Characteristics of children and adolescents with rifampicin-resistant tuberculosis and predictors of treatment outcomes: a systematic review and individual participant data meta-analysis

Abstract

Background:

There are limited data on treatment of children and adolescents with multidrug- or rifampicin-resistant (MDR/RR) tuberculosis, especially with more recently available drugs and regimens. We aimed to describe the characteristics, outcomes and predictors of treatment outcomes in this vulnerable population.

Methods:

This systematic review and individual participant data meta-analysis (Prospero CRD42020187230) included published and unpublished studies of children and adolescents 0 to 19 years treated for confirmed or clinically-diagnosed MDR/RR tuberculosis. Multivariable logistic regression models were used to identify clinical and treatment predictors of treatment outcome.

Findings:

The search identified 23,369 individual participants from 42 studies, mostly from India and South Africa. Overall, 16,825 (72%) were successfully treated (treatment completed or cured), 2,848 (12%) died, 722 (3%) experienced treatment failure and 2,947 (13%) were lost-to-follow-up. In primary analyses, median (IQR) age was 16 (13–18) years, 2,448 (13%) were living with HIV, and 17,707 (89.6%) had a microbiologically confirmed diagnosis. After adjusting for clinical predictors of outcome, the use of two (aOR 1.41, [95 % CI, 1.09 to 1.82], p=0.008) or three (aOR 2.12, [95 % CI, 1.61 to 2.79], p<0.001) WHO-classified Group A drugs (bedaquiline, moxifloxacin, levofloxacin, linezolid) was positively associated with treatment success.

Interpretation:

Younger and clinically diagnosed children were underrepresented and should be a focus for case-finding efforts. Overall treatment outcomes were better than in adults but below international targets of ≥90% successfully treated. Treatment with more Group A drugs was associated with better treatment outcomes in children and adolescents, highlighting the need for more rapid access to these drugs and improved regimens.

Funding:

Unitaid

BACKGROUND

As many as 32,000 children (<15 years) are estimated to develop multidrug-resistant (MDR) or rifampicin-resistant (RR) tuberculosis globally each year.¹ The past decade has witnessed a remarkable transformation in the treatment of MDR/RR tuberculosis, but children and adolescents have not been able to benefit from these advances largely due to age-based exclusions from studies. While children and adolescents represent a heterogenous population when it comes to the spectrum of tuberculosis they present with and their psychosocial needs, there is an urgent need to understand how this group might benefit from the transformative therapeutic approaches that are now the norm in adult MDR/RR tuberculosis. Treatment approaches for children have traditionally been extrapolated from adult data; however, the spectrum of tuberculosis in children differs from that in adults. Paediatric pulmonary tuberculosis tends to be paucibacillary, especially in younger children (<5 years) where the highest burden of disease is concentrated, and less severe compared to adults. Evaluating treatment approaches specifically in children is therefore important. Adolescents (10–19 years of age) are more likely to develop adult-type pulmonary tuberculosis compared to younger children. Older adolescents (15 to 19 years of age) are a particularly high-risk group and may have worse treatment outcomes compared to adults or younger children.^2,3 Given the specific risks and needs of adolescents with tuberculosis, there is increasing recognition that they should be considered separately from adults, and should be included in the larger scope of paediatric tuberculosis research.⁴

In 2014, a systematic review and individual participant data (IPD) meta-analysis of children treated for MDR/RR tuberculosis included 28 studies with 1,009 participants.⁵ The pooled overall treatment success was 78%. Although this was based on all evidence available at the time, the number of included children was still small. Clinical and treatment factors associated with unsuccessful outcome were identified, including severe tuberculosis disease, undernutrition and HIV infection. There was limited use of levofloxacin, moxifloxacin, linezolid and no use of bedaquiline, the current World Health Organization (WHO) drug-resistant tuberculosis Group A drugs, while a substantial proportion of children still received injectable drugs (amikacin, kanamycin, capreomycin). This evidence contributed to the 2016 WHO guidelines which made specific recommendations for the management of MDR/RR tuberculosis in children.⁶

Since then, the treatment landscape for MDR/RR tuberculosis in adults has evolved dramatically. Major changes include recommendations for shortened treatment regimens, widespread use of moxifloxacin and levofloxacin, exclusion of injectable drugs, inclusion of the repurposed drugs clofazimine and linezolid, and prioritisation of the novel agents bedaquiline, delamanid and pretomanid.^7–9 There are limited data on the uptake of these treatment innovations among children and adolescents, and their impact on treatment outcomes, with no data yet available on pretomanid in children due to delayed paediatric evaluations. Robust data are a priority to inform evidence-based approaches to MDR/RR tuberculosis treatment in children and adolescents.

The objectives of this systematic review and IPD meta-analysis were to describe the demographic and clinical characteristics, treatment regimens and outcomes of children and adolescents treated for MDR/RR tuberculosis globally, and to characterise clinical and treatment predictors of outcome.

