Predictive Value of EuroSCORE II in Contemporary Cardiac Surgery : Insights from the EPICARD Registry

Technical Summary: Predictive Value of EuroSCORE II in Contemporary Cardiac Surgery

Problem Statement
Accurate preoperative risk stratification is fundamental to guiding clinical decision-making, patient counseling, and institutional benchmarking in cardiac surgery. The EuroSCORE II model, introduced in 2012 to update the original EuroSCORE for contemporary practice, has been the standard tool for predicting operative mortality. However, cardiac surgery has evolved significantly over the last decade, characterized by an ageing population, increased comorbidity burdens, and advances in perioperative management and surgical techniques (including minimally invasive and hybrid procedures). Concerns have been raised regarding the long-term reliability of EuroSCORE II, specifically whether it retains its calibration accuracy in current populations or if it systematically overestimates or underestimates risk in specific subgroups. While international validation studies exist, contemporary nationwide data evaluating EuroSCORE II performance within the French healthcare system remains scarce.

Methodology
This study conducted a retrospective observational analysis of the EPICARD registry, a nationwide multicenter database prospectively collecting data from French cardiac surgery centers.

• Cohort: The study included 108,412 consecutive adult patients (≥18 years) undergoing cardiac surgery between January 2020 and December 2024.
• Exclusions: Patients undergoing heart transplantation, durable mechanical circulatory support implantation, cardiac tumor surgery, or those with missing EuroSCORE II data or 30-day mortality data were excluded.
• Data Linkage: The registry was linked to the French national mortality database (INSEE) to ensure complete ascertainment of in-hospital and 30-day postoperative mortality.
• Endpoints: The primary endpoint was operative mortality (in-hospital or 30-day death).
• Statistical Analysis:
  • Discrimination: Assessed using the Area Under the Receiver Operating Characteristic Curve (AUC).
  • Calibration: Evaluated using the Hosmer–Lemeshow goodness-of-fit test, calibration plots across risk deciles, and the Observed-to-Expected (O/E) mortality ratio.
  • Subgroups: Analyses were stratified by surgical procedure type (isolated CABG, isolated valvular, acute aortic dissection, combined procedures, etc.).
  • Software: Analyses were performed using R software (v4.4.2 or later).

Key Results

• Overall Performance: The cohort had a median EuroSCORE II of 1.7% and an overall operative mortality of 3.3%. EuroSCORE II demonstrated acceptable discriminative performance with an AUC of 0.83 (95% CI 0.82–0.84).
• Overall Calibration: The overall O/E ratio was 0.93 (95% CI 0.90–0.96), indicating a modest overestimation of mortality by the model. The Hosmer–Lemeshow test indicated statistically significant differences between predicted and observed mortality (p < 0.001), though the authors note this is likely due to the test's sensitivity in large sample sizes.
• Subgroup Variations:
  • Low-Risk/Isolated CABG: The model tended to overestimate mortality, particularly in lower-risk patients (O/E ratio 0.71 for isolated CABG).
  • High-Risk/Emergency: The model consistently underestimated mortality in high-risk settings.
    • Acute Aortic Dissection: Observed mortality significantly exceeded predictions across most risk deciles (O/E ratio 1.47 in the first decile).
    • Combined CABG-Aortic Procedures: Underestimation was observed, particularly in the highest-risk decile (28.9% observed vs. 19.5% predicted; O/E ratio 1.46).
  • Isolated Valvular Surgery: Showed overestimation in low-risk patients but slight underestimation in the highest-risk decile.
  • Combined CABG-Valve: Showed relatively close alignment between predicted and observed mortality (O/E ratio 0.93).

Key Contributions

1. Large-Scale Validation: This study provides the largest contemporary nationwide validation of EuroSCORE II in France, utilizing over 100,000 real-world cases from a prospective registry.
2. Discriminative Robustness: It confirms that EuroSCORE II retains strong discriminative ability (AUC 0.83) even 14 years after its publication, effectively distinguishing between patients who survive and those who die.
3. Calibration Nuance: The study highlights that while discrimination is preserved, calibration is heterogeneous. It identifies specific clinical scenarios where the model fails to accurately predict absolute risk:
   • Overestimation in routine, low-risk elective procedures (e.g., isolated CABG).
   • Underestimation in critical, high-risk emergency scenarios (e.g., acute aortic dissection and complex combined aortic procedures).
4. Registry Utility: The study demonstrates the utility of the EPICARD registry, linked with national mortality data, for rigorous risk model evaluation.

Significance and Claims
The authors claim that EuroSCORE II remains a valuable tool for preoperative risk stratification and patient counseling in contemporary cardiac surgery due to its preserved discriminative power. However, they emphasize that clinicians must interpret the predicted mortality with caution, recognizing that:

• Predicted mortality may be slightly inflated for low-risk elective patients.
• Predicted mortality may be dangerously underestimated for patients undergoing emergency surgery for acute aortic dissection or complex combined aortic procedures.

The paper concludes that these calibration limitations in specific high-risk populations support the need for periodic recalibration of risk models and the development of future models that incorporate contemporary real-world data and procedure-specific risk factors to improve prediction accuracy. The study does not propose a new model but rather calls for the recognition of current model limitations in specific subgroups.