After removal of duplications, the search identified 322 articles. Based on the abstract, 313 studies were excluded for reasons detailed in Fig. 1. Further exclusion based on the full text was performed on five other studies. A total of four studies fulfilled the inclusion criteria and were included [17,18,19,20].
The studies fulfilled most of the quality criteria in the appraisal checklist. The complete appraisal result of these studies can be found in the Electronic Supplementary Material. Some studies only partially addressed uncertainty such as through scenario analyses [19, 20]. However, one study using a complex model mentioned that sensitivity analyses other than a deterministic analysis and scenario analyses would be computationally expensive .
General Characteristics of the Studies
Three of the included studies were model-based economic evaluations [17,18,19], while one was an empirical study based on publicly available and local data . Two of the model-based studies used a decision analytic model, which did not take into account TB transmission process (i.e., a static model) [17, 18]. One study did not specify the type of static model used , while another used a decision tree model . The other study used a model that considers the transmission process; i.e., an individual-based dynamic transmission model (dynamic model) . The model-based studies analyzed the cost effectiveness of adopting a hypothetical new first-line SCR that was given in 4 months compared with the standard, 6-month, first-line regimen. These studies expressed the health outcome of implementing the new first-line SCR in disability-adjusted life-years (DALY) averted [17,18,19].
The empirical study analyzed the cost effectiveness of long-course regimens (LCRs, 20 months or more) and SCRs (9–12 months) containing bedaquiline compared with the regimens containing injectable drugs for drug-resistant TB treatment . This study expressed the health outcome of implementing the regimens in the number of successful treatments .
There was no clinical evidence for the hypothetical first-line SCR analyzed in the model-based studies. Hence, the studies took various assumptions, which were consistent in all studies. The studies assumed that the novel SCR was non-inferior and non-superior to the current regimen. The same risk of mortality during treatment and probability of cure upon treatment completion were applied for both regimens. They also assumed that the default or loss-to-follow-up rate per month during treatment was the same for both regimens. Hence, the studies only analyzed the impact of shortening the duration of treatment . There was also no information about the potential price of the new regimen; hence, the studies performed a price-threshold analysis with 1 USD per day as a baseline price. The study by Knight et al. did not assume a baseline price for the new regimen; however, it looked at the price that resulted in a cost per DALY averted equal to the gross domestic product (GDP) per capita of the study setting, that is, South Africa . The relatively high GDP per capita of South Africa (compared with most of the settings of the other studies) resulted in a regimen price higher than 1 USD per day .
A long-term follow-up of a clinical trial showed different levels of cure following patients’ default from treatment . Based on this information, the model-based studies assumed a possibility of cure after treatment default; depending on how long the individuals had been on treatment before defaulting. They assumed that early defaulters who received the novel first-line SCR would have a higher probability for cure compared with those who received the standard regimen.
In the empirical study, bedaquiline-based regimens were assumed to be more effective than the injectable-based regimens since they shortened the time to sputum smear conversion (an indication of treatment success and reduction of transmission risk). The efficacy of injectable-based SCR was taken from the STREAM 1 clinical trial, while the efficacy of injectable-based LCR referred to the WHO-reported global treatment success rate for multi-drug resistant (MDR) TB . For bedaquiline-based LCR, the efficacy was taken from a meta-analysis, and for bedaquiline-based SCR, it was taken from a modeling study since there was no information from clinical studies .
Dealing with Uncertainties
All model-based studies performed a deterministic sensitivity analysis. The deterministic analysis was mainly performed on parameters that were determined with assumptions, such as default rate, as well as cost parameters, such as treatment delivery costs (i.e., treatment costs excluding the drug regimen costs).
Two of the model-based studies considered the health system dynamic by performing the analysis with different scenarios, i.e., a ‘perfect’ and ‘current’ scenario [17, 19]. The analysis with the ‘perfect’ scenario assumed that the adherence of healthcare providers and patients to the TB treatment guideline was high. The high adherence caused a high rate of TB care utilization, high treatment delivery costs, and low rate of treatment default. The analysis with the ‘current’ scenario applied the actual level of adherence to the TB treatment guideline in the study setting. With this scenario, the rate of TB care utilization and the treatment delivery costs were lower, while the treatment default rate was higher than the rate with the ‘perfect’ scenario. In one of the model-based studies, the analysis was also performed with another scenario, which assumed that TB could only be cured when the treatment was completed.
