Machine Learning to Predict Radiation Enteritis in Patients Undergoing Radical Radiotherapy for Cervical Squamous Cell Carcinoma

Abstract

Abstract Background Radiation enteritis (RE) is an adverse event associated with radical radiotherapy (RT) for cervical carcinoma (CC). However, the risk of RE has not been well predicted. We hypothesized that inflammatory markers of pre-/post-treatment complete blood count (CBC)-derived parameters can improve the predictive accuracy for RE using machine learning. Methods Patients with cervical squamous cell carcinoma of stage IB2-IIIB receiving radical RT in our hospital from January 1, 2013, to December 31, 2015, were included. Inflammatory markers of pre/post-treatment CBC-derived at the initial diagnosis and after RT were analyzed. A machine learning algorithm was used to develop a generalized linear model (GLM) for predicting RE risk. Results A total of 321 patients were eligible, of whom 39.3% (126/321) developed RE2 after RT, whether acute or chronic. The final predictive GLM for RE2 included an inflammatory marker, platelet-to-lymphocyte ratio (PLR1) (P = 0.021); age (P = 0.148); stage (P = 0.017); and RT technique (P = 0.047). A nomogram was constructed based on GLM. Decision curve analysis verified the better predictive power of the model for net clinical benefit. Conclusions This is the first study to determine the relationship between hematological inflammatory parameters and RE2 in patients with radical RT CC and to establish a relevant prediction model using machine learning. PLR1 was significantly associated with RE2. This study developed a comprehensive model integrating hematological inflammatory parameters and clinical variables to predict RE before RT, which provides an opportunity to guide clinicians.