Predicting The Efficacy Of Chemoradiotherapy Based On Clinical,genomic And Radiomic Features In Locally Advanced Esophageal Cancer

Purpose:Definitive chemoradiotherapy is the standard of care for locally advanced unresectable esophageal squamous cell carcinoma(ESCC),but the overall efficacy is poor and individual differences are significant.How to predict the efficacy of chemoradiotherapy individually in patients with ESCC remains a major challenge.This study intended to screen out clinical and radiomics features associated with treatment efficacy in locally advanced ESCC patients receiving definitive chemoradiotherapy.Meanwhile,combined with the gene mutation characteristics of patients,a multi-omics prediction model based on clinical features,radiomics and gene mutation characteristics was constructed to provide comprehensive reference for the prediction of curative effect of definitive chemoradiotherapy.Methods:This study retrospectively included 118 patients with locally advanced ESCC who underwent definitive chemoradiotherapy,randomly separated into training(n=82)and validation(n=36)cohorts.Baseline tumor samples from all patients were tested by solid tumor gene sequencing technology to obtain the characteristics of CHEK2,NOTCH2 and homologous recombination repair(HRR)pathways.The 3D-Slicer software was used to outline the primary tumor as the region of interest(ROI)on the initial CT images of the patients.Radiomics features were extracted from ROI and perform feature dimensionality reduction in the training set to filter out radiomics features associated with efficacy and construct radiomics score(Rad-score).The correlations between clinical,radiomics and genetic characteristics and efficacy were analyzed using univariate and multifactorial COX regression.The area under the receiver operating characteristic curve(AUC)and the concordance index(C-index)were used to evaluate the predictive performance of the prediction models.Results:(1)A total of 851 radiomics features were extracted from the ROI of each patient.Six radiomics features related to progression free survival(PFS)were finally selected after features dimensionality reduction by LASSO method and combined with the corresponding weighting coefficients to construct the Rad-score.Multivariate COX regression analysis revealed that Rad-score(HR:2.052,95%CI:1.109-3.798,P=0.022)and HRR pathway alteration(HR:2.747,95%CI:1.313-5.748,P=0.007)were independent prognostic factors for PFS in patients with locally advanced ESCC treated with definitive chemoradiotherapy.(2)The radiomics prediction model was constructed using Rad-score=0.36 as the cutoff value.In the training set,the AUC values for the prediction of 1-year,2-year,and 3-year PFS probability with this model were 0.604,0.605,and 0.528,respectively,the C-index was 0.587.In the validation set,the corresponding AUC values were 0.618,0.662,and 0.73,respectively,the C-index was 0.625.(3)A genomics prediction model was constructed based on the HRR pathway status,which classified patients into high and low progression risk groups.In the training set,the AUC values for the prediction of 1-year,2-year,and 3-year PFS probability with this model were 0.615,0.593,and 0.589,respectively,the C-index was 0.557.In the validation set,the corresponding AUC values were 0.582,0.642,and 0.621,respectively,the C-index was 0.586.(4)A comprehensive prediction model was constructed based on Rad-score and HRR pathway status,which classified patients into high,intermediate,and low progression risk groups.In the training set,the AUC values for the prediction of 1-year,2-year,and 3-year PFS probability with this model were 0.676,0.662,and 0.594,respectively,the C-index was 0.616.In the validation set,the corresponding AUC values were 0.642,0.709,and 0.756,respectively,and the C-index was 0.649.Conclusion:In this study,we found that Rad-score and HRR pathway status are independent prognostic factors for PFS in locally advanced ESCC patients undergoing definitive chemoradiotherapy.The integrated model combining radiomics and genomics outperformed the radiomics or genomics models,with higher C-index and AUC values for predicting PFS in ESCC patients after definitive chemoradiotherapy.The integrated predictive models may be more useful in screening ESCC patients with different prognoses.