Objective measures of clinical performance are needed before economics or Benchmarking can successfully maximize the efficiency of the health care system. In the Intensive care unit (ICU), mortality is one of the most important clinical outcomes and different tools have been developed to estimate its probability of occurrence (Acute Physiology and Chronic health Evaluation (APACHE), the Simplified Acute Physiology Score (SAPS), and the Mortality Probability Model (MPM)). By assigning probabilities of hospital mortality to each patient, these systems classify patients by severity and are useful for the control of confounding by severity, the discussion of prognosis with patients and their families and in the evaluation of performance. However, if poor fit exists in one particular ICU, this is consistent with differences in both, either performance or patient-mix between this ICU and those used to develop the model. Case mix is one of the most important biases in health care economical evaluations and severity models are still inappropriate to fully adjust for case mix. The objectives of this research were to describe how differences in diagnostic covariate pattern mix affect model fit and to explore adjustment methods for case mix when the ratio of observed to expected deaths is used to compare the performance of a study ICU with the overall performance of other ICUs. The maximum likelihood adjustment of rate ratios and the dummy variable method of adjustment for case mix are useful tools to adjust for changes in patient mix and could be applied to compare ICU quality performance. The proportional sampling method of adjustment for patient mix is not applicable in real life situations because it fails to adjust for patient mix, especially when an ICU has a lower overall mortality ratio (attributable to a particular patient mix), than the developmental data set.