Multi-omics-based prognostic prediction for locally advanced hypopharyngeal cancer treated with surgery and postoperative chemoradiotherapy: a dual-center study

Purpose This study aimed to predict the progression-free survival (PFS) of the patients who were diagnosed with hypopharyngeal cancer and received postoperative chemoradiotherapy by using multi-omics method which integrating clinical factors, dosimetric and radiomic features. Materials and methods This study retrospectively collected the pretreatment T1-weighted MR imaging data of 88 hypopharyngeal cancer patients with postoperative chemoradiotherapy, including 56 cases from one center (training and testing cohorts) and 32 cases from another center (external validation cohort), and the gross tumor volumes (GTV) were countered for all cases. A Python-based library, pyradiomics was used to extract the radiomics features from each GTV. Least absolute shrinkage and selection operator (LASSO) regression was used to identify the most important features for classifier establishment. On the other hand, complete radiotherapy data are retained for 48 patients among them, and the planning tumor volumes (PTV) were countered for radiotherapy planning. The dose distribution features extracted by using pyradiomics and the dosimetric parameters were combined with the radiomics features to establish the classifiers. The probabilities of positive sample calculated from the best classifier, the radiomics and multi-omics signatures were obtained for establish the Cox proportional hazards models. Results The ensemble learning (EL) model was selected as the superior model with the higher area under the receiver operating characteristic curve (AUC) values than other classifier during the radiomics-only analysis, and the EL model with stacking technique showed the best performance, yielding AUC values of 0.93, 0.79, and 0.78 for the training, testing, and external validation cohorts, respectively. Furthermore, the multi-omics analysis integrating radiomics and dosiomics improved the effectiveness of the EL model with AUC values of 0.98 and 0.88 for the training and testing cohorts, respectively. Furthermore, the C-index of the Cox proportional hazards models resulted in a 0.099 improvement in the testing cohort when employing the multi-omics signature versus the radiomics signature. Conclusion Regarding the patients with hypopharyngeal cancer receiving postoperative chemoradiotherapy, the multi-omics-based prognostic prediction could achieve a more robust predictive capability than the radiomics-only study. This approach warrants further validation through prospective studies.