Decision support in healthcare: determining provider influence on treatment outcomes with robust risk adjustment

Abstract

Recent advances in decision support offer a powerful toolkit to monitor, benchmark and improve healthcare quality. Since the underlying demography and prior illness of patients vary across hospitals, performance should be risk-adjusted to allow for fair comparison of providers. While previous research has proposed risk-standardisation models, their estimation still remains a non-trivial task. As a remedy, this paper utilises the least absolute shrinkage operator, LASSO for short, as a Bayesian method as part of a framework to propose a robust technique for estimating the risk-adjusted performance of hospitals. We compare our robust estimation to existing approaches (namely, the case mix approach and hierarchical generalised linear models) and demonstrate its advantages with real data. Our findings show that our method, which is robust and fully automated, can yield useful findings even with small data sets.