Sex differences in dementia risk and risk factors: Individual‐participant data analysis using 21 cohorts across six continents from the COSMIC consortium

Abstract Introduction Sex differences in dementia risk, and risk factor (RF) associations with dementia, remain uncertain across diverse ethno‐regional groups. Methods A total of 29,850 participants (58% women) from 21 cohorts across six continents were included in an individual participant data meta‐analysis. Sex‐specific hazard ratios (HRs), and women‐to‐men ratio of hazard ratios (RHRs) for associations between RFs and all‐cause dementia were derived from mixed‐effect Cox models. Results Incident dementia occurred in 2089 (66% women) participants over 4.6 years (median). Women had higher dementia risk (HR, 1.12 [1.02, 1.23]) than men, particularly in low‐ and lower‐middle‐income economies. Associations between longer education and former alcohol use with dementia risk (RHR, 1.01 [1.00, 1.03] per year, and 0.55 [0.38, 0.79], respectively) were stronger for men than women; otherwise, there were no discernible sex differences in other RFs. Discussion Dementia risk was higher in women than men, with possible variations by country‐level income settings, but most RFs appear to work similarly in women and men.


INTRODUCTION
The number of people living with dementia is projected to exceed 150 million by 2050 worldwide, three times the 50 million estimated in 2019. 1 There are strong social and economic imperatives to address the increasing burden associated with dementia, particularly in low-and middle-income countries (LMICs), where most people with dementia live. 1 Previous research, predominantly conducted in high-income countries, shows that women have a greater lifetime risk of developing dementia than men. [2][3][4] This is partially due to longer survival into older age, 2,3 although this may not fully account for the sex difference in dementia risk. 5 There is a growing recognition of sex as an effect modifier for several diseases as well as their risk factors (RFs), 6 but the evidence remains sparse for dementia. 6 However, many studies have examined RFs for dementia by adjusting for sex as a covariate rather than explicitly testing for sex differences, leaving the question of whether RF effects differ by sex unanswered. 5,7 The 2020 Lancet Commission Report estimated that up to 40% of dementia cases could be attributed to 12 modifiable RFs: less education, hypertension, obesity, diabetes, depression, hearing impairment, smoking, excessive alcohol consumption, physical inactivity, low social contact, traumatic brain injury, and air pollution. 8 Other common RFs, such as unfavorable lipids profile 9 and the presence of apolipoprotein E (APOE) ε4 allele, 5 are also recognized for dementia.
While women appear to be affected disproportionately, the dementia burden is also distributed unevenly by ethnic groups 10 and geographical regions. 1,2,7 Previous studies were also limited by small samples and low generalizability (across countries or continents). The burden of dementia is increasing around the world, particularly in LMICs. 11 High-quality data from LMICs remains scarce, with one crosssectional study suggesting that dementia prevention potential is higher in LMICs, given that the RFs are more common in these regions. 12 However, the sex difference in dementia risk in LMICs is less well known. 7 The present study aimed to estimate the sex-specific risk of all-cause dementia and the sex-specific association between major modifiable and non-modifiable RFs and dementia, using individual participant data from the large-scale, truly globally representative Cohort Studies of Memory in an International Consortium (COSMIC). 13

The international COSMIC consortium
Twenty-one studies of the COSMIC collaboration 14 contributed to the current analysis, comprising 45,628 participants. Details of each cohort are presented in Table S1 in supporting information.
The current analyses excluded those with dementia at study baseline (n = 2559). Table S2 in supporting information describes the dementia definitions used at baseline for each study. Participants without their sex recorded at the study baseline (n = 44), and those without any follow-up (n = 13,175), were also excluded. A total of 28.9% of participants were lost to follow-up; sex-specific attrition rates for each study were calculated and are presented in Table S3 in supporting information, and characteristics by prespecified subgroups of those individuals lost to follow-up are presented in Table S4   clinical interviews (including the use of Clinical Dementia Rating scale) and the process is described in greater detail in Table S5 in supporting information.

Outcome definition and harmonization
In cases for which the exact times of the dementia diagnosis were unknown, given the variations in follow-up intervals between cohorts, they were estimated by taking the mid-point of the interval between the previous study visit and the visit when dementia was first recorded. 16 Those without dementia recorded were censored at the end of study follow-up or at the last study visit.

