BACKGROUND: In a proportion of patients with HIV-associated tuberculosis who develop paradoxical immune reconstitution inflammatory syndrome (IRIS), the clinical course of IRIS is prolonged necessitating substantial health care utilization for diagnostic and therapeutic interventions. Prolonged TB-IRIS has not been prospectively studied to date. We aimed to determine the proportion of patients with prolonged TB-IRIS, as well as the clinical characteristics and risk factors for prolonged TB-IRIS. METHODS: We pooled data from two prospective observational studies and a randomized controlled trial conducted in Cape Town, South Africa, that enrolled patients with paradoxical TB-IRIS. We used the same diagnostic approach and clinical case definitions for TB-IRIS in the 3 studies. Prolonged TB-IRIS was defined as TB-IRIS symptoms lasting > 90 days. Risk factors for prolonged TB-IRIS were analysed using Wilcoxon rank sum test, Fisher's exact test, multivariate logistic regression and Cox proportional hazards models. RESULTS: Two-hundred and sixteen patients with TB-IRIS were included. The median duration of TB-IRIS symptoms was 71.0 days (IQR 41.0-113.2). In 73/181 patients (40.3 %) with adequate follow-up data, IRIS duration was > 90 days. Six patients (3.3 %), mainly with lymph node involvement, had IRIS duration > 1 year. In univariate logistic regression analysis the following were significantly associated with IRIS duration > 90 days: lymph node involvement at initial TB diagnosis, drug-resistant TB, lymph node TB-IRIS, and not being hospitalised at time of TB-IRIS diagnosis. In our multivariate logistic regression model lymph node TB-IRIS (aOR 2.27, 95 % CI 1.13-4.59) and not being hospitalised at time of TB-IRIS diagnosis (aOR for being hospitalised 0.5, 95 % CI 0.25-0.99) remained significantly associated with prolonged TB-IRIS, and drug-resistant TB was of borderline significance (aOR 3.26, 95 % CI 0.97-12.99). The association of not being hospitalised with longer duration of IRIS might be related to 1 of the 3 cohorts in which all patients were hospitalised at ART initiation with close inpatient follow-up. This could have resulted in diagnosis of milder cases and earlier IRIS treatment potentially resulting in shorter TB-IRIS duration in these hospitalised patients. CONCLUSIONS: Around 40 % of patients with TB-IRIS have symptoms for more than 90 days. Involvement of lymph nodes at time of TB-IRIS is an independent risk factor for prolonged TB-IRIS. Future studies should address whether more prompt anti-inflammatory treatment of lymph node TB-IRIS reduces the risk of prolonged TB-IRIS. TRIAL REGISTRATION: The randomized controlled trial was registered with Current Controlled Trials ISRCTN21322548 on 17 August 2005.