Covariate Analysis of Multistate Life Tables: Application to Women’s Contraceptive Dynamics

C. M. Suchindran, University of North Carolina at Chapel Hill
Tzy-Mey (May) Kuo, Research Triangle Institute
Helen P. Koo, Research Triangle Institute

Past research indicates that women often make changes among several contraceptive methods. Multi-state life table method is a useful tool for examining such a dynamic behavior. Currently, the method is used as a descriptive tool lacking capability for covariate analysis of data. This paper fills the gap by performing covariate analysis of multi-state life tables. We derived transition probabilities for multi-state life tables using multinomial logit models via the Generalized Estimating Equations approach to take into account the correlation among multiple observations from a given woman. We also computed summary measures from multi-state life tables. We used contraceptive history data from 1840 women to estimate models containing race, education, age, and gravidity at baseline as covariates. Results indicate that Black women switched their contraceptive methods more often than Whites. Women with less than high school education also made more changes and had more pregnancies than better educated women.