Potential selection bias in (2), as only 69% of participants were household contacts.

Potential misclassification: Eight studies (3–5,7,10,11,13,14) did not indicate whether household contacts with active TB were excluded from the analysis or did not provide sufficient data to calculate the number of household contacts with active TB per age stratum.

High heterogeneity among studies (I² = 94%) probably due to difference in background TB incidence. Risk ratios of two studies (1,5) showed opposite effect.

Small sample size in (5) (n < 50).

Potential misclassification: Seven studies (3,5,6,10,11,13,14) did not indicate whether household contacts with active TB were excluded from the analysis or did not provide sufficient data to calculate the number of household contacts with active TB per age stratum.

High heterogeneity among studies (I² = 97%) probably due to the differences in background TB incidence. Risk ratio of one study (5) showed opposite effect.

Wide confidence interval of pooled risk ratio. Small sample sizes in (5) (n < 50) and (12) (n < 100).

Potential selection bias in (15), as only 89% of participants were household contacts.

High heterogeneity among studies (I² = 93%) probably due to differences in background TB incidence. Risk ratios in three studies showed opposite effects (5,19,21).

Small sample size in (5) and (18) (n < 50).

Potential misclassification: Ten studies (3–5,10,13,14,20,21,23,26) did not indicate whether household contacts with active TB were excluded from the analysis or did not provide sufficient data to calculate the number of household contacts with active TB per age stratum.

High heterogeneity among studies (I² = 98%) probably due to differences in background TB incidence.

Small sample sizes in (5) and (26) (n < 100).

Because of the small number of studies in the other categories, only data from studies with a follow-up of 1–2 years in high TB incidence countries are presented in the table.

Serious inconsistencies due to heterogeneity (I² = 71%): One study showed an increased risk in the age group 5–15 years. This was not observed in the other studies.

Small number of events.

High heterogeneity among studies probably due to differences in background TB incidence and methods used to diagnose active TB (I² = 89.3%).

Owing to the small number of studies in the other categories, only data from studies with a follow-up of 1–2 years in high TB incidence countries are presented in the table.

One outlier (28) was excluded because of uncertainty about the cases included (co-prevalent vs incident cases).

High heterogeneity among studies (I² = 87.6%), probably due to the difference in background TB incidence.

LTBI does not apply to the general population.

Ascertainment bias highly likely, as TB cases in the general population detected passively, while TB cases in contacts detected actively. As a result, the relative and absolute risks might be overestimated. The composition of the general and the study population differed (general population of all ages versus a specific age group).

High heterogeneity among studies (I² = 83.9%), probably due to differences in background TB incidence.

Serious imprecision with a wide confidence interval for the effect estimates, probably due to small study size and number of outcome events.

I² = 72.5%, indicating moderate heterogeneity, probably due to differences in background TB prevalence; however, there is a trend across age groups and studies.

Few events and wide CI.

These comparisons included studies with a maximum follow-up of 24 months; therefore, TB incidence in the general population was multiplied by a factor of 2 to estimate the number of cases occurring during 24 months.

LTBI does not apply to the general population.

Ascertainment bias highly likely: TB cases in the general population detected passively, while TB cases in the contacts detected actively. As a result, relative and absolute risks might be overestimated. The composition of the general and study populations differs (general population of all ages versus a specific age group). TB incidence in the population was estimated by multiplying the yearly notification rate by a factor of 2.

High heterogeneity between studies probably due to difference in background TB incidence (I² = 84.4%).

Few events and wide CI.

I² = 88.1%, indicating high heterogeneity probably due to difference in background TB prevalence; however, there is a trend across age groups and studies.

I² = 16%.

Ascertainment bias highly likely, as TB cases in the general population detected passively, while TB cases in the contacts detected actively. As a result, the relative and absolute risk might be overestimated. The composition of the general and study populations differs (general population of all ages versus a specific age group).

I² = 0%.

Few events and wide CI.

These comparisons are based on studies with a maximum follow-up of 24 months; therefore, TB incidence in the general population was multiplied by a factor of 2 to estimate the number of cases occurring during 24 months.

Ascertainment bias highly likely, as TB cases in the general population detected passively, while TB cases in the contacts detected actively. As a result, the relative and absolute risks might be overestimated. The composition of the general and study populations differs (general population of all ages versus a specific age group). TB incidence in the population was estimated by multiplying the yearly notification rate by a factor of 2.

