The Predictive Role of Symptoms/signs on ACR20 Responses in Rheumatoid Arthritis Analyzed with Data Mining Approaches

Objective: The role of symptoms/signs for American College of Rheumatology 20% response (ACR20) prediction in rheumatoid arthritis (RA) analyzed with decision tree and neuron network was explored. Methods: 489 patients were randomly divided to receive Western medicine (WM) therapy, 247 cases; and traditional Chinese medicine (CM) therapy (TCM), 242 cases. ACR20 response was employed as effectiveness evaluation point. The symptoms/signs at baseline were collected and analyzed for ACR20 response prediction with decision tree and neural network methods, and 75% data were for training and 25% data for verification set. Results: 19 symptoms/signs in CM treated patients and 26 in WM treated patients were obtained from MANTELHAENSZEL test or Fisher´s exact test (p<0.2 as inclusion criteria) for decision tree analysis, and the ACR20 responses were different in the different combinations of the symptoms/signs both in CM and WM. The results were verified in the verification data sets. For neural network analysis, the training data from CM and WM treated patients were put into the neuron network model, and the Lift Chart was created which showed that the total effective rate could be predicted to be 80% if only right 10 percentage of patients where treated based on the chosen symptoms/signs in CM, and to be 98% if 20% percentage of patients where treated based on the chosen symptoms/signs in WM. Conclusion: Symptoms/signs from TCM have predictive roles for ACR20 response evaluation in RA.