| Objective:Accurate evaluation of human epidermal growth factor receptor 2(HER2)status in patients with advanced gastric cancer(AGC)is crucial for achieving personalized medical treatment and improving prognosis.However,the currently used HER2 testing technique involves an invasive and time-consuming diagnostic procedure.Therefore,this study aimed to develop a CT radiomics-based diagnostic model to help physicians identify potential HER2-positive patients from AGC patients.Then the efficacy and generalization ability of the model was verified and evaluated to explore its feasibility as an imaging biomarker.Materials and Methods:A total of 950 patients with pathologically confirmed gastric cancer and met the criteria for advanced-stage disease in our hospital(tertiary health-care institution)were retrospectively recruited and allocated to a training cohort(n=388,underwent conventional CT examinations)or an internal validation cohort(n=325,underwent conventional CT examinations)or an external validation cohort(n=237,underwent spectral CT examinations)according to the time order of diagnosis and CT protocol.All patients had pre-treatment enhanced CT images of the abdomen and accurate information of HER2 expression status.The general clinical data,CT and laboratory characteristics,and venous phase(VP)CT images of each patient at the time of initial diagnosis were collected.In the training cohort,the support vector machine(SVM)algorithm was first used to construct Radscore based on VP CT images of each patient,as a predictor of HER2 status of AGC.Then,a radiomics model was built by combining significant clinical-laboratory variables and Radscore on the basis of univariate analysis and multivariate logistic regression.The diagnostic effectiveness and clinical practicability of the radiomics model were evaluated and determined using the area under curve(AUC)of the receiver operating characteristic curve(ROC),and decision curve analysis(DCA),respectively.The 95%CIs of the AUCs were also calculated.Validation techniques were employed to quantitatively assess the diagnostic performance,efficacy stability,and generalization ability of the radiomics model in internal and external validation cohorts.Stratification analysis and Delong test were conducted to investigate the certain universality and stability of the model.Results:Tumor location,clinical TNM stage(cTNM),carcinoembryonic antigen(CEA),carbohydrate antigen 199(CA199),and Radscore were independent predictors of HER2 status in AGC.Incorporating the above five variables,the constructed radiomics model yielded favorable diagnostic abilities of HER2 status,with AUCs of 0.732(95%CI:0.683-0.781),0.703(95%CI:0.624-0.783),and 0.711(95%CI:0.625-0.798)observed for the training,internal validation,and external validation cohorts,respectively.The DeLong test showed no statistical difference between the AUCs in the three cohorts(P=0.546,0.678,and 0.897),suggesting that the conventional CTbased radiomics model was robust,and can be generalized to AGC patients with spectral CT scan.The DCA revealed that the radiomics model would offer more net benefits for patients in all three study cohorts than the default simple "all HER2targeted therapy" scheme(to consider all patients as having HER2-postive tumors)or the "none HER2-targeted therapy" scheme(to consider all patients as having HER2negative tumors),with the threshold probability of the clinical decision ranging from 10%to 80%,from 20%to 65%,and from 10%to 75%,respectively.Furthermore,the performance of radiomics model was not affected by the age,gender,tumor location,IHC results,and type of tissue for confirmation(Delong test,all P>0.05),indicating its generality on different kinds of AGC population.Conclusion:In this present study,we constructed and validated a conventional CT-based radiomics model for decoding HER2 status in patients with AGC.This model had good discrimination performance and the potential to generalize to spectral CT,which is beneficial to simplify clinical workflow and help clinicians initially identify potential candidates who might derive benefit from HER2-targeted therapy.In a nutshell,the presented radiomics model holds tremendous promise in becoming a noninvasive imaging biomarker for detecting HER2 status in AGC.Objective:Trastuzumab-containing systemic therapy has become the preferred treatment for HER2-positive late-stage metastatic gastric cancer(HER2-p LSMGC).Unfortunately,not all HER2-p LSMGC patients benefit from trastuzumab-containing regimens.Therefore,the exploration of accurate prognostic and predictive biomarkers that can