BACKGROUND: The durability of isoniazid preventive therapy (IPT) in preventing tuberculosis (TB) is limited in high-prevalence settings. The underlying mechanism (reactivation of persistent latent TB or reinfection) is not known. We aimed to investigate the timing of TB incidence during and after IPT and associated risk factors in a very high TB and HIV-prevalence setting, and to compare the observed rate with a modelled estimate of TB incidence rate after IPT due to reinfection. METHODS: In a post-hoc analysis of a cluster-randomized trial of community-wide IPT among South African gold miners, all intervention arm participants that were dispensed IPT for at least one of the intended 9 months were included. An incident TB case was defined as any participant with a positive sputum smear or culture, or with a clinical TB diagnosis assigned by a senior study clinician. Crude TB incidence rates were calculated during and after IPT, overall and by follow-up time. HIV status was not available. Multivariable Cox regression was used to analyse risk factors by follow-up time after IPT. Estimates from a published mathematical model of trial data were used to calculate the average reinfection TB incidence in the first year after IPT. RESULTS: Among 18,520 participants (96% male, mean age 41 years, median follow-up 2.1 years), 708 developed TB. The TB incidence rate during the intended IPT period was 1.3/100 person-years (pyrs; 95% confidence interval (CI), 1.0-1.6) and afterwards 2.3/100 pyrs (95% CI, 1.9-2.7). TB incidence increased within 6 months followed by a stable rate over time. There was no evidence for changing risk factors for TB disease over time after miners stopped IPT. The average TB incidence rate attributable to reinfection in the first year was estimated at 1.3/100 pyrs, compared to an observed rate of 2.2/100 pyrs (95% CI, 1.8-2.7). CONCLUSIONS: The durability of protection by IPT was lost within 6-12 months in this setting with a high HIV prevalence and a high annual risk of M. tuberculosis infection. The observed rate was higher than the modelled rate, suggesting that reactivation of persistent latent infection played a role in the rapid return to baseline TB incidence.