Purpose: Within the evidence-based medicine paradigm, randomized controlled trials represent the "gold standard" to produce reliable evidence. Indeed, planning and implementing randomized controlled trials in critical care medicine presents limitations because of intrinsic and structural problems. As a consequence, observational studies still occur frequently. In these cases, propensity score (PS) (probability of receiving a treatment conditional on observed covariates) is an increasingly used technique to adjust the results. Few studies addressed the specific issue of a PS correction of repeated-measures designs.
Materials and methods: Three techniques for correcting the analysis of nonrandomized designs (matching, stratification, regression adjustment) are presented in a tutorial form and applied to a real case study: the comparison between intravenous and enteral sedative therapy in the intensive care unit setting.
Results: After showing the results before and after the use of PS, we suggest that such a tool allows to partially overcoming the bias associated with the observational nature of the study. It permits to correct the estimates for any observed covariate, while unobserved confounders cannot be controlled for.
Conclusions: Propensity score represents a useful additional tool to estimate the effects of treatments in nonrandomized studies. In the case study, an enteral sedation approach was equally effective to an intravenous regime, allowing for a lower level of sedation and spare of resources.
Copyright © 2012 Elsevier Inc. All rights reserved.