Estimating Population-Level Trends in Cardiometabolic Risk Factors Using Disparate Data Sources

• Activity Number: 613
• Type: Topic Contributed
• Date/Time: Thursday, August 4, 2011 : 8:30 AM to 10:20 AM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Mariel McKenzie Finucane*+ and Christopher Joseph Paciorek and Goodarz Danaei and Majid Ezzati
• Companies: Harvard School of Public Health and Harvard School of Public Health and Harvard School of Public Health and Imperial College London
• Keywords: Bayesian inference ; hierarchical models ; smoothing ; variance components ; out-of-sample prediction ; combining data sources

Abstract

We present a flexible Bayesian model that combines data from disparate sources to estimate population-level trends in cardiometabolic risk factors for 199 analysis countries from 1980 to 2008. The model allows for time and age nonlinearity, and it borrows strength in time, age, covariates, and within and across regional country clusters to make estimates where data are sparse. Uncertainty propagates naturally through the model, allowing inference to reflect fully all sources of variability. Our cross-validated model has good out-of-sample predictive validity, globally as well as by region, age group, levels of covariates, and extent of data missingness, hence substantiating our predictions for country-years without data. This analysis represents the largest-ever analysis of high blood pressure, serum cholesterol, and body mass index, and the first global analysis of trends. In this paper, we use blood pressure as an illustrative example of model development and validation.