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

Figure 1. Flow chart of the study patients. ESKD, end-stage kidney disease; CRRT, continuous renal replacement therapy; IDH, intradialytic hypotension; ML, machine learning.

Figure 2. (a) ROC curves generated by the six models with the outcome defined as a reduction in SBP ≥ 20 mmHg from baseline. (b) ROC curves generated by the six models, with the outcome defined as a reduction in SBP ≥ 30 mmHg from baseline. (c) ROC curves generated by the six models with the outcome defined as a reduction in MAP ≥ 10 mmHg from baseline. SVM, support vector machine; XGBoost, extreme gradient boosting; GBM, gradient boosting machine; RF, random forest; DT, decision tree; LR, logistic regression; ROC, receiver operating characteristic; AUC, the area under the receiver operating characteristic curve; SBP, systolic blood pressure; MAP, mean arterial pressure.

Figure 3. Calibration belts produced by the SVM, XGBoost, GBM, RF, DT, and LR models, with the outcome defined as a reduction in SBP ≥20 mmHg from baseline. The horizontal axis represents the model-predicted probability, and the vertical axis represents the observed event rate. The red diagonal line indicates the ideal calibration line (representing perfect agreement between predicted probability and actual observed outcome). The gray shaded area denotes the 95% confidence interval of each model's calibration curve, calculated based on the standard error of the actual outcome within each bin, reflecting the uncertainty of the model's calibration performance. SVM, support vector machine; XGBoost, extreme gradient boosting; RF, random forest; DT, decision tree; LR, logistic regression; GBM, gradient boosting machine.

Figure 4. DCA results for the six models with the outcome defined as a reduction in SBP ≥20 mmHg from baseline. SVM, support vector machine; XGBoost, extreme gradient boosting; GBM, gradient boosting machine; RF, random forest; DT, decision tree; LR, logistic regression; DCA, decision curve analysis; SBP, systolic blood pressure.

Figure 5. LASSO regularization path graph with the outcome defined as a reduction in SBP ≥20 mmHg from baseline. The log(λ1se) criterion was chosen to determine the variables used to construct models that were concise and had good predictive performance. LASSO, least absolute shrinkage and selection operator.

Figure 6. SHAP summary plot for the SVM model with the outcome defined as a reduction in SBP ≥20 mmHg from baseline. SVM, support vector machine; SBP, systolic blood pressure; DBP, diastolic blood pressure; ACEI/ARB, angiotensin-converting enzyme inhibitor/angiotensin II receptor blocker; SHAP, Shapley additive explanation.

Figure 7. SHAP force plot for the SVM model with the outcome defined as a reduction in SBP ≥ 20 mmHg from baseline. SVM, support vector machine; SBP, systolic blood pressure; DBP, diastolic blood pressure; SHAP, Shapley additive explanation.