Search tips
Search criteria 


Logo of nihpaAbout Author manuscriptsSubmit a manuscriptHHS Public Access; Author Manuscript; Accepted for publication in peer reviewed journal;
Sci Transl Med. Author manuscript; available in PMC 2014 April 10.
Published in final edited form as:
PMCID: PMC3714172

Community-wide isoniazid preventive therapy drives drug-resistant tuberculosis: a model-based analysis


Tuberculosis control has proven especially difficult in settings of high HIV prevalence as HIV co-infection erodes host immunity and leads to a high risk of progression to active tuberculosis. Studies have demonstrated that a 6-month (or longer) course of monotherapy with isoniazid (isoniazid preventative therapy (IPT)) can reduce this risk of progression. The World Health Organization endorses the use of IPT for symptom-free individuals with HIV/tuberculosis co-infection and has placed considerable effort in expanding IPT to entire communities (community-wide IPT). Though previous reviews have not found a statistically significant elevated risk of isoniazid-resistant tuberculosis among those previously treated with IPT, community-wide IPT programs may nonetheless generate substantial selective pressure and increase the burden of drug-resistant TB (DRTB). We develop mathematical models to identify the conditions under which community-wide IPT interventions can increase the burden of isoniazid-resistant M. tuberculosis, even when IPT does not select for resistance among those treated with IPT. In any model that explicitly or implicitly includes forms of inter-strain competition (such as partial immunity conferred by a previous M. tuberculosis infection), community-wide IPT interventions confer an indirect benefit to drug-resistant strains through selective suppression of drug-sensitive infections. We demonstrate that aggressive community-wide IPT can have an impressive impact on reductions of drug-sensitive disease, but at the cost of increasing the selective pressure for resistance. Improving the detection and treatment of DRTB can mitigate this risk. The lack of an observed elevation in the risk of DRTB among those receiving IPT in small-scale studies of limited duration does not imply that the selective pressure imposed by community-wide IPT will not be substantial. Community-wide IPT programs will likely play an important role in disease control in high incidence settings and their roll-out should be accompanied by interventions for the detection and treatment of drug-resistant disease.


WHO guidelines recommend that HIV infected individuals free of symptoms suggestive of tuberculosis receive treatment with isoniazid preventive therapy (IPT) for at least 6 months [1]. The rationale for this recommendation is that individuals with HIV are at high risk of progression to active tuberculosis disease (TB) if infected with M. tuberculosis [2] and multiple studies have documented that IPT reduces the risk of progression to disease among HIV-infected individuals with positive tuberculin skin tests [3]. Despite this strong WHO recommendation, uptake of IPT has been slow with only 178,000 HIV positive individuals receiving IPT in 2010 [4].

While there are many explanations for the delayed scale-up of IPT, one persistently documented concern is that widespread use of IPT may lead to an increasing burden of isoniazid-resistant TB in communities [1]. Given the importance of isoniazid to the efficacy of first-line regimens [5, 6] and the potential that isoniazid-resistance may serve as a stepping-stone to combination resistance (such as multi-drug resistance), understanding the potential for community-wide IPT to increase levels of drug resistant TB (DRTB) is critical.

During the development of the WHO IPT guidelines, a formal review process was undertaken to assess whether IPT can increase the risk of developing DRTB. This review, conducted according to Grading of Recommendations Assessment, Development and Evaluation (GRADE) criteria, included eight studies and a meta-analysis [7]. The summary measure of relative risk of isoniazid resistant TB among those that had previously received IPT compared to those that had not received IPT was 1.87 (95% Confidence Interval [0.65–5.38]). Based on what was judged to be “moderate quality of evidence” the committee made a “strong recommendation” that IPT does not increase the risk of developing DRTB [1].

In this paper, we address the question of whether and how community-wide IPT policies can increase levels of DRTB from a different perspective than the one taken in the GRADE evaluation. The studies included in the GRADE evaluation assess only the direct effect of IPT on the risk of isoniazid-resistant disease among those receiving this intervention. However, because TB is a transmissible disease, we should also consider potential indirect effects of such large-scale interventions. It is possible for IPT to select for resistance at the population level even if it does not cause acquired resistance to isoniazid among those treated (i.e. no direct effect). Here we use mathematical models to demonstrate how this might occur and to identify the assumptions needed to confidently conclude that community-wide IPT interventions would not increase the levels of isoniazid resistant TB in populations with high HIV prevalence.

