The loss of the CD4 lymphocyte is the central pathophysiologic event in the progression of human immunodeficiency virus (HIV) infection. This retrospective study, based on review of data from deceased HIV patients followed in a single HIV clinic, was conducted to determine if the rate of CD4 lymphocyte decline was predictive of survival. Forty of 172 patients met defined criteria for inclusion in this study. For each patient, CD4-cell counts showed approximate exponential decline over time. A Cox regression analysis was used to assess the association of CD4 cell decline (half-life), race, age, gender, initial CD4-cell count, and treatment (anti-Pneumocystis carinii pneumonia prophylaxis and/or zidovudine vs. no therapy) on total survival (from initial CD4 cell count) and on remaining survival time after reaching a CD4 cell count of 100 (estimated). For all patients, the rate of CD4 cell decline was predictive of total survival (p = .009) but not for survival after reaching a count of 100 (p = .6). For patients who had never received therapy (6 patients), however, the CD4 half-life remained associated with survival time from 100 CD4 cells (p < .05) as opposed to the treated patients. Therapy was the single variable most predictive of both survival endpoints, resulting in an increase in median total survival of 27.2 mo (p < .00001) and of 15.4 mo from a CD4 cell count of 100 (p < .00004). Nonwhites had a slight survival disadvantage compared to whites (p = .08 overall; p = .02 from CD4 cell count of 100).(ABSTRACT TRUNCATED AT 250 WORDS)