Comparison of Random Survival Forests for Competing Risks and Regression Models in Determining Mortality Risk Factors in Breast Cancer Patients in Mahdieh Center, Hamedan, Iran

Introduction: Breast cancer is one of the most common cancers among women worldwide. Patients with cancer may die due to disease progression or other types of events. These different event types are called competing risks. This study aimed to determine the factors affecting the survival of patients with breast cancer using three different approaches: cause-specific hazards regression, subdistribution hazards regression, and the random survival forest for competing risks. Methods: A historical cohort study was conducted on 527 breast cancer patients diagnosed in Mahdieh Medical Center, Hamadan, between 2004 and 2015. To determine risk factors for death due to cancer progression or other competing risks, cause-specific hazards and substandard hazards models and a random survival forest for competing risk were fitted. Data analysis was performed with R 3.4.3. Results: Findings showed that for death from the progression of breast cancer, age and number of involved lymph nodes were significant in both models (P < 0.05), and in the random survival forest model for death due to cancer progression, tumor size, number of involved lymph nodes, progesterone, estrogen, and family history were the important identified variables. Conclusion: In the presence of competing risks, when the underlying assumptions of cause-specific and subdistribution hazard regression models are not established, the use of random survival forest for competing events data to determine the risk factors affecting survival according to the coordination index and Brier score is more appropriate.