There are three mechanisms by which community-wide IPT could theoretically cause increasing levels of DRTB in a population:

  1. Failure to detect active TB in an individual before starting IPT can lead to acquired resistance (mutations occur which cause resistance and these are selected due to inadvertent monotherapy during active disease) [8, 9, 10, 11].
  2. Use of IPT during latency may select for sporadically occurring isoniazid-resistant mutants and hence increase the risk of DRTB among those treated [12, 13].
  3. Community-wide use of IPT may confer a competitive advantage to circulating strains of isoniazid-resistant M. tuberculosis strains by exerting selective pressure against isoniazid-susceptible strains.

While the first and second mechanisms are very important to consider, in this paper we focus on elucidating the conditions under which the third mechanism might occur. Whether selective suppression of isoniazid-susceptible strains through community-wide IPT would facilitate higher levels of DRTB depends on how strains of M. tuberculosis compete within and between human hosts. The degree to which strains confer partial immunity to superinfection and/or progression [14, 15, 16, 17] and the specific mechanisms by which strains compete are not yet well understood. Accordingly, we have constructed mathematical models that allow us to evaluate how varying the mechanisms and strength of inter-strain competition affects the projected impact of community-wide IPT on the short and longer-term levels of TB and isoniazid-resistant TB in populations where HIV is highly prevalent. By focusing our analysis on this third mechanism, we can identify whether, under optimal conditions where we target IPT appropriately (only giving IPT to HIV infected individuals without active TB, thereby averting mechanism 1) and do not generate resistance through acquisition (averting mechanism 2), community-wide IPT interventions could still present a strong enough selective pressure to increase the expected burden of INH-resistance.


To determine how inter-strain competition affects the impact of community-wide IPT we use a deterministic model of HIV and TB transmission with two phenotypes of TB: drug sensitive and drug resistant (Figure 1). TB infection is modelled with latent and active classes for each strain and mixed latent classes, as in previous models [18]. HIV infection is represented with an SI model, only a fraction of the population are at risk of HIV infection (as in [19]). We parameterize the model to generate epidemics that qualitatively reflect the trajectories and magnitude of HIV and TB trends in Lesotho, a country with a high burden of both diseases, and thus where large community-wide IPT interventions might be deployed (see Table S2 for the parameters). We simulate IPT when the epidemics are stable and compare the drug resistant and drug sensitive incidence and prevalence 20 and 100 years after IPT, in models with and without the IPT intervention.

Figure 1
Flow chart of the TB model

Community-wide IPT interventions have counteracting effects in our simulations. IPT acts to suppress DSTB but may facilitate the emergence of DRTB solely through the selective suppression of drug-sensitive strains. Figure 2 shows the prevalence of latent infection and incidence of active disease with DSTB and DRTB in the presence and absence of the community-wide IPT interventions. For these simulations, parameter values are set to their baseline values (parameters shown in Table S2). Note that while the expected short-term increase in DRTB over the 10 years after IPT is introduced is minimal, the long-term effect is more pronounced. As we increase the selective pressure of the community IPT intervention (i.e. when the rate of getting placed on IPT (tHIV) is high and the duration of a course of IPT (1/w) is long), we observe much larger short- and longer-term decreases in the incidence of DSTB, but at the expense of increasing the incidence of DRTB (see Figure 3).

Figure 2
Results from the model
Figure 3
The trade-off between drug sensitive and drug resistant TB

The degree to which IPT can increase the levels of drug-resistant infection and disease depends on both the mechanism and strength of inter-strain competition. When there is a low degree of protection conferred by previous infection (low immunity, high values of k and kHIV) we observe more re-infection, and hence more ‘theft’ of latently infected individuals from one strain to another. This occurs through both fast and slow progression. The effect through slow progression is mediated by the fraction of individuals (d) in whom the initial infecting strain remains dominant after re-infection. Weaker immunity also results in more mixed infections. When immunity is low, the two strains are not only competing for susceptible hosts, but are also competing for latently infected hosts. The rapid effects of competition through the fast progression mechanism suggests that changing the magnitude of the immunity term (k) will have a strong early effect and a less dramatic late effect (compare Figure 4A&C).

Figure 4
Varying the competition between strains

As lower levels of immunity lead to higher expected numbers of individuals in the mixed infection states, high values of k may allow for community-wide IPT interventions to have a stronger resistance-promoting effect through the selective action of IPT against drug-sensitive bacilli within individuals with mixed latent infections. However, this effect of IPT is mitigated when superinfection events are less likely to lead to a change in the dominant strain (i.e. when d is high). The increased tendency of an infection to remain the dominant strain after re-infection with another strain (high values of d and dHIV) shifts the focus of inter-strain competition towards fully susceptible hosts. In contrast to variations in partial immunity (k), the effect of changes in d are more modest in the short term and greater in the long term because it takes time for competition via susceptible hosts to play out (Figure 4B&D).

