|
Tianyu Shen , Australian National University
James O'Donnell, Australian National University
Demographic studies on healthy life expectancy mostly rely on the Markov assumption and suffer from the limitation that the duration of exposure to risk is not considered. There are models designed to account for duration dependency, such as the semi-Markov model and the multistate life table with duration dependency (DDMSLT). However, these models cannot be directly used on left-censored survey data as they require knowledge of time spent in the initial state, which is rarely known due to the survey design. This study proposes a flexible approach to utilize this type of survey data in a DDMSLT framework to estimate the multistate life expectancy. The approach involves dropping some left-censored observations but keeping as many as possible by the truncation of a duration length after which duration dependency is negligible. We apply this approach to older adults in the US based on the Health and Retirement Study to compute healthy life expectancy and examine the duration dependency compared to the typical multistate life table with Markov assumption. Our findings suggest that duration dependency is present in transition probabilities. However, the effect on healthy life expectancy is averaged out between the short-term states and the long-term states. As a result, the bias is minimal in the context of this study, and for the simplicity of the model, the Markov assumption is justified when calculating healthy life expectancy.
Presented in Session 105. Measuring Health, Wellbeing and Morbidity