Documenting the optimal model fit among eleven different categorizations of self-reported sleep duration and mortality in a large population-based sample

Abstract

Past researchers have used various categorizations of sleep duration to analyze how sleep duration is associated with mortality. Here we analyze eleven categorizations of sleep duration to analyze the best model fit in relation to mortality for the U.S. population and by gender. Data from the 2004–2018 National Health Interview Survey (n = 420,037) was linked to the National Death Index through 2019 (36,574 deaths). We fit Cox Proportional Hazard models with and without covariates and used Bayesian Information Criterion (BIC) to determine the optimal model for self-reported sleep duration. Different categorizations produced vastly different substantive results. Categorizations A (≤ 4, 5, 6, 7 [ref], 8, 9, or ≥10 h) and E (≤ 5, 6, 7 [ref], 8, 9, or ≥10 h) provided the best model fit. Both of these categorizations were “J-shaped” and there was no difference between 6 and 7 h, but other reported durations were associated with higher hazards of mortality. Overall, we document how different specifications of sleep duration within the same sample may lead to different conclusions regarding the risk of mortality and that the most optimal specification tends to include more hours of sleep and have a “J-shape.” The findings of this study can help researchers, clinicians, and policymakers better understand the relationship between sleep and mortality and clarify the optimal sleep duration(s).