Validation of a model to predict all-cause in-hospital mortality in vascular surgical patients.

ObjectiveTo develop and validate a pre- and postoperative model of all-cause in-hospital mortality in South African vascular surgical patients.MethodsWe carried out a retrospective cohort study. A multivariate analysis using binary logistic regression was conducted on a derivation cohort using clinical, physiological and surgical data. Interaction and colinearity between covariates were investigated. The models were validated using the Homer-Lemeshow goodness-of-fit test.ResultsIndependent predictors of in-hospital mortality in the pre-operative model were: (1) age (per one-year increase) [odds ratio (OR) 1.03, 95% confidence interval (CI) 1.0-1.06), (2) creatinine > 180 micromol.l(-1) (OR 6.43, 95% CI: 3.482-11.86), (3) chronic beta-blocker therapy (OR 2.48, 95% CI: 1.38-4.48), and (4) absence of chronic statin therapy (OR 2.81, 95% CI: 1.15-6.83). Independent predictors of mortality in the postoperative model were: (1) age (per one-year increase) (OR 1.05, 95% CI: 1.02-1.09), (2) creatinine > 180 micromol.l(-1) (OR 5.08, 95% CI: 2.50-10.31), (3) surgery out of hours without statin therapy (OR 8.27, 95% CI: 3.36-20.38), (4) mean daily postoperative heart rate (HR) (OR 1.02, 95% CI: 1.0-1.04), (5) mean daily postoperative HR in the presence of a mean daily systolic blood pressure of less than 100 beats per minute or above 179 mmHg (OR 1.02, 95% CI: 1.01-1.03) and (6) mean daily postoperative HR associated with withdrawal of chronic beta-blockade (OR 1.02, 95% CI: 1.01-1.03). Both models were validated.ConclusionThe pre-operative model may predict the risk of in-hospital mortality associated with vascular surgery. The postoperative model may identify patients whose risk increases as a result of surgical or physiological factors.