A data-driven method for syndrome type identification and classification in traditional Chinese medicine

Area of Science:

• Integrative Medicine
• Computational Biology
• Medical Informatics

Background:

• Accurate classification of patients into Traditional Chinese Medicine (TCM) syndrome types is crucial for effective treatment of Western Medicine (WM) diseases.
• Existing methods like Latent Class Analysis (LCA) have limitations due to strong independence assumptions regarding symptoms.

Purpose of the Study:

• To develop and validate a novel data-driven method for classifying patients into TCM syndrome types.
• To overcome the limitations of traditional statistical methods by incorporating symptom co-occurrence patterns.

Main Methods:

• A generalized Latent Class Analysis (LCA) approach was employed, relaxing independence assumptions.
• Symptom co-occurrence patterns were discovered from unlabeled survey data and used as features for LCA.
• The method involves six steps: data collection, pattern discovery, interpretation, syndrome identification, type identification, and classification.
• A software package, Lantern, was developed to facilitate the method's application.

Main Results:

• The developed method successfully identified and quantified TCM syndrome types based on statistical symptom patterns.
• The approach demonstrated its utility in classifying patients, illustrated with a dataset on vascular mild cognitive impairment.
• The relaxation of independence assumptions improved the robustness of the classification.

Conclusions:

• The novel data-driven method offers a more robust approach to TCM syndrome classification compared to traditional LCA.
• This advancement can lead to improved patient stratification and potentially enhance the efficacy of TCM treatments for WM diseases.
• The Lantern software package provides a practical tool for implementing this advanced classification method.