To study the effect of sequential or additive use of zalcitabine or didanosine on survival in 308 human immunodeficiency virus-infected patients with advanced disease treated with zidovudine, an observational study using time-dependent Cox proportional hazards models was done. Changing to sequential or additive therapy was based on deterioration of a patient's health status, a significant drop in CD4 cell count, or intolerance for zidovudine. The median CD4 cell count at baseline was 110 x 106/L; 42% of patients had AIDS. The median count before a change in therapy was 50 x 106/L. Additive or sequential treatment was associated with an increased risk for death (relative hazard, 1.59; 95% confidence interval ICI], 1.01- 2.49; and 1.58; 95% CI, 1.10-2.37, respectively). Adjustment of the models for prognostic factors failed to substantially affect this observation. Possibly the lack of benefit in this study is because patients switched therapy at advanced stages, whereas the switch may be more effective in early disease.