Mortality prediction models in intensive care: acute physiology and chronic health evaluation II and mortality prediction model compared.

Abstract

To compare the Acute Physiology and Chronic Health Evaluation (APACHE II) score with the Mortality Prediction Model (MPM). A prospective study. A nine-bed ICU in a 300-bed, nonteaching secondary hospital. Three hundred thirty-two consecutive, unselected adults. We found a good correlation between APACHE II and MPM; their performance expressed as area under the receiver operating characteristics curve was nearly the same. Goodness-of-fit between observed and expected occurrences was better for APACHE II than for admission MPM, which overestimated deaths. Because we evaluate patients early, often in the Emergency Department, we felt that a "lead-time bias" could explain this discrepancy. Reevaluation after initial stabilization improved the performance of the MPM model to the level of APACHE II. Our investigation indicates that both APACHE II and MPM are good predictors of hospital outcome in our population, but the level of intensive care services received before conventional ICU admission modifies accuracy of predictive models. In any study of outcome using comparative studies of classification systems, confounding biases should be measured.