Multidimensional Sleep and Mortality in Older Adults: A Machine-Learning Comparison With Other Risk Factors
Abstract Background Sleep characteristics related to duration, timing, continuity, and sleepiness are associated with mortality in older adults, but rarely considered in health recommendations. We applied machine learning to: (i) establish the predictive ability of a multidimensional self-reported sleep domain for all-cause and cardiovascular mortality in older adults relative to other established risk factors and (ii) to identify which sleep characteristics are most predictive. Methods The analytic sample includes N = 8,668 older adults (54% female) aged 65–99 years with self-reported sleep characterization and longitudinal follow-up (≤15.5 years), aggregated from three epidemiological cohorts. We used variable importance (VIMP) metrics from a random survival forest to rank the predictive abilities of 47 measures and domains to which they belong. VIMPs > 0 indicate predictive variables/domains. Results Multidimensional sleep was a significant predictor of all-cause (VIMP [99.9% confidence interval {CI}] = 0.94 [0.60, 1.29]) and cardiovascular (1.98 [1.31, 2.64]) mortality. For all-cause mortality, it ranked below that of the sociodemographic (3.94 [3.02, 4.87]), physical health (3.79 [3.01, 4.57]), and medication (1.33 [0.94, 1.73]) domains but above that of the health behaviors domain (0.22 [0.06, 0.38]). The domains were ranked similarly for cardiovascular mortality. The most predictive individual sleep characteristics across outcomes were time in bed, hours spent napping, and wake-up time. Conclusion Multidimensional sleep is an important predictor of mortality that should be considered among other more routinely used predictors. Future research should develop tools for measuring multidimensional sleep—especially those incorporating time in bed, napping, and timing—and test mechanistic pathways through which these characteristics relate to mortality.