Development of a Web-Based Radiation Toxicity Prediction System Using Metarule-Guided Mining to Predict Radiation Pneumonitis and Esophagitis in Lung Cancer Patients

Abstract

Radiation toxicity grades must be reviewed based on existing knowledge-based clinical evidence or be assessed by referencing the clinical research literature to minimize radiation toxicity associated with radiotherapy. The aim of this study was to develop a radiation toxicity prediction system using metarule-guided mining of the clinical research literature to predict radiation pneumonitis and esophagitis in lung cancer patients. A semantic pattern database was built using 100 clinical research articles. Semantic patterns of prognostic factors and toxicity grades were extracted by metarule-guided mining. Feature analysis and correlation investigation of prognostic factors and toxicity grades were performed by using dendrogram and heatmap. A web-based user interface for the prediction system was designed. Patient prognostic factors were used in this prediction system to predict toxicity grade results. Radiation toxicity grades for patients with radiation pneumonitis and esophagitis were calculated through our prediction system. Age and chemotherapy were prognostic factors as were toxicity grades 1–3 and 5 based on the metarule-guided mining system. The odds ratios had similar trends to those in the existing meta-analysis literature. The radiation toxicity prediction system that we developed can potentially be used as a clinical decision support system for patient-specific radiation treatment after weighting of prognostic factors, performing a correlation analysis, and performing a validity evaluation.