Tuberculosis control in South African gold mines: mathematical modeling of a trial of community-wide isoniazid preventive therapy

Abstract

A recent major cluster randomized trial of screening, active disease treatment, and mass isoniazid preventive therapy for 9 months during 2006-2011 among South African gold miners showed reduced individual-level tuberculosis incidence but no detectable population-level impact. We fitted a dynamic mathematical model to trial data and explored 1) factors contributing to the lack of population-level impact, 2) the best-achievable impact if all implementation characteristics were increased to the highest level achieved during the trial ("optimized intervention"), and 3) how tuberculosis might be better controlled with additional interventions (improving diagnostics, reducing treatment delay, providing isoniazid preventive therapy continuously to human immunodeficiency virus-positive people, or scaling up antiretroviral treatment coverage) individually and in combination. We found the following: 1) The model suggests that a small proportion of latent infections among human immunodeficiency virus-positive people were cured, which could have been a key factor explaining the lack of detectable population-level impact. 2) The optimized implementation increased impact by only 10%. 3) Implementing additional interventions individually and in combination led to up to 30% and 75% reductions, respectively, in tuberculosis incidence after 10 years. Tuberculosis control requires a combination prevention approach, including health systems strengthening to minimize treatment delay, improving diagnostics, increased antiretroviral treatment coverage, and effective preventive treatment regimens.

Keywords: mass community-wide isoniazid preventive therapy; mathematical model; tuberculosis.

Figure 1.

Schematic of the time course of the Thibela TB randomized controlled trial among South African gold miners, 2006–2011. TB, tuberculosis.

Figure 2.

General structure of the dynamic transmission model for an intervention cluster in the Thibela TB randomized controlled trial among South African gold miners, 2006–2011. Because of high levels of transmission in the miners, all gold miners were assumed to have been infected at least once in their lifetime, which accounts for the absence of an uninfected compartment. The dashed lines reflect activities relating to case finding or isoniazid preventive therapy. The arrows out of the compartments, which have no destination, reflect out-migration or death. The small arrows into the compartments, which do not start from any destination, reflect in-migration. The shaded boxes reflect people who are taking IPT. IPT, isoniazid preventive therapy; s−c+, smear-negative, culture-positive; s+c+, smear-positive, culture-positive; TB, tuberculosis.

Figure 3.

Some of the key data used to parameterize the model describing the Thibela TB randomized controlled trial among South African gold miners, 2006–2011. A) Proportion of miners in the baseline and final prevalence surveys that reported ever taking ART; B) proportion of smear-positive and smear-negative miners who had not started TB treatment at different times since detection, according to mining company; C) monthly rates of in- and out-migration. Bars (in part A), 95% confidence interval. ART, antiretroviral therapy; TB, tuberculosis.

Figure 4.

Summary of the best-fitting impact on the weekly measured tuberculosis disease incidence rate (per 100,000 person-years) during the Thibela TB randomized controlled trial among South African gold miners, 2006–2011. The incidence rate is defined as the incidence that would be observed if it were measured weekly. A) Model predictions obtained by assuming that IPT fully cures all infections and protects against reinfection (IPT assumption 1: 100% cure, 100% protection); B) model is permitted to estimate that 6 months of IPT does not cure all infections and also does not give 100% protection against reinfection during IPT (IPT assumption 3: estimated percentage cured, estimated percentage protection). Note that, for all IPT models, the best-fitting values for the disease rates differed slightly (Web Figure 10), leading to differences in the predicted measured incidence before the introduction of IPT. For each plot, the predicted measured incidence increases in the intervention clusters after the start of the trial because of increased case detection, resulting from screening miners on recruitment into the trial. The cross shows the observed incidence in the intervention arm, aggregated for all intervention clusters; the empty square shows the “observed” incidence in the control arm, taken to equal the incidence in the intervention clusters, divided by 0.98 (the point estimate of the trial impact on incidence). Bars, 95% confidence intervals. IPT, isoniazid preventive therapy; TB, tuberculosis.

Figure 5.

Results of the Bayesian melding (resampling 20,000 parameter combinations from 2.28 million parameter combinations using the likelihood of the measured incidence as the weight). Box plot of estimates of the proportion of infections that were cured by 6 months of IPT (A), the protection provided by IPT against reinfection (B), and the impact of the intervention (C). The boxes reflect the interquartile range (IR), the “whiskers” extend to 1.5 times the IR, and the points outside this range are represented with filled circles. The resampling process resulted in 2,028 unique parameter combinations. HIV, human immunodeficiency virus; IPT, isoniazid preventive therapy.

Figure 6.

Impact of different interventions implemented individually (A–C) or in combination (D) predicted for the Thibela TB randomized controlled trial among South African gold miners, 2006–2011. Summary of the predicted impact of different interventions on the number of cases/100,000/year (the true TB incidence rate), after the treatment delay has been reduced. Each panel shows the impact of reduced treatment delay plus in A) preventive treatment, with 1) IPT provided community-wide in an initial round for 9 months of IPT, with coverage at the highest levels seen in Thibela, and 2) IPT provided community-wide in an initial round for 9 months, with coverage at the highest levels seen in Thibela, followed by continuous community-wide IPT with 50% coverage. This is achieved through keeping those who are still on IPT at the end of the initial round on IPT thereafter and providing IPT to 50% of new mining employees, and 3) a single round with a 3-month fully curing regimen provided community-wide (without 9 months of IPT), with coverage at the highest levels seen in Thibela. B) Scale-up of ART, with ART coverage increasing to reach 80% in 2009 in the HIV-positive groups specified in the figure legend; C) improved diagnosis using Xpert MTB/RIF, with 1) radiographs being used to screen at routine medical examinations and for newly employed miners and Xpert MTB/RIF being used to diagnose people with suspected TB, and 2) Xpert MTB/RIF being used to screen and diagnose at routine medical examinations for newly employed miners and on passive presentation; D) combined interventions. Combined impact of introducing reduced treatment delay, screening with Xpert MTB/RIF, ART for 80% of HIV-positive people, and IPT for those on ART. The shaded areas show the incremental impact of adding each intervention, so that the white area reflects the impact of having all interventions in place simultaneously. For the scenario involving Xpert MTB/RIF, Xpert MTB/RIF is used in routine medical examinations, for newly employed miners, and on passive presentation. For both the ART and ART/IPT scenarios, the coverage is increased to reach 80% by 2009. ART, antiretroviral therapy; HIV+, human immunodeficiency virus–positive; IPT, isoniazid preventive therapy; PT, preventive therapy; TB, tuberculosis.