Related Experiment Videos
Validation of two risk models for perioperative mortality in patients undergoing elective abdominal aortic aneurysm
Miklos D Kertai1, Ewout W Steyerberg, Eric Boersma
1Department of Cardiology, Erasmus Medical Center, Rotterdam, The Netherlands.
Abstract:
The authors validated 2 clinical risk models for perioperative mortality in patients scheduled for elective open abdominal aortic aneurysm surgery (AAA surgery). They studied 361 patients who underwent elective AAA surgery between 1991 and 1999 (validation population). Two clinical risk models were validated. The first was developed in 238 patients from Leiden University Hospital (Leiden risk model). The Leiden risk model was modified to provide predictions for the validation population based on 6 predictors including age, gender, a history of previous myocardial infarction, congestive heart failure, renal disease, and pulmonary disease. The second was a recently published simpler risk model developed in 820 patients in the UK Small Aneurysm Trial (UK risk model) and included 3 predictors (age and renal and pulmonary comorbidity). Logistic regression was used to quantify the relationship between predictors and outcome (mortality within 30 days of surgery). Validation further included the concordance statistic (c-statistic) for discriminative ability and the Hosmer-Lemeshow test for model reliability. The perioperative mortality in the validation population was 6.6% (24/361). Predictors had similar odds ratios, with particularly strong effects of congestive heart failure, pulmonary disease, and renal impairment. The Leiden risk model had reasonable good ability (c-statistic 0.72) and showed adequate calibration (chi(2)=3.3, p=0.97). It could particularly identify a low-risk group. The UK risk model did not perform well (c-statistic 0.60), showing statistically significant lack of fit (chi(2)=64.9, p<0.001). This study showed similar predictive ability of previously identified predictors for perioperative mortality. The Leiden risk model could identify a low-risk group, while the UK risk model showed a relatively poor performance. The current study supports the use of the Leiden model for preoperative risk assessment.