Developing machine learning models to predict primary graft dysfunction after lung transplantation.

Area of Science:

• Cardiology
• Pulmonary Medicine
• Transplantation Surgery

Background:

• Primary graft dysfunction (PGD) is a major cause of death post-lung transplant.
• Predicting PGD risk is challenging due to complex donor/recipient factor interactions.
• Current methods lack clarity for donor acceptance decisions.

Purpose of the Study:

• To develop and evaluate machine learning (ML) models for predicting PGD grade 3.
• To identify predictive variables available at donor offer acceptance.
• To improve donor-recipient matching and organ utilization.

Main Methods:

• Retrospective cohort study of 576 bilateral lung recipients.
• Utilized lasso regression to select 11 key predictive variables.
• Developed and validated a K-nearest neighbors (KNN) ML model.

Main Results:

• 30% (173/576) of recipients developed PGD grade 3.
• The KNN model achieved an AUC of 0.65.
• The ML model demonstrated good calibration and performance.

Conclusions:

• Machine learning models can predict PGD grade 3 risk using pre-transplant data.
• This predictive capability can inform donor acceptance decisions.
• ML holds promise for optimizing lung transplant outcomes.