Moderate heterogeneity among studies (I² = 67.1%) probably due to differences in background TB incidence.

Few events and wide CI.

High heterogeneity among studies (I² = 87.5%) probably due to differences in background TB incidence.

Moderate heterogeneity among studies (I² = 72.5%) probably due to differences in background TB incidence.

From references (29,30)

Imprecise estimate for sensitivity. Downgraded by one.

The possibility of publication bias is not excluded, but it was not considered of sufficient concern to downgrade.

From references (31–37)

Limitations in study design (see QUADAS-2): High risk of selection bias in one study (31). In all studies, less than half of participants received the reference standard; accuracy was calculated under the assumption that those who did not receive the reference standard were culture and/or smear negative (no active TB).

Indirectness (see QUADAS-2): Some concern about applicability of reference standard in 2 studies - no downgrading.

Inconsistency: Little heterogeneity for sensitivity and specificity (based on visual inspection of CIs).

Imprecision: Precise estimates for sensitivity and specificity.

Publication bias: Not applicable (the evidence base for publication bias in studies of diagnostic test accuracy is very limited).

From references (31–34,36,38–43)

Limitations in study design (see QUADAS-2): high risk of selection bias in 1 study (den Boon, 2006) and in two studies unclear risk of bias for the reference standard. In 9 of the 11 studies less than half the participants received the reference standard; accuracy was calculated under the assumption that those who did not receive the reference standard were culture and/or smear negative (no active TB).

Indirectness (see QUADAS-2): No major concern about applicability.

Inconsistency: Moderate heterogeneity for sensitivity and significant heterogeneity for specificity (based on visual inspection of CIs) - downgrading on specificity.

Imprecision: Precise estimates for sensitivity and imprecise estimate for specificity.

Publication bias: Not applicable (the evidence base for assessing publication bias in studies of diagnostic test accuracy is very limited.

Notes on GRADE summary table

Overall quality:

All studies start with one point docked off because none were RCTs. The lowest quality score achievable is 1 out of 4; no minus scores are given.

Quality assessment: Based on the relative effect measure (RR or IRR) for both TST and IGRA. Studies not marked down if estimates for both tests score high on a specific GRADE quality item.

Other study quality considerations: Newcastle-Ottawa Scale quality items were considered when assessing the risk of bias. One point will be docked if at least one concern is present.

A1: Risk of bias is possible. Issues in the studies include selection bias, risk of incorporation bias, ascertainment and publication bias. Methods for ascertaining TB included microbiological methods, but not all incident TB cases had a definite culture-confirmed diagnosis of TB. Publication bias not formally assessed, but expected to be likely. Several large prospective studies are ongoing and/or unpublished; their results were not included in this analysis. However, addition of results is not expected to change the overall conclusions of this review.

A2: Serious unexplained inconsistency of RR estimate for TST. Points docked if serious inconsistency identified in either estimate.

A3: Although the number of studies included is small, they involve a range of populations, including adults and children, immunocompromised people and TB contacts, providing direct evidence for these groups.

A4: Serious imprecision of RR estimate for TST. Lower limit of 95% CI indicates lack of predictive utility. Points docked if serious imprecision identified in either estimate.

B1: Risk of bias is possible. Issues include selection bias, risk of incorporation bias, ascertainment and publication bias. Incorporation bias could not be ruled out in the cohort that included antepartum and postpartum women because information was not available; moreover, there are concerns with selection. The ART cohort study reported reference standards that do not account for index tests; however, assessors were not blinded to baseline TST results that were recorded in patient records. Methods for ascertaining TB included microbiological methods, but not all incident TB cases had a definite diagnosis of TB. Publication bias not formally assessed, but expected to be likely. Several large prospective studies are ongoing and/or are unpublished; their results were not included in this analysis. However, addition of results is not expected to change the overall conclusions of this review.

B2: Serious unexplained inconsistency in RR estimates for both TST and IGRA.

B3: This pooled estimate is based on only two studies: one study of HIV-infected people on ART with a median CD4+ approximately 250, and one on HIV-infected antepartum and postpartum women. No direct evidence for treatment-naïve patients and/or HIV-infected patients with high CD4 counts or other sub-populations of HIV-infected individuals (e.g. children).

