3.1 Participant Characteristics
Of the 372,867 dementia-free participants, the mean (SD) age at baseline was 55.61 (8.06) years old. Among them, 202,166 (54.2%) were female, and 354,601 (95.1%) had white racial/ethnic background (Table 1). Older participants were more likely to be less educated, higher in TDI, non-current smokers, to sleep too much or too less, to have hypertension, diabetes, and CVD. Female participants were less likely to be current smokers, to have sufficient physical activity, moderate alcohol intake, hypertension, CVD, and diabetes.
3.2 Sex- and age-specific modifiable risk factors of dementia
During 4,338,030 person-years (mean = 9.5 y) of follow-up, a total of 3,078 dementia cases were reported, among whom 113, 146, 360, 1,087, and 2,002 were from participants with baseline age of 40-<50, 50-<55, 55-<60, 60-<65, and >=65 y, respectively.
In the overall participants, all risk factors, except for suboptimal diet, were associated with a significant higher risk of later-life dementia (Figure 1 & Table S1). The adjusted HRs (95% CI) were generally higher for health condition risk factors, being 1.99 (1.73, 2.28) for depressive symptom, 1.93 (1.74, 2.14) for diabetes, 1.81 (1.65, 1.99) for CVD, and 1.16 (1.08, 1.25) for hypertension. Additionally, lifestyle factors were significantly associated with increased risk of dementia, with HR being 1.31 (1.17, 1.46) for current smoking, 1.29 (1.20, 1.39) for non-moderate alcohol intake, 1.16 (1.08, 1.24) for sleep duration <=6 or >=8 hrs, and 1.13 (1.06, 1.22) for insufficient physical activity. Socioeconomic risk factors, including low education level (1.14 [1.07, 1.23]) and high TDI (1.13 [1.06, 1.21]), were also related to higher risk of dementia.
We observed different associations between several risk factors and incident dementia by age groups (Figure 1). For suboptimal diet, a significantly elevated risk of dementia was mainly observed in younger participants aged 40-<50 y (HR [95%CI] = 1.86 [1.26, 2.74], P-interaction = 0.006). Non-moderate alcohol intake was associated with higher risk of dementia across almost all age groups, with the strongest association in the age group of 50-<55 y (1.90 [1.35, 2.68], P-interaction = 0.049). The associations of chronic diseases, including hypertension, CVD, and diabetes, were all stronger in younger participants than in older adults. For example, CVD was associated with a 4.2-fold hazard of dementia (4.20 [2.15, 8.22]) for the youngest participants (aged 40-< 50 y), but the relation was relatively weaker in older adults (1.64 [1.45, 1.85]). Although the statistical interactions between age groups and other risk factors were non-significant, the trend of stronger assocations in younger participants persisted (Table S1).
Overall, the composite health condition risk score demonstrated the strongest association with dementia (HR [95%CI] = 1.50 [1.44, 1.56] for per unit increment), followed by the lifestyle risk score (1.15 [1.11, 1.19]) and the socioeconomic risk score (1.14 [1.09, 1.20]) (Figure 2 & Table S2). The corresponding associations were similar in men as in women, while stronger associations were consistently observed among younger participants. For socioeconomic risk score, the HRs (95% CI) for per unit increment were 1.74 (1.28, 2.35) for participants age 40-<50 y, and 1.11 (1.04, 1.19) for participants >=65 y. For lifestyle risk score, the HRs (95% CI) were 1.56 (1.33, 1.84) for participants <50 y, and 1.11 (1.07, 1.16) for participants >=65 y. The elevated hazard of dementia associated with health condition risk score was much higher in participants <50 y (2.36 [1.90, 2.95]) than in older adults (1.37 [1.29, 1.45]).
When further adjusted for the APOE genotype, we observed similar results as in primary analysis (Table S3). After excluding participants who developed dementia or died within the first two years after baseline assessment, the magnitudes of the associations between risk factors and dementia was slightly attenuated but remained significant (Table S4). The relations remained similar when excluding participants with stroke history at baseline (Table S5). Taking mortality of other causes into consideration, the estimated HRs were not substantially affected by competing risk of mortality (Table S6). Further, we observed a significant non-linear association (P<0.001) between alcohol intake and dementia (Figure S2), confirming the validity of the definition of non-moderate alcohol intake in this study population.
3.3 Sex- and age-specific population attributable fractions (PAFs)
In the overall participants, the dementia risk were mostly attributable to lifestyle factors (PAF [95% CI]=21.90% [13.33%, 30.04%]), with the leading factors being non-moderate alcohol intake (8.73% [6.19%, 11.40%] and insufficient or excessive sleep (8.65% [4.52%, 12.44%]) (Figure S3). The corresponding PAFs for socioeconomic and health conditions risk factors were 13.39% (6.83%, 20.49%), and 14.31% (10.40%, 18.42%), respectively.
The PAFs for each individual risk factor showed different patterns across sex and age groups (Figure 3 & Table S7). For female participants, non-moderate alcohol intake was the leading factors (11.14% [6.69%, 15.59%]), followed by inadequate or excessive sleep duration (9.25% [3.07%, 15.43%]). For male participants, suboptimal sleep duration (9.13% [3.59%, 14.67%]), CVD (9.12% [7.01%,11.22%]), and education below high school (8.67% [3.19%, 14.15%]) were three primary population attributable factors for dementia. For participants with a baseline age below 50, the risk were mostly attributable to socioeconomic and lifestyle risk factors, including high deprivation (28.25% [6.54%, 49.97%]), lower education (34.67% [11.95%, 57.38%]), suboptimal diet (27.80% [11.26%, 44.35%]), and suboptimal sleep duration (25.77% [3.66%, 47.87%]). In older participants (>=65 y), the leading factors were suboptimal sleep duration (8.97% [3.22%, 14.71%]), lower education(7.11% [1.06%, 13.17%]), and CVD (6.90% [4.95%, 8.86%]).
The PAFs for each risk factor category also demonstrated a specific pattern according to sex and age. Compared to female participants, PAF for health conditions was higher in the male participants (Figure 4A & Table S8). Overall, the PAFs for these factors decreased as the age increased, with the risk attributable to socioeconomic and lifestyle factors decreasing more rapidly (Figure 4B). The PAFs of health conditions were generally constant across age groups and became dominant in older adults.