Machine learning for predicting long-term kidney allograft survival: a scoping review

Area of Science:

• Nephrology
• Medical Informatics
• Biostatistics

Background:

• Supervised machine learning (ML) algorithms are increasingly used for pattern recognition, classification, and prediction.
• Predicting long-term allograft survival is crucial for kidney transplant recipients.

Purpose of the Study:

• To conduct a scoping review on the application of supervised ML algorithms for predicting long-term allograft survival in kidney transplant recipients.
• To synthesize evidence comparing ML techniques with traditional statistical methods.

Main Methods:

• Searched PubMed, CINAHL, and IEEE Xplore databases (inception to November 2019).
• Screened titles, abstracts, and full-text reports to identify relevant studies.
• Extracted data from eleven included studies.

Main Results:

• Decision trees, artificial neural networks (ANN), and Bayesian belief networks were common ML methods.
• Area under the receiver operating curve (AUC), sensitivity, and specificity were primary performance measures.
• Methodological and reporting quality of included studies was poor, with mixed results on predictive potential.

Conclusions:

• Supervised ML shows promise in predicting kidney transplant allograft survival, but evidence is limited by study quality.
• Further high-quality research is needed to validate ML models and compare them rigorously with traditional statistical approaches.