Improving prediction of heart transplantation outcome using deep learning techniques

Area of Science:

• Cardiology
• Transplantation Medicine
• Artificial Intelligence in Healthcare

Background:

• Accurate prediction of post-transplant survival is crucial for managing heart transplant recipients.
• Existing risk models may have limitations in predicting outcomes effectively.

Purpose of the Study:

• To compare the accuracy of the International Heart Transplantation Survival Algorithm (IHTSA), a deep learning model, against the Index for Mortality Prediction After Cardiac Transplantation (IMPACT) model.
• To evaluate the models' ability to predict one-year mortality and long-term survival after heart transplantation.

Main Methods:

• Utilized data from 27,705 adult heart transplant recipients from the UNOS registry (1997-2011).
• Divided data into derivation (pre-2009) and test (post-2009) cohorts.
• Assessed model performance using receiver operating characteristic (ROC) values and C-index for one-year mortality and long-term survival.

Main Results:

• IHTSA achieved higher ROC values (0.654) for one-year mortality compared to IMPACT (0.608).
• IHTSA demonstrated a superior C-index (0.627) for long-term survival versus IMPACT (0.584).
• The deep learning approach resulted in a 12% error reduction for ROC and 10% for C-index.

Conclusions:

• The IHTSA model, utilizing deep learning, exhibits significantly better discriminatory power for predicting one-year mortality and long-term survival after heart transplantation.
• IHTSA represents a more accurate tool for risk stratification in heart transplant recipients compared to the IMPACT model.