Online Program

Using Full Propensity Score Matching to Estimate Causal Effects in a Non-Experimental Study: Effects of a Nursing Intervention on Rehospitalizations of Cognitively Impaired Older Adults
*Alexandra L. Hanlon, University of Pennsylvania 
Luigi Salmaso, University of Padova, Italy 

Keywords: observational study, propensity score matching, bias reduction, nursing research, GEE, Cox regression

Background: Medical research often presents situations where an intervention cannot be randomized, leading to an imbalance in important covariates. Estimating intervention effects from observational data without compensating for the imbalance may lead to biased results.

Objective: To describe and exemplify propensity score (PS) methods, along with syntax and diagnostics implemented in R, used to balance the distribution of observed covariates in study groups, followed by a complete outcome analysis for the dataset associated with optimal bias reduction.

Methods: A quasi-experimental clinical study was conducted in 3 hospitals (n=407) to estimate the effects of an intervention on readmission among cognitively impaired older adults. Study groups were balanced using PS methods; time to first readmission, total number of readmissions and readmission days, were examined using Cox regression and Poisson GEE modeling.

Results: Model diagnostics indicated that full PS matching was associated with optimal bias reduction. Outcome modeling demonstrated that intervention was associated with significantly longer time to readmission, reduced readmissions, and fewer readmission days.