B4: Very serious imprecision of RR estimates for both TST and IGRA. CIs are wide and indicate both significant predictive performance and lack of predictive utility. Studies had few events.

C1: Risk of bias is possible. Issues include selection bias, risk of incorporation bias (no information) and publication bias. Publication bias not formally assessed, but expected to be likely. Several large prospective studies are ongoing and/or are unpublished; their results were not included in this analysis. However, addition of results is not expected to change the overall conclusions of this review.

C2: Inconsistency not assessed.

C3: This single study comprises household case contacts in a high-incidence country. No direct evidence for other subpopulations of case contacts.

C4: Serious imprecision of TST effect estimates. Lower limit of 95% CI indicates lack of predictive utility.

D1: Risk of bias is possible. Issues include selection bias, lack of use of microbiological tools in methods to ascertain TB, incorporation bias and publication bias. Publication bias not formally assessed, but expected to be likely. Several large prospective studies are ongoing and/or are unpublished; their results were not included in this analysis. However, addition of results is not expected to change the overall conclusions of this review.

D2: Inconsistency not assessed.

D3: This single study comprises health care workers at a primary health care clinic. No direct evidence for other subpopulations of health care workers or all settings of health care.

D4: Very serious imprecision of IGRA and TST effect estimates; CIs are wide and indicate both significant predictive performance and lack of predictive utility.

E1: Risk of bias is possible. Issues include selection bias, incorporation of index tests in methods to ascertain incident TB and publication bias. Publication bias not formally assessed, but expected to be likely. Several large prospective studies are ongoing and/or are unpublished; their results were not included in this analysis. However, addition of results is not expected to change the overall conclusions of this review.

E2: Inconsistency not assessed.

E3: This single study comprises adolescents in a high-incidence setting. No direct evidence for other subpopulations of children or adolescents.

E4: No serious imprecision: Few events with large sample size.

From references (49–51)

Although there was a risk of selection bias, the characteristics of the two groups were similar. Patients with poor compliance were not included in the analysis of treatment outcomes. Downgraded by one level.

There was no clinical disease. The outcome reported was new radiographic findings suggesting possible active disease. No data compared with 6H. Downgraded by one level.

A high risk of detection bias due to lack of blinding. The RH group included participants enrolled during the second period, whose characteristics were different; they were not randomized between the RH group and the 9H group. Downgraded by two levels.

No data compared with 6H. Downgraded by one level.

Risk of bias due to poor comparability of the two groups. Downgraded by one level.

Low event rate and wide 95% CI. Downgraded by one level.

Lack of blinding. Medication adherence test was performed at home by parents. Although there was a risk of selection bias, the characteristics of the two groups were similar. Downgraded by one level.

Wide 95% CI. Downgraded by one level.

# The study reported adherence rates; compliance was considered to be poor if no medication was detected in urine strips or if patients did not return for follow-up visits or were lost to follow-up. Poor compliance was considered non-completion in the analysis.

CI: Confidence interval; RR: Risk ratio

Explanations

The GDG decided that for efficacy outcomes the pooled outcomes for phase 2 and phase 3 studies be considered one trial as the same protocol was used for both phases conducted by the same investigating team, even if the number of sites increased in the phase 3 study. Although the quality was not downgraded for this, the GDG noted that Inconsistency could not be judged given that there was only a single trial. Ideally replication by other trials would be desirable. For adverse events the studies can be considered as two separate trials.

Phase 2 (54) and Phase 3 (52) open-label trials conducted in nine countries, assigning adults with latent tuberculosis infection to receive treatment with a 4-month regimen of daily rifampicin or a 9-month regimen of daily isoniazid. The primary outcome in the phase 2 trial was incidence of grade 3 to 5 adverse events (superiority design), with secondary outcomes of treatment completion and incidence of active tuberculosis within 28 months of randomization. The primary outcome of the phase 3 trial was microbiologically confirmed active tuberculosis within 28 months after randomization (non-inferiority design), with secondary outcomes of clinically diagnosed active tuberculosis, grade 3 to 5 adverse events, and treatment completion. Outcomes of active tuberculosis and adverse events were adjudicated by three-member, blinded, independent review panels; treatment completion based on pill counts at routine follow-up visits.

Between the phase 2 and phase 3 trials in adults, there were no significant changes in guidelines or risk profiling of latent TB reactivation in terms of judging ‘increased risk for reactivation’. Randomization in both trials was stratified by site and centrally computer-randomized. Patients were randomized 1:1 in blocks of varying length (2 to 8) to isoniazid or rifampicin.

