We find that implementing a 10-month treatment regimen for MDR-TB is among the most effective means for reducing the impact of this dangerous threat to global TB control in Karakalpakstan, a region with high rates of drug resistance among TB patients. The more rapid throughput of patients leads to an initial transient increase in mortality under the conservative assumption of unchanged treatment outcomes, although this is not a programmatically significant effect and is followed by a quick recovery and consistent decline in disease rates thereafter. Our model did not predict that wider use of the shorter MDR-TB regimen would increase the acquisition of additional drug resistance. The comparator intervention that led to reductions in MDR-TB disease burden most similar to expansion of the short-course regimen was doubling of MDR-TB treatment availability, while the other comparators were less effective. Moreover, synergistic effects could be expected if wider use of shorter MDR-TB regimens is combined with improved case detection.

Our first modelling study of TB transmission aimed to establish a flexible approach to simulating TB transmission dynamics in highly endemic settings within the framework of a deterministic compartmental model, but assumed regimen duration to be fixed for each strain. In this earlier work, we found that MDR-TB became the dominant strain at model equilibrium even in the presence of significant fitness costs, which is attributable to both lower rates of case detection and differences in progression through treatment [14]. In this study, we consider the issues surrounding the diagnostic process in greater detail, distinguishing detection of active TB from the process of determining the extent of drug resistance in the infecting organism and subsequent progress through the treatment regimen. The relative importance of each of these processes is likely to be setting-dependent and programs may act synergistically, as bottlenecks will exist at different points in the complex journey from active disease through to treatment completion.

Our conclusions depend on a number of model assumptions and the local TB epidemiology simulated. In particular, our modelling of a treatment program close to capacity explains the lack of effect observed from improved detection of TB cases and improved identification of MDR-TB patients from those detected. Additional file 1: Figure S2 shows that the reason for the relative ineffectiveness of most comparator interventions (all except increasing availability of MDR-TB treatment) is that they do nothing to relieve the bottleneck of treatment availability, such that numbers of patients awaiting treatment increase rapidly over the intervention period. Although there is no formal limit on MDR-TB treatment availability in Karakalpakstan, we consider expansion of the treatment program to manage the markedly increased patient load to be a programmatic intervention. As expected, doubling treatment availability and decreasing treatment time 2.4-fold had comparable effects on incidence and mortality, although the shorter treatment time had a greater effect on prevalence, as patients are considered prevalent cases until treatment is completed. The small increase in MDR-TB burden through improved detection is due to patients transitioning from being on inappropriate treatment for DS-TB (which is considered to have a partial therapeutic effect) to identified but awaiting treatment (and so untreated). Under- or over-estimation of the absolute or relative burden of each of the TB strains in the Province are likely to affect our conclusions, although only a markedly lower absolute burden of MDR-TB is likely to result in significant attenuation of the benefits from the shorter regimen. Given the complexity of the baseline dynamics simulated, we focused on programmatic parameters in our sensitivity analysis, rather than exploring variations of all parameters.

Our findings are likely to be generalisable to a number of other contexts in which treatment capacity is an important constraint, as the shorter regimen can be used in HIV-positive and paediatric populations. However, the programmatic situation is a key determinant of the regimen’s likely impact, as other factors may limit treatment commencement. For example, if MDR-TB treatment capacity is available but access to drug-susceptibility testing (DST) is limited and many patients are on incorrect regimens, improving access to DST is likely to compare more favourably to other interventions. Such situations may exist in contexts where intense community transmission of MDR-TB occurs, but DST is reserved for retreatment cases only. Alternatively, if extensive pre-health system delays to presentation are important in limiting the rate at which MDR-TB patients commence treatment, active case finding is likely to have a greater effect in reducing the burden of disease attributable to this strain. Therefore, in these situations, the shorter regimen may compare less favourably to these two interventions. Last, if poor treatment outcomes are reported programmatically, the shorter regimen may have a significant impact if improved treatment outcomes can be achieved, rather than by relieving the bottleneck to treatment commencement. Synergistic effects were observed in this study, which is understandable as both the shorter regimen and increased treatment capacity led to unfilled treatment capacity, which could be used if more patients with MDR-TB were detected by the health system and/or correctly classified as MDR-TB (Additional file 1: Figures S2, S4, S5).

