Primary Graft Dysfunction After Lung Transplantation: A Temporal Classification and Machine Learning Clustering

Area of Science:

• Cardiology
• Transplantation Medicine
• Data Science

Background:

• Primary graft dysfunction (PGD) significantly impacts morbidity and mortality post-lung transplantation (LTx).
• Current PGD grading at static time points limits understanding of its dynamic nature.
• Statistical methods may not capture the multifactorial complexity of PGD.

Purpose of the Study:

• To introduce a temporal classification for PGD phenotypes.
• To apply machine learning (ML) for patient clustering based on clinical variables.
• To overcome sample-size limitations using synthetic data generation.

Main Methods:

• Analysis of 794 LTx cases with classification into 4 temporal PGD phenotypes.
• Unsupervised ensemble k-means clustering using 30 clinical variables.
• Generation of synthetic patient data via Wasserstein Generative Adversarial Network with Gradient Penalty.

Main Results:

• Temporal PGD phenotypes showed significantly different 5-year survival rates (P < 0.001).
• K-means clustering identified stable patient subgroups, with optimal clusters at k=5 and k=9.
• Key clustering features included preoperative ICU stay and postoperative chest X-rays.

Conclusions:

• Temporal PGD classification enhances understanding of PGD's dynamic impact on survival.
• ML techniques, including clustering and synthetic data, show promise for unraveling complex clinical factor interactions.
• These methods can overcome sample-size limitations in PGD research.