Predicting graft survival in paediatric kidney transplant recipients using machine learning

Area of Science:

• Nephrology
• Transplantation immunology
• Artificial intelligence in medicine

Background:

• Graft survival in kidney transplantation is crucial for reducing mortality.
• Artificial intelligence (AI) offers impartial evaluation of clinical factors impacting transplant outcomes.
• Understanding these factors is vital for improving graft longevity, especially in pediatric cases.

Purpose of the Study:

• To utilize machine learning (ML) to identify critical factors affecting graft survival in pediatric kidney transplant recipients.
• To develop predictive models for graft survival using a comprehensive dataset.

Main Methods:

• Retrospective analysis of 465 pediatric kidney transplant recipients (1994-2021) with >12 months follow-up.
• Application of ML algorithms (Naive Bayes, logistic regression, SVM, multi-layer perceptron, XGBoost) for predicting graft survival.
• Data imputation using the nearest neighbor method and model evaluation via accuracy and F1 score.

Main Results:

• Key predictors of graft survival identified by ML include antibody-mediated rejection, acute cellular rejection, estimated glomerular filtration rate (eGFR) at 3 and 5 years, pre-transplant peritoneal dialysis, and cadaveric donor.
• Logistic regression and Support Vector Machine (SVM) models demonstrated comparable performance.
• High model performance achieved with an F1 score of 91.9% and accuracy of 96.5%.

Conclusions:

• Machine learning effectively identifies significant factors influencing kidney transplant graft survival.
• These findings can inform the creation of pre-transplant risk stratification tools.
• Further research can expand these models to generate predictive risk maps for personalized transplantation strategies.