Development and validation of a nomogram for predicting the overall survival in non-small cell lung cancer patients with liver metastasis.

Abstract

Among all metastatic lesions in non-small cell lung cancer (NSCLC), liver metastasis (LM) is the most lethal site with a median survival of less than 5 months. Few studies exclusively report on prognostic factors for these unique patients. We aimed to construct and validate a practical model to predict the prognosis of NSCLC patients with LM. Cases of NSCLC with LM diagnosed between 2010 and 2015 were collected from the Surveillance, Epidemiology, and End Results (SEER) database, and were randomly split into training and validation cohort (7:3). The overall survival (OS) was measured from diagnosis until date of death or last follow-up. Cox regression analyses were performed to identify potential predictors of the model. A nomogram incorporating those independent factors was constructed and validated by the concordance index (C-index) and calibration plots. The decision curve analysis (DCA) and a risk stratification system were used to evaluate its clinical value. A total of 2,367 cases were selected for analysis and randomized to the training cohort (n=1,677) and the validation cohort (n=690). The patients were mainly male (59.3%), married (83.1%) and White (77.3%). Apart from LM, 54.2%, 26.7%, and 36.7% of patients also present with bone, brain, and lung metastases, respectively. The median follow-up was 4.0 months for all patients and 23 months for alive cases. The median OS was 5 months [interquartile range (IQR), 2-11 months]. Sex, age, race, grade, T stage, bone metastasis, brain metastasis, surgery, and chemotherapy were identified as the independent risk factors of the OS and used to develop the nomogram. The calibration curves exhibited excellent agreement between the predicted and actual survival in both the training and validation set, with a C-index of 0.700 [95% confidence interval (CI): 0.684-0.716] and 0.677 (95% CI: 0.653-0.701), respectively. The DCA and the risk classification system further supported that the prediction model was clinically effective. This is the first study to build a prediction model for NSCLC patients with LM. It aids in treatment decisions, focused care, and physician-patient communication. The global prospective data is needed to further improve this model.