Previous phase I-II clinical trials have shown that recombinant human erythropoietin (rHuEpo) can ameliorate anemia in a portion of patients with multiple myeloma (MM) and non-Hodgkin's lymphoma (NHL). Therefore, we performed a randomized controlled multicenter study to define the optimal initial dosage and to identify predictors of response to rHuEpo. A total of 146 patients who had hemoglobin (Hb) levels < or = 11 g/dL and who had no need for transfusion at the time of enrollment entered this trial. Patients were randomized to receive 1,000 U (n = 31), 2,000 U (n = 29), 5,000 U (n = 31), or 10,000 U (n = 26) of rHuEpo daily subcutaneously for 8 weeks or to receive no therapy (n = 29). Of the patients, 84 suffered from MM and 62 from low- to intermediate-grade NHL, including chronic lymphocytic leukemia; 116 of 146 (79%) received chemotherapy during the study. The mean baseline Hb level was 9.4 +/- 1.0 g/dL. The median serum Epo level was 32 mU/mL, and endogenous Epo production was found to be defective in 77% of the patients, as judged by a value for the ratio of observed-to-predicted serum Epo levels (O/P ratio) of < or = 0.9. An intention-to-treat analysis was performed to evaluate treatment efficacy. The median average increase in Hb levels per week was 0.04 g/dL in the control group and -0.04 (P = .57), 0.22 (P = .05), 0.43 (P = .01), and 0.58 (P = .0001) g/dL in the 1,000 U, 2,000 U, 5,000 U, and 10,000 U groups, respectively (P values versus control). The probability of response (delta Hb > or = 2 g/dL) increased steadily and, after 8 weeks, reached 31% (2,000 U), 61% (5,000 U), and 62% (10,000 U), respectively. Regression analysis using Cox's proportional hazard model and classification and regression tree analysis showed that serum Epo levels and the O/P ratio were the most important factors predicting response in patients receiving 5,000 or 10,000 U. Approximately three quarters of patients presenting with Epo levels inappropriately low for the degree of anemia responded to rHuEpo, whereas only one quarter of those with adequate Epo levels did so. Classification and regression tree analysis also showed that doses of 2,000 U daily were effective in patients with an average platelet count greater than 150 x 10(9)/L. About 50% of these patients are expected to respond to rHuEpo. Thus, rHuEpo was safe and effective in ameliorating the anemia of MM and NHL patients who showed defective endogenous Epo production. From a practical point of view, we conclude that the decision to use rHuEpo in an individual anemic patient with MM or NHL should be based on serum Epo levels, whereas the choice of the initial dosage should be based on residual marrow function.