Nomograms for predicting early death in elderly colorectal cancer patients based on the SEER database

AbstractBackground: Elderly colorectal cancer (ECRC) patients are generally in poor physical condition due to long-term tumor consumption, and are prone to severe complications or treatment failure. This study found out the risk factors that are related to early death in ECRC patients. A predictive model was then developed. This model can be used to calculate the risk of all-cause and cancer-specific early death in ECRC patients. Methods: Data were obtained from the Surveillance, Epidemiology and End Results (SEER) database. Patients with ECRC between 2010 and 2015 were included, and cases with complete data were screened using established criteria. The study then used univariate logistic regression analyses and multivariate logistic regression analyses (stepwise selection method) to identify the most relevant factors among the many variables associated with early death in ECRC patients. On this basis, nomogram prediction models were constructed. These models can be used to predict the risk of early death in ECRC patients. Finally, the model was evaluated in the experiment using receiver operating characteristic (ROC) curves, decision curve analysis (DCA), calibration curves, net reclassification improvement (NRI), and integrated discrimination improvement (IDI). Results: 16,512 ECRC patients were selected for study from SEER. Of these, 3443 patients died early ( death within 3 months of initial diagnosis). The early deaths of 2387 patients were cancer-specific early deaths. Race, Grade, AJCC stage, T stage, N stage, surgery, chemotherapy, radiotherapy, bone metastasis, lung metastasis, and primary site were independent risk factors for predicting all-cause early death in ECRC patients. Race, Grade, AJCC stage, T-stage, N-stage, surgery, chemotherapy, radiotherapy, bone, brain, liver and lung metastases and tumour size were independent risk factors to predict cancer-specific early death in ECRC patients. Then, the nomogram predictive models were built, using these identified variables. These models showed good concordance and accuracy in early mortality risk. Conclusion: Nomogram predictive models developed in this research is a practical tool. This model can help clinicians easily and quickly recognise high-risk ECRC patients. It can also be used as a reference for developing personalised treatment plans for patients.