Objective: To examine whether the serial measurement of plasma cytokine levels can assist in the early recognition of AIDS/tuberculosis patients with poor response to anti-tuberculosis treatment. Design: Longitudinal, prospective cohort study. Setting: A university hospital, the largest centre for HIV/AIDS patients in Taiwan. Methods: Between January 1997 and September 1998, 25 consecutive patients with advanced HIV infection and suspected tuberculosis were enrolled in the study. Plasma samples were obtained on day 1 (baseline), 3, 7 and 14 of anti-tuberculosis treatment and the levels of tumour necrosis factor-alpha (TNF-alpha) were measured. Patients were classified as either responders or non-responders according to the results of assessment of symptoms and follow-up cultures during the sixth and eighth week of anti-tuberculosis treatment. Thirty consecutive HIV-negative tuberculosis patients were also enrolled in the study. Results: The data of a total of 16 AIDS patients (median CD4 cell count 16 x 10(6)/l; 12 responders and four non-responders) and 21 HIV-negative patients (16 responders and five non-responders), whose tuberculosis was culture-proven, were included for analysis. In responders, TNF-alpha levels declined remarkably within the first week of anti-tuberculosis treatment; however, the decline of TNF-alpha levels in non-responders was significantly less [the median ratio of TNF-alpha level on day 7 to that at baseline was 0.32 versus 0.85 (P < 0.001) in AIDS patients; 0.34 versus 0.80 (P = 0.001) in HIV-negative patients). The lack of a greater than or equal to 50% reduction in pretreatment TNF-alpha levels during the first week of treatment was strongly associated with a poor response to anti-tuberculosis treatment (P = 0.001 in AIDS patients; P < 0.001 in HIV-negative patients). Conclusion: Serial measurement of plasma TNF-alpha levels may help to assess the response to anti-tuberculosis treatment in AIDS patients, in spite of very low CD4 cell counts. Failure of TNF-alpha levels to decline by greater than or equal to 50 % of pre-treatment levels in the first week of treatment may be an early surrogate marker of a poor response. (C) Lippincott Williams & Wilkins.