Survival analysis in acute myeloid leukemia: A retrospective cohort study and simulation study of small events per independent variable / Law Kian Boon

Intensive chemotherapy for acute myeloid leukemia (AML) is used in both induction and consolidation treatments. The combination of fludarabine, high dose cytarabine and granulocyte colony-stimulating factor (FLAG) has been proven effective and safe as an induction treatment for refractory and poor risk AML, but not as a consolidation treatment especially in older AML patients. Hence, a retrospective cohort study was conducted to investigate the role of the FLAG regimen as consolidation treatment in older AML patients. Survival data of 41 eligible older patients were analyzed by using Kaplan-Meier method, log-rank test and Cox model. The results showed that patients consolidated with the FLAG regimen had a longer overall survival (OS) and disease free survival (DFS) when compared to non-FLAG regimens. The primary treatment variable (FLAG) was strongly associated with the survival outcomes with statistically different OS (log-rank, p = 0.0025) and DFS (log-rank, p = 0.0026). However, the regression analysis was performed at low events per independent variable (EPV) condition. The validity of Cox coefficient estimate requires at least 10 to 20 EPV, which can be difficult to achieve in a small study. Therefore, a simulation study was performed to assess the performance of Cox coefficient estimate for the primary treatment variable at low EPV spectrum. Our results showed that 3 and 4 EPV were associated with highest level of bias and disparity in accuracy, precision and statistical properties. At 5 to 6 EPV, the performance of Cox model started to gain stability. Above 6 EPV, increasing the number of events was less likely to improve the overall performance of the Cox model. The FLAG regimen should be used as part of consolidation for AML as the regimen improved both OS and DFS among older AML patients. The EPV rule has exceeded the number of outcome events required by a variable of strong association to the survival outcomes