We gave 56 patients with newly diagnosed acute myelogenous leukemia (AML) granulocyte-macrophage colony-stimulating factor (GM-CSF) 20 or 125 micrograms/m2 once daily subcutaneously before (for up to 8 days or until GM-CSF-related complications developed) and during, or only during (patients presenting with blast counts greater than 50,000 or other leukemia-related complications) ara-C (1.5 g/m2 daily x 4 by continuous infusion) and daunorubicin (45 mg/m2 daily x 3) chemotherapy. Because results seemed independent of GM-CSF schedule, we compared results in these 56 patients with results in 176 patients with newly diagnosed AML given the same dose and schedule of ara-C without GM-CSF (110 patients ara-C alone, 66 patients ara-C + amsacrine or mitoxantrone). Comparison involved fitting a logistic regression model predicting probability of complete remission (CR) and a Cox regression model to predict survival (most patients in all three studies were dead) with treatment included as a covariate in both analyses. After adjusting for other prognostically significant covariates [presence of an antecedent hematologic disorder, an Inv (16), t(8;21), or abnormalities of chromosomes 5 and/or 7, performance status, age, bilirubin], treatment with ara-C + daunorubicin + GM-CSF was predictive of both a lower CR rate and a lower survival probability. There were no treatment-covariate interactions, suggesting that the negative effect of this GM-CSF treatment regime was not an artifact of some imbalance in patient characteristics. The unadjusted Kaplan-Meier hazard rate of the ara-C + daunorubicin + GM-CSF group was not uniquely high during the initial 4 weeks after start of therapy, but was highest among the three treatment groups throughout weeks 5 to 16, suggesting that the negative effect of this treatment was not caused by acute toxicity. Patients who did not enter CR with this treatment tended to have persistent leukemia rather than prolonged marrow aplasia, suggesting that this treatment and, in particular, GM-CSF may increase resistance of myeloid leukemia cells to chemotherapy. To date, relapse rates are similar in all three groups (P = .43) (as are survival rates once patients are in CR) but much of the remission duration data is heavily censored, unlike the survival data. Our results suggest caution in the use of GM-CSF to sensitize myeloid leukemia cells to daunorubicin + ara-C chemotherapy.