Methods

We conducted a Markov model-based analysis to estimate the incremental cost-effectiveness of the three BPaL-based regimens investigated in the TB-PRACTECAL trial (described in S1 Table) for treatment of TB that is at least resistant to rifampicin, as compared with the current standard of care (SOC) in Philippines, South Africa, Georgia, and India (Table 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Country characteristics.

https://doi.org/10.1371/journal.pgph.0001337.t001

Incremental costs from the provider perspective and disability-adjusted life years (DALYs) averted were modelled over a 20-year time horizon. This model-based analysis was conducted to inform growing policy questions following the early termination of the TB-PRACTECAL trial and WHO recommendations for programmatic uptake of BPaL-based regimens [9]. Methods for the economic evaluation were adapted from the planned per-protocol economic analysis [10]. This is a model-based analysis using previously published data and did not require ethical approval.

Patient and public involvement

The research primarily targeted policy makers, and the study question was driven by the Resource Considerations requirement in the WHO guidelines development process. An economic evaluation is considered critical to the development of public health guidelines and this research study fulfilled this need. Patients were not involved in the design of this modelling study, however the aspect of this study that involves clinical outcomes was designed and implemented with extensive engagement of communities and patients [16]. Results of the study will be shared in a lay summary format with community advisory boards and distributed in clinics where the study took place. Compensation for participation to the main trial was defined in close collaboration and approval from the local ethics committees.

Study population and settings

We developed models to reflect programmatic uptake of each regimen among all patients with RR-TB in four countries with a range of characteristics potentially relevant to global decision-making context, including burden of MDR/RR-TB, burden of HIV among people with TB, and current mix of long vs. short SOC regimens for MDR-TB. Table 1 shows relevant characteristics for each country.

Interventions and comparators

The TB-PRACTECAL trial evaluated three six-month treatment regimens; arm 3 included bedaquiline, pretomanid and tapered-dose linezolid (600mg daily for 16 weeks then 300mg daily for the remaining 8 weeks) (BPaL). Arm 1 included the BPaL backbone with addition of moxifloxacin (400mg) (BPaLM); arm 2 included the BPaL backbone with addition of clofazimine (50-100mg) (BPaLC). Each TB-PRACTECAL regimen was compared with the current mix of SOC regimens, which was estimated from the reported number of patients enrolled on short vs. long MDR-TB regimens in data underlying the most recent Global TB Report [2]. Specifications for short and long SOC regimens were taken from the national MDR treatment guidelines for each country [12–15].

Model description

Our Markov model was structured to represent progression of a cohort of patients through different health states and activities under each treatment option (S1 Fig). Patients were assumed to enter the model with active RR-TB at the point of treatment initiation. At model entry, mean age was assumed to be 35, the average age of PRACTECAL participants [17] (S2 Table). At the end of each month, patients could transition to the next month of treatment, loss to follow-up, treatment failure, or death.

Treatment duration depended on the treatment regimen; we assumed the standardized BPaL-containing regimens lasted 24 weeks in line with the TB-PRACTECAL protocol, the standard short SOC regimen lasted 36 weeks and long SOC regimens lasted 80 weeks. Patients who were lost to follow-up (LTFU) before treatment completion had a chance of re-entering treatment each month in the first two years from the start of treatment. Following failure from the initial treatment regimen, we assumed all patients were put on a long SOC ‘rescue’ regimen regardless of their initial treatment regimen. Patients who had treatment failure on the rescue regimen moved to an ‘end of life’ state. Patients successfully completing treatment had a chance of experiencing relapse in the first four years following treatment completion; reinfection was not included in the model. Patients who relapsed had a chance of re-entering treatment following national TB treatment coverage rates. We did not model any potential reduction of onward transmission of TB due to a shorter treatment period. Key model parameters are listed in S2 Table.

Treatment outcomes

Outcomes for SOC regimens in the model were based on national treatment outcomes from the 2019 treatment cohort in each country, as reported to WHO [2]. We assumed that there was no significant difference in treatment failure or death between standardized long and short SOC regimens [18]. Following Walker et al. [19] we assumed variable risk of loss to follow-up according to the time on treatment, and adjusted treatment success rates to account for higher loss to follow-up in longer SOC regimens.

