Machine learning-driven clinical decision support for radical local consolidative therapy in synchronous oligometastatic NSCLC: A SEER population-based analysis of 17 cancer registries (2018-2021).

[BACKGROUND] Synchronous oligometastatic nonsmall cell lung cancer (NSCLC) is a unique clinical entity with potential benefit from radical local consolidative therapy (LCT). However, robust tools to stratify patients most likely to benefit from radical LCT are lacking.

[MATERIALS AND METHODS] We identified patients with synchronous oligometastatic NSCLC from the Surveillance, Epidemiology, and End Results 17 registry (2018-2021). Radical LCT was defined as surgical resection of the primary tumor (±metastasectomy) and/or radiotherapy. Patients were stratified accordingly. Survival was analyzed by Kaplan-Meier and Cox models with inverse probability weighting to adjust for confounding. A weighted random survival forest (RSF) model and SHAP analysis were applied to capture nonlinear interactions and key prognostic determinants. Model discrimination was assessed by C-index, AUC, and integrated Brier score (IBS).

[RESULTS] Among 5377 patients, 221 received radical LCT. The radical LCT group was younger, had earlier T stage, smaller tumors, and more frequent brain metastases. Median overall survival (mOS) was significantly longer in the radical LCT group than the no-LCT group (27 vs. 7 months; P < 0.001), with maximal benefit among those also receiving chemotherapy (mOS: 30 months). The RSF model outperformed Cox regression in training (C-index: 0.752 vs. 0.735) with similar validation results. Chemotherapy (HR = 0.41; 95% CI, 0.36-0.46; P < 0.001) and radical LCT (HR = 0.59; 95% CI, 0.38-0.91; P = 0.018) were independent favorable factors. SHAP analysis demonstrated that radical LCT and chemotherapy contributed most to risk reduction in the survival prediction model, with consistent protective effects observed across different age, T stage, and histological subgroups.

[CONCLUSION] This multicenter, population-based study is the first to establish and internally validate a machine learning-based risk stratification tool for patients with synchronous oligometastatic NSCLC. Radical LCT, particularly when combined with chemotherapy, was associated with improved OS in this cohort. The model may facilitate preliminary risk stratification and assist multidisciplinary management, but external validation and integration of key clinical and molecular variables are needed before clinical application. Prospective multicenter validation is warranted.