A NEW machine learning model developed at King Abdullah International Medical Research Center offers a promising tool for predicting 30-day hospital readmission risk following renal transplantation, and it comes with something many AI tools lack: explainability.
The study created a supervised learning model that integrates explainable artificial intelligence (XAI) techniques to enhance clinical interpretability. Using data from 588 kidney transplant recipients, predominantly living donor cases, researchers built a model through a four-stage pipeline including data processing, feature preparation, model development, and validation.
Of the algorithms tested, gradient boosting delivered the best performance, achieving an AUC of 0.837 (95% CI: 0.802–0.872) and an overall accuracy of nearly 80%. Critically, the researchers applied SHAP and LIME analyses to make the model’s predictions transparent for clinicians.
Key risk factors identified included length of hospital stay and post-transplant systolic blood pressure, with notable differences between living and deceased donor transplant cases. For instance, pre-transplant BMI was more predictive in deceased donor recipients, while HbA1c and eGFR were more influential among living donor recipients.
Although the observed readmission rate of 88.9% in the cohort was higher than typically reported, the authors highlight that it reflects specific local practices and underscores the need for center-specific models.
The team has implemented the model as a web-based tool for transplant physicians, offering real-time, donor-specific risk stratification. However, they stress that external validation in diverse transplant centres will be essential before broader adoption.
By combining high predictive performance with clinical explainability, this new tool paves the way for more personalized care strategies and improved patient outcomes after kidney transplantation.
Aleksandra Zurowska, EMJ
Reference
Alnazari N et al. Development of explainable artificial intelligence based machine learning model for predicting 30-day hospital readmission after renal transplantation. BMC Nephrol. 2025;DOI: 10.1186/s12882-025-04128-w.
