Prediction recurrence in stage I epidermal growth factor receptor-mutated non-small cell lung cancer using...

Integrated recurrence prediction models that combine clinical, imaging, and genetic data are lacking for epidermal growth factor receptor (EGFR)-mutated stage I non-small cell lung cancer (NSCLC). We developed a recurrence prediction model for Stage I EGFR-mutated NSCLC by integrating clinical, radiological, and whole-exome sequencing (WES) data.

A total of 306 patients with Stage I EGFR-mutated NSCLC were stratified into training (n = 206) and validation (n = 100) cohorts using stratified random sampling. Cox proportional hazards models combined binary labels of clinical and radiological factors from univariate analysis, machine-learning-derived MultiGenes labels (determined by multiple gene mutations) from WES data, and high-impact gene mutations. Predictive performance was assessed using the concordance index (C-index), time-dependent area under the curve (AUC), and receiver operating characteristic curves. The top three models by category were evaluated using a survival analysis.

Three optimal models were identified; the clinicoradiological model (Model 17) achieved a C-index of 0.70, the model incorporating clinicoradiological factors and MultiGenes (Model 28) achieved 0.69, and the clinicoradiological model with TP53 mutations (Model 39) demonstrated the best performance, with 0.73. In Model 17, the 60-month recurrence-free survival (RFS) rates were 59.1 % for the high-risk group and 83.2 % for the low-risk group (hazard ratio [HR] = 3.47; 95 % confidence interval [CI]: 1.60-8.00). Model 39, which incorporated TP53 mutations, demonstrated superior performance, with 60-month RFS rates of 57.1 % for the high-risk and 87.1 % for the low-risk groups (HR = 4.79; 95 %CI: 1.96-11.69).

Clinical and radiological factors are effective predictors of recurrence risk in Stage I EGFR-mutated NSCLC, and incorporating TP53 mutation data further improves the prognostic accuracy.