Machine learning models based on clinical and PET/CT imaging biomarkers for predicting prognostic value of concurrent chemoradiation therapy in stage Ⅲ NSCLC patients
Objective To explore the potential value of machine learning models based on PET/CT images for predicting prognosis in patients with stage Ⅲ non-small cell lung cancer(NSCLC)receiving concurrent chemoradiotherapy(CCRT).Methods Clinical data of 100 NSCLC patients treated with CCRT at Shandong Cancer Hospital from January 1,2019,to December 31,2021,were retrospectively analyzed and randomly divided into training cohort(n=70)and valida-tion cohort(n=30)in a ratio of 7∶3.Clinical features were selected by univariate and multivariate Cox proportional haz-ards model.The Least Absolute Shrinkage and Selection Operator(LASSO)algorithm was used to screen PET and CT imaging biomarkers.Clinical models,radiomics models,combined models,and nomograms were constructed.The receiver operating characteristic curves,decision curves,Kaplan Meier curves and calibration curves were used to evaluate their per-formance.Results The median value of the radiomics score(RS)in the training cohort was 0.21.Survival curve analysis revealed that the high-risk group with RS≥0.21 had worse prognosis than the low-risk group with RS＜0.21 in the train-ing cohort,and the difference was statistically significant(x2=8.661,P=0.003).Similarly,a statistically significant difference in survival was observed between the high-risk group(RS≥0.21)and the low-risk group(RS＜0.21)in the validation cohort,with X2=5.833 and P=0.016.Multivariate Cox regression analysis showed that smoking history(HR=0.405,95％CI:0.205-0.800,P=0.009),lymph node status(HR=1.717,95％CI:1.013-2.908,P=0.045),and consolidation immunotherapy(HR=0.443,95％CI:0.223-0.877,P=0.019)were independent prognostic factors for progression-free survival(PFS)in patients with stage Ⅲ NSCLC.A clinical model was constructed using the three se-lected clinical characteristics.Stratified by consolidation immunotherapy,univariate Cox regression analysis revealed an as-sociation between the lung immune prognostic index(LIPI)and PFS in patients with stage Ⅲ NSCLC who received con-solidation immunotherapy(P=0.028).The area under the curve(AUC)for predicting PFS in the training cohort was 0.674 for the clinical model,0.771 for the radiomics model,and 0.803 for the combined model.In the validation cohort,the AUC for predicting PFS was 0.655 for the clinical model,0.670 for the radiomics model,and 0.710 for the combined model.Decision curve analysis demonstrated that the combined model was superior to the clinical or radiomics models in predicting survival in patients with stage Ⅲ NSCLC.Conclusion PET/CT radiomics can predict the survival of patients with stage Ⅲ NSCLC after chemoradiotherapy with good performance.The combined model has the best predictive per-formance,and the visual nomogram can assist in clinical decision-making.
NETL
NSTL
万方数据
