Background Long-term treatment-related outcomes in patients with extensively drug-resistant (XDR) tuberculosis are unknown. We followed up a cohort of patients to address knowledge gaps. Methods Between March, 2008, and August, 2012, we prospectively followed up a cohort of 107 patients from three provinces in South Africa, who had been diagnosed with XDR tuberculosis between August 2002, and February, 2008. Available isolates from 56 patients were genotyped to establish strain type and used for extended susceptibility testing. Findings All patients were treated empirically as inpatients with a median of eight drugs (IQR six to ten). 44 patients (41%) had HIV. 36 (64%) of 56 isolates were resistant to at least eight drugs, and resistance to an increasing number of drugs was associated with the Beijing genotype (p=0.01). After 24 months of follow-up, 17 patients (16%) had a favourable outcome (ie, treatment cure or completion), 49 (46%) had died, seven (7%) had defaulted (interruption of treatment for at least 2 consecutive months), and 25 (23%) had failed treatment. At 60 months, 12 patients (11%) had a favourable outcome, 78 (73%) had died, four (4%) had defaulted, and 11 (10%) had failed treatment. 45 patients were discharged from hospital, of whom 26 (58%) had achieved sputum culture conversion and 19 (42%) had failed treatment. Median survival of patients who had failed treatment from time of discharge was 19.84 months (IQR 4.16-26.04). Clustering of cases and transmission within families containing a patient who had failed treatment and been discharged were shown with genotypic methods. Net sputum culture conversion occurred in 22 patients (21%) and median time to net culture conversion was 8.7 months (IQR 5.6-26.4). Independent predictors of probability of net culture conversion were no history of multidrug-resistant tuberculosis (p=0.0007) and use of clofazamine (p=0.0069). Independent overall predictors of survival were net culture conversion (p<0.0001) and treatment with clofazamine (p=0.021). Antiretroviral therapy was also a predictor of survival in patients with HIV (p=0.003). Interpretation In South Africa, long-term outcomes in patients with XDR tuberculosis are poor, irrespective of HIV status. Because appropriate long-stay or palliative care facilities are scarce, substantial numbers of patients with XDR tuberculosis who have failed treatment and have positive sputum cultures are being discharged from hospital and are likely to transmit disease into the wider community. Testing of new combined regimens is needed urgently and policy makers should implement interventions to minimise disease spread by patients who fail treatment. Funding European and Developing Countries Clinical Trials Partnership, South African National Research Foundation (SARChI), and the South African Medical Research Council.