A previous programmatic application of a similar model to Western Province of Papua New Guinea found a smaller impact of the short-course regimen [16]. However, in this earlier study, we considered treatment commencement to be dependent on the rate at which MDR-TB patients were detected, but independent of treatment availability. We also previously considered an extended period of hospitalisation to be necessary for implementation of the shorter regimen, due to the number of drugs employed during the intensive phase of treatment. Although this consideration is not explicitly modelled here, the community-based approach to treatment currently employed in Karakalpakstan would make scale-up of treatment (e.g. Scenario 6) less resource-intensive and more feasible than in settings where hospitalisation is deemed essential throughout the intensive phase.

We do not present an economic analysis and the comparator interventions are not intended to be equivalent in terms of resource consumption or expense. However, several may be considerably more difficult to implement and many of the resources already in place to provide the standard WHO regimen could be adapted to short-course treatment. In fact, the short-course intervention is likely to be significantly cost-saving, as we estimate the expense of the short-course regimen at around 760 Euros in Karakalpakstan by comparison to over 3000 Euros for the standard regimen (personal communication MSF), which is consistent with estimates from elsewhere [10, 18]. Therefore, even under scenarios that achieve a higher throughput of patients as a result of faster treatment completion, the short-course intervention should be cost saving due to its lower cost per unit time on treatment. By contrast, for programs such as active case finding, improved MDR-TB identification, patient monitoring for response to treatment and support for adverse effects, and increased treatment availability, significant additional resources are likely to be required.

The short-course regimen we consider is based on analysis of sequential cohorts of patients enrolled into treatment in Bangladesh and elsewhere [6–9]. The study and subsequent follow-up has demonstrated favourable outcomes sustained after treatment completion without significant amplification of resistance [9]. There has been debate over whether a regimen based on this form of evidence, rather than the gold standard of the randomised controlled trial, should be accepted for programmatic use. Therefore, a multi-centre, non-inferiority randomised controlled trial has been initiated to better determine the efficacy of safety of the regimen [12]. Such evidence will be of great use in determining the extent and speed with which such regimens should be adopted, particularly given that the standard regimens are based on very low quality evidence [19] and that meta-analyses of standardised regimens estimate treatment success rates around 50% (when including patients ineligible for shorter regimens) [20].

The recent WHO guidelines provide a conditional recommendation supporting the use of the shorter regimen in the context of further research, although the broader epidemiological impact of the regimen has not yet been observed. Modelling the likely effect of such a programmatic change is important in this context. This study aims to realistically simulate the introduction of short-course regimens for a similar patient group to that recommended for treatment by the WHO guidelines [10].

Local patterns of drug resistance are also an important consideration, as it is important to limit treatment to patients infected with strains susceptible to the constituents of the regimen as much as possible, to avoid further exacerbating drug resistance problems. Although evidence for the effectiveness of short-course MDR-TB regimens is now emerging from a range of settings [7], our study is not intended to determine the regimen’s efficacy, but rather to estimate likely improvements in MDR-TB control through shortening treatment duration.

Although new agents are now available for the treatment of MDR-TB [21, 22], these are largely intended to strengthen conventional MDR-TB regimens [23]. More important than the development of single agents is the formulation of new regimens to reduce treatment duration at the programmatic level. Therefore, trials of new shorter regimens, such as STAND (Shortening Treatment by Advancing Novel Drugs, NCT02342886), Practecal and Nix-TB, hold promise for improvements in TB treatment effectiveness [24–26]. Moreover, STREAM II (The Evaluation of a Standard Treatment Regimen of Anti-Tuberculosis Drugs for Patients with Multidrug-Resistant Tuberculosis, ISRCTN18148631), which considers several short-course regimens, includes a treatment arm in which injectable agents are avoided entirely [27]. Although proving the efficacy of these regimens is essential, it is also important to demonstrate that programmatic benefits are achievable to argue for their introduction.