METHODS

The review is registered in Prospero [CRD42020187230].

Search strategy and selection criteria

Controlled and non-controlled retrospective and prospective studies were eligible, including case series, case-control studies, cohort studies, non-randomised experimental studies, and randomised controlled trials. Unpublished data, including programmatic data, were included as long as collection was approved by the ethics board of the originating institution and could be shared with permission. To be eligible, studies must have included more than five children or adolescents (0–19 years of age) treated for microbiologically confirmed or clinically-diagnosed MDR/RR tuberculosis within a defined treatment cohort, and reported on regimen composition and treatment outcomes.

A comprehensive electronic search was carried out without language or publication restrictions, based on the previously completed IPD meta-analysis in children which reviewed data through 1 October 2014.⁵ PubMed, Scopus, and The Cochrane Library were searched from 1 October 2014 through 30 March 2020 (see Supplemental Materials for detailed search strategy). Individual studies conducted outside of the search time frame were also included if they consisted of unpublished data. Conference abstracts and bibliographies of retrieved articles were searched, in addition to contacting paediatric tuberculosis experts (Supplemental Materials).

After de-duplication, abstracts were screened independently by two authors to identify potentially eligible records. Full texts of these records were reviewed by two independent authors to identify studies meeting the eligibility criteria (additional details in Supplemental Materials).

For studies and cohorts that met required inclusion criteria, IPD were requested from authors and persons responsible for individual cohorts; anonymised IPD was securely transferred and included in the final dataset if eligible. All studies from the previous IPD analysis were included with permission from the primary authors. Variables extracted from each study’s IPD were combined in a final dataset (Supplemental Material).

The risk of bias for included studies, using the case series checklist, was assessed independently by two reviewing authors using the Joanna Briggs critical appraisal tools.¹⁰

Inclusion of individual participant data

Individual level data were included into an overall cohort (Figure 1). Participants with missing data on second-line antituberculosis treatment regimens, isoniazid-monoresistant tuberculosis, age >19 years, or missing or implausible treatment duration (equal to 0 or greater than 100 months) were excluded. Individuals who were lost to follow-up (LTFU) were included in the overall cohort given its importance as a programmatic outcome, and to allow for comparison with previous reports.

Figure 1. PRISMA Flow diagram for individual participant data meta-analysis of children and adolescents treated for multidrug- or rifampicin-resistant tuberculosis.

Detailed descriptions of included studies are shown in Supplemental Table 1.

Following this process, we identified a primary analysis cohort that excluded those individuals with an outcome of LTFU (Figure 1). LTFU represents a heterogenous group of participants who have interrupted treatment, by definition have uncertain clinical outcomes, and have the potential to bias our understanding of treatment effects, which was a core objective of this review. This approach is consistent with previously reported IPD meta-analyses in both children and adults with MDR/RR tuberculosis that have informed WHO guidelines.^5,11

Statistical analysis

Baseline clinical and demographic characteristics, treatment characteristics (individual drugs administered during the treatment period and total treatment duration), and treatment outcomes were described for both the overall and the analysis cohorts.

The analysis cohort was used for subsequent analyses. The primary effectiveness outcome was treatment success (defined as cure or treatment completion) versus unfavourable outcome (defined as treatment failure or death), using WHO definitions. Treatment characteristics (including the use of individual drugs, number of drugs in each WHO drug group used at any point during the treatment regimen, and overall duration of treatment) were described to explore any trends in MDR/RR tuberculosis treatment overall, over time, and also focusing on major contributing countries.

We assessed baseline characteristics and number of drugs used in each WHO drug group as predictors of treatment success versus unfavourable treatment outcome using mixed-effects logistic regression models adjusted for study site variability (with a random intercept estimated by study site). First, we analysed univariable associations of baseline clinical predictors with treatment success. Thereafter, we built a multivariable logistic regression model including the combination of relevant clinical predictors of treatment success, after being carefully evaluated for inclusion, individually and in combination, in terms of clinical relevance, physiological plausibility and statistical significance. A p-value less than 0.05 was taken as the reference for the limit of statistical significance for the forward inclusion for multivariable model building. Finally, the number of drugs in each WHO group was included as predictor of treatment success, adjusting for the combination of clinical characteristics included in the clinical multivariable logistic regression model. During model building, we assessed key assumptions of the mixed-effects logistic regression model, including the absence of multicollinearity and the linearity of quantitative predictors. These models were built with the aim of characterizing the combination of clinical and treatment characteristics associated with tuberculosis treatment success. Models were developed for the entire analysis dataset, and for subsets of data stratified by age group, HIV status, and major countries contributing to the data, to explore patterns of association between treatment features and treatment outcome, across strata.

