Machine learning model for predicting hypotension following continuous renal replacement therapy initiation in end-stage kidney disease patients: a SHAP-interpretable approach.

Background: Early prediction of intradialytic hypotension (IDH) after starting continuous renal replacement therapy (CRRT) is critical for timely intervention. However, effective models for predicting the risk of IDH in patients with end-stage kidney disease (ESKD) undergoing CRRT are currently lacking. Therefore, the aim of this study was to develop a machine learning (ML)-based predictive model to facilitate the early identification of high-risk patients and to support clinical decision-making. Methods: Adult patients with ESKD who underwent CRRT were enrolled in this study and randomly divided into training (70%) and testing sets (30%). IDH was defined as a reduction in systolic blood pressure (SBP) ≥ 20 mmHg from baseline within 6 h after CRRT initiation; supplementary definitions were also applied, including a decrease in SBP ≥ 30 mmHg or in mean arterial pressure (MAP) ≥ 10 mmHg from baseline. Demographic characteristics, medication use, laboratory parameters, and treatment-related variables were collected. Multiple ML algorithms-including gradient boosting machine (GBM), extreme gradient boosting (XGBoost), decision tree (DT), support vector machine (SVM), random forest (RF), and logistic regression (LR)-were used to develop predictive models. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC) and other performance metrics. Shapley additive explanations (SHAP) was applied to quantify the contribution of each feature to the model predictions. Results: Overall, 1,103 patients were included. Within the dataset used in this study, the SVM model consistently outperformed the other compared algorithms across all definitions of IDH, with an AUC of 0.805, indicating good calibration and clinical utility. The optimized simplified model retained stable predictive capacity (AUC = 0.809). SHAP analysis revealed that SBP was the most important feature for predicting IDH. Conclusion: The results of this study demonstrate the effectiveness of several ML algorithms in predicting the risk of IDH following the initiation of CRRT in patients with ESKD. The SVM model yielded the most favorable predictive performance in our comparative analysis. SBP was identified as a key predictor of IDH. The proposed model can assist in the clinical identification of high-risk patients and facilitate timely interventions.