Maximizing Survival in Pediatric Congenital Cardiac Surgery Using Machine Learning, Explainability, and Simulation...

Area of Science:

• Medical Informatics
• Surgical Outcomes Research
• Machine Learning in Healthcare

Background:

• Pediatric and congenital heart surgery (PCHS) carries significant risks, often stemming from disease severity or suboptimal timing of procedures.
• Existing prognostic models aid surgical decision-making but cannot actively reverse adverse outcomes.
• There is a critical need for advanced methods to improve survival probabilities in PCHS.

Purpose of the Study:

• To develop and validate an innovative approach integrating machine learning (ML), explainability techniques (ET), and simulation to reverse negative prognoses in PCHS.
• To enhance the accuracy of predicting surgical outcomes and identify key risk factors.
• To create a framework for designing personalized health scenarios to improve patient survival.

Main Methods:

• Utilized machine learning (ML) models for predicting mortality and survival in PCHS patients.
• Employed an explainability technique (ET), specifically LIME, to identify and quantify the impact of major risk factors.
• Integrated a simulation method to model potential health scenarios aimed at reversing negative prognoses.

Main Results:

• Achieved 96% accuracy in predicting mortality and survival using a dataset of 565 PCHS patients and 10 risk factors.
• Case studies confirmed LIME's explanations align with clinical observations.
• Successfully reversed an initial prognosis of death to survival in a simulated real-world case.

Conclusions:

• An integrated method combining ML, ET, and simulation effectively reverses negative prognoses in PCHS.
• The approach provides valuable insights for medical decision-making, supporting personalized patient care.
• Experimental validation demonstrates the potential to significantly improve outcomes in high-risk pediatric cardiac surgeries.