Machine learning to build and validate a model for radiation pneumonitis prediction in patients with non–small cell lung cancer

Purpose: Radiation pneumonitis is an important adverse Results: A total of 131 patients were eligible and 17 (13.0%) event in patients with non–small cell lung cancer (NSCLC) developed RP2. IL8 and CCL2 had significantly (Bonferroni) receiving thoracic radiotherapy. However, the risk of radiation lower expression levels in patients with RP2 than without RP2. pneumonitis grade 2 (RP2) has not been well predicted. This But none of the changes in cytokine levels during radiotherapy study hypothesized that inflammatory cytokines or the was significantly associated with RP2. The final predictive GLM dynamic changes during radiotherapy can improve predictive model for RP2 was established, including IL8 and CCL2 at accuracy for RP2. baseline level and two clinical variables. Nomogram was Experimental Design: Levels of 30 inflammatory cyto-constructed based on the GLM model. The model's predicting kines and clinical information in patients with stages I–ability was validated in the completely independent test set III NSCLC treated with radiotherapy were from our (AUC ¼ 0.863, accuracy ¼ 80.0%, sensitivity ¼ 100%, spec-prospective studies. Statistical analysis was used to select ificity ¼ 76.5%). predictive cytokine candidates and clinical covariates for Conclusions: By machine learning, this study has developed adjustment. Machine learning algorithm was used to and validated a comprehensive model integrating inflamma-develop the generalized linear model for predicting risk tory cytokines with clinical variables to predict RP2 before RP2. radiotherapy that provides an opportunity to guide clinicians.