Absolute and relative risk prediction in patients candidate to isolated aortic valve replacement: should we change our mind?

Abstract

Objective: The European System for Cardiac Operative Risk Evaluation (EuroSCORE) calculator performance in 30-day outcome prediction after isolated aortic valve replacement (AVR) was evaluated to assess its absolute reliability and usefulness as selection criteria to percutaneous aortic valve implantation (PAVI).

Methods: We carried out a retrospective statistical analysis on 379 patients (group 0) consecutively submitted to isolated AVR in the past 10 years of surgical activity. We discriminated two periods of 5 years each, so we considered two subgroups of patients: group 1 (200 patients operated during 1999-2003); group 2 (179 patients operated during 2004-2008). We used receiver operating characteristics (ROC) curves for discriminatory power analysis. Model calibration was evaluated with the Hosmer-Lemeshow goodness-of-fit test and Pseudo R(2) analysis.

Results: The overall expected mortality rate at the logistic calculator was 9.37% compared with an observed 10-year mortality of 5.2% (p=0.006). Absolute risk prediction in group 1 fitted the observed outcome (p=0.24) while expected mortality in group 2 was significantly higher than observed (p=0.005). Applying threshold values used as PAVI selection criteria (logistic EuroSCORE >20 or >15), against 29% and 24.3% expected mortality rate, respectively, we registered a significant difference in the observed values (11.4%, p=0.022; 8.6%, p=0.005, respectively). The Hosmer-Lemeshow test demonstrated a lack of model fit in the overall group (p=0.019). ROC analysis revealed a sufficient discriminatory power for either total population (logistic area under curve (AUROC) 0.706; 95% confidence interval (CI): 0.604-0.809; p=0.002) and group 1 (logistic AUROC 0.752; 95% CI: 0.643-0.860; p=0.002). Group 2 showed a lack of risk stratification (logistic AUROC 0.613; 95% CI: 0.401-0.824; p=0.348).

Conclusions: EuroSCORE appears to be an invalid model in absolute and relative risk prediction for isolated AVR. On this basis, its use in selecting candidates to PAVI should be carefully weighted. Correct stratification and sufficient calibration of absolute risk estimate of high-risk patients are, therefore, mandatory in the aim of assigning those patients who show risk factors really responsible for the worst surgical outcome to new techniques. The goal should be reached by exploring the weight of each independent predictor of death in each single institution involved in PAVI procedures, evaluating local surgical results in terms of absolute risk and analysing those variables significantly affecting relative risk.