.34 (0.93 – 1.92)	35.3	8.1	1.30 (0.76 – 2.23)	35.3	5.7	1.98 (1.10 – 3.56) *	35.3	17.0
Smoking	1.63 (1.12 – 2.38) †	49.1	17.8	1.47 (0.85 – 2.53)	48.7	11.1	1.73 (0.97 – 3.08)	50.0	17.7
Stroke	1.67 (0.94 – 2.97)	9.1	4.3	2.44 (1.11 – 5.35) *	7.8	6.0	0.95 (0.38 – 2.34)	11.8	–
TBI	0.66 (0.31 – 1.41)	8.4	–	0.56 (0.19 – 1.61)	9.0	–	1.29 (0.41 – 4.01)	7.2	1.4
Vision impairment	1.56 (0.97 – 2.49)	16.6	6.4	1.77 (0.81 – 3.87)	15.0	6.2	1.31 (0.67 – 2.55)	20.0	3.8
Total weighted PAF (%)	42.9			42.5			51.5

Note: Associations were derived from Cox proportional hazard models adjusted for age, race and APOE ε4 status. Weighted PAF values that are negative or zero are not included in the table, as they excluded when calculating the total weighted PAF.

Abbreviations: CI, confidence interval; HR, hazard ratio; MAP, Memory and Aging ProjectMCI, mild cognitive impairment; NCI, no cognitive impairment; PAF, population attributable fraction; TBI, traumatic brain injury.

^* p≤0.05.

^† p≤0.01.

^‡ p≤0.0.

FIGURE 2.

PAFs for modifiable dementia risk factors in males and females with no cognitive impairment. Outer circles represent the prevalence of each risk factor with larger, darker circles denoting higher prevalence. Gray lines connecting each outer circle to its corresponding inner circle indicate the hazard ratio for dementia. Inner circles represent the individual PAF for each factor with size proportional to PAF magnitude. PAF, population attributable fraction.

FIGURE 3.

PAFs for modifiable dementia risk factors in males and females with mild cognitive impairment. Outer circles represent the prevalence of each risk factor with larger, darker circles denoting higher prevalence. Gray lines connecting each outer circle to its corresponding inner circle indicate the hazard ratio for dementia. Inner circles represent the individual PAF for each factor with size proportional to PAF magnitude. PAF, population attributable fraction; TBI, traumatic brain injury.

3.4. PAFs of modifiable dementia risk factors in NCI and MCI subgroups for females

For females, the most common risk factors in both NCI and MCI subgroups were unmarried marital status (prevalences ranging from 67.3% to 81.3%) while the least common was TBI affecting less than 6% of females (prevalences ranging from 5.7% to 6.3%) (Table 5). Among the 14 risk factors, depression and social isolation had statistically significant positive associations with dementia risk in the NCI subgroup, while seven risk factors were positively associated with dementia risk but were not statistically significant, and six had hazard ratios at or below 1.00. In the MCI subgroup, none of the risk factors were significantly associated with dementia risk though hypertension showed a significant negative association, four had hazard ratios for dementia above 1.00, and eleven had hazard ratios at or below 1.00. The overall weighted PAF for all positively associated risk factors was 25.1% in the NCI subsample and 12.4% in the MCI subsample. In the NCI subsample, weighted PAFs ranged from 0.5% (excessive alcohol) to 5.5% (unmarried). In the MCI subsample, weighted PAFs ranged from 1.7% (low education) to 4.8% (vision impairment) (Figures 2 and 3).

TABLE 5.

Hazard ratio, prevalence and weighted PAF of females in MAP cohort stratified by NCI and MCI subgroups.