Model evaluation was performed using the area under the receiver operating characteristic curve (ROC-AUC), and marginal and conditional R² values were calculated. Given the extent of missing information, missing baseline clinical characteristics of interest (country income, age, year of treatment, sex, data on HIV status, tuberculosis disease spectrum, previous tuberculosis treatment, site of the disease and acid-fast bacilli [AFB]-smear status) were imputed using multivariate imputation via chained equations (MICE package, R software). MICE imputes missing values by leveraging observed values from other variables, preserving the dataset’s inherent relationships among variables. The imputation model included the treatment outcome variable together with the baseline clinical variables of interest (country income, age, year of treatment, sex, data on HIV status, tuberculosis disease spectrum, previous tuberculosis treatment, site of the disease and acid-fast bacilli [AFB]-smear status). Those characteristics missing in more than 70% of the population were omitted from the analysis. We performed 25 Gibb’s sampling iterations to generate each of the 20 imputed datasets; original data were preserved within each imputed dataset and imputed data were only used where data were missing. The proportion of included patients with missing data for the different baseline clinical characteristics are reported in Table 2. Seventy-one percent of participants had at least one missing predictor in the multivariable regression model used. Complete case analysis results are included in Supplementary material. The sixteen participants with missing age data were assumed to be under 19 years of age and were included in the analysis and the imputation process.

Table 2.

Demographic and clinical characteristics of children and adolescents (0 to 19 years of age) treated for rifampicin-resistant tuberculosis excluding those with an outcome of lost to follow-up in the analysis cohort

	Analysis cohort (N=20395)	Treatment outcome
		Treatment success (N=16825)	Death (N=2848)	Treatment failure (N=722)
Age (years)
Median (IQR)	16 (13,18)	16 (13,18)	17 (15,18)	17 (15,18)
N (%) missing	16 (<0.1%)	13 (<0.1%)	2 (<0.1%)	1 (0.1%)
Age group
<5 years	1526 (7.5%) [100%]	1439 (8.6%) [94%]	77 (2.7%) [5%]	10 (1.4%) [1%]
5– <10 years	1195 (5.9%) [100%]	1070 (6.4%) [89%]	103 (3.6%) [9%]	22 (3.1%) [2%]
10–<15 years	3779 (18.5%) [100%]	3181 (18.9%) [84%]	475 (16.7%) [13%]	123 (17.1%) [3%]
15–19 years	13879 (68.1%) [100%]	11122 (66.2%) [80%]	2191 (77.0%) [4%]	566 (78.5%) [16%]
N (%) missing	16 (<0.1%)	13 (<0.1%)	2 (<0.1%)	1 (0.1%)
Sex
Male	12001 (58.9%) [100%]	9766 (58.1%) [81%]	1814 (63.7%) [15%]	421 (58.3%) [4%]
Female	8384 (41.1%) [100%]	7051 (41.9%) [84%]	1032 (36.3%) [12%]	301 (41.7%) [4%]
N (%) missing	10 (<0.1%)	8 (<0.1%)	2 (<0.1%)	0
HIV status
HIV positive	2448 (13.8%) [100%]	1821 (12.3%) [74%]	526 (22.2%) [21%]	101 (15.8%) [4%]
HIV negative	15316 (86.2%) [100%]	12936 (87.7%) [84%]	1840 (77.8%) [12%]	540 (84.2%) [4%]
N (%) missing	2631 (13%)	2068 (12%)	482 (17%)	81 (11%)
Income country
Low or lower-middle	13772 (67.8%) [100%]	11230 (66.9%) [82%]	2104 (74.0%) [15%]	438 (60.7%) [3%]
Upper-middle or high	6571 (32.2%) [100%]	5547 (33.1%) [84%]	740 (26.0%) [11%]	284 (39.3%) [4%]
N (%) missing	52 (0.2%)	48 (0.3%)	4 (0.1%)	0
Failed First-Line Treatment
Yes	260 (4.2%) [100%]	196 (3.7%) [75%]	49 (6.9%) [19%]	15 (5.6%) [6%]
No	5958 (95.8%) [100%]	5041 (96.3%) [85%]	663 (93.1%) [11%]	254 (94.4%) [4%]
N (%) missing	14177 (70%)	11588 (69%)	2136 (75%)	453 (63%)
Any previous TB treatment
Yes	3707 (18.3%) [100%]	2823 (16.9%) [76%]	657 (23.3%) [18%]	227 (31.8%) [6%]
No	16505 (81.7%) [100%]	13854 (83.1%) [84%]	2164 (76.7%) [13%]	487 (68.2%) [3%]
N (%) missing	183 (0.9%)	148 (0.9%)	27 (0.9%)	8 (1%)
TB status
Microbiologically confirmed	17707 (89.6%) [100%]	14613 (89.9%) [83%]	2425 (86.8%) [14%]	669 (93.2%) [4%]
Clinically diagnosed	2057 (10.4%) [100%]	1640 (10.1%) [80%]	368 (13.2%) [18%]	49 (6.8%) [2%]
N (%) missing	631 (3%)	572 (3%)	55 (2%)	4 (0.6%)
Site of disease
Pulmonary	17398 (90.1%) [100%]	14216 (88.9%) [82%]	2528 (95.8%) [14%]	654 (95.9%) [4%]
Extrapulmonary	1710 (8.9%) [100%]	1589 (9.9%) [93%]	98 (3.7%) [6%]	23 (3.4%) [1%]
Both	141 (0.7%) [100%]	124 (0.8%) [88%]	13 (0.5%) [9%]	4 (0.6%) [3%]
N (%) missing	1077 (5%)	829 (5%)	208 (7%)	40 (6%)
AFB smear
Positive	3532 (52.6%) [100%]	2778 (48.8%) [79%]	481 (70.0%) [13%]	273 (83.2%) [8%]
Negative	3180 (47.4%) [100%]	2919 (51.2%) [92%]	206 (30.0%) [6%]	55 (16.8%) [2%]
N (%) missing	13683 (67%)	11128 (66%)	2161 (76%)	394 (55%)
Resistance to FQN/SLI
None	1436 (49.9%) [100%]	1241 (54.5%) [86%]	128 (32.2%) [9%]	67 (33.0%) [5%]
FQN	489 (17.0%) [100%]	359 (15.8%) [73%]	86 (21.6%) [18%]	44 (21.7%) [9%]
SLI	321 (11.2%) [100%]	235 (10.3%) [73%]	59 (14.8%) [18%]	27 (13.3%) [9%]
FQN and SLI	632 (22.0%) [100%]	442 (19.4%) [70%]	125 (31.4%) [20%]	65 (32.0%) [10%]
N (%) missing	17517 (86%)	14548 (86%)	2450 (86%)	519 (72%)
Treatment year
Median (IQR)	2018 (2016,2019)	2018 (2016,2018)	2018 (2017,2019)	2017 (2014,2018)
N (%) missing	18 (<0.1%)	14 (<0.1%)	4 (0.1%)	0
Treatment duration (months)
Median (IQR)	16 (9,23)	19 (9,24)	5 (2,10)	10 (7,20)