An effective increase in the number of available hosts arises in two ways in all TB models that incorporate interactions between two or more TB strains. The first is that individuals exposed to one strain of TB retain some partial immunity protecting them from re-infection, and hence ultimately from progression and active TB disease with the other strain. If the prevalence of one strain is reduced, then the number of hosts with protective immunity is reduced. Depending on whether immunity is specific or not, this can effectively raise the number of hosts susceptible to (re-)infection with the other strain. The second is that when individuals are re-infected, models assume that they enter either a latent or active state of infection with the re-infecting strain. This means that re-infection replaces some or all of the risk of progression with the first strain with risk of progression with the newly-infecting strain.

We conducted uncertainty and sensitivity analyses of the model using parameter sets selected by Latin hypercube sampling (LHS). While there was substantial variation in outcomes, all LHS parameters sets resulted in predicted levels of DRTB that were higher in scenarios with IPT. The sensitivity analysis revealed that the partial immunity term, kHIV, is most correlated with DRTB, implying that reducing our degree of uncertainty about this parameter is most critical for establishing the impact of IPT. Further details are in the Supplementary Material.

In the model results presented up to this point, we assumed optimal program conditions in which individuals with active TB were not inadvertently prescribed monotherapy and also that treatment of active TB does not generate acquired resistance. Relaxing the strong assumption that treatment of active TB does not generate new resistance through acquisition reveals that over ranges of reasonable risk of acquired resistance, the long term effect of selective pressure of community-wide IPT on the emergence of DRTB remains (Figure 5). Over shorter time periods, when the risk of acquired resistance during treatment is quite high (≈10%, see Figure 5 where blue solid line dips below blue dotted line), the beneficial effect of preventing cases of latent infection from progressing to active TB can outweigh the selective pressure of IPT. This implies that in programs where treatment of active DSTB tends to generate substantial DRTB, prevention of progression from latency may initially mitigate the emergence of DRTB, but this effect may be reversed over longer time horizons as the selective pressure of community-wide IPT takes effect.

Figure 5
The effect of including acquisition of active DRTB from treatment of active DSTB

To confirm that our results are governed by inter-strain competition, we developed a model in which the strains do not compete with each other (see Fig. S6). This requires not only that there not be immunity (an unrealistic assumption for TB [17]), but also that re-infection not alter the risk of progression or disease with an existing infection. In this case, community-wide IPT reduces drug-sensitive disease without increasing resistance levels. Figure 6 shows the prevalence and incidence of latent and active DSTB and DRTB in a model without competition and illustrates the striking difference between these results and those of the competition model (Figure 2). However, if immunity is re-introduced into the model (or if we assume that individuals with mixed infection do not experience a doubled rate of slow progression to disease) the resistance-promoting effect of community-wide IPT returns.

Figure 6
Results from the completely non-competing model


There is a compelling argument that IPT should not cause acquired resistance among eligible individuals receiving appropriate treatment. Latent infection has been considered a state characterized by relatively low bacterial levels and relative physiologic inactivity of the bacteria that are present. Based on previous estimates of mutation rates conferring resistance to isoniazid (on the order of 10−8 mutation per bacterium per generation [20]), the risk of selecting for resistant mutants during latency should be low [21].

However, more recently, TB researchers have proposed a model of latency that suggests that latent infection is a continuous spectrum of disease states characterized by heterogeneous bacterial populations in different physiological states [22]. Others investigations have found that bacterial division [12] and mutation [13] may occur quite commonly during states that would be clinically characterized as latency. These studies collectively suggest that use of monotherapy during latency could result in acquired resistance. However, the data summarized during the GRADE review for the 2011 WHO IPT guidelines [1] did not find a statistically significant increased risk of isoniazid-resistant disease among patients previously treated with IPT; this was interpreted as an absence of evidence that IPT causes acquired resistance to isoniazid. While it is encouraging that this systematic review did not detect a statistically significant effect, it should be noted that the point estimate is consistent with a greater than 80% increased risk and that only an extremely large study (or group of studies) of IPT would be capable of detecting a statistically significant effect.

The implementation of community-wide IPT interventions requires that health care workers can rule out active disease among HIV-seropositive patients. Screening tools with good operating characteristics have been published [23], but in some settings these screening tools alone may not be adequate to rule out subclinical TB disease [8, 24, 9, 10, 11]. There is therefore a risk that as IPT interventions are rolled out in new settings, isoniazid resistance might arise as a result of inadvertent treatment of active disease with a single agent.

