Methods/Statistics
Sociodemographic, health-related, and clinical characteristics and their associations with mortality among All of Us participants compared with the United States general population Jingxuan Wang* Jingxuan Wang Erin L. Ferguson Peter Buto Ruijia Chen Anna Pederson Minhyuk Choi Deborah Blacker M. Maria Glymour
Background: The large and diverse All of Us Research Program offers tremendous opportunities for health research. However, results may not be generalizable to the US population due to the program’s targeted recruitment efforts.
Methods: We compared All of Us participants to those from the nationally representative continuous 1999-2018 National Health and Nutrition Examination Survey (NHANES) with respect to: overall mortality rates; the distribution of sociodemographic, health-related, and clinical characteristics; the association of each characteristic with mortality (estimated using Cox proportional-hazards models); and population attributable fraction (PAF) for mortality for each characteristic.
Results: All of Us participants were older, less likely to be Non-Hispanic White, had more years of education, and had a higher prevalence of major chronic conditions than NHANES. Mortality rates were generally lower for All of Us participants, especially at older ages. For example, the mortality rates for participants aged 50-60 were 4.4 per 1,000 person-years in All of Us, compared to 5.5 in NHANES. For those aged 60-70, the rates were 6.1 in All of Us and 11.7 in NHANES; mortality of participants aged 80 and above was 17.5 in All of Us and 93.0 in NHANES. The direction of associations in All of Us and NHANES matched for almost all comparisons, but differed in magnitude for some conditions, primarily clinical diagnoses. For example, in All of Us, mortality among participants with a prevalent cancer diagnosis was 2.79 (95% CI: 2.59 to 3.01) times higher than among participants without cancer; in NHANES the hazard ratio was only 1.24 (95% CI: 1.16 to 1.33). PAFs were generally higher in All of Us.
Conclusions: Predictors of mortality in All of Us do not consistently generalize to the US population. Analytical approaches are needed to address non-representativeness and mitigate potential biases associated with the selection into the All of Us cohort.
