QUANTIFYING THE IMPACT OF SOCIAL DETERMINANTS ON THE DEMAND FOR HEALTHCARE SERVICES

While there is a general understanding that Social Determinants of Health have an impact on health outcomes, it has been difficult to quantify this relationship and build useful predictive models. This paper describes an approach that uses transactional data from local government as measurable indicators of Social Determinants of Health (SDOH) and applies machine learning techniques to determine a quantifiable relationship between social and economic factors and the demand for healthcare services and develop a predictive model. Our study uses data from the New York City government (citizen complaints and crime statistics), census data (median household income) as indicators for SDOH. The health outcome is Emergency Room visits for asthma. The data was compiled and analyzed for zip codes within New York City. Various machine learning algorithms were attempted, and a Random Forest Regressor had the best results. We found that our regression model was able to explain 80% of the variance in ER visits, which we consider a very good result. The three most influential predictors in our model were: complaints regarding electrical issues, misdemeanor crime incidents, and median household income.