Background:  Tuberculosis (TB) and HIV co-infection is common in South Africa. The mainstay of the diagnosis of TB is by sputum positivity on microscopy but many patients with suggestive symptoms are smear-negative. Clinicians are faced with dilemma:  Introducing ART may cause an IRIS but delaying ART, while awaiting investigation puts the patient at risk for further opportunistic infections. We assessed the validity of a clinically driven diagnostic paradigm for smear-negative TB

Methods: All 650 adult patients being screened for a number of clinical ART trials in Johannesburg were asked if they had a cough, weight loss, anorexia, night sweats, or chest pain. If present, a chest X-ray was done. If appropriate, sputum was collected and antibiotics were prescribed. If they did not respond to antibiotics, then a trial of TB treatment was undertaken. Patients with acid-fast bacilli in their sputum on microscopy were excluded.

Results:  In 38 patients, there was a suspicion of TB at screening on symptoms:  2/3 (n=26) were female with a mean Cd4⁺ count of 150.6 (SD = 84); 1 patient had a grade 3 (<7.4 g/dL); 2 patients had grade 3 transaminitis; all had radiological abnormalities, but not diagnostic of TB. Based on the clinical paradigm listed above, we started 25 patients on TB treatment; 13 patients were not started on TB treatment. There was no difference between the mean CD4⁺ count of patients given TB treatment and those not given TB treatment (p = 0.59, Student’s t-test, 1 df). Patients who were given TB treatment had more symptoms than those who were not given TB treatment. For the TB cases, we considered the outcome as either microbiological confirmation on sputum culture (n = 14) or a positive trial of treatment, defined as improvement of symptoms (n = 10). One of the cases started on TB treatment did not respond to it and was referred for further investigation of possible bronchiectasis. In the 13 cases for whom TB treatment was not started, 2 patient developed TB within 2 weeks are starting ART, and 1 had an initial clinical response to antibiotics. Another patient, presenting with chest pain, was not given antibiotic and went on to have an IRIS.

Conclusions:  Using the above methods the diagnostic paradigm had a sensitivity of 96% and a specificity of 85%.