Therapeutic Resistance Prediction of Tyrosine Kinase Inhibitors

The investigators develop a multi-CT-phenotypic-feature-based classifier to predict TKI benefit and therapeutic resistance for stage IV EGFR-mutant non-small cell lung cancer (NSCLC). The investigators also compared its prognostic and predictive efficacy with single features and clinicopathological risk factors. An individualized nomogram integrated the classifier and three clinicopathological risk factors was built for clinical use. The prognostic accuracy of the proposed model was evaluated in two independent validation sets. Nearly 500 patients will be enrolled in this clinical trial. Eligible patients were diagnosed with NSCLC, and stage IV according to the TNM system classification of the American Joint Committee on Cancer, presence of activating EGFR mutations, age 20 years or older, and no history of systemic anticancer therapy for advanced disease. Patients who underwent first-line or second-line EGFR TKIs were eligible for inclusion. All patients had to be capable of undergoing contrast-enhanced CT, and pretreatment CT was strictly controlled in two weeks before the EGFR TKIs starts. Patients who underwent resection for local advanced or metastatic disease were withdrawn from the study. Therapeutic resistance was measured by PFS, as the time from the initiation of EGFR TKIs therapy to the date of confirmed disease progression or death. PFS was censored at the date of death from other causes, or the date of the last follow-up visit for progression-free patients. The investigators will use extracted 1000 phenotypic features on the region of interest manually segmented by radiologists. The Lasso Cox regression model and Nomogram will be used to build a prognosis model for the therapeutic resistance prediction of EGFR TKIs for stage IV EGFR-mutant NSCLC. The Harrell's concordance index(C-index) of the proposed nomogram will be used to quantify the discrimination performance.