In a new Social Science and Medicine journal article, Hilltop data scientists Leigh Goetschius, PhD, Morgan Henderson, PhD, and Fei Han, PhD—with co-authors Dillon Mahmoudi, PhD, Chad Perman, MPP, Howard Haft, MD, and Ian Stockwell, PhD—present their findings on accounting for area-level social determinants of health in risk prediction. The study team investigated whether and to what extent the use of more granular area-level risk factors (e.g., Census Tract-level poverty rates) strengthens clinical predictive models relative to less granular measures (e.g., ZIP code tabulation area-level poverty rates).
The authors found that while increasing the granularity of these measures did not lead to substantial improvement in overall model predictive power, it did change which features were retained during variable selection and reduce the risk attributed to certain demographic predictors.
This study is an important contribution to ongoing debates regarding best practices for the inclusion of social determinants of health in risk prediction models.
Read the article.
Learn more about Hilltop’s predictive modeling.