Predictive Algorithms for Brain Metastases in Non-small Cell Lung Cancer: Clinically Impactful Insights from a Systematic Review and Independent Cohort Validation Analysis

Introduction: Upwards of 40% of non-small cell lung cancer (NSCLC) patients will develop brain metastases (BM). Validated predictive algorithms facilitate development of brain-directed prophylaxis. We undertook a systematic review to identify BM predictive algorithms in NSCLC. We used our independent dataset, one of the largest to date, to validate the top-performing algorithm and develop a novel independent predictor.

Methods: We performed a systematic review of studies reporting statistical models to predict BM occurrence in NSCLC, then benchmarked these models using an independent cohort of NSCLC patients in electronic medical records (2011-2020) from Penn State Health. For modeling, patient data were randomly split into 70% training, used to train L1-regularized (Lasso) logistic regression models, and 30% testing. Performances were evaluated using receiver operating characteristic (ROC) analyses and summarized using area under the ROC curve (AUROC).

Results: Twenty-two of 1,643 publications met our inclusion criteria. Twelve studies (572,258 patients) included variables consistently available in patient charts. Our internal cohort included 1,088 NSCLC patients; median age at diagnosis was 68; 19.9% developed BM. Of the feasible models, Zhang 2021 performed highest within our cohort (AUROC, 0.91; 95% CI, 0.86-0.96); age at diagnosis; surgical, chemotherapy, and radiation status; T and N stage; histological grade; and number of organs with metastases were significant independent factors for predicting BM. Among our patients, the most informative predictors using linear interpolation were number of organs with metastases (OR [95% CI], 2.71 [2.17-3.39]), age at diagnosis (OR [95% CI], 0.97 [0.96-0.99]), N-stage 3 at diagnosis (OR [95% CI], 2.26 [1.23-4.17]), and lung radiation treatment (OR [95% CI], 1.64 [1.09-4.17]).

Conclusion : Our robust approach, including systematic review and one of the largest-to-date independent institutional datasets, proposes a clinically feasible, novel algorithm for predicting BM risk in NSCLC. Upon further validation, these parameters could inform screening and surveillance guidelines for BM in this patient population.