Optima TB: A tool to help optimally allocate tuberculosis spending

Approximately 85% of tuberculosis (TB) related deaths occur in low- and middle-income countries where health resources are scarce. Effective priority setting is required to maximise the impact of limited budgets. The Optima TB tool has been developed to support analytical capacity and inform evidence-based priority setting processes for TB health benefits package design. This paper outlines the Optima TB framework and how it was applied in Belarus, an upper-middle income country in Eastern Europe with a relatively high burden of TB. Optima TB is a population-based disease transmission model, with programmatic cost functions and an optimisation algorithm. Modelled populations include age-differentiated general populations and higher-risk populations such as people living with HIV. Populations and prospective interventions are defined in consultation with local stakeholders. In partnership with the latter, demographic, epidemiological, programmatic, as well as cost and spending data for these populations and interventions are then collated. An optimisation analysis of TB spending was conducted in Belarus, using program objectives and constraints defined in collaboration with local stakeholders, which included experts, decision makers, funders and organisations involved in service delivery, support and technical assistance. These analyses show that it is possible to improve health impact by redistributing current TB spending in Belarus. Specifically, shifting funding from inpatient- to outpatient-focused care models, and from mass screening to active case finding strategies, could reduce TB prevalence and mortality by up to 45% and 50%, respectively, by 2035. In addition, an optimised allocation of TB spending could lead to a reduction in drug-resistant TB infections by 40% over this period. This would support progress towards national TB targets without additional financial resources. The case study in Belarus demonstrates how reallocations of spending across existing and new interventions could have a substantial impact on TB outcomes. This highlights the potential for Optima TB and similar modelling tools to support evidence-based priority setting.

Numbered indices correspond to the aforementioned six-fold split; X 1 = X SP DS , X 2 = X SP M DR , X 3 = X SP XDR , X 4 = X SN DS , X 5 = X SN M DR , and X 6 = X SN XDR . This applies only to D u i , D d i and T i for i = 1, . . . , 6. Parameter definitions can be found in Table A.

Vaccination
Vaccination against TB is generally part of the national childhood immunisation programme in most of the countries where TB is endemic, reaching >80% of neonates and infants. Bacille Calmette-Guérin (BCG) vaccine has limited effect in preventing primary infection [4] and, if administered to already-infected individuals, does not prevent reactivation of latent pulmonary infection [13]. If administered to uninfected individuals, it can slow down the progression to active disease (following infection with the bacteria). Roy et al. [4] estimated limited protection against infection, with a protective efficacy of 19-27%, compared with 71% protective efficacy against reactivation to active TB, while protection by BCG against progression from infection to active disease, calculated by using infected individuals, was 58%.

LTBI progression
Results from a recent review [14] show the importance of distinguishing between people of varying disease-development rates by adding more than one latent compartment to TB models. Optima TB includes two compartments, early and late infection; all newly infected individuals enter the early latent stage and, if they do not develop active disease in the first 5 years after infection, move to the late latent compartment. This is in line with reports showing that 95% of active cases develop in the first 2 to 5 years after infection, with the other 5% over the rest of an individual's lifetime [15][16][17].
Parametrising the latent tuberculosis infection (LTBI) components of a model is one of the main issues in deterministic TB modelling, because countries do not collect these data and prevalence studies of latent infection are scarce. We initialise the LTBI compartments using estimates of annual trends [18]. We then calculate latency progression by applying a simple survival model with constant hazard, CI = 1 − exp(−IR × duration), to the number of infectious cases [8], where the cumulative incidence (CI) is a function of incidence rate (IR) and time. These studies, however, are from the pre-chemotherapy era; these values are tentative starting points for calibration, to be adjusted when required.

LTBI treatment
Studies have shown that LTBI treatment is important towards the eradication of TB [19][20][21][22], because a quarter of the world population is estimated to be currently infected [18]. However, because of the widespread infection and the 10% lifetime probability of progressing to active TB, mass treatment would not be a cost-effective strategy. Moreover, even in high-burden circumscribed areas, mass screening and treatment have been proved to have no significant effect on tuberculosis due to the extremely high force of infection and loss of protection from infection once treatment is over [23]. For these reasons, WHO established a ≥ 90% preventive treatment target for 2025 only in high risk populations, such as people living with HIV or adult and child contacts of pulmonary TB cases [24]. In countries with high or upper-middle income and low TB burden (i.e. TB incidence of <100 per 100000 per year), WHO guidelines recommend testing and treatment of LTBI for people living with HIV, adult and child contacts of pulmonary TB cases, people initiating anti-tumour necrosis factor treatment, patients receiving dialysis, patients preparing for organ or haematological transplantation, and patients with silicosis. In prisoners, healthcare workers, immigrants from high TB burden countries, homeless persons and illicit drug users, systematic testing and treatment of LTBI is conditionally recommended, according to TB epidemiology and resource availability [25]. To test for LTBI, either commercial interferon-gamma release assays (IGRA) or Mantoux tuberculin skin testing (TST) could be used. Due to low TST specificity, BCG vaccination may result in false positive results and an IGRA test would be required to give more precise results. Due to the higher costs of IGRA, however, TST is often the only test available. Before LTBI treatment, a chest radiography should be performed to rule out active TB disease. Recommended treatment regimens for LTBI include: 6 or 9 month isoniazid, 12 week rifapentine plus isoniazid, 3-4 month isoniazid plus rifampicin, or 3-4 month rifampicin alone [25].
In the model we include a secondary latency pathway collecting all individuals who have been previously treated for LTBI or TB and re-infected or infected after vaccination. This distinction is useful because these individuals will always result positive to TST testing and will not generally be treated for LTBI. The progression toward active TB from these compartment is reduced only in the 0-15 age group because of the effects of vaccination in children, while it is the same as the progression from the primary latent pathway for the older age groups because treatment does not entail immunisation.

