Machine learning model for predicting the cold-heat pattern in Kampo medicine: a multicenter prospective...

Area of Science:

• Computational Medicine
• Traditional East Asian Medicine
• Machine Learning Applications

Background:

• Cold-heat patterns are a key diagnostic concept in Kampo medicine.
• Accurate identification of these patterns is crucial for effective treatment.
• Existing diagnostic methods may benefit from computational approaches.

Purpose of the Study:

• To develop and validate a machine learning model for predicting the four cold-heat patterns.
• To apply artificial intelligence to subjective symptoms described in the International Classification of Diseases Traditional Medicine Conditions - Module 1.
• To enhance the diagnostic accuracy of cold-heat patterns in Kampo outpatients.

Main Methods:

• A multicenter prospective observational study involving 622 first-visit Kampo outpatients.
• Development of separate cold and heat pattern prediction models using 148 symptoms, BMI, blood pressure, age, and sex.
• Classification included single cold/heat, tangled heat/cold, and moderate (heat/cold) patterns.

Main Results:

• The combined prediction models achieved high performance metrics: 96.7% accuracy, 93.2% macro-recall, 85.6% precision, and 88.5% F1-score.
• Key predictive features aligned with established definitions of cold-heat patterns.
• The model demonstrated robust predictive capabilities on the study cohort.

Conclusions:

• A machine learning model was successfully developed to predict cold-heat patterns.
• The model utilizes patient-reported symptoms that align with physician diagnoses.
• This approach offers a promising tool for objective cold-heat pattern prediction in clinical practice.