Comparison of a Vascular Study Group of New England risk prediction model with established risk prediction models of in-hospital mortality after elective abdominal aortic aneurysm repair.
Eslami, Mohammad H. MD a,*; Rybin, Denis MS b; Doros, Gheorghe PhD b; Kalish, Jeffrey A. MD b; Farber, Alik MD a; for the Vascular Study Group of New England
[Article]
Journal of Vascular Surgery.
62(5):1125-1133e2, November 2015.
(Format: HTML, PDF)
Background: A certain number of deaths may result from elective abdominal aortic aneurysm (AAA) repair due to inherent risks of operation; however, no agreement exists about which predictive model for in-hospital mortality is most accurate in predicting these events. This study developed a risk prediction model using Vascular Study Group of New England (VSGNE) data and compared it with established models.
Methods: VSGNE data (2003-2013) were queried for patients undergoing elective AAA repair by open or endovascular techniques. Clinical variables and known predictors of mortality were included in a full prediction model. Backward elimination with [alpha] = .2 was used to construct a parsimonious model. This VSGNE model was compared with established models-Medicare, Glasgow Aneurysm Score (GAS), and Vascular Governance North West (VGNW)-based on the scope of VSGNE data collection. Model fit was compared with the Vuong test. Model discrimination was compared in equally sized risk-group VSGNE terciles.
Results: The overall mortality rate for 4431 elective AAA patients was 1.4%. The discriminating ability of the VSGNE model was high (C statistic = 0.822) and corrected slightly to 0.779 after internal validation. Vuong tests yielded significant overall fit difference favoring the VSGNE model over the Medicare (C statistic = 0.769), VGNW (C statistic = 0.767), and GAS (C statistic = 0.685) models. The VGNW and Medicare models performed better than GAS in predicting mortality among risk-group terciles.
Conclusions: The VSGNE risk prediction model is best at forecasting mortality among this patient population. The Medicare and VGNW models showed good discrimination.
(C) 2015Elsevier, Inc.