Development of Machine‐Learning‐Based Models for Detection of Cognitive Impairment in Patients Receiving Maintenance Hemodialysis

FIGURE 1. Classification Tree for Cognitive impairment and Risk Stratification for the Derivation Cohort. Classification tree stratifies hemodialysis patients into high‐ and low‐risk groups for cognitive impairment based on four key features (ORIEN18, STM12, MMSE4, and MMSE11) from Cognitive Ability Screening Instrument (CASI) and Mini‐Mental State Examination (MMSE). Each node shows the number of patients and the percentage with CASI scores below the 20th percentile, with color coding indicating high or low risk level. *Percent of participants with CASI < 20th percentile in each node.

FIGURE 2. Importance of Each Predictor in the (a) Random Forest and (b) Extreme Gradient‐Boosted Decision Tree Models. Variable importance rankings for (a) Random Forest and (b) XGBoost models, showing the relative contribution of Mini‐Mental State Examination (MMSE) and Cognitive Ability Screening Instrument (CASI) features (e.g., ORIEN18 and STM12) to predicting cognitive impairment. Higher values indicate greater predictive importance.

FIGURE 3. Receiver operating characteristic (ROC) curves of the six Machine Learning Models. ROCs comparing the discriminative ability of six machine learning models. The area under the curve (AUC) values indicate model performance, with random forest (RF), extreme gradient boost (XGBoost), and artificial neural network (ANN) achieving the highest AUCs (0.95). ANN, Artificial Neural Network; CART, Classification and Regression Tree; LASSO, logistic least absolute shrinkage and selection operator regression; RF, Random Forest; SVM, Support Vector Machine; Xgboost, EXtreme Gradient Boosting.

FIGURE 4. Decision curve analysis of the six machine learning models. Decision curve analysis evaluates the net benefit of six machine learning models across various threshold probabilities for cognitive impairment screening. Random Forest (RF) and Support Vector Machine (SVM) show superior net benefit at most thresholds, balancing true positives and false positives. ANN, Artificial Neural Network; CART, Classification and Regression Tree; LASSO, logistic least absolute shrinkage and selection operator regression; RF, Random Forest; SVM, Support Vector Machine; Xgboost, EXtreme Gradient Boosting.