2.2 Summary of evidence

This section provides the PICO questions posed, the data and studies considered to answer the questions, the methods used for analysis and data synthesis, a summary of evidence on desirable and undesirable effects and certainty of evidence, and a summary of other evidence considered during development of the recommendation. Additional detail on the evidence is available in the web annexes containing the GRADE evidence summary tables (Web Annex 3) and GRADE evidence-to-decision tables (Web Annex 4).

PICO questions

The following PICO question was used for the evidence assessment in 2019 that led to the conditional recommendation for use of the all-oral bedaquiline-containing 9-month regimen.

  • PICO question 2–2019 (MDR/RR-TB, 2019): In MDR/RR-TB patients, does an all-oral treatment regimen lasting 9–12 months and including bedaquiline safely improve outcomes when compared with other regimens conforming to WHO guidelines?

The following PICO question (split into two sub-PICO questions because of different comparators) guided the analyses and the assessment, and eventually led to a summary recommendation:

  • PICO question 1–2022 (MDR/RR-TB, 2022): Should a shorter all-oral regimen (less than 12 months) containing at least three Group A medicines24 be used in patients with MDR/RR-TB and with fluoroquinolone resistance excluded?

Data and studies considered

In 2019, for the WHO guideline update, the South African Department of Health provided WHO with access to programmatic data on injectable-free regimens that had been used in South Africa since 2017, when most eligible patients were enrolled on a shorter regimen, with bedaquiline replacing the injectable (42). In August 2019, WHO issued a public call for IPD on the use of all-oral shorter regimens of 9–12 months (43), but this call yielded no additional evidence on the implementation of such regimens. Consequently, the evidence review on injectable-free regimens in 2019 was based primarily on programmatic data from South Africa, recorded in the Electronic Drug-Resistant Tuberculosis Register (EDRWeb). Secondary comparative analyses were carried out using the IPD, to balance the assumptions and adequacy of the data, and adding to the generalizability of findings – in particular, the applicability to a global population. The IPD used at that time was a global dataset of the records of individual patients who have been treated for MDR/RR-TB; as of November 2019, it contained 13 273 records from 55 studies or centres in 38 countries. The evidence reviews focused on the performance of a standardized shorter regimen in which the injectable agent was replaced by bedaquiline, in combination with levofloxacin (or moxifloxacin), clofazimine, and high-dose isoniazid, ethambutol, pyrazinamide and ethionamide (or prothionamide). Patients on this regimen did not receive any injectable agents, nor were they administered cycloserine, terizidone, p-aminosalicylic acid, delamanid or linezolid. According to the clinical guidance issued by the South African Department of Health, at the time of regimen roll-out patients were not enrolled on the all-oral shorter regimen if they had extensive disease, severe extrapulmonary TB, fluoroquinolone resistance, previous exposure to second-line treatment for more than 1 month or genotypic DST showing mutations in both inhA and katG genes.

In June 2021, WHO issued a public call (44) for IPD on the treatment of DR-TB. The call for individual patients’ data on bacteriologically confirmed MDR/RR-TB patients (including MDR/RR-TB, MDR/RR-TB with additional resistance to second-line TB drugs, and patients with pre-XDR-TB or XDR-TB) included the following specifics:

  • use of the modified shorter (<12 months) all-oral regimens using at least bedaquiline and linezolid;
  • use of the WHO-recommended shorter all-oral bedaquiline-containing regimen (9–11 months) in the following combination: 4 or 6 months of bedaquiline (used for at least 6 months), levofloxacin (or moxifloxacin), clofazimine, pyrazinamide, ethionamide, ethambutol and high-dose isoniazid, followed by 5 months of levofloxacin (or moxifloxacin), clofazimine, pyrazinamide and ethambutol; and
  • use of the WHO-recommended longer all-oral treatment regimen containing at least bedaquiline and linezolid.

The South African Department of Health provided WHO with the programmatic data from 2018 to 2019 on the use of a modified 9-month regimen in which ethionamide was replaced by linezolid. Several country programmes that provided WHO with IPD on the use of longer regimens according to WHO recommendations are listed in the Introduction (See Scope of the 2022 update and available evidence).

Once again, in 2021, the evidence review was based on programmatic data from South Africa on treatment outcomes of patients treated with the 9-month regimen (with either ethionamide or linezolid), recorded in the EDRWeb. Both datasets from South Africa (2017 and 2018–2019) with the 9-month regimens systematically excluded patients with extensive TB disease (extensive bilateral pulmonary cavitations), severe forms of extrapulmonary TB (meningitis, osteoarticular TB, pericardial effusion and abdominal TB), fluoroquinolone resistance, previous exposure to second-line treatment for more than 1 month or with genotypic DST showing mutations in both inhA and katG genes. In addition, comparative analyses were carried out using the 2021 IPD, which was compiled for the review and analyses in preparation for the GDG 2022; this IPD was of individual patients who had been treated for MDR/RR-TB. The evidence review focused on the performance of a standardized shorter regimen in which the injectable agent was replaced by bedaquiline (used for 6 months), in combination with levofloxacin/moxifloxacin, ethionamide, ethambutol, isoniazid (high-dose), pyrazinamide and clofazimine for 4 months (with the possibility of extending to 6 months if the patient remained sputum smear positive at the end of 4 months), followed by 5 months of treatment with levofloxacin/moxifloxacin, clofazimine, ethambutol and pyrazinamide. The comparators used included a nearly identical regimen where ethionamide was replaced by 2 months of linezolid (600 mg once daily) and the set of longer regimens designed based on the 2020 WHO recommendations.

