Herbal medicine for inflammatory bowel diseases: development of pattern identification algorithms by retrospective analysis of case series data

Introduction: Herbal prescriptions are frequently sought as complementary medicine treatment for inflammatory bowel disease (IBD). However, variability in pattern identification of Traditional Medicine has been criticised. Using patient data on clinical remission, we aimed to develop pattern identification (PI) algorithms refined by identified pattern and key symptoms which practitioners can easily differentiate. Methods: Symptoms of IBD patients with remission were divided into Large intestine, Water-dampness, Respiratory, Upper gastrointestinal (GI) tract, and Coldness patterns. Using the term frequency-inverse document frequency (TF-IDF) method, symptoms described as indications of the herbal prescriptions and 5 patterns were matched. Decision tree modeling was used for prediction of relevant patterns from symptoms. Five-fold cross validation and sensitivity analysis was conducted to validate the model. Results: Incomplete feeling of bowel emptying for Large intestine pattern, water dampness for Water-dampness pattern, chronic rhinitis for Respiratory pattern, gastric stuffiness for Upper GI tract pattern, and abdominal pain for Coldness pattern were selected by TF-IDF analysis. Overall accuracy of the decision tree was 64.4 %. In the treatment algorithm by the decision tree modeling, associations emerged between presence of incomplete feeling of bowel emptying and Large intestine type (100 %), chronic rhinitis and Respiratory type (100 %), water dampness symptoms and Water-dampness type (78.4 %), and gastric stuffiness and Upper GI tract type (58.3 %). Conclusions: The PI algorithm we suggest can help clinicians determine patterns of IBD. Future studies with a large sample could yield an improved algorithm that predicts not only patterns but also corresponding herbal medicine prescriptions.