Risk Model for Adult Congenital Cardiac Reoperations: Analysis of the Society of Thoracic Surgeons Adult Cardiac...

Accurate risk prediction is essential for informed clinician-patient decision making in adults with congenital heart disease (ACHD). We performed extended validation of an institutional risk model for composite morbidity and mortality in reoperative ACHD surgery.

Risk models developed in a high-volume ACHD surgical program using multivariable logistic regression and machine learning (ML) were externally validated in the Society of Thoracic Surgeons Adult Cardiac Surgery Database (STS-ACSD) among reoperative ACHD patients 7/1/2017-12/31/2023. The outcome was a composite of operative mortality and major postoperative morbidity including mechanical circulatory support, dialysis, unplanned noncardiac reoperation, neurologic deficit, and cardiac arrest. Predictor importance was examined using Shapley additive explanations and model discrimination was assessed with area under the receiver operating characteristic curve (AUROC).

The primary outcome was more common in STS-ACSD than the institutional cohort (16.7% vs 8.8%), attributable in part to a higher proportion of urgent/emergent procedures (26.6% vs 5.4%). Institutional risk model discrimination was attenuated in STS-ACSD with systematic underestimation of risk. A de-novo model derived from the 15 most influential STS-ACSD variables (aortic procedure, status, hematocrit, ejection fraction, creatinine, age, white blood cell count, hypertension, connective tissue disorders, platelets, symptoms, reoperation number, heart failure, primary payor, body mass index) demonstrated good discrimination with mean, cross-validated AUROC 0.74 for extreme gradient boosting and AUROC 0.73 for penalized logistic regression.

Reoperative ACHD surgery remains high risk nationally, underscoring the need for refined risk stratification to support care. Clinically interpretable risk models integrating ML and regression offer a practical path forward.