Subclinical TB
The progression from LTBI to active TB can involve a state of asymptomatic disease [26][27][28]. Not much is known of this subclinical TB in terms of onset and duration. However, the undiagnosed individuals are potentially already infectious at this stage. Following subclinical TB, symptoms (which can include fever, weight loss and prolonged cough) start to appear and a diagnosis may follow. In the model D u e is a 'junction' compartment that merely collects and redistributes cases of newly activated TB across smears and strains; it does not contribute to the dynamics of the system. Thus,

Smear status
The presence of acid-fast-bacilli (AFB) or other specimens on a sputum smear often indicates TB disease, and a posterior-anterior chest radiography is used to assess the possibility of pulmonary tuberculosis. Several studies have analysed and discussed specificity and sensitivity of different blood and/or sputum based TB tests [29][30][31]. smear-positive cases are generally assumed to be more infectious than smear-negative because of the higher amount of bacilli they expectorate [32]. The reduction of infectiousness in smear-negative cases with respect to smear-positive seems to be around 80%, however, studies analysing the number of secondary cases produced by primary active cases are either old and circumscribed to a delimited area [2,33] or from the pre-chemotherapy era and unreproducible nowadays due to ethics.

Diagnosed vs undiagnosed TB
From a modelling perspective, this aspect of TB epidemiology is best supported by current data-availability. In conjunction with TB diagnosis, countries usually collect data on cases notified, drug-resistance status and co-infection with HIV. Data on smear status may also be available. Moreover, WHO publishes an annual TB report showing reports and estimates of incidence, diagnostic coverage, treatment uptake and deaths by geographic areas. The amount of time between disease onset and both diagnosis and treatment initiation, or, more specifically, the amount of time an individual stays infectious, combined with the rates of diagnosis and treatment uptake, is the main factor behind infection transmission. Styblo estimated that about 10 secondary infections arise annually from one untreated smear-positive case [32,34]. Our model is informed by nationally collected data on notified cases, treatment initiations and completion, loss to follow-up and number of deaths. Unsurprisingly, highest TB-related death rates usually occur in patients aged 65 and above because of more frequent drug-related adverse events and increased co-morbidity [35,36]. Outputs from the model are calibrated against prevalence and incidence estimations annually calculated by WHO, which we disaggregate by age group according to local demographic rates. Similarly as for latency progression, problems arise in the calculation of natural recovery and death rates in the undiagnosed population, because they are inferred by using a simple survival model to very old datasets [12]. The resulting values should be considered as default values and calibrated, if needed, when analysing current epidemics.

Drug sensitivity
Once TB is diagnosed, one of the most significant pieces of information is the drug sensitivity of the strain. Multi-drug resistant (MDR) strains are resistant to the two main anti-TB drugs, isoniazid and rifampicin, while extensively drug resistant (XDR) strains are also resistant to any fluoroquinolone and at least one of three second-line injectable drugs (capreomycin, kanamycin and amikacin).
Recent WHO global reports [37,38] show that in 2016 there were 600,000 new cases of rifampicin-resistant (RR) TB globally, 490,000 of which were MDR-TB, challenging the prospect of ending TB by 2035. Treatment for drug-sensitive TB usually involves taking isoniazid and rifampicin for six months supplemented by ethambutol and pyrazinamide in the first two months [39], while treatment for drug resistant TB is longer and less efficacious and can last from 9 months to 2 years [37]. While it is possible for drug resistance to arise following improper treatment, it has been suggested that transmission of drug-resistant strains has higher overall impact than primary escalation [40], i.e. mutations conferring resistance while on treatment. Moreover, very little information is currently available on strain escalation, countries do not usually collect this type of information. Thus, while the model includes strain escalation, the corresponding parameter default values are currently set to zero; they can be adjusted in case of data availability.

Recurrent TB
Finally, an important aspect of TB is its recurrence following treatment completion, but related data are not usually available. Several studies have analysed recurrence in TB [41][42][43][44][45][46][47][48][49][50], usually analysing small cohorts and reporting very different results, showing that numbers will vary according to several factors such as HIV status, drug resistance, and geography or, more precisely, different forces of infection in different areas. The WHO recording and reporting system distinguishes between new and re-treatment cases, however it does not discriminate between true relapse and reinfection, and it designates as relapse any recurrence of TB [51]. Since the majority of relapses arises during the first months up to 1-2 years after the conclusion of the first treatment [5-7], in order to differentiate between the two, we assume that cases of recurrence during the first 2 years following treatment completion are due to relapse, while later recurrences are due to reinfection. In the first case people move back to the undiagnosed TB compartments D u according to a data-informed relapse-rate while, in the second case, move first to the successfully treated compartment J where they can proceed to the secondary latent pathway if re-infected.

Additional Results
Projections of active TB incidence ( Figure