Development and validation of nomograms to predict early death in non-small cell lung cancer patients with brain metastasis: a retrospective study in the SEER database.

Background: Throughout the course of non-small cell lung cancer (NSCLC), a lot of patients would develop brain metastasis (BM) associated with the poor prognosis and high rate of mortality. However, there have been few models to predict early death (ED) from NSCLC patients with BM. We aimed to develop nomograms to predict ED in NSCLC patients with BM. Methods: The NSCLC patients with BM between 2010 and 2015 were selected from the Surveillance, Epidemiology, and End Result (SEER) database. Our inclusion criteria were as follows: (I) patients were pathologically diagnosed as NSCLC; (II) patients who suffered from BM. The patients were randomly divided into 2 cohorts at the ratio of 7:3, for training and validation cohorts, respectively. The univariate and multivariate logistic regression methods were managed to identify risk factors for ED in NSCLC patients with BM. Two nomograms were established and validated by calibration curves, receiver operating characteristic (ROC) curves, and decision curve analysis (DCA). The follow-up data included survival months, causes of death, vital status. Death that occurred within 3 months of initial diagnosis is defined as ED and the endpoints were all-cause ED and cancer-specific ED. Results: A total of 4,920 NSCLC patients with BM were included and randomly divided into 2 cohorts (7:3), including the training (n=3,444) and validation (n=1,476) cohorts. The independent prognostic factors for all-cause ED and cancer-specific ED included age, sex, race, tumor size, histology, T stage, N stage, grade, surgical operation, radiotherapy, chemotherapy, bone metastasis, and liver metastasis. All these variables were used to establish the nomograms. In the nomograms of all-cause and cancer-specific ED, the areas under the ROC curves were 0.813 (95% CI: 0.799-0.837) and 0.808 (95% CI: 0.791-0.830) for the training dataset as well as 0.835 (95% CI: 0.805-0.862) and 0.824 (95% CI: 0.790-0.849) for the validation dataset, respectively. Besides, the calibration curves proved that the predicted ED was consistent with the actual value. DCA suggested a good clinical application. Conclusions: The nomograms can be used to predict the specific probability of a patient's death, which aids in treatment decisions and focused care, as well as in physician-patient communication.