Predicting Primary Graft Dysfunction in Lung Transplantation: Machine Learning-Guided Biomarker Discovery.

Area of Science:

• Transplant immunology
• Biomarker discovery
• Machine learning in medicine

Background:

• Primary graft dysfunction (PGD) is a significant complication following lung transplantation.
• Understanding PGD pathophysiology is essential for developing predictive point-of-care diagnostics.
• Multiplex analysis of biological samples can identify PGD risk factors.

Purpose of the Study:

• To identify predictive biomarkers for primary graft dysfunction (PGD) using a multiplex approach.
• To evaluate different biological sample sources (donor lung perfusate, bronchoalveolar lavage fluid, plasma) for PGD prediction.
• To develop a machine learning model for early PGD risk stratification.

Main Methods:

• Collected patient-matched biospecimens from donor lung perfusate, bronchoalveolar lavage (BAL) fluid (2h), and plasma (2h and 24h) from bilateral lung transplant recipients.
• Utilized a 71-analyte multiplex panel on all biospecimens.
• Applied cross-validated logistic regression (LR) and random forest (RF) machine learning models to discriminate between PGD grades 0 and 3.

Main Results:

• Bronchoalveolar lavage (BAL) fluid at 2 hours post-transplant demonstrated the highest predictive performance for PGD (LR: 0.825, RF: 0.919).
• Combined analysis of clinical data, BAL fluid, and plasma achieved perfect discrimination (LR: 1.000, RF: 1.000).
• Interleukin-1 receptor antagonist (IL-1RA), B-cell-attracting chemokine 1 (BCA-1), and Fractalkine were identified as key predictors of severe PGD.

Conclusions:

• Bronchoalveolar lavage (BAL) fluid collected 2 hours post-transplant is a strong predictor of severe PGD.
• Machine learning identified novel cytokines associated with PGD and potential analytes for a point-of-care diagnostic panel.
• This approach facilitates early identification of patients at high risk for developing severe PGD.