Congenital Heart Surgery Machine Learning-Derived In-Depth Benchmarking Tool

Area of Science:

• Cardiovascular Surgery
• Machine Learning
• Health Services Research

Background:

• Optimal Classification Trees (OCTs) previously demonstrated accuracy in predicting risk and assessing performance in congenital heart surgery.
• This methodology is extended to provide comprehensive, interpretable, and actionable hospital performance analysis across all procedures.

Purpose of the Study:

• To extend machine learning-based OCT methodologies for interpretable and actionable hospital performance analysis in congenital heart surgery.
• To establish case-adjusted benchmarking standards using a 'virtual hospital' concept.

Main Methods:

• Analysis of 172,888 congenital cardiac surgical procedures from the European Congenital Heart Surgeons Association database (1989-2022).
• Development of OCT models to predict hospital mortality (AUC, 0.866), prolonged mechanical ventilation (AUC, 0.851), and length of stay (AUC, 0.818).
• Creation of an online, interactive tool for hospital self-assessment by risk-matched patient cohorts.

Main Results:

• OCT models established case-adjusted benchmarks against a 'virtual hospital' aggregate.
• 20.5% of 146 centers statistically overperformed and 20.5% underperformed their predicted hospital mortality benchmark.
• An interactive tool reveals 14 hospital-specific patient cohorts for performance assessment.

Conclusions:

• Machine learning-based OCT benchmarking offers automatic, case-adjusted performance assessment for hospitals performing congenital heart surgery.
• Analysis extends beyond overall performance to include specific, risk-matched patient cohorts.
• The user-accessible online platform facilitates hospital self-assessment and performance improvement.