Integrating machine learning and human use experience to identify personalized pharmacotherapy in Traditional Chinese Medicine: a case study on resistant hypertension.

Abstract

Objective: To enhance the understanding of identifying personalized pharmacotherapy options in Traditional Chinese Medicine (TCM), and further support the registration of new TCM drugs.

Methods: Generalized Boosted Models and XGBoost were employed to construct a classification model to identify the bad prognosis factors in resistant hypertension (RH) patients. Furthermore, we used association analysis to explore the rules of "symptom-syndrome" and "symptom-herb" for the major influencing factors, in order to summarize prescription pattern and applicable patients of TCM.

Results: Patients with major adverse cardiac events mostly have complex symptoms of phlegm, stasis, deficiency and fire intermingled with each other, and finally summarized the human experience of using Chinese herbal medicine to precisely intervene in some symptoms of RH patients on the basis of conventional Western medical treatment.

Conclusions: Machine learning algorithms can make full use of human use experience and evidence to save clinical trial resources and accelerate the development of TCM varieties.