Exposure variables
From the twelve RFs identified in the Lancet Commission 2020 report, 8 we were able to harmonize nine RFs in the present analyses: baseline education years (

Statistical analysis
Characteristics of women and men were summarized as number (percentage) for categorical variables and as mean (standard deviation) for continuous variables, unless they had skewed distributions when median (interquartile interval) was used.
Sex-specific Kaplan-Meier survival curves were constructed using the survminer R package to assess the dementia-free survival probability, and the log-rank test was performed to compare the difference between the two survival curves. Only age and education were available for all individuals. Missing values ( Pooled sex-specific, age-and education-adjusted HRs were calculated for the associations between each RF and the risk of all-cause dementia, using the one-stage meta-analysis approach using mixedeffect Cox regression models accounting for the study as a random effect. Similarly, women-to-men ratio of hazard ratios (RHRs), were obtained by fitting the interaction term between each RF and sex. 19 In a further sensitivity analysis, a complete case analysis was also applied.
We additionally estimated the sex-specific, age-adjusted incidence rates per 1000 person-years, using Poisson regression models.
Subgroup analyses were conducted to explore whether sex differences in dementia risk and rate varied by prespecified subgroups: age at baseline (≥ or < 80 years; given the sex difference in dementia incidence has been reported to diverge after 80 years of age 5,7,20  Additional analysis for APOE genotype examined the effect modifications on the sex differences by age and region.
Multiple-adjusted pooled estimates were also produced from mixed-effect Cox regression models, with different prespecified adjustment sets which excluded potential mediators, for each exposure.
Given the heterogeneity in incident dementia definitions, with the most common definition being DSM-based across the cohorts (15 out of 21), sensitivity analysis was conducted by excluding the cohorts not using DSM criteria for defining dementia. The overall sex difference in dementia risk, and the sex differences in RF associations with dementia were evaluated.
All analyses were performed using R version 4.1.0.

Standard protocol approvals, registrations, and patient consent
This project was approved by the University of New South Wales Human Research Ethics Committee (HC 17292). Individual contributing cohorts obtained prior ethics approvals (Table S15 in supporting information).
The mean age at baseline was 71.6 years (range: 24 to 120 years; and 72.0 years for women and 71.0 years for men; Table 1). On average, women had fewer years of education, and were more likely to have ever had depression; while men were more likely to have ever smoked, be currently consuming alcohol, and engaged in high physical activity, than women. Baseline characteristics for individual cohorts are presented in Table S16 in supporting information.

Sex differences in the rates of all-cause dementia
Kaplan-Meier survival curves indicated that men survived longer without a diagnosis of dementia than women (log-rank P-value < 0.0001; Figure 1).
The age-adjusted incidence rates (95% CIs) per 1000 person-years for dementia were 16.4 (15.2, 17.6) in women and 12.3 (11.1, 13.5) in men (Table 2). Consistently, the incidence rates were higher in women than men across all the categories of the subgroups considered.
Comparison between the subgroup categories revealed that higher incidence rates were observed among those ≥80 years than those Interestingly, when stratified by country-level income, incidence rates were the highest among women from low to LMICs, but rates were similar across country-level incomes for men (Table 2).

Sex differences in the risk of all-cause dementia
The pooled risk of all-cause dementia was higher in women than men after adjusting for age and education (HR, 1.

Sex differences in risk factor associations
In women and men, older age, diabetes, depression, hearing impairment, and APOE ε4 carriage were associated with a greater risk of dementia ( Figure 4); while more years of education, higher hip circumference, current alcohol use (vs. never), and high physical activity (vs. none to minimal) were associated with a lower risk of dementia.