FQN, fluoroqouinolone; SD, standard deviation; IQR, interquantile range; SLI, second-line injectable; TB, tuberculosis

Income country calculated based on World Bank country classification.

Percentages in italics represent the percentage of missing data for each variable.

Percentages in bold and in brackets represent the percentage of patients with treatment success, death and treatment failure in each clinical characteristic category.

Intrathoracic lymph node disease was considered pulmonary TB.

In order to better understand the impact of excluding LTFU participants from the primary analyses, sensitivity analyses were conducted evaluating clinical and treatment predictors of outcome with participants with a LTFU outcome considered either to have a successful or an unfavourable tuberculosis treatment outcome.

All statistical analyses were performed with R software (version 4.1.0).

Ethics statement

The overall review was approved by the Health Research Ethics Committee of Stellenbosch University (X20/02/007), and ethics approval was obtained by the corresponding authors of the primary data in accordance with local policies.

Role of the funding source

The funder did not have input on the study design, collection, analysis or interpretation of data, writing of the report or the decision to submit the paper for publication.

FINDINGS

The results of the search and selection of studies and IPD are shown in Figure 1. Characteristics of included and excluded studies are shown in Supplemental Tables 1 and 2. All data were from observational studies or routine programmatic data. Most studies were judged as having moderate to low risk of bias; however, all data included were from case series or observational data and are inherently lower quality than data from other designs such as clinical trials (Supplemental materials Table 3 for risk of bias assessments).

After initial exclusions (Figure 1), 23,369 individual participants from 42 studies were included in the overall cohort. Baseline and treatment characteristics of this group are shown in Supplemental Table 4. Overall, 16,825 (72%) were successfully treated (completed treatment or cured), 2,848 died (12%), 722 (3%) had treatment failure and 2,974 (13%) were LTFU. Among participants <15 years of age (n=7115), 5,690 (80%) were successfully treated, 655 (9.2%) died, 155 (2.2%) had treatment failure and 615 (8.6%) were LTFU.

After excluding participants with a LTFU outcome (Figure 1), 20,395 individual participants were included in the primary analysis cohort. Included studies represented data from 19 countries with broad geographic and income level distribution; 19 of the 30 high-burden MDR/RR tuberculosis countries were represented (Table 1). Most participants were from India (n=12,898, 63%) and South Africa (n=5,954, 29%). There were 18,090 participants from programmatic data sources (n=5,320 South Africa, n=12,770 India).

Table 1.

