Comprehensive nomogram models for prediction of checkpoint inhibitors pneumonia

Abstract Introduction: Checkpoint inhibitors pneumonia (CIP) is a common type of irAEs with poor clinical prognosis. Currently, there is a lack of effective biomarkers and predictive models to predict the occurrence of CIP.Materials and Methods: The study retrospectively included 327 patients who received immunotherapy for the first time as a training cohort and 167 patients as a validation cohort. Patients were assessed for the occurrence of CIP, and were divided into any grade CIP cohort, grade ≥2 CIP cohort, and grade ≥3 CIP cohort. Multivariate logistic regression analyses were used to determine the independent risk factors. Based on that, predictive nomograms were constructed and validated to predict the risk of CIP. Results: The overall incidence of CIP was 18.61%. Based on the results of multivariate analysis, we established nomogram models for predicting any grade CIP and grade ≥2 CIP, respectively. For the nomogram to predict any grade CIP, the C index of the model in the training cohort and validation cohort were 0.814 (95%CI=0.760-0.870) and 0.919 (95%CI=0.842-0.941), respectively. Similarly, as for the nomogram of grade 2 or higher CIP, the C index of the model in the training cohort and validation cohort were 0.879 (95%CI=0.81-0.92) and 0.936 (95%CI=0.812-0.982), respectively. Conclusions: After internal verification and external verification, the predictive power of nomogram models is satisfactory and they are expected to be a convenient, visual, and personalized clinical tool for assessing the risk of developing CIP in patients receiving ICIs treatment.