While it is possible that IPT can cause resistance among those with latent infection, and resistance is likely to occur among those with active TB that improperly receive IPT, in this paper, we developed models that assume that IPT does not lead to resistance through either of these direct mechanisms. Instead, we explored whether IPT could affect resistance levels indirectly through inter-strain competition. Accordingly, the assumptions of our model are consistent with data such as those reported in [25], in which the risk of isoniazid resistance among those receiving IPT was no greater than the baseline levels in the community, within the 2.5 years of the study. Our results indicate that an increased risk of DRTB need not occur among individuals receiving IPT in the short term in order for IPT to present a risk of driving resistance through population-level inter-strain competition. The fact that isoniazid-resistant disease is already prevalent in most settings where IPT would be implemented increases the relevance of this investigation [26]. We also note that even though in our model we assume that IPT does not cause acquired resistance, it allows individuals with either latent susceptible or latent mixed susceptible and resistant infections to clear only their sensitive isolates; in observational studies this effect of IPT would not be distinguishable from true acquired resistance and thus may explain some of the (non-statistically significant) increased risk in the GRADE evaluation.

It is possible to develop multi-strain TB models that assume that strains do not compete for hosts through competition either for latently infected or susceptible hosts (see Fig. S6). Without competition, infection with one strain does not affect the risk of infection (ie no partial immunity) or progression with the other. The latter requires mixed infection states where the total progression rate is the sum of the progression rates of the two strains. In such a system, IPT given to mixed latently infected individuals would select for resistant bacilli when given to individuals with mixed infections. This effect would be stronger than in the model we have used because without partial immunity there would be more of these individuals. Paradoxically, when strains are not competing, the relative risk of progression to drug-resistant compared to drug-sensitive disease would be increased for individuals receiving IPT (because some of them would harbour mixed infections) but selectively treating drug-sensitive infections would not drive resistance (as shown in Figure 6). However, if strains are in competition for hosts and the number of mixed infections is low, an increased individual-level risk of DRTB among those given IPT would not be detected, because the increases in DRTB due to the combination of competition and a reduction of DSTB would not occur among the treated individuals, but rather would occur in the population at large. Therefore, individual-level studies that do not show increased risk of DRTB among those given IPT should not be taken to imply that community-wide IPT will not have long-term population level consequences. We emphasize that the assumptions needed for a model without inter-strain competition are not consistent with current understanding of TB. For example, there is ample evidence for partial immunity conferred by previous infection [17]. However, developing a model that is devoid of competition between resistant and sensitive strains is the only way in which we could fully remove the population-level effect of IPT on the spread of resistance.

While it is clear that human hosts can harbor multiple M. tuberculosis strains [27, 28, 29, 30, 31], little is known about how these strains interact [32, 33, 34]. The consequences of such forms of within-host interaction can have important impacts on TB epidemiology and drug-resistance in particular [35]. As shown in our uncertainty analysis, the protection conferred by previous infection and the interaction of strains in the mixed latent classes can have a large impact on the balance of DSTB and DRTB strains in the population. We believe that currently collected epidemiological data can currently provide only limited information about the mechanisms and magnitude of interstrain competition. In most populations, the fraction of M. tuberculosis isolates that is typed is low and the time course over which the effects of strain competition are expected to manifest is long. While these practical issues challenge the use of routinely collected data to inform our understanding of interstrain competition, the introduction of whole genome sequencing as a routine typing tool will permit the development and application of genomic approaches for identifying the effects of this type of competition from M. tuberculosis sequences collected from communities. In the meantime, we believe that animal models may serve as a very useful tool for investigating the effects of previous M. tuberculosis exposure on subsequent infection or risk of progression. These types of in vivo studies, in which model organisms can be exposed to particular M. tuberculosis simultaneously or in sequence [33, 34, 36], can provide key information about competition effects, though the generalizability of these findings to humans will remain in question. Further in vivo studies of inter-strain competition are needed to improve our understanding of these interactions and their dependence on host immune type and strain type in order to predict the effects of interventions such as community-wide IPT.

Wargo and colleagues reported that in a mixed strain (drug-sensitive and drug-resistant) rodent malaria model selective removal of drug-sensitive parasites through chemotherapy resulted in “competitive release” of the previously constrained drug-resistant parasites [37], giving within-host densities of drug-resistant parasites above the level observed in the absence of the drug-sensitive competitor. Consistent with this, Harrington and colleagues found that preventative intermittent treatment of malaria in pregnant women with sulfadoxine-pyrimethamine was associated with a higher fraction of cases with a resistance allele and increased levels of parasitemia [38]. Whether similar mechanisms apply to M. tuberculosis is not known.

