3.1 Sample size and baseline characteristics
Out of the initial sample of 30,828 individuals (ELSA 10,274; HRS 20,544) who participated in the baseline survey, a total of 13,652 individuals had complete baseline data, fall status, and at least one reassessment of cognitive function (Figure 1). The ELSA cohort contributed 5,400 participants, with a mean age of 70.6 years (SD 7.2). Among these, 45.1% (2,437 out of 5,400) were men, and the median follow-up duration was 8 years (IQR, 6–8). The HRS cohort included 8,252 participants, with a mean age of 75.3 years (SD 6.7). Among the HRS participants, 40.3% (3,323 out of 8,252) were men, and the median follow-up duration was 6 years (IQR, 4–8). Baseline falls and participant covariates are detailed in Table 1.
3.2 Association between baseline falls and cognitive decline in specific domains
Table 2 presents the longitudinal findings from both cohorts and the pooled estimates of cognitive decline in specific domains. Over a follow-up of 85,000 person-years, individuals with a single fall (pooled β = -0.007; 95% CI, -0.013 to -0.002; P = 0.006) and those with multiple falls (pooled β = -0.021; 95% CI, -0.026 to -0.016; P <0.001) experienced a faster decline in global cognitive z scores compared to the reference group (no falls). These results were adjusted for the aforementioned covariates.
When examining individual cognitive domains, similar patterns were observed across memory, executive, and orientation functions. Specifically, individuals who had experienced a single fall or multiple falls exhibited a faster decline compared to the reference group. In the memory domain, individuals with a single fall showed a more rapid decline in z scores (pooled β = -0.009; 95% CI, -0.016 to -0.003; P = 0.006), as did those with multiple falls (pooled β =-0.015; 95% CI, -0.022 to -0.009; P <0.001). Similar observations were found in the executive domain, with z scores declining at a faster rate for individuals with a single fall (pooled β = -0.009; 95% CI, -0.015 to -0.003; P = 0.004) and those with multiple falls (pooled β = -0.019; 95% CI, -0.025 to -0.013; P <0.001) compared to the reference group. Furthermore, in the orientation domain, individuals with a single fall (pooled β = -0.012; 95% CI, -0.022 to -0.002; P = 0.016) and those with multiple falls (pooled β = -0.030; 95% CI, -0.040 to -0.020; P <0.001) also exhibited a more rapid decline in z scores relative to the reference group. (Table 2 and Figure 2)
3.3 Sensitivity Analyses
The first sensitivity analysis excluded participants with baseline global cognitive scores in the lowest 15%, and the results were fully consistent with the main analysis (Table S1 and Figure S1). Secondly, the sex sensitivity analyses confirmed the overall findings of the main analyses, supporting a link between falls and cognitive decline in both men and women (Table S2 and Figure S2, and Table S3 and Figure S3). Thirdly, the racial sensitivity analyses largely corroborated the main findings, suggesting an association between falls and cognitive impairment in both White and People of Color (Table S4 and Figure S4, and Table S5 and Figure S5). Fourthly, the sensitivity analysis results without multiple imputations aligned closely with the main analyses (Table S6 and Figure S6). Although some associations from the sex sensitivity analysis, the sensitivity analyses of People of Color and the sensitivity analysis without multiple imputation, particularly within the single fall group, did not reach statistical significance for individual domains, the overall trends remained consistent.
3.4 Nonresponse Analyses
Among the baseline population of 16,472 individuals, a total of 2,820 individuals (17%) were excluded from the analyses due to a loss of follow-up. This included 1,599 individuals from the ELSA cohort and 1,221 individuals from the HRS cohort. When comparing the characteristics of the included participants, it was observed that the excluded participants were older, had experienced more falls, had a higher prevalence of chronic diseases, were more likely to be smokers, reported worse self-reported general health, and had lower memory and orientation scores at baseline. These differences in characteristics between the included and excluded participants were observed in both the ELSA and HRS cohorts. For further details, please refer to Tables S7, S8, and S9.