The European System for Cardiac Operative Risk Evaluation (EuroSCORE) is not appropriate for withholding surgery in high-risk patients with aortic stenosis: a retrospective cohort study

Aortic valve replacement (AVR) is established therapy for the treatment of severe, symptomatic aortic stenosis (AS) and has been shown to relieve symptoms and significantly improve survival [1‐4]. Conventional AVR involves median sternotomy and the institution of cardiopulmonary bypass, excision of the stenotic native valve, and replacement with a biologic or mechanical prosthesis. Recently, catheter-based transfemoral and transapical approaches to aortic valve implantation have been developed for patients with severe AS that are thought to be inoperable because of advanced age or significant comorbidities, with acceptable early results [5‐13].

In these studies, patient selection for transcatheter therapy was determined largely on the basis of the European System for Cardiac Operative Risk Evaluation (EuroSCORE). The EuroSCORE is a surgical risk scoring system developed in 1999 from a multinational European database [14]. The model provides an estimate of a patient's anticipated 30-day mortality according to the patient's demographic characteristics, cardiovascular and non-cardiovasculr risk factors, and procedural variables. Its predictive accuracy has previously been validated in a variety of clinical settings [15‐17]. However, concerns have been raised about the EuroSCORE's ability to accurately predict outcomes of patients at the extremes of risk and patients undergoing valve surgery [18‐22].

In order to assess the accuracy of the prognostic risk prediction provided by EuroSCORE, we examined the operative and long-term mortality of a contemporary, real-world cohort of patients with a pre-procedural logistic EuroSCORE > 20 (which represents a predicted operative mortality risk of 20%) that have undergone conventional AVR.

The purpose of this study was to determine the early and medium-term outcomes of a contemporary cohort of patients with severe symptomatic AS undergoing AVR and to evaluate the accuracy of the EuroSCORE in predicting operative mortality. We have shown that in a tertiary care surgical practice in Canada a significant proportion of patients requiring AVR for severe symptomatic AS (16.7%) were found to have a high predicted operative mortality based on logistic EuroSCORE. We have arbitrarily chosen a logistic EuroSCORE > 20, which corresponds to a predicted in-hospital or 30-day mortality > 20%, as our cutoff of high risk based on published reports that have suggested that such patients may be considered inoperable and potentially eligible for alternative catheter-based therapy. Among the patients in our cohort with logistic EuroSCORE > 20, the mean operative mortality predicted by EuroSCORE was 38.7%. However, the actual mortality observed was 11.4%, significantly lower than that predicted by EuroSCORE (P < 0.05). In addition, the EuroSCORE appeared to overestimate operative mortality within all strata of predicted risk. The discrepancy between observed and expected mortality widened as preoperative risk determined by EuroSCORE increased. Although long-term mortality from any cause is significantly higher among patients with EuroSCORE > 20 (log rank P = 0.0001), approximately 60% are alive at five years after operation. Similar findings were recently reported by Brown et al [22]. However, unlike our study, no data are given on longitudinal outcomes.

The EuroSCORE predictive model for cardiac surgery has been extensively studied and found to accurately predict outcomes in a variety of jurisdictions and clinical practices [14‐16, 30]. The initial additive EuroSCORE model was simple and reproducible and it could readily be used at the patient's bedside for rapid assessment of operative risk. However, what it gained in parsimony, interpretability, and ease of use, it lost in accuracy as it was found to have a tendency to over-predict mortality in low-risk patients and under-predict mortality in high-risk patients [31]. The later-appearing logistic EuroSCORE, utilizing the complete regression equation of the multivariate model to predict mortality risk, was more stable at the extremes of risk and had greater predictive accuracy [26]. However, we have shown that the prognostic performance of the logistic EuroSCORE is poor in patients undergoing AVR, and progressively worsens as risk increases. Similar findings were reported by others in octogenarians undergoing valve operations [21] and in patients undergoing coronary artery bypass graft surgery [32].

It has been shown that subtle inter-institutional differences in the definitions of variables, in the sensitivity of risk factor screening tools, and in the potential presence of various degrees of interdependence and collinearity among covariates assumed to be independent may profoundly affect the performance of a risk prediction model such as the EuroSCORE [33]. However, these influences are likely to be minor in our study given the availability of a nuanced dataset at our institution in which patient information on over 200 variables was prospectively collected at the time of surgery with minimal retrospective chart review, allowing us to adhere to the EuroSCORE definitions of variables as faithfully as possible and to score patients accurately. In no instance was risk factor status imputed secondary to incomplete or missing data fields.

Recent reports suggest that 30–40% of patients in which AVR is indicated are denied surgical treatment, mostly secondary to advanced age and left ventricular dysfunction [34, 35]. Less invasive, catheter-based approaches to relieving the obstruction to left ventricular outflow in severe AS were developed in an attempt to target the subset of patients who were not felt to be candidates for conventional open AVR in order to potentially favorably alter their dismal prognosis. Initially, transcatheter AVR was offered to patients on compassionate grounds and was largely reserved for moribund patients with end-stage AS in whom risk factors were felt to be prohibitive for traditional surgery. As experience with transcatheter aortic valve implantation increased with attendant improvements in feasibility, safety, and efficacy profiles, patient selection criteria were expanded and investigators are increasingly relying on EuroSCORE to determine eligibility for catheter-based intervention on the aortic valve. Recent reports evaluating retrograde transfemoral and transapical aortic valve implantation have included patients with mean logistic EuroSCORE 11–35 [6‐8, 10‐13]. In these studies, 30-day mortality rates range from 7 to 22% and the incidence of stroke ranges from 0 to 10%. However, these studies are limited by small numbers of patients and lack long-term data. In our study, the incidence of early stroke was 5.1% in the subgroup of patients with logistic EuroSCORE > 20; 30-day mortality in this high-risk group of patients was 11.4%. Furthermore, five-year survival was acceptable at nearly 60% despite a predicted operative mortality of approximately 40% in this cohort of patients. Taken further, this very same cohort of patients could have been potentially denied life-prolonging operation under current EuroSCORE criteria. However, this study is not a comparison of the effectiveness of conventional AVR and transcatheter aortic valve implantation, nor is it an analysis of the relative merits of one approach over the other. Importantly, our study does not attempt to question the value of important novel approaches to valvular heart disease but questions the validity of using EuroSCORE as a predictive model for high-risk patients with severe AS and suggests that caution should be exercised when using the EuroSCORE risk model to make important management decisions in this complex patient population. This is particularly important when one attempts to make inter-institutional comparisons, compare individual surgeons with different case mixes, predict risk for any single individual patient, or determine patient eligibility for surgery.

In our consecutive series of patients with severe AS undergoing AVR, we found that the logistic EuroSCORE was not an accurate risk assessment tool in all categories of risk but especially in high-risk patients. Therefore, this predictive model should not be used to determine procedural risk in patients with severe AS. Furthermore, the utilization of the logistic EuroSCORE in the assessment of operability in patients with severe AS may not be appropriate. More accurate risk prediction models are needed for risk-stratifying patients with severe AS.