Is APACHE II a useful tool for clinical research? - Critical Care Science (CCS)

The population of patients admitted to the intensive care unit (ICU) is quite heterogeneous. Overall, the outcome of ICU treatment depends on the site, cause of admission, age, prior comorbidities, and acute physiological changes at admission and during the first several hours of treatment. Predictions of the in-hospital mortality of ICU patients play important roles with respect to inclusion/exclusion criteria in clinical trials, comparisons of observed mortality with predicted mortality using a score, and estimations of standardized mortality ratios in populations of critical patients. The need for such predictions has led many researchers to develop equations to calculate probabilities of associated mortality. Although prognostic scores have been used since the 1950s (such as the Apgar⁽⁾ score for newborns, which was developed by Virginia Apgar), their use for critically ill patients was established only in 1985, when Knaus et al. published the second version of the Acute Physiology and Chronic Health Evaluation (APACHE II),⁽⁾ which quickly became the most widely used prognostic index in ICUs and clinical trials worldwide. The original description of APACHE II is the most cited study in the intensive medicine literature to date.⁽⁾

The ability of a prognostic index to predict an outcome (in this case, in-hospital mortality) is assessed based on its calibration and discrimination. Calibration refers to the correspondence between expected mortality predicted using the index and observed mortality in the examined population. Typically, calibration is evaluated by comparing observed and predicted mortality in given predicted risk groups (e.g., deciles, which are used in the Hosmer-Lemeshow test).⁽⁾ The calibration of a prognostic model generally deteriorates over time due to changes in ICU admission and discharge criteria, the evolution of support, and variations in the availability and effectiveness of different treatments for particular conditions. Thus, technological and scientific developments in intensive medicine over the last 30 years have rendered APACHE II obsolete. At present, this model generally overestimates mortality in many scenarios in which it is applied. Subsequent versions of this model, such as the most recent variant, APACHE IV,⁽⁾ correct this problem, at least in part. As described by Soares et al.,⁽⁾ APACHE II should not be used as a benchmarking tool in the ICU because almost any ICU today would be considered “high performance” based on having hospital mortality much lower than that expected in 1985.

[…]