Systematic review of perioperative mortality risk prediction models for adults undergoing inpatient non-cardiac surgery

Background: Risk prediction tools can be used in the perioperative setting to identify high-risk patients who may benefit from increased surveillance and monitoring in the postoperative period, to aid shared decision-making, and to benchmark risk-adjusted hospital performance. We evaluated perioperative risk prediction tools relevant to an Australian context. Methods: A systematic review of perioperative mortality risk prediction tools used for adults undergoing inpatient noncardiac surgery, published between 2011 and 2019 (following an earlier systematic review). We searched Medline via OVID using medical subject headings consistent with the three main areas of risk, surgery and mortality/morbidity. A similar search was conducted in Embase. Tools predicting morbidity but not mortality were excluded, as were those predicting a composite outcome that did not report predictive performance for mortality separately. Tools were also excluded if they were specifically designed for use in cardiac or other highly specialized surgery, emergency surgery, paediatrics or elderly patients. Results: Literature search identified 2568 studies for screening, of which 19 studies identified 21 risk prediction tools for inclusion. Conclusion: Four tools are candidates for adapting in the Australian context, including the Surgical Mortality Probability Model (SMPM), Preoperative Score to Predict Postoperative Mortality (POSPOM), Surgical Outcome Risk Tool (SORT) and NZRISK. SORT has similar predictive performance to POSPOM, using only six variables instead of 17, contains all variables of the SMPM, and the original model developed in the UK has already been successfully adapted in New Zealand as NZRISK. Collecting the SORT and NZRISK variables in a national surgical outcomes study in Australia would present an opportunity to simultaneously investigate three of our four shortlisted models and to develop a locally valid perioperative mortality risk prediction model with high predictive performance.