Country of treatment by World Health Organization Region of children and adolescents with rifampicin-resistant tuberculosis in the analysis cohort (n=20,395)

Country	N (%)
African Region (AFR)
South Africa	5,954 (29.2%)
Country not specified*	23 (0.1%)
AFR Total	5,977
Region of the Americas (AMR)	AMR Total:
Brazil	125 (0.6%)
Peru	234 (1.1%)
United States	9 (<0.1%)
Country not specified*	31 (0.2%)
AMR Total	399
South-East Asia Region (SEAR)
India	12,898 (63.2 %)
Country not specified*	29 (0.1%)
SEAR Total	12927
European Region (EUR)
Austria	8 (<0.1%)
Belarus	5 (<0.1%)
Georgia	42 (0.2%)
Germany	10 (<0.1%)
Latvia	52 (0.3%)
Netherlands	3 (<0.1%)
Russia	53 (0.3%)
Spain	25 (0.1%)
Tajikistan	250 (1.2%)
Ukraine	6 (<0.1%)
Uzbekistan	327 (1.6%)
Country not specified*	43 (0.2%)
EUR Total	824
Eastern Mediterranean Region (EMR)
Pakistan	210 (1.0%)
Country not specified*	50 (0.2%)
EMR Total	260
Western Pacific Region (WPR)
South Korea	3 (<0.1%)
Taiwan	4 (<0.1%)
Country not specified*	1 (<0.1%)
WPR Total	8

^*Country not specified indicates that the dataset did not report country of treatment in order to maintain anonymity but did report WHO region of treatment.

Demographic and clinical characteristics of included participants are shown in Table 2. The median age was 16 years (interquartile range [IQR] 13, 18 years), with only 1,526 (7.5%) of participants <5 years of age, compared to 13,879 (68%) older adolescents (15–19 years). Overall, 17,707 (89.6%) had microbiologically confirmed tuberculosis. Among children under 5 years of age, 75% (n=920) had microbiologically confirmed tuberculosis. This proportion increased with age, with 86% (n=935) of children aged 5 to 10 years, 91% (n=3376) of children aged 10 to 15 years, and 91% (n=12,461) of adolescents aged 15 to 19 years having microbiological confirmation of tuberculosis. Of the 17,764 (87%) participants with reported HIV status, 2,448 (13.8%) were living with HIV. There was a high degree of missing data for baseline weight (n=17,560, 86.1%) and height (n=12,420, 60.9%), so these variables could not be assessed.

Table 3 shows the individual medications used in treatment regimens, and the number of medications used by WHO drug grouping. Tables of individual drugs used stratified by age, HIV status and country are shown in Supplemental Tables 5–10. Nearly 90% of children (n=18,552) received a late-generation fluoroquinolone, either moxifloxacin (n=10,039, 49.2%) or levofloxacin (n=8,513, 41.7%). Fewer received bedaquiline (n=1,521, 7.5%) or linezolid (n=1,309, 6.4%) and even fewer delamanid (n=543, 2.7%). Supplemental Figure S3 shows the frequency of individual antituberculosis drug use overall and trends over time for South Africa and India which represent 92% of the analysis cohort. The median (IQR) duration of treatment was 16 (9, 23) months overall, and 17 (9, 24) months for participants with treatment success treated after 2014 versus 21 (18, 24) months for those treated before 2014.

Table 3.

Medications used in the treatment of children and adolescents with rifampicin-resistant tuberculosis in the analysis cohort

