Prediction of dialysis adequacy using data-driven machine learning algorithms

Area of Science:

• Nephrology
• Biomedical Informatics
• Machine Learning

Background:

• Hemodialysis (HD) adequacy, measured by spKt/V, is crucial for patient outcomes.
• Current methods require pre- and postdialysis blood samples, limiting frequent assessment.
• Developing alternative methods for assessing HD adequacy is essential.

Purpose of the Study:

• To develop and validate machine learning (ML) models for predicting hemodialysis adequacy (spKt/V > 1.4).
• To assess the feasibility of using noninvasive, readily available variables for prediction.
• To identify key variables that predict dialysis adequacy.

Main Methods:

• Retrospective analysis of 1869 HD sessions from 373 end-stage renal disease (ESKD) patients.
• Data preprocessing to select relevant general, intradialytic, and laboratory variables.
• Development and evaluation of six binary classification models, including Random Forest.

Main Results:

• The Random Forest model achieved high prediction accuracy (AUROC 0.873).
• A model using only noninvasive variables demonstrated strong performance (AUROC 0.868).
• Key predictors included vascular access, gender, BMI, ultrafiltration volume, and dialysis duration.

Conclusions:

• Machine learning models can accurately predict dialysis adequacy using noninvasive data.
• This approach offers a feasible method for noninvasive clinical assessment of HD adequacy.
• The identified predictive variables can guide clinical practice and patient management.