Despite the anticipated population benefits of IPT, this intervention has only been implemented in a limited number of settings and in 2007, a mere 0.1 per cent of eligible individuals actually began a course of IPT [31
]. Possible reasons for the slow pace of adoption of IPT interventions include the challenges associated with administering daily single drug preventive therapy for a very large number of individuals without symptomatic TB disease [32
] and the difficulty in ruling out the presence of active TB disease in HIV co-infected individuals [33
]. Increasing the uptake of IPT in communities with high HIV prevalence may require projections of the cost-effectiveness of such interventions and further demonstration that available diagnostic tools can effectively rule out subclinical TB disease before monotherapy is used. To encourage access and usage of IPT, the World Health Organization (WHO) recently revised IPT guidelines to suggest that the absence of symptoms (current cough, fever, weight loss or night sweats) is sufficient for ruling out active disease and initiating preventive therapy [34
]. They strongly recommend that eligible individuals complete a six month course of IPT, and in high-transmission areas, a 36 month course is advised.
However, recent studies aiming to assess the effectiveness and optimal duration of IPT have come to varying conclusions. A study in Botswana of HIV-positive individuals compared six and 36 months of IPT. Among those who tested positive for latent TB, in the six month cohort, the TB rate was 2.22 per 100 person years compared with 0.57 per 100 person years in the 36 month cohort. In individuals who tested negative for latent TB, there was little difference between the six and 36 month courses [7
]. By contrast, a trial in India of HIV-positive patients without active TB concluded that six and 36 month courses of IPT were equally effective [35
Our work suggests that differences between respiratory contact patterns and TB incidence between these settings may help explain why IPT strategies may have variable impact in different settings. While IPT treatment given to individuals prevents progression to disease, the impact of IPT at the community level depends on (i) the prevalence of TB infection among an individual's respiratory contacts, and (ii) the incidence of disease among those contacts, resulting in re-infection occurring after the IPT course is complete. The first of these is related to the extent of clustering of TB among shared groups of respiratory contacts, together with the overall prevalence of TB infection in the population. The second is mediated by the level of HIV/TB co-infection and the level of IPT coverage.
We have shown that the effect of IPT is highly variable in locally clustered networks compared with more globally connected ones. This is because in locally clustered networks, latent and incident TB infections occur in clusters, so that some areas suffer very high rates of incidence and re-infection, while other clusters are effectively protected from disease. In these protected clusters, treatment of a sufficient number of TB cases results in a reduction or elimination of onward TB transmission in the area. This occurs as a result of IPT when (i) a sufficient portion of the local TB burden is among HIV-infected individuals, and (ii) IPT coverage of eligible individuals is sufficiently high. We would therefore expect IPT to be most undermined by network and dynamical effects when TB infections are clustered among local groups of shared respiratory contacts, when HIV/TB co-infection is low among prevalent TB cases and/or when IPT coverage levels are too low to prevent onward transmission. This effect is exacerbated by local clustering of respiratory contacts, because it is in these circumstances that individuals who have been infected with TB once (whether or not they received IPT) are likely to be re-infected rapidly after their IPT treatment ends.
In terms of policy guidelines, our results imply that IPT will perform best if entire clusters of infected individuals at high risk of progression can be treated simultaneously. The benefits of IPT may be eroded if infection is not cleared from these clusters. This provides support to the conclusions from studies on clustering of TB cases in Africa [16
], advising that cluster detection techniques (of both high and low case numbers) may help TB control programmes use their resources efficiently and effectively. Contact patterns and clustering are difficult to measure, particularly because relationships are dynamic and contacts may occur in casual settings where individuals are not known. However, techniques such as contact tracing from known TB cases have proved useful for identifying possible clusters (covering both past and current contacts) [37
]; these efforts to identify clustered risk may be supported by social network analysis [17
] and targeted investigation within settings of possible transmission such as bars [38
It has previously been observed that clustering of contacts reduces disease spread over a network [39
]; we also saw this effect in our simulations. Indeed, a high degree of clustering reduces disease spread compared with what would be expected from the initial growth rate of an epidemic or outbreak [41
]. Because epidemics spread first within a community and then (typically later) between them, networks with community structure can maintain epidemics at quite low incidence levels overall, but with a large number of generations of infectives over a relatively long period of time. By contrast, globally connected networks tend to have relatively fast, high-incidence, but short epidemics [42
]. This can make estimating the efficiency of control measures very difficult.
Our original hypothesis was that high levels of re-infection in locally clustered networks would undermine IPT programmes by allowing those individuals who had recently received IPT to be at risk of TB disease despite their treatment. However, instead we saw that the same mechanism that exposed some individuals to high levels of re-infection (clustering) also worked to protect other individuals from re-infection, via some clusters enjoying the near-elimination of TB after the introduction of IPT. As the protective effect of clustering is much weaker on globally connected networks, IPT performance on local networks can actually be better than on global ones at the population level, despite substantial regional variability.
The model is limited by the network structures we assumed. It is difficult to obtain comprehensive network data for populations, so our networks are caricatures of the structures of contact networks. In addition, for simplicity, we assumed that the networks were static; this assumption probably alters the transmission dynamics across the population. However, these limitations do not alter the overall qualitative insights from the model.
In this model, we have investigated single, fixed-length courses of IPT, which is consistent with current policy guidelines. We showed that the network structure affects both the frequency and distribution of TB re-infection, and thereby may modify the expected effectiveness of the impact of these types of IPT interventions. To mitigate the effect of re-infection, it may be useful to consider continuous or multiple courses of IPT, as suggested by some studies [7
]. These policies could potentially improve the impact of IPT by providing a longer period of protection from re-infection and progression, but there are concerns that longer or multiple courses of IPT may provide increased selective pressure and may facilitate the emergence of drug-resistant strains of TB (despite there being no evidence that IPT causes acquired resistance in individual patients [10
]). In the recent publication of TB guidelines, the WHO concluded that IPT does not increase the risk of developing drug-resistant TB, so this should not be a barrier to its provision. Drug-resistant strains have been identified in most high-TB incidence areas [43
], and IPT would be expected to exert selective pressure at the population level if given in the context of co-circulating resistance and sensitive strains [46
]. Our model assumes only one, drug-sensitive strain of TB and does not consider the evolution of new drug-resistant strains or the effect on existing levels of resistance.
In conclusion, we have developed a new multi-network model for HIV and TB and find that local network structure can induce high levels of repeated re-infection that may undermine the projected effectiveness of IPT. At the same time, strong local clustering may provide a protective effect wherein TB is essentially eliminated from some areas and re-introduction of TB is limited by a scarcity of long-range connections. The combination of these dynamics means that IPT's effectiveness in clustered networks may be variable; this insight will be useful as we attempt to interpret and reconcile results from ongoing IPT trials.