Background: To compare hospital outcomes of aortic aneurysm surgery, casemix correction for preoperative variables is essential. Most of these variables can be deduced from mortality risk prediction models. Our aim was to identify the optimal set of preoperative variables associated with mortality to establish a relevant and efficient casemix model.
Methods: All patients prospectively registered between 2013 and 2016 in the Dutch Surgical Aneurysm Audit (DSAA) were included for the analysis. After multiple imputation for missing variables, predictors for mortality following univariable logistic regression were analyzed in a manual backward multivariable logistic regression model and compared with three standard mortality risk prediction models: Glasgow Aneurysm Score (GAS, mainly clinical parameters), Vascular Biochemical and Haematological Outcome Model (VBHOM, mainly laboratory parameters), and Dutch Aneurysm Score (DAS, both clinical and laboratory parameters). Discrimination and calibration were tested and considered good with a C-statistic > 0.8 and Hosmer-Lemeshow (H-L) P > 0.05.
Results: There were 12,401 patients: 9,537 (76.9%) elective patients (EAAA), 913 (7.4%) acute symptomatic patients (SAAA), and 1,951 (15.7%) patients with acute rupture (RAAA). Overall postoperative mortality was 6.5%; 1.8% after EAAA surgery, 6.6% after SAAA, and 29.6% after RAAA surgery. The optimal set of independent variables associated with mortality was a mix of clinical and laboratory parameters: gender, age, pulmonary comorbidity, operative setting, creatinine, aneurysm size, hemoglobin, Glasgow coma scale, electrocardiography, and systolic blood pressure (C-statistic 0.871). External validation overall of VBHOM, DAS, and GAS revealed C-statistics of 0.836, 0.782, and 0.761, with an H-L of 0.028, 0.00, and 0.128, respectively.
Conclusions: The optimal set of variables for casemix correction in the DSAA comprises both clinical and laboratory parameters, which can be collected easily from electronic patient files and will lead to an efficient casemix model.