Predicting anemia treatment outcomes in maintenance hemodialysis patients using multiple machine learning models

Area of Science:

• Nephrology
• Biostatistics
• Machine Learning

Background:

• Anemia is a common complication in patients undergoing maintenance hemodialysis (MHD).
• Predicting and optimizing anemia treatment attainment in MHD patients is crucial for improving outcomes.
• Machine learning offers novel approaches to identify predictors for personalized anemia management.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting anemia treatment attainment in MHD patients.
• To identify significant predictors of renal anemia treatment outcomes.
• To provide a basis for personalized anemia treatment strategies in MHD patients.

Main Methods:

• Clinical data from 222 MHD patients were collected.
• Multiple machine learning models, including logistic regression and support vector classification, were employed.
• Model performance was evaluated using discrimination, calibration, and clinical utility metrics.

Main Results:

• Support vector classification demonstrated high specificity (0.914).
• Logistic regression achieved the highest AUC (0.713).
• Stacking models showed strong precision (0.871) and recall (0.710) with a low Brier score (0.087).
• The top five predictors identified were albumin, total cholesterol, transferrin saturation, high-density lipoprotein cholesterol, and C-reactive protein.

Conclusions:

• The identified predictors offer valuable insights for guiding renal anemia treatment.
• Personalized treatment strategies based on these factors may improve patient outcomes.
• Machine learning models can effectively predict anemia treatment attainment in MHD patients.