National risk prediction model for elective abdominal aortic aneurysm repair

Abstract

Abstract Background Mortality results for elective abdominal aortic aneurysm (AAA) repair are published by the Vascular Society of Great Britain and Ireland. These mortality results are not currently risk-adjusted. The objective of this study was to develop a national risk prediction model for elective AAA repair. Methods Data for consecutive patients undergoing elective AAA repair from the National Vascular Database between April 2008 and March 2011 were analysed. Multiple logistic regression and backwards model selection were used for model development. The study outcome measure was in-hospital mortality. Model calibration and discrimination were assessed for all AAA repairs, and separately for open repair and endovascular aneurysm repair (EVAR) subgroups. Results There were 312 in-hospital deaths among 11 423 AAA repairs (2·7 (95 per cent confidence interval (c.i.) 2·4 to 3·0) per cent): 230 after 4940 open AAA repairs (4·7 (4·1 to 5·3) per cent) and 82 after 6483 EVARs (1·3 (1·0 to 1·6) per cent). Variables associated with in-hospital death included in the final model were: open repair, increasing age, female sex, serum creatinine level over 120 µmol/l, cardiac disease, abnormal electrocardiogram, previous aortic surgery or stent, abnormal white cell count, abnormal serum sodium level, AAA diameter and American Society of Anesthesiologists fitness grade. The area under the receiver operating characteristic (ROC) curve was 0·781 (95 per cent c.i. 0·756 to 0·806) with a bias-corrected value of 0·774. Model calibration was good (P = 0·963) based on the Hosmer–Lemeshow goodness-of-fit test, (bias-corrected) calibration curves, risk group assessment and recalibration regression. Conclusion This multivariable model for elective AAA repair can be used to risk-adjust outcome analyses and provide patient-specific estimates of in-hospital mortality risk for open AAA repair or EVAR.