	OVERALL (N = 1,495)			NCI (N = 1,136)			MCI (N = 359)
Variables	HR (95% CI)	Prevalence (%)	Weighted PAF (%)	HR (95% CI)	Prevalence (%)	Weighted PAF (%)	HR (95% CI)	Prevalence (%)	Weighted PAF (%)
Excessive alcohol	1.27 (0.88 – 1.83)	7.4	1.8	1.08 (0.67 – 1.76)	7.6	0.5	1.60 (0.89 – 2.90)	6.8	3.7
Underweight	1.05 (0.82 – 1.33)	20.3	0.9	1.17 (0.85 – 1.60)	19.8	2.8	0.87 (0.59 – 1.27)	22.0	–
Depression	1.61 (1.15 – 2.24) †	8.2	4.3	1.71 (1.12 – 2.60) †	8.0	4.6	1.34 (0.75 – 2.38)	9.0	2.8
Diabetes	0.83 (0.57 – 1.21)	11.3	–	0.88 (0.55 – 1.40)	11.3	–	0.79 (0.41 – 1.53)	11.4	–
Low education	1.18 (0.82 ‐1.71)	6.7	1.1	1.33 (0.85 – 2.07)	6.6	1.9	1.26 (0.62 – 2.54)	7.0	1.7
Heart attack	0.94 (0.62 – 1.42)	6.7	–	0.94 (0.53 – 1.67)	6.0	–	0.69 (0.37 – 1.28)	9.2	–
Hypertension	0.93 (0.76 – 1.13)	54.5	–	1.08 (0.83 – 1.41)	53.9	3.6	0.71 (0.51 – 0.98) *	56.5	–
Social isolation	1.56 (1.14 – 2.14) †	10.3	4.9	1.63 (0.99 – 2.66) *	8.3	4.3	0.94 (0.60 – 1.46)	16.7	–
Unmarried	1.08 (0.84 – 1.39)	70.6	4.9	1.10 (0.80 – 1.50)	67.3	5.5	0.80 (0.51 – 1.25)	81.3	–
Physical inactivity	1.01 (0.83 – 1.23)	41.4	0.4	1.10 (0.85 – 1.42)	40.3	3.4	0.92 (0.66 – 1.29)	45.0	–
Smoking	0.91 (0.74 – 1.12)	39.5	–	0.88 (0.67 – 1.16)	39.3	–	0.95 (0.67 – 1.34)	40.4	–
Stroke	1.21 (0.86 – 1.70)	8.8	1.7	1.29 (0.82 – 2.04)	7.2	1.8	0.98 (0.57 – 1.69)	13.9	–
TBI	0.81 (0.53 – 1.25)	5.8	–	0.83 (0.49 – 1.40)	5.7	–	1.00 (0.46 – 2.16)	6.3	–
Vision impairment	1.08 (0.83 – 1.40)	15.9	1.1	0.94 (0.65 – 1.35)	14.0	–	1.24 (0.83 – 1.83)	22.3	4.8
Total weighted PAF (%)	19.3			25.1			12.4

Note: Associations were derived from Cox proportional hazard models adjusted for age, race, and APOE ε4 status. Weighted PAF values that are negative or zero are not included in the table, as they excluded when calculating the total weighted PAF.

Abbreviations: MAP, Memory and Aging Project; MCI, mild cognitive impairment; NCI, no cognitive impairment; PAF, population attributable fraction; HR, hazard ratio; TBI, traumatic brain injury.

^* p≤0.05.

^† p≤0.01.

4. DISCUSSION

This study explored differences in the PAFs for dementia for 14 modifiable risk factors in mostly White, older adults in Northeastern Illinois. Our analyses provide new information for the development of dementia risk‐reduction strategies for individuals diverse in cognitive status and sex. Our main findings were: (1) the proportion of potentially preventable dementia cases was not influenced by cognitive status; (2) in both NCI and MCI subgroups, the proportion of potentially preventable cases was considerably higher in males than in females; and (3) dementia risk factor profiles differed between males and females. These findings underscore the importance of considering cognitive status and sex into precision medicine approaches to dementia risk reduction. By recognizing and addressing these differences, tailored interventions can be developed to more efficiently and effectively reduce dementia risk.

Our findings suggest that the magnitude of potential dementia risk reduction remains consistent across cognitive stages but differs between males and females. Our study reported similar overall PAFs for both NCI (18.2%) and MCI (18.6%) subgroups. The similar PAFs suggest that dementia risk reduction efforts could be equally impactful across both cognitive stages. This finding partially aligns with the findings from a recent study, in which PAFs were comparable in both NCI and MCI subgroups in an analysis of a Greek cohort but differed in an analysis of a US‐based cohort. ¹⁴ This discrepancy may be attributed to differences in study design and recruitment strategy, as the US‐based study used a clinical cohort, whereas the Greek study used random population sampling and MAP used a community‐based cohort. The consistency in PAF trends between the Greek cohort and MAP, despite their geographic and demographic differences, underscores the value of diverse community and population‐based studies in informing public health policies and prevention strategies. Additionally, the MCI stage often represents the first point of contact with healthcare systems for cognitive evaluation and, given the potential impact of modifiable risk factors at this stage, may serve as a pivotal opportunity for early intervention.

