A Nomogram Model To Predict The Response And Prognosis Of Immunotherapy For Non-small Cell Lung Cancer Was Established Based On Radiomics And Clinical Features

Objective:Patients with advanced non-small cell lung cancer(NSCLC)who have lost the opportunity for surgery due to distant metastasis or other reasons have a poor prognosis.Currently,single-agent therapy using programmed cell death protein 1(PD-1)and its ligand(programmed death-ligand 1,PD-L1)inhibitors,as well as combination therapy with PD-1/PD-L1 inhibitors and chemotherapy,have become the standard treatment for advanced NSCLC in China.However,not all patients can benefit from immunotherapy.Therefore,it is crucial to identify predictive biomarkers of treatment response in order to more accurately screen potential beneficiaries.To this end,we have established a clinical-radiomics nomogram based on the baseline clinical characteristics and CT radiomics features of patients as core indicators.The model can predict the efficacy and survival outcomes of NSCLC patients receiving targeted PD-1/PD-L1 immunotherapy more accurately.Methods:This study included 187 patients with advanced non-small-cell lung cancer who received immunotherapy at Shandong Provincial Hospital and met the inclusion criteria from January 2019 to July 2021.The follow-up period focused on the endpoints of progression-free survival(PFS)and overall survival(OS).Clinical information and baseline chest-enhanced CT images,including gender,age,tumor histology,smoking history,etc.,were collected.Patients were randomly divided into a training set and a validation set at a 7:3 ratio.The region of interest(ROI)was delineated using the 3D-Slicer software,and the PyRadiomics package was used to extract image features.The best features related to treatment response were selected using the Least absolute shrinkage and selection operator(LASSO)algorithm in the training set,and a radiomics signature was constructed.Important clinical features related to immunotherapy response were screened by using univariate logistic regression analysis.A multivariate logistic regression model for predicting immunotherapy response was established by combining radiomics signature and clinical features,and a visualization nomogram was created.The performance of the model was validated by using the area under the receiver operating characteristic curve(AUC)and evaluated by using calibration and clinical decision curves.Based on the cutoff point obtained from the training set ROC curve,patients in the validation set were divided into high and low risk groups,and the PFS and OS of the two groups were compared by using Kaplan-Meier curves.Furthermore,the clinical-radiomic nomogram model was analyzed in subgroups of first-line,second-line,and later-line immunotherapy to evaluate its applicability across treatment lines.Results:1.Out of 187 patients,132 exhibited a therapeutic response or disease stabilization after receiving PD-1/PD-L1 immune therapy.During a median follow-up period of 633.5 days(ranging from 61 to 1328 days),disease progression occurred in 154 cases and 72 patients died.2.A total of 851 imaging features were extracted from the chest-enhanced CT.And there were 363 robust image features screened by intraclass correlation coefficient(ICC).3.Using Lasso analysis on the training set,9 image features were identified as the most relevant to the response to immunotherapy.Based on these features,radiomics signatures were constructed and radiomics score(Radscore)were computed.4.In the training set,serum levels of CRP,CEA,and metastasis status were identified as three significant clinical features associated with immune therapy response through the screening of univariate logistic regression.5.Multivariate logistic regression was used to construct a predictive immunotherapy response nomogram model with Radscore and clinical features,which had good results(training set AUC:0.881;validation set AUC:0.850)and was higher than the radiomics model alone(training set AUC:0.839;validation set AUC:0.774)and the clinical factors model alone(training set AUC:0.667;validation set AUC:0.727).The calibration curves of the training and validation sets showed that the model predicted immunotherapy response with a high agreement with the actual immunotherapy response.6.Decision curve analysis yielded a greater net benefit of the combined clinical and radiomics nomogram model than the no radiomics model.Kaplan-Meier survival analysis showed shorter PFS and OS in the high-risk group in the validation set compared to the low-risk group(PFS:log-rank p=0.023,OS:log-rank p=0.01).7.In the three subgroups of immunotherapy as first-line,second-line and later-line,the AUC values of the clinical-radiomics Nomogram for predicting response to immunotherapy were 0.932,0.875 and 0.824,respectively,indicating that this clinical-radiomics Nomogram has good predictive efficacy for different lines of immunotherapy.Conclusion:1.A model based on CT radiomics and clinical features can predict response and survival outcomes in patients with advanced NSCLC treated with PD-1/PD-L1 immunotherapy.2.In immunotherapy as first-line,second-line and later-line subgroups,the clinical-radiomics nomogram had good predictive efficacy for different lines of immunotherapy.3.This clinical-radiomics nomogram can be used as a surrogate marker for predicting the efficacy of immunotherapy,providing an important decision-making tool for planning immunotherapy.