Hospital efficiency analysis depends largely on the model specifications. This study discusses the importance of the case-mix index (CMI) to homogenize the sample of inpatient discharges. It proposes a new index where they are classified by service, since it is usual to have lack of data to compute the CMI and this can influence the credibility of results. Data from the Portuguese national diagnosis-related group (DRG) database was utilized. Three different approaches are developed in this paper, based on locally convex order-m method as well as on translog functions. The first one correlates the efficiency with different inpatients weighting schemes, by using the Nadaraya-Watson method. The second approach compares different frontiers that have been computed using the different weighting schemes. Finally, by using bootstrap, the paper investigates whether the inclusion of severity/ complexity-related variables in the model statistically modifies the results. It has been shown that, under the Portuguese healthcare framework, if the model is environment corrected (which should include epidemiological and main political/ structural health reforms variables), then the severity adjustment of inpatients is pointless. The employment of an inpatient-weighting scheme, such as the CMI, may introduce significant frontier shift, thus its absence is not recommended in productivity evolution analyzes. The CMI shifts the efficiency frontier, but not the relative position of units against it (the last scenario if exogenous variables are present).
Keywords: Case-mix index; Internment services efficiency; Locally convex order-m method; Service-mix index; Translog function.