Drugs used in the regimen	Analysis cohort (N=20395)	Treatment outcome
		Treatment success (N=16825)	Death (N=2848)	Treatment failure (N=722)
Group A
Moxifloxacin	10039 (49.2%)	8181 (48.6%)	1434 (50.4%)	424 (58.7%)
Levofloxacin	8513 (41.7%)	7150 (42.5%)	1144 (40.2%)	219 (30.3%)
Linezolid	1309 (6.4%)	1147 (6.8%)	108 (3.8%)	54 (7.5%)
Bedaquiline	1521 (7.5%)	1261 (7.5%)	199 (7.0%)	61 (8.4%)
Group B
Cycloserine	8601 (42.2%)	6867 (40.8%)	1476 (51.8%)	258 (35.7%)
Terizidone	4653 (22.8%)	3862 (23.0%)	562 (19.7%)	229 (31.7%)
Clofazimine	9076 (44.5%)	7427 (44.1%)	1318 (46.3%)	331 (45.8%)
Group C
Ethambutol	16545 (81.1%)	13830 (82.2%)	2198 (77.2%)	517 (71.6%)
Delamanid	543 (2.7%)	405 (2.4%)	105 (3.7%)	33 (4.6%)
Pyrazinamide	19718 (96.7%)	16267 (96.7%)	2756 (96.8%)	695 (96.3%)
Amikacin	6919 (33.9%)	5907 (35.1%)	776 (27.2%)	236 (32.7%)
Ethionamide	18180 (89.1%)	14869 (88.4%)	2676 (94.0%)	635 (88.0%)
Prothionamide	708 (3.5%)	638 (3.8%)	30 (1.1%)	40 (5.5%)
Para-aminosalicylic acid	1805 (8.9%)	1524 (9.1%)	191 (6.7%)	90 (12.5%)
Streptomycin	67 (0.3%)	63 (0.4%)	4 (0.1%)	0 (0.0%)
Imipenem	23 (0.1%)	1 (0.0%)	2 (0.3%)	23 (0.1%)
Meropenem	5 (0.0%)	4 (0.0%)	1 (0.0%)	0 (0.0%)
Other
Isoniazid (standard)	725 (3.6%)	655 (3.9%)	49 (1.7%)	21 (2.9%)
High-dose Isoniazid	7417 (36.4%)	6374 (37.9%)	809 (28.4%)	234 (32.4%)
Kanamycin	15397 (75.5%)	12658 (75.2%)	2214 (77.7%)	525 (72.7%)
Capreomycin	1611 (7.9%)	1196 (7.1%)	323 (11.3%)	92 (12.7%)
Ofloxacin	1255 (6.2%)	1076 (6.4%)	122 (4.3%)	57 (7.9%)
Ciprofloxacin	266 (1.3%)	227 (1.3%)	23 (0.8%)	16 (2.2%)
Use of Group A drugs
None	1778 (8.7%)	1478 (8.8%)	218 (7.7%)	82 (11.4%)
One	16957 (83.1%)	13917 (82.7%)	2471 (86.8%)	569 (78.8%)
Two	903 (4.4%)	772 (4.6%)	89 (3.1%)	42 (5.8%)
Three	757 (3.7%)	658 (3.9%)	70 (2.5%)	29 (4.0%)
Use of Group B drugs
None	550 (2.7%)	473 (2.8%)	59 (2.1%)	18 (2.5%)
One	17377 (85.2%)	14560 (86.5%)	2225 (78.1%)	592 (82.0%)
Two	2468 (12.1%)	1792 (10.7%)	564 (19.8%)	112 (15.5%)
Use of Group C drugs
None	50 (0.2%)	40 (0.2%)	7 (0.2%)	3 (0.4%)
One	376 (1.8%)	305 (1.8%)	50 (1.8%)	21 (2.9%)
Two	2937 (14.4%)	2300 (13.7%)	505 (17.7%)	132 (18.3%)
Three	10028 (49.2%)	8239 (49.0%)	1478 (51.9%)	311 (43.1%)
Four or more	7004 (34.3%)	5941 (35.3%)	808 (28.4%)	255 (35.3%)

Within the analysis cohort, 18,825 (82.5%) participants had a successful treatment outcome, 2,848 (14.0%) died and 722 (3.5%) had treatment failure (see Figure 2 for forest plot).

Figure 2. Forest plot demonstrating proportion of children and adolescents with rifampicin-resistant tuberculosis with successful treatment outcome vs. treatment failure or death.

Events refers to successful treatment outcome. Total refers to treatment success and unfavourable outcome (which included treatment failure or death), and proportion were calculated accordingly. Logit transformation was used. Cohorts with <5 participants were included in the previously published paediatric individual patient data meta-analysis from 2014 and were thus included here.

Table 4 shows clinical predictors of successful treatment outcomes versus death or treatment failure. Older age, male sex, living with HIV, previous tuberculosis treatment, having pulmonary tuberculosis and AFB-smear positivity were all negatively associated with treatment success in multivariable regression (ROC-AUC 0.67 [95% CI, 0.66. to 0.68]; Marginal R²/Conditional R² 0.1/0.35). Table 5 additionally shows results of this analysis stratified by age group.

Table 4.

Clinical predictors of treatment success versus death and treatment failure among children and adolescents treated for rifampicin-resistant tuberculosis

	Treatment success versus unfavourable outcomes (death/treatment failure)
	Univariable logistic regression models		Multivariable logistic regression model
Clinical predictor	Crude OR (95% CI)	p-value	Adjusted OR (95% CI)	p-value
Upper-middle or high-income country	1.94 (0.93–4.1)	0.079	-	-
Age (years)	0.94 (0.93–0.95)	<0.0001	0.96 (0.94–0.97)	<0.0001
Male sex	0.85 (0.79–0.91)	<0.0001	0.84 (0.78–0.91)	<0.0001
HIV positive	0.39 (0.34–0.44)	<0.0001	0.42 (0.36–0.48)	<0.0001
Microbiologically confirmed tuberculosis	1.41 (1.25–1.59)	<0.0001	-	-
Previous tuberculosis treatment	0.44 (0.4–0.49)	<0.0001	0.52 (0.46–0.58)	<0.0001
Pulmonary tuberculosis	0.4 (0.26–0.6)	<0.0001	0.53 (0.34–0.81)	0.0034
Extrapulmonary tuberculosis	1.61 (1.24–2.11)	0.0010	-	-
AFB smear positive	0.43 (0.37–0.49)	<0.0001	0.46 (0.4–0.53)	<0.0001