Open label design but endpoints of active TB and adverse events adjudicated by three-member, independent, blinded review panels. There were 18 per protocol exclusions among those randomized to isoniazid and 19 per protocol exclusions among those randomized to rifampicin. These per protocol exclusions were due to being a household contact of a tuberculosis patient with resistance to isoniazid or rifampicin (proven post-randomization). There were nine individuals randomized to isoniazid and five individuals randomized to rifampicin who withdrew consent post-randomization. The GDG decided to downgrade by one level because of the open label design possibly leading to performance bias.

Among those randomized to isoniazid and forming the modified intention-to-treat population, there were 260 individuals lost to follow-up. Among those randomized to rifampicin and forming the modified intention-to-treat population, there were 245 individuals lost to follow-up. Among all persons forming the modified intention-to-treat population, 7.4% of individuals were lost to follow-up.

The quality was not downgraded for Indirectness, but the GDG noted that the trial compared 4R with 9H and therefore did not cover all other comparisons of the PICO, especially 6H, the most widespread standard of care in TB preventive treatment. Some study sites were low TB incidence settings for which a WHO recommendation for use of 4R already exists.

All active TB events occurred within the phase 3 trial (52).

Unadjusted estimate.

The rate difference was estimated by a Poisson model with the use of generalized estimating equations with a log link and the inclusion of the log of person-time as an offset. An exchangeable correlation structure with robust standard errors was used to account for the correlation of participants coming from the same household.

Values reported as per Table 3 of (52). Values include Phase 2 results (54) as well.

Denominators are representative of the combined safety population of phase 2 (54) and phase 3 (52) as indicated in supplemental tables S2 and S3 of the phase 3 publication. From the phase 2 trial, 396 patients receiving isoniazid and 393 patients receiving rifampicin formed the safety population; from the phase 3 trial, 2809 patients receiving isoniazid and 2887 patients receiving rifampicin formed the safety population.

All mortality events occurred in the phase 3 trial (52).

A zero cell correction of 0.5 has been used to calculate the risk ratio.

The risk difference was estimated by a binomial distribution model with an identity link and generalized estimating equations. An exchangeable correlation structure and robust standard errors were used to account for correlation of patients coming from the same family. If no events occurred in one or both arms, confidence intervals were calculated based on (56).

Among adverse events from the phase 2 trial (54), 10 patients receiving rifampicin experienced grade 3–5 adverse events which led to permanent discontinuation of the medication, of which 7 were deemed possibly/probably related to study drug; 19 patients receiving isoniazid experienced grade 3–5 adverse events which led to permanent discontinuation of the medication, of which 16 were deemed possibly/probably related to study drug. Among adverse events from the phase 3 trial (52), 43 patients receiving rifampicin experienced grade 3–5 adverse events which led to permanent discontinuation of the medication, of which 24 were deemed possibly/probably related to study drug; 100 patients receiving isoniazid experienced grade 3–5 adverse events which led to permanent discontinuation of the medication, of which 59 were deemed possibly/probably related to study drug.

Also included is the phase 1 trial (55), a single center, open-label randomized trial assessing superiority of four months of daily rifampicin to nine-months of daily isoniazid for treatment completion.

Open label trial, unblinded assessment of compliance judged on the basis of pill counts at monthly follow-up visits.

Numerator and denominator values are derived from the Phase 1 trial (55), Phase 2 trial (54), and Phase 3 trial (52). Treatment completion was defined as taking at least 80% of prescribed doses (i.e., at least 96 pills of rifampicin or 216 pills of isoniazid). In the phase 1 trial, 44 of 58 individuals randomized to isoniazid and 53 of 58 individuals randomized to rifampicin completed treatment. In the phase 2 trial, 254 of 427 individuals randomized to isoniazid and 328 of 420 individuals randomized to rifampicin completed treatment. In the phase 3 trial, 1890 of 2989 individuals randomized to isoniazid and 2382 of 3023 individuals randomized to rifampicin completed treatment.

