Comprehensive validation of machine learning models predicting chemotherapy related electrolyte disorders in a...

Area of Science:

• Oncology
• Medical Informatics
• Biostatistics

Background:

• Electrolyte abnormalities are frequent and serious complications in cancer patients receiving chemotherapy.
• These abnormalities can lead to treatment delays and poorer health outcomes.
• Predictive tools are needed to manage these risks proactively.

Purpose of the Study:

• To develop and validate machine learning (ML) models for predicting eight common electrolyte abnormalities.
• To identify key predictors of these abnormalities in cancer patients undergoing chemotherapy.
• To assess the models' performance using rigorous validation and interpretability methods.

Main Methods:

• Retrospective analysis of 11,227 cancer patient records from two Korean tertiary hospitals.
• Application of four ML algorithms to predict electrolyte abnormalities within four weeks of chemotherapy initiation.
• Comprehensive validation (internal, external, temporal) and interpretability analysis using Shapley additive explanations (SHAP).

Main Results:

• Electrolyte abnormalities occurred in 74.0% (internal) and 84.7% (external) of patients, with high in-hospital mortality rates (35.9% and 32.8%).
• The best ML models achieved an average AUC of 0.798, with a slight performance drop during external validation.
• Key predictors included serum albumin, heart rate, and estimated glomerular filtration rate (eGFR).
• High-risk patients showed significantly increased odds of abnormalities and mortality.

Conclusions:

• Developed ML models offer a robust method for predicting chemotherapy-induced electrolyte abnormalities.
• These models can aid in risk stratification and prioritizing patient monitoring.
• Further validation in diverse populations and integration into clinical decision support are recommended.