^a

Performance evaluation for univariable models:

• Predictor Income country: ROC-AUC 0.54 [95 % CI, 0.53 to 0.56]; Marginal R²/Conditional R² 0.02/0.32
• Predictor Age: ROC-AUC 0.58 [95 % CI, 0.57 to 0.59]; Marginal R²/Conditional R² 0.02/0.25
• Predictor Sex: ROC-AUC 0.55 [95 % CI, 0.54 to 0.56]; Marginal R²/Conditional R² 0.001/0.34
• Predictor HIV status: ROC-AUC 0.59 [95 % CI, 0.58 to 0.6]; Marginal R²/Conditional R² 0.02/0.35
• Predictor TB status: ROC-AUC 0.55 [95 % CI, 0.54 to 0.56]; Marginal R²/Conditional R² 0.002/0.38
• Predictor Any previous TB treatment: ROC-AUC 0.59 [95 % CI, 0.58 to 0.6]; Marginal R²/Conditional R² 0.02/0.32
• Predictor Pulmonary TB: ROC-AUC 0.54 [95 % CI, 0.536 to 0.0.55]; Marginal R²/Conditional R² 0.004/0.37
• Predictor Extrapulmonary TB: ROC-AUC 0.57 [95 % CI, 0.56 to 0.58]; Marginal R²/Conditional R² 0.01/0.36
• Predictor AFB smear status: ROC-AUC 0.63 [95 % CI, 0.62 to 0.64]; Marginal R²/Conditional R² 0.04/0.33

^bMultivariable model adjusted for the following baseline clinical characteristics: age, sex, HIV status, history of any previous tuberculosis treatment, having pulmonary disease and AFB smear positive.

^cPerformance evaluation for multivariable model: ROC-AUC 0.67 [95 % CI, 0.66 to 0.68]; Marginal R²/Conditional R² 0.1/0.35

Note: Complete case analysis results are reported in Supplementary Table S18.

AFB, acid-fast bacilli; OR, odds ratio

Table 5.

Clinical predictors of treatment success versus death and treatment failure among children and adolescents treated for rifampicin-resistant tuberculosis stratified by age

Clinical predictor	Treatment success versus unfavourable outcomes (death/treatment failure)
	Age < 5 years				Age 5 to <10 years				Age 10 to <15 years				Age 15–19 years
	Univariable logistic regression models		Multivariable logistic regression model		Univariable logistic regression models		Multivariable logistic regression model		Univariable logistic regression models		Multivariable logistic regression model		Univariable logistic regression models		Multivariable logistic regression model
	Crude OR (95% CI)	p-value	aOR (95% CI)	p-value	Crude OR (95% CI)	p-value	aOR (95% CI)	p-value	Crude OR (95% CI)	p-value	aOR (95% CI)	p-value	Crude OR (95% CI)	p-value	aOR (95% CI)	p-value
Upper-middle or high-income country	5.24 (0.1.5–18.3)	0.0095	-	-	2.37 (0.85–6.59)	0.097	-	-	1.5 (0.83–2.72)	0.17	-	-	0.82 (0.55–1.22)	0.33	-	-
Age (years)	1.16 (0.96–1.4)	0.13	1.17 (0.97–1.42)	0.11	0.94 (0.82–1.07)	0.37	1.01 (0.87–1.16)	0.94	0.90 (0.84–0.96)	0.0025	0.91 (0.85–0.98)	0.011	0.99 (0.95–1)	0.37	0.99 (0.96–1)	0.82
Male sex	0.87 (0.56–1.37)	0.55	0.84 (0.53–1.35)	0.47	1.07 (0.74–1.56)	0.72	1.08 (0.73–1.59)	0.71	0.75 (0.61–0.91)	0.0045	0.75 (0.61–0.92)	0.0067	0.86 (0.79–0.93)	0.00038	0.87 (0.79–0.95)	0.0015
HIV positive	0.29 (0.16–0.52)	<0.0001	0.3 (0.16–0.56)	0.00016	0.34 (0.19–0.59)	0.00013	0.49 (0.27–0.89)	0.019	0.39 (0.29–0.53)	<0.0001	0.42 (0.30–0.58)	<0.0001	0.4 (0.34–0.47)	<0.0001	0.39 (0.33–0.46)	<0.0001
Microbiologically confirmed tuberculosis	0.68 (0.29–1.6)	0.37	-	-	0.37 (0.17–0.81)	0.012	-	-	1.32 (0.98–1.78)	0.070	-	-	1.6 (1.4–1.83)	<0.0001	-	-
Previous tuberculosis treatment	0.76 (0.39–1.5)	0.42	0.96 (0.47–1.94)	0.47	0.51 (0.33–0.80)	0.0036	0.64 (0.39–1.05)	0.080	0.45 (0.36–0.57)	<0.0001	0.51 (0.40–0.66)	0.011	0.48 (0.43–0.55)	<0.0001	0.51 (0.45–0.58)	<0.0001
Pulmonary tuberculosis	0.5 (0.14–1.82)	0.30	0.6 (0.16–2.26)	0.44	0.27 (0.08–0.96)	0.043	0.30 (0.08–1.11)	0.071	0.45 (0.16–1.24)	0.45	0.52 (0.18–1.47)	0.21	0.59 (0.36–0.96)	0.032	0.53 (0.32–0.89)	0.017
Extrapulmonary tuberculosis	1.61 (0.51–2.64)	0.72	-	-	1.51 (0.75–3.04)	0.25	-	-	1.84 (1.26–2.68)	0.0019	-	-	1.5 (1.12–2)	0.0081	-	-
AFB smear positive	0.33 (0.18–0.59)	0.00022	0.34 (0.19–0.63)	0.00054	0.31 (0.20–0.48)	<0.0001	0.35 (0.21–0.60)	<0.0001	0.47 (0.37–0.59)	<0.0001	0.48 (0.38–0.6)	<0.0001	0.47 (0.4–0.56)	<0.0001	0.46 (0.39–0.54)	<0.0001