Open-label, non-inferiority trial conducted in seven countries, assigning children with latent tuberculosis infection to receive treatment with a 4-month regimen of rifampicin or a 9-month regimen of isoniazid for the incidence of grade 3 to 5 adverse events during treatment. Secondary outcomes were the incidence of microbiologically confirmed active tuberculosis within 16 months after randomization and completion of the treatment regimen. Outcomes of active TB and adverse events were adjudicated by two- or three-member, blinded, independent review panels; treatment completion based on pill counts at routine follow-up visits (53).

Randomization in the paediatric trial was stratified by country and centrally computer-randomized. Patients were randomized 1:1 in blocks of varying length (2 to 8) to isoniazid or rifampicin. Enrollment and randomization in this trial was completely separate from the adult trials.

Among those randomized to isoniazid and forming the modified intention-to-treat population, there were 6 individuals lost to follow-up. Among those randomized to rifampicin and forming the modified intention-to-treat population, there were 5 individuals lost to follow-up. Among all children forming the modified intention-to-treat population, 1.3% of individuals were lost to follow-up.

Open label design but endpoints of active TB and adverse events adjudicated by two-member and three-member, respectively, independent, blinded review panels. There were 9 per protocol exclusions among those randomized to isoniazid and 6 per protocol exclusions among those randomized to rifampicin. These per protocol exclusions were due to being tuberculin skin test negative at the end of the window period (two months after exposure). GDG decided to downgrade by one level because of the open label design and because some sites were not high burden.

A zero cell correction of 0.5 has been used to calculate the rate ratio.

Values as reported in the text of the paediatric trial (53).

Values as reported in Table 3 of the paediatric trial (53).

Values reported in Table 2 of the paediatric trial (53).

Subgroup analysis within randomized trials that involved relatively small numbers of HIV-infected patients when compared to all patients included in the trials.

Denominators include HIV-positive patients known at the time of randomization as reported in Supplemental Table S1 of the phase 3 adult trial (52), as well as patients diagnosed post randomization as a result of baseline assessment. This includes 130 patients and 8 patients receiving isoniazid with an HIV-diagnosis at time of randomization and post-randomization, respectively, and 125 patients and 7 patients receiving rifampicin with an HIV-diagnosis at time of randomization and post-randomization, respectively. This resulted in modified intention to treat population sizes of 132 for rifampicin and 138 for isoniazid. Among HIV-positive patients randomized to rifampicin, 2 did not receive a dose of therapy. Thus, the safety population sizes were 130 for rifampicin and 138 for isoniazid.

Unadjusted absolute estimate.

Among patients receiving rifampicin included in the safety population, 6 patients were HIV-positive in the phase 2 trial and 124 patients were HIV-positive in the phase 3 trial. All grade 3–5 adverse events among patients receiving rifampicin occurred in the phase 3 trial. Two patients experienced a grade 3–5 adverse event with rifampicin that resulted in permanent discontinuation of the study drug, only 1 was deemed possibly/probably related to the study drug. Among patients receiving isoniazid included in the safety population, 7 patients were HIV-positive in the phase 2 trial and 131 were HIV-positive in the phase 3 trial. One patient in the phase 2 trial and 7 patients in the phase 3 trial receiving isoniazid experienced a grade 3–5 adverse event resulting in permanent discontinuation of the study medication. The events were deemed possibly/probably related to the study drug for the one patient from the phase 2 trial and for 4 patients from the phase 3 trial.

CI: Confidence interval; HR: Hazard Ratio; RR: Risk ratio

Explanations

Randomized, open-label, phase 3 noninferiority trial comparing the efficacy and safety of a 1-month regimen of daily rifapentine plus isoniazid (1-month group) with 9 months of isoniazid alone (9-month group) in HIV-infected patients who were living in areas of high tuberculosis prevalence or who had evidence of latent tuberculosis infection. Primary end point was the first diagnosis of TB or death from TB or an unknown cause. Noninferiority would be shown if the upper limit of the 95% confidence interval for the between-group difference in the number of events per 100 person-years was less than 1.25. LTBI was not confirmed in about 80% of participants. Enrolment restricted to individuals ≥13 years old who were not pregnant or breastfeeding. Overall TB incidence observed in the trial was lower than expected. The number of patients with a CD4+ <250 cells per cu mm was small, and neither inferiority nor noninferiority of the 1-month regimen was shown in this stratum.