Recently, Basu et al [39] used a model to predict the effects of IPT. Their model assumes strong competition between DRTB and DSTB that occurs both because of partial immunity and because they assume re-infection results in strain replacement. Their approach to modeling the selective effect of IPT was quite conservative and consequently only showed small benefits of IPT and a relatively small increase in DRTB over 50 years. The role that IPT plays in the emergence of resistance results from the combined effect of the individual-level and population-level effects we have outlined. We anticipate that under scenarios in which IPT has a larger benefit in their model, it will also cause increased resistance due to the population-level effects we report here.

Our model makes simplifying assumptions that limit our ability to predict the impact of community-wide IPT on the future trajectory of the TB epidemic. We considered a simple model of HIV infection with two states (infected/not infected) and our IPT intervention and treatment strategies were modelled in a simple way. Furthermore, in focusing the effects of competition in the absence of direct effects of IPT interventions (mechanisms 1 and 2 of the introduction) we have not modelled the acquisition of drug resistance among those with latent drug-sensitive infections. We have not aimed to use the model for quantitive prediction of the effects of IPT for these reasons. Rather we have focused our investigation on the mechanisms by which community-wide IPT may accelerate the spread of DRTB even within perfectly operating programs.

Our analyses demonstrate that community-wide IPT can be an effective intervention against DSTB in settings of high HIV prevalence, but this intervention can also increase the absolute burden of DRTB. We expect that the most aggressive IPT programs will produce the greatest selective pressure for drug-resistant strains that are already circulating in these communities. We therefore emphasize the crucial role of conducting surveillance for DRTB, especially as community-wide IPT programs go to scale. Our simulations indicate that the rate at which we expect IPT to drive up drug resistance is fairly slow, so there should be ample opportunity to introduce universal drug susceptibility testing to ensure that patients receive effective therapy for drug-resistant disease. Ensuring access to rapid testing and treatment for drug-resistant disease should mitigate unintended consequences of community-wide IPT.

As new TB drugs are introduced to the market [40], the opportunity to use these drugs as monotherapy among contacts of individuals with highly drug-resistant disease has been suggested [41]. While this approach may well help protect those at risk of developing disease with very difficult (or nearly impossible) to treat disease, it also may risk driving up the burden of disease resistant to these newest agents. Ideally, we would have enough new agents to reserve some for use as prevention and others for treatment of active disease.

Materials and Methods

We use a deterministic model of TB and HIV transmission with two phenotypes of TB: drug-sensitive (DSTB) and drug-resistant (DRTB). The TB model structure is similar to previous models (see flowchart in Figure 1): after infection with M. tuberculosis, individuals remain latently infected (LSi or LRi) or rapidly progress to active TB disease (ISi or IRi). Latently infected individuals may slowly progress to active TB disease and, whilst latent, can be re-infected by strains of TB circulating in the community. We assume partial immunity for those latently infected [14, 15, 16, 17]; for simplicity we assume that this immunity acts to reduce the rate at which previously-infected individuals are successfully re-infected upon re-exposure. Individuals who are re-infected can either progress rapidly to active disease with the superinfecting strain, or can retain both infections and suffer a slow rate of progression to active disease characterized by whichever strain is dominant (LSRi and LRSi) (as in Cohen et al. [18]). We refer to states of latency where an individual harbors infection with both types of strains as “mixed strain infections” [27, 28, 29, 30]. IPT given to individuals with mixed strain infections will efficiently select for isoniazid resistant bacilli resulting in a resistant latent infection.

We repeat this core TB model structure for three types of individuals indexed by HIV status (as in [19]): 1) individuals possessing behavioral or social characteristics that mean that they are not at risk of HIV infection; 2) those at risk of HIV, but as yet uninfected; and 3) those with HIV infection. We indicate HIV status within TB infection model states with the use of subscript i. In the model, we assume that HIV infection increases the risk of rapid progression to TB after infection, the reactivation of latent infections, as well as mortality. A complete model description, equations, parameters and implementation details are in the Supplementary Material.

We parameterize the model to generate epidemics that qualitatively reflect the trajectories and magnitude of HIV and TB trends in Lesotho, a country with a high burden of both diseases, and thus where large community-wide IPT interventions might be deployed (see the Supplementary Material). We introduce IPT in the model when the prevalence of active TB is approximately 475/100,000 and the prevalence of HIV is approximately 28%.

During IPT individuals may become infected with DRTB but not DSTB. Figure 1B shows details of how the IPT intervention is modeled. We explore the effect of increasing the rate that individuals receive IPT and the duration of the course of IPT on both the DSTB and the DRTB epidemic.