assist physicians in identifying which patients will respond to the trastuzumab-specific therapy can guide the application of HER2-targeted drugs more accurately and meet the requirements of precision medicine.In this study,we aimed to establish a CT radiomics-based imaging biomarker that could predict the prognostic of HER2-p LSMGC patients and attempted to make trastuzumab-based decision recommendations according to the risk stratification of patients with different treatment regiments.Materials and Methods:A total of 123 patients with pathologically confirmed HER2-p gastric cancer and met the criteria for late-stage metastatic disease in hospital 1(tertiary health-care institution)were retrospectively selected.After propensity score matching patients with different treatment regimens based on age,gender,and clinical TNM staging,116 HER2-p LSMGC patients were enrolled and assigned into a training cohort.Meanwhile,56 consecutive gastric cancer patients meeting inclusion criteria in hospital 2(tertiary medical institution)were recruited as an independent external validation cohort.All patients had pre-treatment enhanced CT images and received systematic therapy with or without trastuzumab at either of the two hospitals.The general clinical data,CT and laboratory characteristics,follow-up information,and venous phase(VP)CT images of each patient at the time of initial diagnosis were collected.The primary endpoint of interest was overall survival(OS).In the training cohort,by conducting the univariate Cox regression,multivariate Cox regression analysis,and random survival forest,the radiomics signature(Radscore)based on VP CT images,the clinical model based on clinical traits,and the CT radiomics model incorporated Radscore and the clinical score of the clinical model was respectively built,as the candidate biomarkers for the prognosis of HER2-p LSMGC patients.Kaplan-Meier estimator and Log-rank test were performed to generate OS and assess the statistical significance of observed differences in survival.Harrell’s C-index and Delong test were employed to evaluate and compare the prognostic performance of the three candidate biomarkers,further selecting the optimal prognostic biomarkers.With the selected prognostic biomarker,patients who received the trastuzumab-containing regimens and the trastuzumab-free regimens were analyzed for risk stratification,and treatment decisions were accordingly recommended.External validation techniques were employed to evaluate the value of the prognostic markers in guiding treatment decisions in other participant data.The certain universality and additional prognostic generalization of prognostic biomarkers were also investigated.Results:The Radscore,clinical model,and radiomics model were correlated with the OS of patients with HER2-p LSMGC,with C-index of 0.749[95%confidence interval(CI):0.692-0.805;P<0.001],0.650(95%CI:0.583-0.717;P<0.001)and 0.756(95%CI:0.701-0.810;P<0.001).Delong test suggested no significant difference between the prognostic efficacy of the Radscore and the radiomics model in patients with HER2-p LSMGC(P>0.05).By balancing the performance and clinical practicability,the Radscore was finally selected as an effective biomarker for predicting the prognosis of HER2-p LSMGC.Stratified analysis showed that,among the high prognostic risk population assessed with Radscore,HER2-p LSMGC patients treated with trastuzumab had a significantly better outcome than that without trastuzumab,whether in the training or external validation cohorts(all P<0.05).However,there was no statistical significance in OS between patients treated with trastuzumab and those treated without trastuzumab in low prognostic risk group determined by Radscore(all P>0.05).It is suggested that trastuzumab may be recommended for high-risk populations stratified by Radscore,and systematic therapy without trastuzumab can be considered appropriate for the low-risk population.Moreover,Radscore can successfully classify HER2-p LSMGC patients≥43 years of age and recruited from an intentionally different setting into high-risk and low-risk groups.Furthermore,Radscore was not affected by the gender and treatment schemes of the patients(all P<0,05),indicating that the established Radscore has certain prognostic universality and generalization ability.Conclusion:In this present study,we built and externally validated a Radscore based on pretreatment CT that can be used to predict the prognosis of HER2-p LSMGC and guide trastuzumab-based treatment decisions.The CT-based Radscore had certain universality and good generalization ability in different