Prediction of delayed graft function after kidney transplantation: comparison between logistic regression and machine...

Area of Science:

• Nephrology
• Transplantation Medicine
• Machine Learning in Healthcare

Background:

• Delayed graft function (DGF) is a common complication after kidney transplantation.
• Traditional predictive models for DGF often rely on logistic regression.
• Evaluating advanced machine learning methods for DGF prediction is crucial.

Purpose of the Study:

• To assess the efficacy of various machine learning models in predicting DGF post-kidney transplantation.
• To compare the performance of machine learning algorithms against logistic regression for DGF prediction.

Main Methods:

• A retrospective analysis of 497 kidney transplantations was conducted.
• Feature selection identified 20 optimal parameters for model development.
• Nine predictive models were trained and evaluated, including logistic regression, discriminant analyses, SVMs, decision trees, random forest, and stochastic gradient boosting.
• Model performance was assessed using sensitivity, positive predictive values, and AUROC with 10-fold cross-validation.

Main Results:

• The incidence of DGF was observed to be 12.5%.
• Linear SVM achieved the highest discriminative capacity (AUROC 84.3%), outperforming logistic regression.
• Several machine learning models, including LDA, radial SVM, and polynomial SVM, demonstrated strong predictive performance (AUROC > 80%).
• Decision tree models showed inferior performance in predicting DGF.

Conclusions:

• Linear SVM is identified as the most appropriate model for predicting DGF due to its high discriminative capacity and sensitivity.
• The study highlights the superiority of certain machine learning models over traditional logistic regression for DGF prediction.
• Further validation of linear SVM in predicting DGF is warranted.