Regency B
A Comparison of Propensity Score Methods for a Diabetes Study at a Community Health Center (303952)
Edith Kieffer, University of Michigan
Gloria Palmisano, CHASS Center
Gretchen Piatt, University of Michigan
*Brandy Sinco, University of Michigan
Michael Spencer, University of Washington
Keywords: propensity score, longitudinal analysis, community health research
Background: Diabetes patients at a community health center, who received a community health worker intervention, were compared with patients who received usual care. The treatment group had higher baseline blood sugar (HbA1c) and was younger than the comparison group.
Objective and Methods: Compare propensity score adjustment by optimal weighting and by stratification to balance the differences between the treatment and comparison groups. Evaluate these two propensity score methods with standardized mean differences, in addition to box, bar, and cloud plots before looking at changes in HbA1c over time.
Results. The stratification method balanced the covariate distributions better than optimal weighting. With the stratification method, the differences between HbA1c and age were in the same direction, but no longer significant. The graphical diagnostics confirmed similar distributions in the treatment and comparison groups after propensity score matching with stratification.
Stratification may have worked better than weighting due to only 8 variables common to the treatment and comparison groups. Longitudinal change in HbA1c remained significant after propensity score adjustment.