Competition between DSTB and DRTB arises in two ways in this model. The first relates to immunity: individuals with latent TB infection retain partial immunity that reduces their risk of re-infection with either strain of TB. When interventions act preferentially against one strain, its force of infection is expected to decrease. This results in an increased availability of susceptible hosts who may be infected by the competing strain. We assume that partial immunity is lower for individuals infected with HIV (kHIV) and individuals who have had a TB infection and been treated with treated with IPT retain some partial immunity (kIPT).

The second form of competition relates to how multiple strains interact within a host. In the model, a proportion of individuals with a latent infection who are re-infected will progress rapidly to active infection with the newly infecting strain. Of those that do not, a fraction d move to a mixed latent class with their original strain remaining dominant while the remainder (1−d) move to a mixed latent class with the new strain becoming dominant. Each is at risk of progression with the dominant strain only (but remain at risk from subsequent reinfection). We assume that individuals with mixed latent infection have the same progression rates as latently infected hosts harboring only a single strain. Accordingly, re-infection with a new strain effectively substitutes a portion of risk of progression with the original strain with a similar risk of progression with the new strain. Higher values of d reflect more competition in the model at the level of susceptible hosts, since with higher d, re-infected individuals retain risk of slow progression with their original infection, reducing each strains’ ability to compete for latent hosts.

Overall, competition between strains in the model is mediated by the parameters d, dHIV, k and kHIV. We explore the effects of changes in these parameters, keeping other variables fixed at baseline values. We assume dHIV = d and kHIV=1-1-k2.

We show time trends and report the effects of IPT at 20 and 100 years after the initiation of this intervention. We have assumed that the operational characteristics of tuberculosis treatment programs do not change over the course of these simulations. While this is unrealistic (e.g. new diagnostics, drugs, vaccines and treatment approaches will likely become available), this approach allows us to to provide qualitative insight into the potential early and late effects of community-wide IPT on drug-susceptible and DRTB. We compare trends in the incidence of active DRTB and the prevalence of latent drug-resistant infections in the presence and absence of community-wide IPT. Since many of the values of parameters are not known or easily measured, we conduct uncertainty and sensitivity analyses using a latin hypercube sampling (LHS) approach. Further details of the LHS are provided in the Supplementary Material.

In addition to our main modeling results, we have also developed an alternative model structure in which the two strains of TB do not compete with one another (we refer to this as the non-competing model, see Fig. S6)). In a model without competition, an increase in either strain should not have any explicit or implicit effect on the duration of latency, risk of progression, or duration of infectiousness of the other strain. In the Discussion, we argue why we think this non-competing model is unrealistic; nevertheless, the use of this model allows us to clearly demonstrate the importance of inter-strain competition to the expected effects of community-wide IPT interventions.

Supplementary Material

List of supplementary materials:

Supplementary Material includes:

  • Supplementary text
  • Supplementary Figure 1: The model fit to prevalence data from Lesotho.
  • Supplementary Figure 2: The results of the uncertainty analysis.
  • Supplementary Figure 3: Histograms of the differences between scenarios with and without IPT in the LHS uncertainty analysis.
  • Supplementary Figure 4: Sensitivity analysis of the models.
  • Supplementary Figure 5: Results from the model with improved targetting of IPT
  • Supplementary Figure 6: Flow chart of the completely non-competing model.
  • Supplementary Table 1: Description of the states in the model.
  • Supplementary Table 2: Table of parameters for the model.


We thank the anonymous reviewers for their helpful comments.

Funding: HLM was supported by the Bristol Centre for Complexity Sciences and EPSRC grant EP/5011214. TC received support from NIH grant DP2OD006663. The content is solely the responsibility of the authors and does not necessarily represent the official views of the Office of the Director or the National Institutes of Health. CC received support from EPSRC grant EP/I03626/1.


Author contributions

HLM, TC and CC designed the model. HLM wrote and ran the model. HLM, TC and CC analysed the results and wrote the paper.

Competing interests

The authors declare no competing interests.