Sensitivity analyses
Estimates from a complete case analysis were similar to those using imputed data (

Missing data
The variables included in the "md.pattern" function were age, edu-

DISCUSSION
In this individual participant data study from the international COSMIC consortium, the risk of all-cause dementia was higher in women than men, although not uniform across cohorts. The age-adjusted incidence rate per 1000 person-years for dementia was higher in women than men, and the greater risk in women was more pronounced in LMICs, and in Africa or South America. There was some evidence indicating that the associations for more years spent in education and former alcohol use with dementia risk appear to be stronger for men than women. observed a higher incidence in women than men, with these differences most pronounced in those of 80 years of age and over. 5,7,20 Our study found the age-adjusted rate of dementia to be the highest among low-to lower-middle-income countries, and higher in women than men.

Heterogeneity in sex-specific risk of dementia
Comparably, the 10/66 Study, with participants recruited from a number of LMICs (China, Cuba, Dominican Republic, Mexico, Peru, and Venezuela), also reported a higher age-adjusted incidence in women. 38 Women, particularly from LMICs, have not had equal educational and occupational opportunities, 5,7 and higher educational attainment and mentally stimulating occupations have been shown to be protective against dementia. 39

F I G U R E 2
Age-and education-adjusted study-specific and pooled hazard ratios for all-cause dementia by sex (women vs. men). CI, confidence interval; HR, hazard ratio After examining by birth year, which incorporated information on age and time periods, we found the sex difference in dementia risk was largely restricted to those born before 1925. While this could be a chance finding, the importance of examining dementia incidence by birth year has been previously highlighted, [42][43][44] and the influence of early-life environment and societal factors embedded in historical context need to be taken into consideration. 45 Women and men may experience parenthood 46 and financial hardship 5,7 differently, possibly mediated by the aforementioned differences in educational and occupational factors. 5,7 These different experiences may be more extreme in those born in the early 20th century, given the adverse global events which took place during that time. 42,44 Nevertheless, population brain health may be shifting around the world, 47 akin to progress in achieving sex and gender equity, 48 and increased effort in equitable cardiovascular disease prevention 21,47 and improved social welfare on a wider population level.

Sex differences in risk factor associations
The current study found similar associations between most RFs and dementia risk in men and women, and there was only moderate evidence indicating that APOE ε4 carriage was more strongly associated with dementia risk in men than women in our diverse populations. In contrast, previous studies from the United States reported a stronger association between APOE ε4 and dementia risk for women than men. 49,50 In our global dataset, we found no evidence indicating an effect modification by age or region, on the sex differences in the effect of APOE ε4 carriage on all-cause dementia. A previous meta-analysis of 27 independent studies, with data representative of non-Hispanic White individuals in North America and Europe, found that the effect of a single copy of APOE ε4 on Alzheimer's disease (AD) was similar for women and men. The study also found some evidence for effect modification by age in those with one copy of APOE ε4, such that women aged 65 to 75 years were at an increased risk of AD than men of the same age group. 51 Nevertheless, particular caution is warranted when interpreting sex-specific genetic associations, given the differential sex survival distributions and that the APOE gene may have pleiotropic effects (potentially influencing both the risk of dementia and mortality/longevity), as well as the high level of missingness for APOE genotypes, which can introduce spurious associations, even if there are no true sex differences present. 52 Another COSMIC study examined the relationship between alcohol and dementia, 53 and found that abstinence from alcohol was associated with a greater risk of all-cause dementia, similarly for women and men. Our finding that former, but not current, alcohol use is F I G U R E 3 Age-and education-adjusted pooled hazard ratios for sex (women vs. men) in association with all-cause dementia, stratified by subgroups. APOE, apolipoprotein E; CI, confidence interval; HR, hazard ratio more strongly related to dementia in men than women is intriguing and requires further study, but could be due to self-deception in self-reported alcohol use.
Notwithstanding the absence of a sex difference in most of the RFs associated with dementia studied here, which is largely inconsistent with a previous study in the Nordic populations, 5 sex differences in the prevalence of the RFs may also drive the sex differences seen in dementia risk. A COSMIC study found that early life education and adulthood occupational complexity were linked to dementia independently. 40 However, the study did not find any sex difference, which may be, at least in part, due to the exclusion of participants without lifetime occupation. 40 As previously discussed, low education and social disadvantage remain more pervasive among women than men around the world. 7 There was moderate evidence indicating more years spent in education were more protective for men than women in terms of dementia risk, with previous studies also showing corroborating findings, 54,55 and the extent to which this could be attributed to sex difference in cognitive reserve needs further investigations. 5 Given that education has been consistently linked to dementia for both women and men, this represents a significant missed opportunity in dementia prevention, given the unequal access to education in many parts of the world, particularly in girls. 12 Importantly, how early-life socioeconomic disadvantage is manifested in brain health later in life, 10 and how it interacts with cardiovascular RFs, and may be related to sex differences in dementia risk need to be better contextualized.
Sex-specific RFs, including reproductive factors and the hormonal milieu, may also be pertinent in understanding the differential contributions to the sex-specific risk of dementia. 56 These should be considered when interpreting the findings from the current study, given that the women included are likely to be post-menopausal. A recent study using UK Biobank data also found several reproductive events related to shorter cumulative exposure to endogenous estrogen were linked to a greater risk of dementia. 46 Future studies with repeated measures of sex hormones, together with reliable documentation of sex-specific factors and exogenous hormone use through the life course would be necessary to understand their contributions to the sex-specific risk of dementia.

