Risk factors and time-to-event prediction of brain metastasis in patients with lung adenocarcinoma: a retrospective cohort study

Background Brain metastasis (BM) is a clinically important complication of lung adenocarcinoma (LUAD), yet prediction is complicated by variable follow-up, censoring, treatment effects, and competing mortality. This study evaluated time-dependent BM risk and developed an exploratory prediction model in patients with LUAD without BM at baseline. Methods Consecutive patients with pathologically confirmed LUAD treated between 2018 and 2024 were retrospectively screened. Patients entered the cohort at the date of a negative baseline contrast-enhanced brain MRI. Incident BM was analyzed using Cox proportional hazards regression. Candidate predictors were prespecified, collinearity was assessed, and LASSO-penalized Cox regression was used for predictor reduction. Internal validation used 1,000 bootstrap resamples. Competing mortality was evaluated using Fine–Gray analysis, and EGFR-TKI exposure was examined as a time-dependent covariate. Results Of 94 screened patients, 66 were included; 26 developed incident BM during a median follow-up of 27.8 months. LASSO retained T classification, N classification, EGFR mutation status, and serum CEA. In the parsimonious Cox model, T3–4 disease, N1–3 involvement, and EGFR mutation were associated with higher BM hazard, while the CEA estimate was less precise. The apparent C-index was 0.781 and the optimism-corrected C-index was 0.742. Time-dependent AUCs were 0.773 at 12 months and 0.758 at 24 months. Competing-risk and treatment-adjusted analyses showed directionally consistent findings. Conclusion A Cox-based model integrating T stage, N stage, EGFR status, and CEA showed moderate internally validated performance for estimating 12- and 24-month BM-free probabilities. External validation is required before clinical implementation.