References and Notes

1. World Health Organization. Guidelines for intensified tuberculosis case-finding and iso-niazid preventive therapy for people living with HIV in resource- constrained settings. 2011 Online.
2. Corbett E, et al. The growing burden of tuberculosis: global trends and interactions with the HIV epidemic. Arch Intern Med. 2003;163:1009. [PubMed]
3. Akolo C, Adetifa I, Shepperd S, Volmink J. Treatment of latent tuberculosis infection in HIV infected persons. Cochrane Database Syst Rev. 2010;1 [PubMed]
4. World Health Organization. [Accessed April 2012];Global Tuberculosis Control Report 2011. 2011 Online.
5. Lew W, Pai M, Oxlade O, Martin D, Menzies D. Initial drug resistance and tuberculosis treatment outcomes: systematic review and meta-analysis. Ann Intern Med. 2008;149:123. [PubMed]
6. Menzies D, et al. Standardized treatment of active tuberculosis in patients with previous treatment and/or with mono-resistance to isoniazid: a systematic review and meta-analysis. PLoS Med. 2009;6:e1000150. [PMC free article] [PubMed]
7. Balcells M, Thomas S, Godfrey-Faussett P, Grant A. Isoniazid preventive therapy and risk for resistant tuberculosis. Emerg Infect Dis. 2006;12:744. [PMC free article] [PubMed]
8. Day J, et al. Screening for tuberculosis prior to isoniazid preventive therapy among HIV-infected gold miners in South Africa. Int J Tuberc Lung Dis. 2006;10:523. [PubMed]
9. Bassett IV, et al. Intensive tuberculosis screening for HIV-infected patients starting antiretroviral therapy in Durban, South Africa. Clin Infect Dis. 2010;51:823. [PMC free article] [PubMed]
10. Churchyard G, et al. Symptom and chest radiographic screening for infectious tuberculosis prior to starting isoniazid preventive therapy: yield and proportion missed at screening. AIDS. 2010;24:S19. [PubMed]
11. Kufa T, et al. Undiagnosed tuberculosis among hiv clinic attendees: Association with antiretroviral therapy and implications for intensified case finding, isoniazid preventive therapy, and infection control. J Acquir Immune Defic Syndr. 2012;60:e22. [PubMed]
12. Gill WP, et al. A replication clock for Mycobacterium tuberculosis. Nat Med. 2009;15:211. [PMC free article] [PubMed]
13. Ford CB, et al. Use of whole genome sequencing to estimate the mutation rate of Mycobacterium tuberculosis during latent infection. Nat Genet. 2011;43:482. [PMC free article] [PubMed]
14. Sutherland I, Svandova E, Radhakrishna S. The development of clinical tuberculosis following infection with tubercle bacilli:: 1. A theoretical model for the development of clinical tuberculosis following infection, linking from data on the risk of tuberculous infection and the incidence of clinical tuberculosis in the Netherlands. Tubercle. 1982;63:255. [PubMed]
15. Vynnycky E, Fine P. The natural history of tuberculosis: the implications of age-dependent risks of disease and the role of reinfection. Epidemiol Infect. 1997;119:183. [PubMed]
16. Brooks-Pollock E, Becerra MC, Goldstein E, Cohen T, Murray MB. Epidemiologic inference from the distribution of tuberculosis cases in households in Lima, Peru. J Infect Dis. 2011;203:1582. [PMC free article] [PubMed]
17. Andrews JR, et al. Risk of Progression to Active Tuberculosis Following Reinfection With Mycobacterium tuberculosis. Clin Infect Dis. 2012;54:784. [PMC free article] [PubMed]
18. Cohen T, Lipsitch M, Walensky R, Murray M. Beneficial and perverse effects of isoni-azid preventive therapy for latent tuberculosis infection in HIV–tuberculosis coinfected populations. Proc Natl Acad Sci USA. 2006;103:7042. [PubMed]
19. Ghys PD, et al. The UNAIDS Estimation and Projection Package: a software package to estimate and project national HIV epidemics. Sex Transm Infect. 2004;80(Suppl 1):i5. [PMC free article] [PubMed]
20. David H. Probability distribution of drug-resistant mutants in unselected populations of mycobacterium tuberculosis. Appl Microbiol. 1970;20:810. [PMC free article] [PubMed]
21. Tufariello J, Chan J, Flynn J. Latent tuberculosis: mechanisms of host and bacillus that contribute to persistent infection. Lancet Infect Dis. 2003;3:578. [PubMed]
22. Barry C, et al. The spectrum of latent tuberculosis: rethinking the biology and intervention strategies. Nat Rev Microbiol. 2009;7:845. [PubMed]
23. Getahun H, et al. Development of a standardized screening rule for tuberculosis in people living with HIV in resource-constrained settings: individual participant data meta-analysis of observational studies. PLoS Med. 2011;8:e1000391. [PMC free article] [PubMed]
