To determine the following: (1) the feasibility of combining the antiretroviral didanosine (ddl) with a 96-hour continuous intravenous (IV) infusion of cyclophosphamide (800 mg/m2), doxorubicin (50 mg/m2), and etoposide (240 mg/m2) (CDE) plus filgrastim in patients with non-Hodgkin's lymphoma (NHL) associated with human immunodeficiency virus (HIV) infection; (2) the effect of ddl on CDE-induced myelosuppression and CD4 lymphopenia; and (3) the effect of CDE on serum p24 antigen and quantitative HIV blood cultures. Twenty-five patients with HIV-related NHL received CDE every 28 or more days. Consecutive patients were assigned in an alternating fashion to group A (ddl given at a standard dose during cycles one, two, five, and six) or group B (ddl given during cycles three, four, five, and six). ddl use was associated with less leukopenia (mean nadir, 3.33 v 1.49 x 10(3)/microL; p = .03), neutropenia (2.38 v 1.07 x 10(3)/microL; p = .03), and thrombocytopenia (76 v 48 x 10(3)/microL; p = .059), and fewer RBC (1.6 v 3.1 per cycle; p < .01) and platelet transfusions (0.7 v 1.5 per cycle; p < .01), but had no significant effect on CD4 lymphopenia. Furthermore, lymphomatous bone marrow involvement and low CD4 count were associated with significantly greater myelosuppression. Although there was no substantial change in serum p24 antigen, the HIV blood culture became quantitatively more positive or converted from negative to positive in seven patients (64%). Complete response (CR) occurred in 58% of patients (95% confidence interval, 38% to 78%), median CR duration exceeded 18 months, tumor-related mortality was 20%, and median survival was 18.4 months. Our results suggest that the CDE and filgrastim regimen is tolerable and effective for patients with HIV-associated NHL, and that combination with ddl is feasible and may result in less myelosuppression.