Methods used for analysis and data synthesis

For comparisons between dataset or cohorts, outcomes were presented as unadjusted RRs and aRRs; the latter were calculated using a log-binomial generalized linear regression (binomial error distribution with log link function). Confounders were adjusted for using inverse probability propensity score weighting. No convergence issues with the log-binomial model arose. When outcome rates were close to the boundary (<5 positive or negative cases) aRRs were not calculated and unadjusted RRs alone were presented. For outcomes where the number of outcome events was zero, an unadjusted RD was calculated. For unadjusted RDs or RRs, the score method was used for calculating CIs. These approaches applied where one arm of a randomized trial was being compared with an external population, and in randomized trials in which subgroup analyses were performed (including by fluoroquinolone resistance status). Covariate selection for calculation of propensity scores was based on data availability and clinical knowledge. The covariates considered for inclusion in the propensity scores analysis included age, gender, baseline smear result, HIV status (including antiretroviral treatment status), prior treatment history (including whether previous infection was drug resistant), body mass index, smoking status, diabetes diagnosis, cavitation at baseline, disease site and presence of bilateral disease. For the calculation of aRRs, multiple imputation by chain equations using the “within” propensity score approach was used to account for missing data in potential confounders when the proportion of missing values for a confounder was less than 45%.

Summary of evidence on desirable and undesirable effects and certainty of evidence

PICO 1–2019

The primary analysis performed in 2019 using programmatic data from South Africa indicated that the use of a shorter all-oral bedaquiline-containing regimen in patients with MDR/RR-TB was associated with:

  • higher treatment success rates (73% all-oral versus 60% standardized shorter regimen success rates, adjusted odds ratio [aOR] for success versus failure or recurrence: 2.1, 95% CI: 1.1–4.0; aOR success versus death: 1.6, 95% CI: 1.2–2.1; aOR success versus failure, recurrence or death: 1.7, 95% CI: 1.3–2.2; and aOR success versus all unfavourable outcomes: 1.9, 95% CI: 1.6–2.4); and
  • lower loss to follow-up than a standardized shorter regimen in which an injectable agent was used (aOR loss to follow-up versus all other outcomes: 0.5, 95% CI: 0.4–0.7).

A similar effect for subgroups of patients with acid-fast bacilli (AFB) smear-positive sputum and People with HIV and HIV-negative patients was observed with the use of the shorter all-oral bedaquiline-containing regimen.

The analysis also indicated that when the shorter all-oral bedaquiline-containing regimen was compared with an injectable-free longer regimen containing bedaquiline, there seemed to be no marked differences in the outcomes observed. However, relatively modest beneficial effects were noted in the direction of the intervention; in particular, success versus failure or recurrence (aOR: 3.9, 95% CI: 1.7–9.1), success versus all unfavourable outcomes (aOR: 1.6, 95% CI: 1.2–2.2) and loss to follow-up (aOR: 0.5, 95% CI: 0.4–0.8), all favouring the use of the all-oral shorter regimen. Further subgroup analysis suggested consistent differences in treatment outcomes, as observed in primary analyses among subgroups, in particular among AFB smear-positive patients and in People with HIV on ART; however, differences in treatment outcomes in all-oral shorter and longer regimens were no longer significant when looking at outcomes for HIV-negative individuals, with the exception of loss to follow-up, which favoured the intervention. The additional comparison also illustrated the effect of a shorter all-oral bedaquiline-containing regimen in comparison with longer regimens without any new drugs. The all-oral shorter regimen performed significantly better across all outcomes and all subgroups in this comparison.

PICO 1–2022

For the assessment performed in preparation for the 2022 GDG, 8 653 records of patients with MDR/RR-TB initiating TB treatment at any time between January and December 2017 were considered, of which the following were included for analyses: 4 244 patients treated with a shorter regimen that included linezolid (used in South Africa in 2019) (intervention), 880 patients who received a shorter all-oral bedaquiline-containing 9-month regimen with ethionamide (used in South Africa in 2017) (comparator), and 850 patients treated with longer regimens that included at least bedaquiline and linezolid.

Sub-PICO 1.1

In sub-PICO 1.1, two observational studies were compared – the 9-month regimen with linezolid (used in South Africa in 2019) (intervention) and the 9-month regimen with ethionamide (used in South Africa in 2017) (comparator). Both datasets were obtained from a programmatic setting in South Africa.

