The objective of this study was to identify prognostic factors for survival in patients with pretreatment CD4 < or =50 cells/mm3 treated with nucleoside analogs, and to develop and validate a mortality risk model based on these factors. The design of the study consisted of retrospective analysis of AIDS Clinical Trials Group (ACTG) protocols 116a, 116b/117, 155, and 118. The setting was the multicenter AIDS Clinical Trials Group. The patients were HIV-infected with pretreatment CD4 < or =50 cells/mm3 and various degrees of prior zidovudine (ZDV) use. Double-blind, three-arm randomized control trials ACTG 116a and ACTG 116b/117 compared ZDV with didanosine (ddI). ACTG 155 compared ZDV with zalcitabine or combination therapy. Our validation study, ACTG 118, compared the effects of three different doses of ddI on survival. The main outcome measures were survival and mortality. The three studies combined enrolled 699 patients with entry CD4 T-lymphocyte counts of < or =50 cells/mm3. Forty percent of patients died during follow-up, with a median survival of 19.7 months. Multivariate analysis showed shorter survival at p < 0.0001 with lower CD4 count (relative hazard [RH] = 0.98) and lower hemoglobin level (RH = 0.81). Other factors included older age (RH = 1.03), male gender (RH = 1.70), Hispanic ethnicity (RH = 1.68), and symptomatic disease stage (RH = 2.06). Our predictive mortality risk model differentiated well patients with differing risks of mortality. When the risk model was applied to ACTG 118, the validation data set, the identified prognostic factors could distinguish patients with varying risks of death (p < 0.001, stratified log-rank test). These results demonstrate that CD4 T-lymphocytes counts < or =50 cells/mm3 should not be considered a precursor of imminent death; considerable variability in survival exists in severely immunocompromised patients. Our identification of prognostic indicators for survival can aid clinicians and patients in management of their disease and researchers in design of future clinical trials.