This study was conducted to assess if neighborhood-level social determinants of health improve model performance of predicting preventable hospitalization. Using medical and pharmacy claims and neighborhood-level social determinants and the split sample method (67% training with balanced sample and 33% validation), the authors developed prospective modeling for preventable hospital use, defined as hospitalization for ambulatory care sensitive conditions (Agency for Healthcare Research and Quality Prevention Quality Indicators 90 and 92) and preventable emergency department (ED) use (based on Billing's algorithm). Performance of age-gender only or age-gender with administrative claims models were compared to models with the addition of social determinants. Adding social determinants to age-gender only models and claim history models improves model performance as measured by Brier score, C statistics, and area under the precision-recall curve for preventable ED use measures while it leads to similar performance for predicting preventable hospital use compared to models without social determinants. Adding neighborhood-level social determinants improved prediction for preventable ED use in the absence of individual-level social determinants, regardless of the availability of full administrative claims history.
Keywords: avoidable emergency department use; predictive model; preventable hospitalization; social determinants of health.