The empirical study also performed several scenario analyses. However, the implications of the different scenarios were limited to capacity utilization and regimens’ price. One scenario assumed a high prevalence of extensive drug-resistant TB, which increased the utilization of inpatient care and hospitalization costs. Another scenario assumed local procurement for the regimens, which decreased the price of injectable-based regimens.
PSA was only performed by one study, under the assumption that the regimen price was 1 USD per day, and the willingness-to-pay threshold (cost per DALY averted) was equivalent to the GDP per capita of the settings (see Table 2 for details on the study settings) . Standard deviation and 95% uncertainty ranges of the ICER were estimated from the PSA (see Table 2). The uncertainty range covers the true ICER of implementing the new regimen. Thus, the wider the range, the higher the uncertainties surrounding the ICER. The threshold also helps in interpreting the range. Implementing the new regimen is not cost saving but cost effective when the ICER lies below the threshold.
Information for Decision Making
The results from the three model-based studies were summarized in Table 2. The studies with static models showed that the new SCR for drug-sensitive TB would be cost saving in most settings [17, 18]. Despite causing an increase in drug acquisition costs, the new first-line SCR decreased the treatment delivery costs (i.e., treatment costs excluding the drug regimen costs) due to the short period of treatment. The first-line SCR was not cost saving but still cost effective in settings with low treatment delivery costs. However, the novel first-line SCR was not cost effective at a price of 1 USD per day in Bangladesh due to the significantly low treatment delivery costs in the country . The study that used a dynamic model showed the small impact of the new first-line SCR on TB prevalence due to its limited effect on the TB transmission process .
The empirical study showed that the cost per treatment success for MDR TB with injectable-based SCR was lower compared with the cost with injectable-based LCR (around 30% lower) . The use of injectable-based SCR instead of LCR mainly reduced the drug acquisition and hospitalization costs. Switching to a bedaquiline-based regimen further reduced the hospitalization costs for SCR and LCR (Fig. 2). It also eliminated the costs of injectable-related adverse event management for both regimens. Overall, switching to bedaquiline-based SCR and LCR was cost saving. However, the cost saving was higher for LCR than for SCR in all settings.
Results of the Deterministic Sensitivity and Scenario Analysis
The deterministic sensitivity analysis of the three model-based studies identified the main drivers for the cost-effectiveness outcomes of the new first-line SCR, that is, treatment default rate and the treatment delivery costs. The health benefit of implementing new first-line SCR was large when the default rate was high. Implementing the new first-line SCR could result in a significant cost saving when the treatment delivery cost was high. Hence, the price threshold for the hypothetical first-line SCR could be high if the treatment delivery costs were high.
One study showed that, with a ‘perfect’ scenario, implementing the new first-line SCR instead of the current regimen would not result in any additional health benefit . However, another study showed that implementing the new first-line SCR resulted in high cost saving due to the high treatment delivery costs with the ‘perfect’ scenario . The cost saving with the ‘current’ scenario was lower compared with the saving with the ‘perfect’ scenario . One study also showed that assuming no cure for patients who defaulted from TB treatment slightly increased the cost effectiveness of the new first-line SCR .
In all scenarios, bedaquiline-based regimens for MDR TB treatment significantly reduced hospitalization costs and eliminated the costs to manage injectable-related adverse events. Thus, these regimens were always cost saving despite a lower cost for the injectable drugs, and higher utilization rate of TB care.
Results of the Probabilistic Sensitivity Analysis (PSA)
The PSA performed in one model-based study showed that the likelihood of the new first-line SCR being cost saving or cost effective was higher in settings with a relatively strong health system (i.e., a system with high adherence to TB treatment guidelines, high TB care utilization rate, and high treatment delivery cost), compared with those with a weak health system (i.e., a system with low adherence to TB treatment guidelines, low TB care utilization rate, and low treatment delivery cost). The PSA result in Table 2 showed that the 95% uncertainty range was wider when the analysis was performed with the ‘perfect’ scenario than with the ‘current’ scenario.