Selection of the best endodontic treatment option using data mining: A decision tree approach

Introduction: The presence of postendodontic pain is an important issue, which can affect the patients' quality of life. Appropriate treatment selection, based on specific characteristics (e.g., clinical test results and patients' demographics), may reduce postendodontic pain. We aimed to evaluate the relationship of data mining algorithms in longitudinal data of postendodontic pain and treatment allocation to predict the best treatment option. Materials and Methods: The pain data of an original multicenter randomized clinical trial with two study arms, pulpotomy with mineral trioxide aggregate (PMTA) (n = 188) and root canal therapy (RCT) (n = 168), were used. The linear mixed-effects model and predictive algorithms were fitted in accordance with the personal characteristics of patients and diagnostic test results to determine the best treatment option. Using SPSS 23, SAS 9.1, and WEKA 3.6.9, the preferred treatment was identified via comparing the areas below the receiver operating characteristic curves and identifying the most appropriate algorithm. In addition, a decision tree was used to allocate the best type of treatment modality to reduce posttreatment pain. Results: For <18-year-old patients with irreversible pulpitis (IP) based on cold test and >18-year-old patients whose electrical pulp test (EPT) exhibited IP, the chosen treatment would be RCT (P < 0.05). However, for >18-year-old patients with IP based on cold test and <18-year-old patients whose EPT revealed IP, the recommended treatment would be PMTA (P < 0.05). Conclusions: The decision tree model seems to be able to predict the reduction of postendodontic pain in ~65% of patients if they receive optimal treatment.