types of patients.Our Radscore may act as a potential tool for promoting the personalized treatment and optimizing management of HER2-p LSMGC.Objective:Accurate risk stratification for the prognosis of patients with HER2-positive surgically resectable advanced gastric cancer(HER2-p SRAGC)can provide a basis for strengthening follow-up monitoring of recurrence and metastasis in high-risk groups,thus facilitating early intervention of HER2-specific drugs and improving the prognosis of patients.Therefore,this study aimed to develop and externally validate a CT radiomics model for predicting the outcome of HER2-p SRAGC before treatment.Moreover,the prognostic efficacy of the individualized radiomics model and the 8th edition TNM staging system was quantitatively compared,and the added value of the model for outcome prediction of the TNM staging system was investigated.Materials and Methods:A total of 473 patients with pathologically confirmed HER2-p SRAGC admitted to hospital 1(tertiary health-care institution)were retrospectively selected and randomly divided into a training cohort(n=330)and an internal validation cohort(n=143).Meanwhile,148 consecutive gastric cancer patients meeting the above criteria in hospital 2(tertiary health-care institution)were recruited as an independent external validation cohort.All patients had pre-treatment enhanced CT images and received either radical gastrectomy or radical gastrectomy after neoadjuvant therapy or chemotherapy at either of the two hospitals.The general clinical data,CT and laboratory characteristics,follow-up information,and venous phase(VP)CT images of each patient at the time of initial diagnosis were collected.The primary endpoint of interest was overall survival(OS).In the training cohort,by conducting the LASSOCox,random survival forest,univariate Cox regression,and multivariate Cox regression analysis based on VP CT images,general CT traits,and laboratory data,the radiomics signature(Radscore),ct_score,and lab_score were respectively built as predictors for the prognosis of HER2-p SRAGC.Incorporating the Radscore,ct_score,and lab_score,an individualized radiomics model was developed using multivariate Cox regression analysis.Kaplan-Meier estimator and Log-rank test were performed to generate OS and assess the statistical significance of observed differences in survival.Harrell’s C-index and validation techniques were employed to evaluate the prognostic efficacy and generalization ability of the radiomics model.Stratification analysis was applied to analyze the prognostic efficacy of the radiomics model in HER2-p SRAGC patients receiving different treatment regimens and compared it with corresponding clinical TNM(cTNM),pathological staging(pTNM),and post-neoadjuvant staging(ypTNM),with the gain effect of the radiomics model on TNM staging,quantified.Results:The radiomics model based on the Radscore,ct_score,and lab_score was significantly correlated with the OS of patients with HER2-p SRAGC.The C-index yielded was 0.711[95%confidence interval(CI):0.666-0.756;P<0.001],0.669(95%CI:0.585-0.753;P=0.009)and 0.693(95%CI:0.597-0.789;P=0.015)for the training,internal,and external validation cohorts,respectively.In all study cohorts,with a cutoff value of 0.048,the radiomics model successfully stratified patients into high-risk and low-risk groups.And the prediction efficiency and stability of the model were better than those of Radscore(C-index of 0.685,0.634 and 0.616,respectively),ct_score(C-index of 0.626,0.575 and 0.593,respectively)and lab_score(C-index of 0.579,0.617,and 0.655,respectively).Stratification analysis showed that the model was better than pTNM(C-index:0.685 VS 0.636)and cTNM(C-index:0.711 VS 0.591),but was inferior to ypTNM(C-index:0.502 VS 0.750),in predicting OS in patients with different treatment regimens.Meanwhile,the radiomics model had added value to the prognostic efficacy of pTNM,and the combined C-index of the two can reach 0.710.In addition,the model was not affected by the age and gender of the patients at first diagnosis.Conclusion:In this study,a radiomics model integrating Radscore,ct_score,and lab_score was constructed and externally verified for individualized prediction of the prognosis of HER2-p SRAGC patients.The efficacy of this model in predicting OS of HER2-p SRAGC was superior to the clinical scores based on clinical data alone,as well as the two TNM staging of the new TNM staging system.Moreover,the model had a valueadded effect on the prognosis of the pTNM staging system and had certain universality in different types of patients. |
PDF Full Text Request