Varying probability of the development of acute nonlymphoblastic leukemia in refractory anemia patients with an excess of blasts.

The cumulative leukemia-free survival rate of refractory anemia with excess of blasts was fitted to three parametric failure time models, i.e., the usual exponential and Weibull, and exponential mixture models. Among the three, the best fit model was the exponential mixture model, which was 119 times more likely to assume that samples came from the exponential mixture distribution than to assume that those were from the usual Weibull distribution. This strongly suggests that refractory anemia with excess of blasts consists of subgroups with very high and very low probability to develop acute nonlymphoblastic leukemia. The estimated proportion of the very low probability group was about 30%. Analysis by the exponential mixture model with covariates revealed that the probability of a patient to develop acute nonlymphoblastic leukemia could be estimated by three covariates, i.e., bone marrow blast percentage; abnormal granules of granulocytes; and mononuclear large megakaryocytes. The estimated probabilities ranged from 17 to 99%, according to the model.