Construction and Validation of Prediction Models of Risk Factors for Early Death in Patients With Metastatic Melanoma

Abstract

To construct nomogram models to predict the risk factors for early death in patients with metastatic melanoma (MM). The study covered 2138 cases from the Surveillance, Epidemiology, and End Results Program (SEER) database and all these patients were diagnosed with MM between 2010 and 2015. Logistic regression was performed to identify independent risk factors affecting early death in MM patients. These risk factors were then used to construct nomograms of all-cause early death and cancer-specific early death. The efficacy of the model was assessed with receiver operating characteristic (ROC) curves, calibration curves, and decision curve analysis (DCA). In addition, external validation of the model was performed with clinicopathologic data of 105 patients diagnosed with MM at Sichuan Cancer Hospital between January 2015 and January 2020. According to the results of logistic regression, marital status, the primary site, N staging, surgery, chemotherapy, bone metastases, liver metastases, lung metastases, and brain metastases could be defined as independent predictive factors for early death. Based on these factors, 2 nomograms were plotted to predict the risks of all-cause early death and cancer-specific early death, respectively. For the models for all-cause and cancer-specific early death, the areas under the curve (AUCs) for the training group were 0.751 (95% confidence interval [CI]: 0.726-0.776) and 0.740 (95% CI: 0.714-0.765), respectively. The AUCs for the internal validation group were 0.759 (95% CI: 0.722-0.797) and 0.757 (95% CI: 0.718-0.780), respectively, while the AUCs for the external validation group were 0.750 (95% CI: 0.649-0.850) and 0.741 (95% CI: 0.644-0.838), respectively. The calibration curves showed high agreement between the predicted and the observed probabilities. DCA analysis indicated high clinical application value of the models. The nomogram models demonstrated good performance in predicting early death in MM patients and can be used to help clinical oncologists develop more individualized treatment strategies.