Multi-omics models predict treatment response and overall survival for non-small cell lung cancer patients following chemo-radiotherapy: A multi-center study

Abstract Background: Varying chemoradiotherapy outcomes in individuals arose from the intricate physical conditions and tumor heterogeneity characteristic of non-small cell lung cancer patients. This study aimed to develop and validate multi-omics models based on the radiomics, pathomics, dosiomics and clinical information for illustrating the heterogeneity and predicting treatment response and overall survival of non-small cell lung cancer patients. Methods: This retrospective study including 220 non-small cell lung cancer patients treated with chemoradiotherapy from three hospitals for overall survival prediction, with 142 of these patients specifically assessed for treatment response prediction. Radiomics and dosiomcis features were obtained from the region of interest, including first-order and texture features. Pathomics features were derived from whole slide images by Resnet34 network. Lasso regression, random forest, and extreme gradient boosting were employed for treatment response prediction to identify the most predictive biomarkers, with model performance evaluated through area under the curve and box plots. Overall survival analysis also involved three different feature selection methods, and model evaluation incorporated area under the curve, concordance index, Kaplan-Meier curves, and calibration curves. The shapley values calculated the contribution of different modality features to the models. Results: Multi-omics models consistently exhibited superior discriminative ability compared to single-modality models in predicting treatment response and overall survival. For treatment response, the multi-omics model achieved area under the curve values of 0.85, 0.81, and 0.87 in the training set, internal validation set, and external validation set, respectively. In the analysis of overall survival, the area under the curve and concordance index of the all-modalities model were 0.83/0.79, 0.74/0.74, and 0.73/0.72 in the training set, internal validation set, and external validation set, respectively. Conclusion: Multi-omics prediction models demonstrated superior predictive ability with robustness and strong biological interpretability. By predicting treatment response and overall survival in non-small cell lung cancer patients, these models had the potential to assist clinician optimizing treatment plans, supporting individualized treatment strategies, further improving tumor control probability and prolonging the patients’ survival.