Background: Delayed graft function (DGF) is a major complication of kidney transplantation that adversely affects long-term graft survival. This study aimed to develop and validate deep learning-based predictive models for DGF risk assessment in deceased donor kidney transplant recipients. Materials and methods: We retrospectively analyzed 670 consecutive patients who underwent deceased donor kidney transplantation at a single center between March 2018 and November 2023. The cohort was randomly divided into training (70%) and validation (30%) datasets. The class imbalance in the training set was addressed using a Synthetic Minority Oversampling Technique. Five deep learning algorithms were employed: bidirectional gated recurrent unit (BiGRU), Convolutional bidirectional long short-term memory, convolutional gated recurrent unit, convolutional neural network (CNN)-BiGRU, and CNN-bidirectional long short-term memory. The model performance was evaluated using receiver operating characteristic curve analysis with area under the curve (AUC), Matthews correlation coefficient, and F1 score metrics. Internal validation was performed using 1000 bootstrap iterations. Results: The study population comprised 670 deceased donor kidney transplant recipients with a mean age of 47.7 ± 11.2 years and a median preoperative serum creatinine of 907.1 (702.5–1113.8) μmol/L. The overall incidence of DGF was 21.8% (n = 146). Synthetic minority oversampling technique successfully addresses class imbalance in the training dataset. Among the 5 models evaluated, the CNN-BiGRU hybrid architecture demonstrated superior predictive performance with an AUC of 0.848 (95% CI, 0.798–0.899), Matthews correlation coefficient of 0.614 (95% CI, 0.609–0.619), and F1 score of 0.816 (95% CI, 0.789–0.843). The model achieved balanced discrimination, with 82.1% sensitivity and 83.5% specificity. Conclusions: The CNN-BiGRU-based prediction model demonstrated excellent performance in identifying patients at a high risk of DGF development. This artificial intelligence-powered tool offers the potential to assist the entire nephrology, urology, and transplant service communities in implementing personalized risk stratification and optimizing posttransplant management strategies to improve patient outcomes.
He et al. (Mon,) studied this question.