• Joseph A Johnston.
• Lilly Research Laboratories, Indianapolis, IN 46285, USA. johnstonja@lilly.com
• Med Decis Making. 2005 Jul 1; 25 (4): 374-86.

ObjectiveTo evaluate the relative importance of predictors of in-hospital mortality in severe sepsis and compare the performance of generic and disease-specific mortality prediction models.MethodsThe author used data from all 826 patients receiving placebo in the Recombinant Human Activated Protein C Worldwide Evaluation in Severe Sepsis (PROWESS) trial. After a variety of clinical factors were examined for their univariate association with in-hospital mortality, logistic regression models incorporating successively more inclusive sets of predictors were created and compared. For each model, discrimination was assessed and the relative contribution of each model component to overall model explanatory power evaluated. The accuracy of using the Acute Physiology and Chronic Health Evaluation (APACHE) II score in isolation as an indicator of "high risk" was assessed by comparing model predictions from APACHE-only models to those of disease-specific models.ResultsAge, a number of laboratory values, and APACHE II score were significant univariate predictors of mortality. In multivariable models, age and laboratory values contributed the most information to model predictions; the contribution of the APACHE II score, in particular, the acute physiology component, was modest at best. A risk model including only the total APACHE II score had a c-statistic of 0.686, whereas the best performing disease-specific model had a c-statistic of 0.787. Use of the APACHE II score alone to establish high risk versus low risk resulted in misclassification of 26% of patients.ConclusionsIndividual severe sepsis patient outcomes depend on an array of clinical predictors. Models incorporating sepsis disease-specific risk factors may predict mortality more accurately than generic ICU severity measures.

19 keywords...

hide…