Early identification of potentially reversible cancer cachexia using explainable machine learning driven by body...

Area of Science:

• Oncology
• Machine Learning
• Biostatistics

Background:

• Cancer cachexia is linked to poor patient outcomes and poses clinical challenges.
• Early identification of reversible cachexia is crucial for effective management.

Purpose of the Study:

• To develop and validate a machine learning (ML) model for identifying potentially reversible cancer cachexia (PRCC).

Main Methods:

• A multicenter cohort study retrospectively diagnosed cachexia using Fearon's framework.
• ML models were trained on patient admission body weight dynamics to predict PRCC.
• The optimal model was evaluated for interpretability, clinical usefulness, and external validation.

Main Results:

• The ML model achieved high performance in predicting PRCC (AUC 0.887 in test set, 0.863 in external validation).
• Weight change in the month preceding baseline was the most impactful predictor.
• Potentially reversible cancer cachexia (PRCC) was identified in 52.6% of patients, with breast cancer showing the highest rate.

Conclusions:

• An explainable ML model effectively identifies PRCC using accessible body weight dynamics.
• This approach shows promise for improving cancer cachexia management and patient outcomes.