Unknown cause of death censored in this analysis, which may cause bias in incidence rate difference if some of these deaths were related to TB (dependent censoring)

The GDG decided to downgrade by one level because of the open label design possibly leading to performance bias. The quality was not downgraded for Indirectness, but the GDG noted that the trial compared 1HP with 9H and therefore did not cover all other comparisons of the PICO, especially 6H, the most widespread standard of care in TB preventive treatment. The GDG noted that Inconsistency could not be judged given that there was only a single trial; results from more trials would be desirable.

Trial conducted only in PLHIV and not in all people at risk of active TB.

Probable TB diagnoses and deaths with non-bacteriologically confirmed TB censored at the time of event

When cause of death was determined to be unknown or not related to TB by blinded external reviewers, these were treated as a competing risk rather than endpoint. Some of these may have actually been due to TB, which may bias estimate.

The proportion of events among controls

Per-protocol population consisted of all participants who completed treatment, or who had died or received a TB diagnosis while they were receiving treatment.

Deaths were reviewed by blinded external reviewers. Unknown causes of death were included as an endpoint, but misclassification of cause of death may bias estimate

There were 21 deaths in the one-month arm, 3 related to TB. There were 28 deaths in the nine-month arm, 3 related to TB.

Small number of events

Incidence Rate Difference per 100 person-years of 0.00 (-0.10 to 0.10)

Assessed via participant self-report at clinic visits

Resistance may be non-emergent and coming from infecting strain

Small sample of bacteriologically confirmed TB who had drug susceptibility test results

Although one of the trials was conducted in low TB incidence countries, this is unlikely to affect the relative effect of rifapentine + isoniazid compared with isoniazid monotherapy. Not downgraded.

95% CIs of both relative and absolute effect include appreciable benefit and harm with 3HP.

Both trials were open-label, which may have introduced bias in ascertainment of adverse events.

Although the trials were open-label, this is unlikely to affect detection of hepatotoxicity, which is usually done by objective measurement (i.e. blood tests.). Not downgraded.

Very low event rates. Upper limit of 95% CI of both relative and absolute effect include appreciable harm with 3HP. Downgraded by two levels.

No comparison with 6 months of isoniazid. The study included 2.7% HIV-positive participants. Although the trial was conducted in low TB incidence countries, this is unlikely to affect relative effect of rifapentine + isoniazid compared with isoniazid monotherapy. Downgraded by one level.

Although the 95% CI of RR is wide, the number of events was small and the CI of absolute effect is narrow. The result also met pre-stated non-inferiority margin. Not downgraded.

Although the 95% CI of RR is wide, the number of events was small and the CI of absolute effect is narrow. Not downgraded.

An open-label design of the trial may have introduced ascertainment bias.

Although the trial was open-label, this is unlikely to affect detection of hepatotoxicity, which is usually done by objective measurement (i.e. blood tests). Not downgraded.

No comparison against 6 months of isoniazid. Although the trial was conducted in low TB incidence countries, this is unlikely to affect relative effect of rifapentine + isoniazid compared with isoniazid monotherapy. Downgraded by one level.

Although the 95% CI of the RR is wide, the number of events was small and the CI of absolute effect is narrow. The result also met pre-stated non-inferiority margin. Not downgraded.

Although the 95% CI of the RR is wide, the number of events was small and the CI of absolute effect is narrow. Not downgraded.

An open-label design of the trial may have introduced ascertainment bias.

Although the trial was open-label, this is unlikely to affect detection of hepatotoxicity, which is usually done by objective measurement (i.e. blood tests). Not downgraded.

CI: Confidence interval; OR: Odds ratio; RR: Risk ratio

Optimal information size not met.

Bias due to confounding is considered serious. Important confounders are not fully accounted for.

Large CI including both appreciable benefits and harms and very few events

CI includes both appreciable benefits and harms

Confounding was not accounted for. Bias due to measurement of hepatotoxicity is considered serious since liver function tests were performed only if clinically indicated, which was likely to be influenced by knowledge of the receipt of IPT.

Very large sample size and CI of absolute effect is very narrow.

Optimal information size not met.

Large CI including both appreciable benefits and harms. Very few events.

CI includes both appreciable benefit and harm.

Risk of bias in selection of the control group, and none of the studies adjusted for confounders. Downgraded by two levels.

The study by Shaaf et al. was excluded, as the incidence of MDR-TB was not reported.

Small sample sizes and wide 95% CIs. Downgraded by two levels.

Reference (68)

Reference (66)

Reference (67)

Reference (69)