Treatment success for TB-PRACTECAL regimens were estimated using country-specific SOC treatment success, multiplied by the ratio of the probability of treatment success at 72 weeks in the interventional arms of the TB-PRACTECAL trial as compared with the SOC arm (S2 Table). Outcomes from the per-protocol population were used as this was the most conservative approach. Treatment success in the model was defined as having no unfavourable outcome (including loss to follow-up, treatment failure, recurrence, or death) at 72 weeks. Although the trial outcomes showed reduced deaths in the investigational regimens, we conservatively assumed no impact on deaths, treatment failures, or recurrence, and instead assumed improvements in treatment success resulted from reductions in loss to follow-up. This conservative approach was taken as the very low number of failures and deaths in both interventional and SOC populations observed in the TB-PRACTECAL trial may not be achievable in programmatic settings, and is tested in univariate sensitivity analysis.

Monthly transition probabilities for all outcomes were calculated using the average treatment duration for each regimen. Patient costs and outcomes were tallied over a lifetime horizon of 20 years, with cycles of one month. We chose 20 years as this was adequate to capture all relevant costs and outcomes.

The measure of effectiveness for our incremental cost-effectiveness estimates was in terms of DALYs averted. DALY weights for each health state were sourced from the most recent Global Burden of Disease study, and are reported in S3 Table [20]. Following evidence that there is a lifelong post-TB health burden even in patients with treatment success, we used a DALY weight of 0.053 for the ‘post-TB’ state [21]. We used expert opinion to map adverse event symptoms to equivalent health states where adverse events were not specified in the Global Burden of Disease database. We used a multiplicative model to generate a combined disability weight for patients with adverse events [22]. All DALYs were discounted at 3% in the base case [23].

Costs

Treatment costs were estimated from a providers’ perspective using a combination of country MDR treatment guidelines, previously published estimates of prices and quantities of different services, and expert opinion (S4 Table). A provider perspective was chosen due to a lack of high-quality cost data from the patient perspective for the included countries. For Georgia, India, and Philippines, we derived the unit costs of outpatient visits, inpatient bed-days, community and other visits, and laboratory tests from the Value TB dataset, the largest multi-country TB costing study to date [24]. Unit costs for health care services in South Africa, and for procedures, medications and fluids for adverse events, were sourced from the literature [25, 26]. All costs are reported in 2019 USD. Costs reported in USD from previous years were inflated using the United States gross domestic product (GDP) deflator [11].

We estimated the quantities of outpatient visits, inpatient bed-days, community- and other visits, and lab tests per month using the Value TB dataset, national country guidelines, and validated these through expert opinion (S2 Fig) [12, 15, 24]. Where data were available, we used the ‘real world’ quantities of a wide range of services, as delivered in health facilities observed in the Value TB study. This often differed from country guidelines but was a better representation of total costs incurred during treatment. Quantities were estimated separately for intensive and continuation phases for all SOC regimens, including a two-month inpatient treatment period for all regimens in Georgia. Quantities of services per month for investigational regimens were assumed to follow the pattern of the first six months of care for the short SOC regimen. Expert opinion was used to estimate the quantities of outpatient visits and inpatient bed-days per average patient for different adverse events. Patients who were lost to follow-up (LTFU) were assumed to incur costs of one LTFU tracing call or home visit per month until death or return to care.

Drug prices were sourced from the Global Drug Facility Catalogue for all countries [27]. Drug quantities were modelled according to each country’s recommended SOC treatment regimen using standard recommended dosage for an adult weighing 51-70kg. All costs were discounted at a rate of 3% in the base case [23].

Uncertainty and sensitivity analyses

Incremental costs and effects were estimated using mean parameter values in the base case. Parameter uncertainty was assessed in a probabilistic sensitivity analysis, varying all parameters over 200 simulations. Risk ratios for treatment success were varied following a log normal distribution. Unit costs were varied using a gamma distribution, and all DALY weights were varied following a beta distribution. We also conducted a univariate sensitivity analysis to test model sensitivity to changes in key parameters, including non-drugs costs, prices for bedaquiline and pretomanid, likelihood of return to care after loss to follow-up, discount rate, and valuation of the post-TB health state. We estimated the incremental costs and effects of interventional regimens with each of these key parameters halved and doubled to represent extreme possible values. We also tested the base-case assumption that improvements in effectiveness from the TB-PRACTECAL regimens resulted from reductions in losses to follow-up; we ran alternative scenarios with reductions in deaths and treatment failures instead. We did not characterize heterogeneity or distributional effects as subgroup analyses were not performed on these preliminary outcomes from the trial.

Role of the funding source

TB-PRACTECAL was supported fully by MSF. Study sponsors were involved in the analysis and interpretation of the data, writing of the manuscript, and the decision to submit the paper for publication.