Objective: The purpose of the present study was to compare important patient questionnaire items by creating a random forest model for predicting deficiency-excess pattern diagnosis in six Kampo specialty clinics.
Design: A multi-centre prospective observational study.
Setting: Participants who visited six Kampo specialty clinics in Japan from 2012 to 2015.
Main outcome measure: Deficiency-excess pattern diagnosis made by board-certified Kampo experts.
Methods: To predict the deficiency-excess pattern diagnosis by Kampo experts, we used 153 items as independent variables, namely, age, sex, body mass index, systolic and diastolic blood pressures, and 148 subjective symptoms recorded through a questionnaire. We extracted the 30 most important items in each clinic's random forest model and selected items that were common among the clinics. We integrated participating clinics' data to construct a prediction model in the same manner. We calculated the discriminant ratio using this prediction model for the total six clinics' data and each clinic's independent data.
Results: Fifteen items were commonly listed in top 30 items in each random forest model. The discriminant ratio of the total six clinics' data was 82.3%; moreover, with the exception of one clinic, the independent discriminant ratio of each clinic was approximately 80% each.
Conclusions: We identified common important items in diagnosing a deficiency-excess pattern among six Japanese Kampo clinics. We constructed the integrated prediction model of deficiency-excess pattern.
Keywords: Decision support system; Machine learning; The 11th version of the international classification of diseases (ICD-11); Traditional medicine pattern ((TM1)).
Copyright © 2019 Elsevier Ltd. All rights reserved.