Estimation of treatment effect of asthma case management using propensity score methods

Alameda County has the second highest rate of asthma hospitalizations for children and teenagers among all counties in California. For Blacks under age 18 in Oakland (Alameda County's largest city), the asthma hospitalization rate is over 1,000/100,000/year. We sought to evaluate a case management program, Oakland Kicks Asthma (OKA), to understand if a broad-based community intervention project could help reduce asthma morbidity in Oakland. In this study, we use healthcare records data from a local managed care Medicaid provider. Since the counterfactuals for participants in OKA are not realized, we employ propensity scores to estimate those factors associated with participation. We use a cross-validation DSA (Deletion Substitution Addition) algorithm to select the best predictive model of individual participation from three domains: healthcare use (e.g. emergency department visits for asthma), socio-demographics, (e.g. race) and provider and plan data (e.g. healthcare plan type). To create balance between the treatment and control groups, we performed a k:k optimal match on Mahalanobis distances calculated from the DSA selected determinants of participation with propensity scores and a caliper. Bivariate analyses among the matched sets show that participants in OKA had higher odds of preventative health behaviors including use of control medication or pulmonary testing in the six month follow up period and any flu or pneumonia vaccine in a one year follow up period (statistically significant at the 5% level); however, participants also had higher odds of any rescue medication and any urgent/emergency care in the 6 months after treatment.