A methodology for measuring the relative effectiveness of healthcare services

In this paper we propose a methodology for measuring the ‘relative effectiveness’ of healthcare services (i.e. the effect of hospital care on patients) under general conditions, in which: α) a healthcare outcome underlies qualitative and quantitative observable indicators; β) we are interested in studying the simultaneous dependency of multiple outcomes on covariates (where the outcomes can also be correlated to each other); γ) the relative effectiveness is adjusted for hospital-specific covariates; δ) we hypothesise a general distribution for random disturbances and the random parameters of relative effectiveness. For this topic, a generalisation of the SURE (seemingly unrelated regression equations) multilevel model is proposed. The solutions are obtained by means of Bayesian inference methods. Since there is currently no software available to estimate this model, an SAS procedure based on Markov Chain Monte Carlo methods has been developed by the authors, in line with Goldstein & Spiegelhalter (1996, J. R. Stat. Soc. Ser. A, 159, 385–443), Spiegelhalter et al. (1996, Bayesian Using Gibbs Sampling Manual. Cambridge: MRC Biostatistic Unit, Institute of Public Health) and Albert & Chib (1997, J. Am. Stat. Assoc., 92, 916–925). In addition, a new theoretical result regarding the joint posterior distribution for the parameters is provided. The model proposed has been implemented for an effectiveness study of a selection of Lombard hospitals.