How to optimize and use predictive models for postoperative pulmonary complications

Abstract

Pulmonary complications are a source of greater postoperative morbidity and mortality and longer hospital stays. Although many factors have been implicated as predictors, few models have been developed with the rigorous methodology required for clinically useful tools. In this article we attempt to describe what to look for when developing or assessing a newly proposed predictive tool and to discuss what must be taken into consideration on incorporating a model into clinical practice. Above all, we stress that we still lack evidence for the clinical and cost effectiveness of many measures proposed for reducing risk or for managing complications perioperatively. For a good predictive model to truly prove its utility in clinical decision-making, such evidence is required.