Predicting remission outcome in acute non-lymphocytic leukemia: general principles and their application to residual...

The objective of this paper is to demonstrate the importance of a rigorous and comprehensive evaluation of predictive tests in acute leukemia, especially when these may influence therapeutic decisions. To illustrate these issues, we have chosen an emerging test, early residual bone marrow leukemia, as the example and we have presented actual data from Leukemia Intergroup Studies of adults with first relapse acute non-lymphocytic leukemia induced with high-dose cytosine arabinoside. We show how the choice of a suitable cut-off value to discriminate between patients predicted to have different remission rates is influenced by several variables including the reasons for remission failure, the estimated efficacy and toxicity of the intervention to be employed, and considerations of resource utilization. We suggest that before applying a new predictive test in practice, clinicians should carefully evaluate its performance, and the consequences of its use in their own clinical settings.