24. Aït-Khaled N, et al. Isoniazid preventive therapy for people living with HIV: public health challenges and implementation issues. Int J Tuberc Lung Dis. 2009;13:927. [PubMed]
25. van Halsema CL, et al. Tuberculosis outcomes and drug susceptibility in individuals exposed to isoniazid preventive therapy in a high HIV prevalence setting. AIDS (London, England) 2010;24:1051. [PubMed]
26. Jenkins H, Zignol M, Cohen T. Quantifying the burden and trends of isoniazid resistant tuberculosis, 1994–2009. PLoS One. 2011;6:e22927. [PMC free article] [PubMed]
27. Richardson M, et al. Multiple Mycobacterium tuberculosis strains in early cultures from patients in a high-incidence community setting. J Clin Microbiol. 2002;40:2750. [PMC free article] [PubMed]
28. Warren R, et al. Patients with active tuberculosis often have different strains in the same sputum specimen. Am J Respir Crit Care Med. 2004;169:610. [PubMed]
29. de Viedma D, Marín M, Ruiz M, Bouza E. Analysis of clonal composition of my-cobacterium tuberculosis isolates in primary infections in children. J Clin Microbiol. 2004;42:3415. [PMC free article] [PubMed]
30. Shamputa I, et al. Genotypic and phenotypic heterogeneity among Mycobacterium tuberculosis isolates from pulmonary tuberculosis patients. J Clin Microbiol. 2004;42:5528. [PMC free article] [PubMed]
31. Cohen T, et al. Mixed-Strain Mycobacterium tuberculosis Infections and the Implications for Tuberculosis Treatment and Control. Clinical Microbiology Reviews. 2012;25:708. [PMC free article] [PubMed]
32. Du Plessis D, Warren R, Richardson M, Joubert J, van Helden P. Demonstration of reinfection and reactivation in HIV-negative autopsied cases of secondary tuberculosis: multilesional genotyping of Mycobacterium tuberculosis utilizing IS 6110 and other repetitive element-based DNA fingerprinting. Tuberculosis. 2001;81:211. [PubMed]
33. Cosma C, Humbert O, Ramakrishnan L. Superinfecting mycobacteria home to established tuberculous granulomas. Nat Immunol. 2004;5:828. [PubMed]
34. Cosma C, Humbert O, Sherman D, Ramakrishnan L. Trafficking of superinfecting mycobacterium organisms into established granulomas occurs in mammals and is independent of the erp and esx-1 mycobacterial virulence loci. J Infect Dis. 2008;198:1851. [PMC free article] [PubMed]
35. Sergeev R, Colijn C, Cohen T. Models to understand the population-level impact of mixed strain m. tuberculosis infections. J Theor Biol. 2011 [PMC free article] [PubMed]
36. Henao-Tamayo M, et al. A mouse model of tuberculosis reinfection. Tuberculosis. 2012
37. Wargo AR, Huijben S, De Roode JC, Shepherd J, Read AF. Competitive release and facilitation of drug-resistant parasites after therapeutic chemotherapy in a rodent malaria model. Proc Natl Acad Sci USA. 2007;104:19914. [PubMed]
38. Harrington W, et al. Competitive facilitation of drug-resistant Plasmodium falciparum malaria parasites in pregnant women who receive preventive treatment. Proc Natl Acad Sci USA. 2009;106:9027. [PubMed]
39. Basu S, Maru D, Poolman E, Galvani A. Primary and secondary tuberculosis preventive treatment in hiv clinics: simulating alternative strategies. Int J Tuberc Lung Dis. 2009;13:652. [PubMed]
40. Cohen J. Approval of novel tb drug celebratedwith restraint. Science. 2013;339:130. [PubMed]
41. Churchyard GJ, Friedland G, Fielding K, Nardell E. Opportunities afforded by new drugs for tuberculosis. The Lancet infectious diseases. 2010;10:368. [PubMed]
42. Blower SM, Dowlatabadi H. Sensitivity and Uncertainty Analysis of Complex Models of Disease Transmission: An HIV Model, as an Example. International Statistical Review Revue Internationale de Statistique. 1994;62:229.
43. Blower SM, et al. The intrinsic transmission dynamics of tuberculosis epidemics. Nat Med. 1995;1:815. [PubMed]
44. Dye C, Espinal M. Will tuberculosis become resistant to all antibiotics? Proceedings of the Royal Society of London Series B: Biological Sciences. 2001;268:45. [PMC free article] [PubMed]
45. Samandari T, et al. 6-month versus 36-month isoniazid preventive treatment for tuberculosis in adults with HIV infection in Botswana: a randomised, double-blind, placebo-controlled trial. The Lancet. 2011 [PubMed]
46. Springett V. Ten-year results during the introduction of chemotherapy for tuberculosis. Tubercle. 1971;52:73. [PubMed]
47. Hughes G, Currie C, Corbett E. Modeling tuberculosis in areas of high HIV prevalence. Simulation Conference, 2006. WSC 06. Proceedings of the Winter; IEEE. 2006. pp. 459–465.