Artificial Intelligence Helps Predict 5-year Mortality and Graft Failure in Patients Undergoing Orthotopic Heart Transplantation

Abstract

Orthotopic heart transplantation (OHT) is a life-saving procedure for advanced end-stage heart failure patients. Given the conservative allocation policy and mismatch of donors and wait-listed patients, the prediction of post OHT outcomes including survival and graft failure (GF) can help optimize organ allocation. We aimed to develop a risk prediction model using machine learning algorithm to predict survival and GF 5 years after OHT. Using International society of heart and lung transplant (ISHLT) registry data, we retrospectively analyzed 15236 patients who underwent OHT from January 2005 through December 2009. 342 variables including pre/ post OHT, discharge and follow up information were extracted and used to develop a risk prediction model utilizing a gradient boosted classification tree algorithm (GBM) to predict the risk of graft failure and mortality 5 yrs. after hospital discharge. Variables with missing observations were handled internally by the GBM algorithm. A 10 fold cross-validation repeated 5 times was used to estimate the model's external performance. Receiver operator curve (ROC) for GBM model was calculated to predict survival and GF 5 years post OHT. The mean duration of follow up of 4.7 years (median 5 years). The mortality and graft failure 5 years post- OHT was 27.3% (n = 4161) & 28 %( n=4276) respectively. The area under ROC (AUC) to predict 5 yr. mortality and GF are 0.717 (95% CI 0.696- 0.737) & 0.661 (95% CI 0.640-0.683) respectively. Length of stay, recipient and donor age, recipient body mass index and ischemic time had the highest weight in predicting 5-year mortality and graft failure. The subgroup of patients with age>60yrs had a higher AUC for both the models. The GBM model has an excellent accuracy in predicting 5 yr.-mortality and graft failure in patients undergoing OHT.