Including Covariates and Constructing Interval Estimates of Multistate Life Table Quantities Using Cross-Sectional Data: An Alternative to Sullivan's Method

Scott M. Lynch, Princeton University
J. Scott Brown, Miami University

Healthy life expectancy (HLE) is an important measure of population health. Sullivan's method is the most commonly used method for estimating HLE, because its data requirements are less stringent than those for multistate methods: Sullivan's method requires only cross-sectional mortality rates and cross-sectional health prevalence rates, whereas multistate methods require panel data. Sullivan’s method, however, is limited in its ability to produce HLE estimates for subpopulations due to limited disaggregation of cross-sectional mortality rates and small cell sizes in health surveys. We develop an alternative method that allows the inclusion of covariates in HLE estimation and the construction of interval estimates of HLE for any desired subpopulation. The method involves using Gibbs sampling to sample parameters from a hazard model, using an ecological inference model for producing transition probability matrices from the Gibbs samples, and using standard multistate equations to convert the transition probability matrices into HLE estimates.