Participants with MDR/RR-TB with fluoroquinolone susceptibility receiving the 9-month regimen with linezolid (n=4 244) compared with participants receiving the 9-month regimen with ethionamide (n=880) experienced:

  • lower levels of treatment success (64% vs 66%); that is, a 4% relative reduction (aRR=0.96, 95% CI: 0.91 to 1.01);
  • lower levels of failure and recurrence (1.1% vs 1.4%); that is, a 20% relative reduction (aRR=0.80, 95% CI: 0.42 to 1.53);
  • higher levels of deaths (20% vs 21%); that is, a 3% relative increase (aRR=1.03, 95% CI: 0.89 to 1.20);
  • higher levels of loss to follow-up (15% vs 12%); that is, a 19% relative increase (aRR=1.19, 95% CI: 0.98 to 1.45); and
  • higher levels of amplification of drug resistance (0.6% vs 0%); that is, a 1% absolute increase (RD=0.01, 95% CI: 0.00 to 0.01).

Adverse events were noted in 5% of participants receiving the 9-month regimen with linezolid but no comparisons could be made because no data were available for participants receiving the 9-month regimen with ethionamide.

The GDG judged the benefits of the 9-month regimen with linezolid to be small and the undesirable effects to be moderate compared with the 9-month regimen with ethionamide. The certainty of evidence was judged to be very low. Based on this, the GDG judged that the balance of health effects does not favour either the 9-month regimen with linezolid or the 9-month regimen with ethionamide.

Conclusion

The use of either the 9-month regimen with linezolid or the 9-month regimen with ethionamide is suggested in people with pulmonary MDR/RR-TB without fluoroquinolone resistance (conditional recommendation, very low certainty of evidence).

Sub-PICO 1.2

In sub-PICO 1.2, two observational datasets were compared – the 9-month regimen with linezolid (used in South Africa in 2019) (intervention) and the all-oral longer regimens containing bedaquiline from the 2021 IPD dataset.

Participants with MDR/RR-TB with fluoroquinolone susceptibility receiving the 9-month regimen with linezolid (n=4 244) compared with participants receiving longer regimens for MDR/RR-TB (n=850) experienced:

  • lower levels of treatment success (64% vs 74%); that is, a 10% relative reduction (aRR=0.90, 95% CI: 0.83 to 0.98);
  • lower levels of failure and recurrence (1.1% vs 3.4%); that is, a 71% relative reduction (aRR=0.29, 95% CI: 0.14 to 0.58);
  • higher levels of deaths (20% vs 11%); that is, a 38% relative increase (aRR=1.38, 95% CI: 1.00 to 1.91);
  • higher levels of loss to follow-up (15% vs 12%); that is, a 33% relative increase (aRR=1.33, 95% CI: 0.97 to 1.81);
  • similar levels of adverse events (5.0% vs 4.7%), (aRR=1.00, 95% CI: 0.59 to 1.69); and
  • lower levels of amplification of drug resistance (0.6% vs 1.4%); that is, a 73% relative reduction (aRR=0.27, 95% CI: 0.12 to 0.61).

The GDG judged both the benefits of the 9-month regimen with linezolid and the undesirable effects to be moderate compared with the longer regimens. The certainty of evidence was judged to be very low. Based on this, the GDG judged that the balance of health effects did not favour either the 9-month regimen with linezolid or the longer regimens. The panel judged that although the balance of effects did not favour either the intervention or the comparator, several other criteria in the GRADE evidence-to-decision tables (e.g. resources, acceptability, equity and feasibility) favoured the 9-month regimen.

Conclusion

The use of either the 9-month regimen with linezolid or the longer (18-month) regimens is suggested in people with pulmonary MDR/RR-TB without fluoroquinolone resistance (conditional recommendation, very low certainty of evidence).

Summary of other evidence

During assessment of sub-PICO 1.1, the panel noted that the cost of component medicines is likely to be similar because both regimens are of the same duration and use the same component medicines except for one – linezolid instead of ethionamide. The duration of linezolid use is 2 months compared with 4 months for ethionamide. Based on GDF prices (45) the cost difference was negligible (2 months of linezolid at 600 mg/day US$ 21, and 4 months of ethionamide at 450 mg/day US$ 32).

The health care costs are also likely to be similar because the two regimens are of the same duration and have the same component medicines, except for one – linezolid instead of ethionamide.

The panel also assumed no difference in DST needs. Both regimens are indicated for patients with MDR/RR-TB and without fluoroquinolone resistance. These patients are usually tested for rifampicin and fluoroquinolone resistance – rapid DSTs for both of these medicines are available. It might also be useful to perform genotypic DST because mutations in the inhA gene also confer resistance to ethionamide.

24 The three medicines included in Group A used for classification of second-line medicines are bedaquiline, fluoroquinolones and linezolid.

Book navigation