Testing a predictive regression model for response in adult acute myeloid leukemia comparing two induction regimens...

Pretreatment characteristics of 295 adults with acute myeloid leukemia who were treated in a large cooperative group of the E.O.R.T.C. between 1976 and 1982 (AML-5) have been evaluated to assess their value as prognostic indicators. Logistic regression methods were applied to derive a model for prediction of achievement of CR. The model was tested prospectively in an independent group of 274 subsequent patients treated with a different but similar induction regimen (AML-6), who were matched in all eligibility criteria. The concordance between the observed percentage of response in the AML-6 trial population and the expected response given by the model built on the AML-5 study was very close for the good response subgroups (WBC less than 50,000/mm3 and age less than 60 yrs), while the discrepancies in the poor prognostic groups were considerable. Furthermore, our study shows that the prognostic value of factors may shift as treatment strategies change and that caution is necessary in applying conclusions from a preceding trial to a current patient population.