F I G U R E 4
Age-and education-adjusted women-to-men ratio of hazard ratios for all-cause dementia by risk factor. APOE, apolipoprotein E; BMI, body mass index; CI, confidence interval; HDL, high density lipoprotein; HR, hazard ratio; LDL, low density lipoprotein; RHR, ratio of hazard ratios

Strengths and limitations
This study was strengthened by the large, ethno-regionally diverse underlying populations, with data from 18 countries across six continents, with many from traditionally underrepresented research populations and study settings, enhancing the potential generalizability in a broader geographical context. Another major strength was the harmonization of RFs across the cohorts, which allowed us to systematically examine sex differences for a comprehensive list of RFs.
Limitations of the study should be acknowledged. Given the variations in assessment intervals across the studies, the exact time of diagnosis of dementia may not be reliably estimated in each study.
While we excluded all dementia cases, as defined by the original study, at study baseline, mild cognitive impairment was not excluded as it is not uniformly recorded for all cohorts included. Hence, some prodromal dementia cases may be missed. Another limitation is the lack of examination by dementia subtypes, and because sex differences have been reported in previous literature by AD and vascular dementia, the combination of all-cause dementia may mask the emergence of some of these differences. Reverse causation in some of the risk exposures is inherent in many observational studies, and it can be challenging when the study outcome is dementia, given its insidious onset, and hence our findings do not imply causality; nevertheless, the exclusion of baseline dementia allowed us to minimize the effect of this bias. Other unmeasured factors such as air pollution, social interactions, frailty, and the use of post-menopausal hormone therapy mean residual confounding could be possible, and raise the possibility of ascertainment bias, particularly in LMICs. Further, despite our best effort, certain factors are difficult to harmonize; for example, the measure of years spent in education may mean different things depending on whether compulsory schooling is in place in a country. There was also heterogeneity in dementia definition across the cohorts. A major limitation is that several COSMIC cohorts relied on using cognitive screening tests to define dementia, rather than establishing a clinical diagnosis. This is likely to introduce potential misclassification of dementia cases. In further sensitivity analyses with cohorts using DSM criteria for defining dementia cases, the greater risk of dementia in women compared to men was no longer significant. Another notable limitation was that most of the exposures were measured in mid-life and late life, though factors involved in early life may also contribute to differential risk for women and men. Future studies should consider taking a life-course approach to study the RFs for dementia. 57 Last, we could not account for the competing risk of mortality in our analysis.

CONCLUSION
In summary, in this study of diverse global populations, a greater risk of dementia for women was observed, contributing to the accruing evidence for sex differences in dementia risk. Although there was almost no evidence of sex differences in most RFs for dementia, the excess risk in women was more pronounced in poorer countries, suggesting possible geographical variations following sex disparity in general terms.
Given the diverse socio-political, socioeconomic, and cultural contexts across the world, these findings justify ongoing efforts to support programs to improve sex and gender equity in brain health throughout the life course, particularly in the populace from underrepresented settings.

AUTHOR CONTRIBUTIONS
Conception All authors critically revised the manuscript and were involved in data interpretation.