Rheumatoid arthritis (RA) is a systematic chronic inflammatory disease. The disease mechanism remains unclear and may have resulted from autoimmune problems caused by genetic predisposing and pathogen infection. In clinical practice, selection of the initial treatment is based on the degree of disease activity, and treatment plans will be added gradually according to increased severity of the disease. However, treatment results can be unclear and treatment process uncertain and ambiguous, which can cause healthcare quality to become worse. This study attempts to combine expert opinions to construct various classifiers using a number of data mining techniques to analyze the different prognosis of two patient groups, by predicting whether the inflammatory indicator erythrocyte sedimentation rates of these two groups will be within the normal range with different medication strategies. Clinical data were collected for construction of different classifiers and we evaluate the prediction accuracy rate of each classifier afterwards. The optimum prediction model is selected from these classifiers to predict the prognosis of RA within these treatment strategies and analyze various results. The results show the accuracy rate of the prediction model by Logistic, SVM and DT module were 0.7927, 07829 and 0.9094, respectively. In the RA complications dataset, the accuracy rate of were 0.9393, 0.9290 and 0.9812, respectively. Futhermore, gain ratio was used to further analyze the rules and to discover which branch nodes are the most importance factor. The results of this study are helpful for formulation and development of guidelines for clinical RA treatments, and implementation of a decision support system by using the prediction model can assist medical staff to make correct decisions in the disease's early stage.
Keywords: decision tree; logic regression; rheumatoid arthritis (RA), data mining techniques; support vector machines.