Prediction of Efficacy of Taeumjowi-Tang for Treatment of Metabolic Risk Factors Based on Machine Learning

Herbal medicine is widely prescribed worldwide. To date, however, studies on the prediction of efficacy of herbal medicine based on machine learning have very rarely been reported. The objectives of this study are to predict the efficacy of Taeumjowi-tang (one of herbal medicines) and evaluate the prediction model in treating metabolic abnormalities. Subjects were divided into an improvement group and a non-improvement group based on the difference before and after oral administration of an herbal medicine. Efficacy models of triglyceride level, high-density lipoprotein (HDL) cholesterol level, systolic blood pressure (SBP), and diastolic blood pressure (DBP) were built using a least absolute shrinkage and selection operator (LASSO) based on variables extracted from face shape, face colors, body circumference, questionnaire, voice, and tongue color. In predicting efficacy for four metabolic risk factors, the efficacy model of HDL cholesterol level showed the best the area under the receiver operating characteristic curve (AUC) value among the four models (AUC = 0.785 (confidence interval = 0.693, 0.877)). The AUC value of the efficacy model of triglyceride level was 0.659 (0.551, 0.768). Efficacy models of DBP and SBP showed AUC values of 0.665 (0.551, 0.78) and 0.54 (0.385, 0.694), respectively. The results may provide a clue to predict whether a drug will be effective for each subject with phenotypic information and to reduce the use of an ineffective drug or its side effects.