A nomogram to predict development of brain metastasis in non-small cell lung cancer patients: a retrospective analysis using routinely available medical records.

Background: Brain metastases (BrM) are a frequent complication among patients with non-small cell lung cancer (NSCLC). While guidelines exist for baseline CNS screening in advanced NSCLC, surveillance strategies for early-stage disease remain limited. This study aimed to develop a time-dependent BrM risk prediction nomogram using readily available clinical information. Methods: We analyzed a retrospective cohort of NSCLC patients at Penn State Health. Our objectives were to (1) systematically evaluate the performance of existing BrM risk prediction algorithms and (2) construct novel nomograms for BrM risk prediction in NSCLC. Using Cox-proportional hazard models with L1-regularization, we predicted BrM risk at 6-month, 1-year, and 2-year follow-up intervals. Findings: The patient cohort included 1904 patients (median age 68 years, range 38-94 years, BrM incidence 22.8%). The cohort included 1059 males (55.6%) and 845 females (44.4%). Of the cohort, 92.8% of patients identified as White (n = 1766), 1.0% as Asian (n = 19), 4.0% as Black (n = 77), and 2.2% as another race (n = 42). The Zhang 2021 model demonstrated the highest performance in predicting BrM incidence in our cohort, achieving an AUROC of 0.91 (95% CI: 0.87, 0.95). Two novel models were developed: a baseline model incorporating clinical and imaging data at diagnosis (cTNM stage, age at diagnosis), and an extended model including additional clinical and treatment data (number of extracranial metastatic sites, prior radiotherapy, chemotherapy, surgery, and histology) (https://nmikolajewicz.shinyapps.io/nomogram_wilding2024/). While both models showed similar short-term performance, the extended model demonstrated superior predictive capacity (AUROC 0.91 at 3-years) for longer-term outcomes. Our nomograms rely exclusively on clinical features routinely documented in patient records, thereby requiring no additional investigations. Interpretation: These clinically accessible nomograms for BrM prediction will facilitate prognostic modeling, risk stratification, refinement of CNS screening guidelines, and patient counseling. Funding: None.