Prediction of EGFR mutation status in non-small cell lung cancer based on CT radiomic features combined with clinical characteristics

To investigate the predictive value of combined radiomic features derived from chest CT scans with clinical characteristics for epidermal growth factor receptor (EGFR) gene mutations in non-small cell lung cancer (NSCLC).

A multi-center case-control study was conducted on the clinical data and CT images of 1070 NSCLC patients from the radiology departments of the 3 medical institutions between January 2013 and October 2023. The 719 NSCLC patients from the First Affiliated Hospital of Army Medical University were randomly divided into a training set and an internal validation set in a ratio of 7∶3; The 173 patients in the Eastern Theatre General Hospital and the 178 patients in Army Medical Centre of PLA were assigned into the external validation set 1 and 2, respectively. Least absolute shrinkage and selection operator (LASSO) regression was employed to identify the optimal radiomic features, which were subsequently used to construct a radiomics model. Univariate and multivariate logistic regression analyses were applied to identify clinical features associated with EGFR mutation, thereby developing a clinical model. The radiomic and clinical features were subsequently combined to develop a comprehensive model. All the 3 classification models were built using random forest (RF) machine learning. The area under curve (AUC), accuracy, sensitivity and specificity were utilized to evaluate the predictive performance of the models. Calibration curve was plotted to assess the goodness of fit of the comprehensive model, while decision curve analysis was performed to assess the clinical utility of the model.

The AUC value of the radiomics model was 0.7624 (95%CI: 0.6924~0.8251), 0.7454 (95%CI: 0.6711~0.8143), and 0.7247 (95%CI: 0.6397~0.8016), respectively, in the internal validation set, external validation set 1, and external validation set 2; The AUC value of the clinical prediction model was 0.6917 (95%CI: 0.6279~0.7576), 0.6525 (95%CI: 0.5767~0.7291), and 0.7792 (95%CI: 0.7125~0.8473), respectively in the above sets in turn; The comprehensive model constructed based on clinical features and radiomic features showed the best predictive efficacy, with an AUC value of 0.8180 (95%CI: 0.7577~0.8743), 0.7824 (95%CI: 0.7031~ 0.8482), and 0.7966 (95%CI: 0.7181~0.8686), respectively in the above sets. Calibration curve analysis indicated that the comprehensive model had a good fit, while decision curve analysis revealed that the model provided a favorable net benefit.

Our comprehensive model constructed based on chest CT radiomic features and clinical characteristics shows superior predictive performance for EGFR gene mutations in NSCLC across multiple center datasets, which may be helpful for clinical decision-making for treatment strategies.