Segmenting the Population and Estimating Transition Probabilities Using Data on Health and Health-Related Social Service Needs from the US Health and Retirement Study

Background Simulation modeling is a promising tool to help policy makers and providers make evidence-based decisions when evaluating integrated care programs. The functionality of such models, however, depends on 2 prerequisites: 1) the analytical segmentation of populations to capture both health and health-related social service (HASS) needs and 2) the precise estimation of transition probabilities among the various states of need. Methods We took a validated instrument for segmenting the population by HASS needs and adapted it to the Health and Retirement Study, a nationally representative survey dataset from the US population older than 50 y. We then estimated the transition probabilities across all 10 need states and death using multistate modeling. A need state was defined as a combination of any of the 5 ordinal global impression segments and a complicating factor status. Results Kaplan-Meier survival curves, log-rank tests, and c-indices were used to assess predictive validity in relation to mortality. The Markov traces, using the estimated transition probability to replicate 2 closed cohorts, resembled the proportion of individuals per health state across subsequent waves well enough to indicate adequate fit of the estimated transition probabilities. Conclusions This article provides a population segmentation approach that incorporates HASS needs for the US population and 1-y transition probabilities across HASS need states and death. This is the first application of HASS segmentation that can estimate transitions between all 10 HASS need states, facilitating novel analysis of policy decisions related to integrated care. Implications Our results will be used as input for a simulation model that performs scenario analysis on the long-term effects of various integrated care policies on population health. Highlights We took a validated tool for segmenting the population according to health and health-related social service (HASS) needs and adapted it to the Health and Retirement Study, a nationally representative survey dataset from the US population over the age of 50 y. We estimated the 1-y transition probabilities across all 10 HASS segments and death. This is the first application of a version of this HASS segmentation tool that includes HASSs in the various need states when estimating transition probabilities. Our results will be used as input for a simulation model that performs scenario analysis on the long-term effects of various integrated care policies on population health.