Determinants of treatment-related paradoxical reactions during anti-tuberculosis therapy: a case control study

Background: Inflammatory response following initial improvement with anti-tuberculosis (TB) treatment has been termed a paradoxical reaction (PR). HIV co-infection is a recognised risk, yet little is known about other predictors of PR, although some biochemical markers have appeared predictive. We report our findings in an ethnically diverse population of HIV-infected and uninfected adults. Methods: Prospective and retrospective clinical and laboratory data were collected on TB patients seen between January 1999 December 2008 at four UK centres selected to represent a wide ethnic and socio-economic mix of TB patients. Data on ethnicity and HIV status were obtained for all individuals. The associations between other potential risk factors and PR were assessed in a nested case-control study. All PR cases were matched two-to-one to controls by calendar time and centre. Results: Of 1817 TB patients, 82 (4.5 %, 95 % CI 3.6-5.5 %) were identified as having a PR event. The frequency of PR was 14.4 % (18/125; 95 % CI 8.2-20.6 %) and 3.8 % (64/1692; 2.9-4.7) for HIV-positive and HIV-negative individuals respectively. There were no differences observed in PR frequency according to ethnicity, although the site was more likely to be pulmonary in those of black and white ethnicity, and lymph node disease in those of Asian ethnicity. In multivariate analysis of the case-control cohort, HIV-positive patients had five times the odds of developing PR (aOR = 5.05; 95 % CI 1.28-19.85, p = 0.028), whilst other immunosuppression e.g. diabetes, significantly reduced the odds of PR (aOR = 0.01; 0.00-0.27, p = 0.002). Patients with positive TB culture had higher odds of developing PR (aOR = 6.87; 1.31-36.04, p = 0.045) compared to those with a negative culture or those in whom no material was sent for culture. Peripheral lymph node disease increased the odds of a PR over 60-fold 4(9.60-431.25, p < 0.001). Conclusion: HIV was strongly associated with PR. The increased potential for PR in people with culture positive TB suggests that host mycobacterial burden might be relevant. The increased risk with TB lymphadenitis may in part arise from the visibility of clinical signs at this site. Non-HIV immunosuppression may have a protective effect. This study highlights the difficulties in predicting PR using routinely available demographic details, clinical symptoms or biochemical markers.