Utilization of machine learning to model the effect of blood product transfusion on short-term lung transplant outcomes.

The objective of this study was to identify the relationship between blood product transfusion and short-term morbidity and mortality following lung transplantation utilizing machine learning. Preoperative recipient characterstics, procedural variables, perioperative blood product transfusions, and donor charactersitics were included in the model. The primary composite outcome was occurrence on any of the following six endpoints: mortality during index hospitalization; primary graft dysfunction at 72 h post-transplant or the need for postoperative circulatory support; neurological complications (seizure, stroke, or major encephalopathy); perioperative acute coronary syndrome or cardiac arrest; and renal dysfunction requiring renal replacement therapy. The cohort included 369 patients, with the composite outcome occurring in 125 cases (33.9%). Elastic net regression analysis identified 11 significant predictors of composite morbidity: higher packed red blood cell, platelet, cryoprecipitate and plasma volume from the critical period, preoperative functional dependence, any preoperative blood transfusion, VV ECMO bridge to transplant, and antifibrinolytic therapy were associated with higher risk of morbidity. Preoperative steroids, taller height, and primary chest closure were protective against composite morbidity.