A multiomics approach-based prediction of radiation pneumonia in lung cancer patients: impact on survival outcome

Abstract

Purpose: To predict the risk of radiation pneumonitis (RP), a multiomics model was built to stratify lung cancer patients. Our study also investigated the impact of RP on survival.

Methods: This study retrospectively collected 100 RP and 99 matched non-RP lung cancer patients treated with radiotherapy from two independent centres. They were divided into training (n = 175) and validation cohorts (n = 24). The radiomics, dosiomics and clinical features were extracted from planning CT and electronic medical records and were analysed by LASSO Cox regression. A multiomics prediction model was developed by the optimal algorithm. Overall survival (OS) between the RP, non-RP, mild RP, and severe RP groups was analysed by the Kaplan‒Meier method.

Results: Sixteen radiomics features, two dosiomics features, and one clinical feature were selected to build the best multiomics model. The optimal performance for predicting RP was the area under the receiver operating characteristic curve (AUC) of the testing set (0.94) and validation set (0.92). The RP patients were divided into mild (≤ 2 grade) and severe (> 2 grade) RP groups. The median OS was 31 months for the non-RP group compared with 49 months for the RP group (HR = 0.53, p = 0.0022). Among the RP subgroup, the median OS was 57 months for the mild RP group and 25 months for the severe RP group (HR = 3.72, p < 0.0001).

Conclusions: The multiomics model contributed to improving the accuracy of RP prediction. Compared with the non-RP patients, the RP patients displayed longer OS, especially the mild RP patients.

Keywords: Lung cancer; Machine learning; Multiomics; Radiation pneumonitis; Survival outcome.