Sex differences in PAFs suggest that the consequences of attending to modifiable dementia risk factors may differ between males and females. Our results indicate that more dementia may be preventable in males compared to females, particularly at the MCI stage. Although previous studies have not identified sex differences in the overall rate of conversion from MCI to dementia, ²⁰ , ²⁴ our findings suggest that males convert to dementia from MCI at a marginally higher rate than females (11.4 vs. 9.6 cases per 100 person‐years, respectively). Some of this excess risk in males may be attributable to modifiable factors, such as smoking and physical activity, which were strong contributors to PAFs in males. In contrast, females exhibited a different risk profile, with depression being the only risk factor common at both the NCI and MCI stages. This aligns with previous research where males were found to be more impacted by lifestyle‐related health conditions, while females were more affected by psychosocial factors. ³⁷ , ³⁸ , ³⁹ , ⁴⁰ The prevalence of psychosocial risk factors and their influence on dementia risk in females is substantial, potentially reflecting gender‐specific social roles and life experiences that may shape cognitive trajectories differently in females than males. ⁴¹ , ⁴² The lower number of contributing risk factors and the reduced percentage of potentially preventable dementia cases in females, particularly at the MCI stage, suggest that prevention strategies may need to be targeted at an earlier stage of life for this group. Overall, in this older cohort, addressing psychosocial factors such as depression and social isolation may be particularly effective in mitigating dementia risk among females. In contrast, for males, interventions targeting lifestyle‐related risk factors, such as promoting physical activity and smoking cessation, reinforce the notion that implementing risk reduction strategies remains beneficial even in later life.

Compared with the 45% global PAF estimated by the 2024 Lancet Commission's meta‐analysis, ⁵ our PAF in the overall sample was lower at 22%. The difference seems to be primarily driven by the lower PAFs observed in females (19.3%). In contrast, the PAF for males in our study (42.9%) was similar to the Lancet Commission's estimate, suggesting that much of the discrepancy lies in the female subgroup. One explanation may relate to the characteristics of our sample, which was older and geographically and racially homogenous, factors that may influence both the prevalence of risk factors and the extent to which they contribute to dementia. Additionally, the limited number of contributing risk factors in females relative to males, may indicate underestimation of modifiable risk, although this is unlikely to be due to missing female‐specific risk factors, since these were also not included in the Lancet Commission analysis. Emerging evidence suggests several female‐specific exposures, such as menopausal transition, reproductive history, and hormone replacement therapy are linked to amyloid accumulation, vascular ageing, and cognitive decline in females. ⁴³ , ⁴⁴ , ⁴⁵ These findings reinforce the need for sex‐stratified analyses and for future studies to incorporate sex‐specific risk exposure to more accurately estimate PAFs and inform equitable dementia prevention strategies.

The key strengths of our study include utilizing the well‐characterized MAP cohort, which allowed for an extensive examination of a wide range of modifiable dementia risk factors in both NCI and MCI subgroups. This comprehensive approach provides valuable insights into sex differences in potentially modifiable dementia risk factors. However, several limitations must also be considered. First, due to the participant sample being drawn exclusively from Northeastern Illinois, the generalizability of our findings is limited and may not be representative of other regions or countries. Second, the smaller sample size, coupled with a higher representation of females, could impact the overall applicability and potentially exaggerate or understate observed differences in PAFs between sexes. Although the higher female representation offers a detailed understanding of risk factors in this group, it may also bias the comparison between sexes. Third, despite using a longitudinal approach, establishing causal relationships between modifiable risk factors and dementia outcomes remains challenging due to the observational design. Factors such as depressive symptoms and lifestyle choices might have bidirectional associations, confounding their relationship with dementia. ⁵ , ⁴⁶ Fourth, overall PAFs were calculated using risk factors that were not necessarily associated with dementia risk at a statistically significantly level. Despite this, a similar approach has been applied elsewhere, ¹⁴ and it offers the advantage of deriving weights directly from our sample rather than relying on external data. However, the smaller sample size in some subgroups limited the statistical power of our analysis. Additionally, while PAFs represent the proportional reduction in dementia cases if risk factors were eliminated, complete elimination is unrealistic. Still, PAFs provide valuable targets for risk reduction.

In conclusion, our study highlights the benefits and underscores the importance of considering cognitive status and sex differences in dementia risk reduction strategies. These findings highlight the importance of sex‐specific and cognitive status‐specific approaches in dementia risk reduction efforts. Understanding these sex differences can facilitate the advancement of personalized medicine, where risk reduction strategies are tailored to the individual, ultimately improving the effectiveness of dementia prevention efforts across diverse populations. Future research should focus on younger cohorts, as midlife risk factors may yield different and potentially more relevant insights, particularly for females. They should also prioritize collecting female‐specific risk factors and integrate these variables into dementia risk models to accurately capture dementia risk in females. Additionally, replicating these findings in more diverse populations will be crucial to validate and refine these approaches.

CONFLICT OF INTEREST STATEMENT

The authors have no conflicts of interest to report. Author disclosures are available in the supporting information.

CONSENT STATEMENT

Each participant signed an informed consent, Anatomic Gift Act, and an RADC Repository consent allowing her data and biospecimens to be repurposed.

Supporting information

Supporting Information

Supporting Information

Martin J, Gordon EH, Reid N, Hubbard RE, Ward DD. Sex differences in modifiable dementia risk factors: Findings from the Rush Memory and Aging Project. Alzheimer's Dement. 2025;21:e70506. 10.1002/alz.70506