Maarten J. Bijlsma, Max Planck Institute for Demographic Research
Alpha Oumar Diallo , University of North Carolina at Chapel Hill
Nikkil Sudharsanan, Heidelberg University
A key part of health-decision making is estimating how proposed interventions will affect the mortality of future cohorts. These questions are typically answered using modeling studies that have three important limitations: (1) they draw estimates from multiple sources, assuming that that the effect of an intervention on mortality from one population can be transported to other countries and populations; (2) they generally require comprehensive mortality registration information, which is often unattainable in developing countries; and (3) they make strong stationarity assumptions and assume that period mortality and health conditions accurately represent the dynamic experience of an aging cohort of individuals. In this paper, we propose an alternative approach that overcomes some of these limitations using longitudinal survey data and the parametric g-formula -- an epidemiological dynamic causal inference model. Specifically, we first estimate mortality and risk factor transitions as a function of age and potential confounders from a real cohort of individuals. We then use this information to project the covariate trajectories of the cohort beyond ages observed in the data and then complete their life course by estimating future mortality as a function of these covariate trajectories. This allows us to estimate the impact of population-policies on cohort life expectancies without having to transport estimates from one context to another while also relaxing stationarity assumptions by incorporating projected cohort covariate trajectories into future predictions of mortality. We describe and demonstrate this approach using a worked example of blood pressure control in South Africa.
Presented in Session 22. Population Dynamics and Mortality