AFB, acid-fast bacilli; OR, odds ratio; aOR, adjusted OR

After adjusting for significant clinical predictors, the use of two (aOR 1.41, [95 % CI, 1.09 to 1.82], p=0.008) or three (aOR 2.12, [95% CI, 1.61 to 2.79], p<0.001) group A drugs compared to the use of no group A drugs at all was associated with higher adjusted odds (aOR) of treatment success versus treatment failure or death (Table 6). The use of two group B drugs compared to the use of no group B drugs was associated with a reduced adjusted odds of treatment success (aOR 0.59 [95 % CI, 0.44 to 0.79]) (ROC-AUC 0.69 [95 % CI, 0.68 to 0.70]; Marginal R²/Conditional R² 0.11/0.31). Table 7 shows results of this same analysis stratified by age group.

Table 6.

Summary of association between the use of number of WHO group drugs used with treatment success in children and adolescents with rifampicin-resistant tuberculosis.

Use of number of drugs of each WHO group	Treatment success versus unfavourable outcomes (death/treatment failure)
	Univariable logistic regression modelsa		Multivariable logistic regression modelsb,c
	Crude OR (95% CI)	p-value	aOR (95% CI)	p-value
Number of drugs group A d
1	1.36 (1.16–1.58)	0.00010	1.06 (0.90–1.25)	0.51
2	1.52 (1.20–1.93)	0.00056	1.41 (1.09–1.82)	0.0080
3	1.77 (1.37–2.28)	<0.0001	2.12 (1.61–2.79)	<0.0001
Number of drugs group B e
1	1.29 (0.99–1.67)	0.052	1.27 (0.96–1.69)	0.094
2	0.62 (0.47–0.81)	0.00050	0.59 (0.44–0.79)	0.00052
Number of drugs group C f
2	0.99 (0.76–1.29)	0.93	1.07 (0.80–1.44)	0.65
3	1.43 (1.10–1.86)	0.0067	1.14 (0.86–1.52)	0.35
4 or more	1.8 (1.38–2.35)	<0.0001	1.29 (0.96–1.73)	0.086

^a

Performance evaluation for univariable models:

• Predictor Number of drugs group A: ROC-AUC 0.55 [95 % CI, 0.54 to 0.56]; Marginal R²/Conditional R² 0.002/0.37
• Predictor Number of drugs group B: ROC-AUC 0.57 [95 % CI, 0.56 to 0.58]; Marginal R²/Conditional R² 0.012/0.35
• Predictor Number of drugs group C: ROC-AUC 0.58 [95 % CI, 0.57 to 0.90]; Marginal R²/Conditional R² 0.008/0.33

^bMultivariable model, including all predictors listed (Number of drugs group A, Number of drugs group B, Number of drugs group C), and adjusted for the following baseline clinical characteristics: age, sex, HIV status, history of any past tuberculosis treatment, having pulmonary disease and AFB smear positive.

^cPerformance evaluation for multivariable model: ROC-AUC 0.69 [95 % CI, 0.68 to 0.70]; Marginal R²/Conditional R² 0.11/0.31

^dUse of no drugs from group A, reference for predictor Number of drugs group A.

^eUse of no drugs from group B, reference for predictor Number of drugs group B.

^fUse of one or no drugs from group C, reference for predictor Number of drugs group C.

Abbreviations: aOR, adjusted odds ratio. CI, confidence interval.

Note: Complete case analysis results are reported in Supplementary Table S19.

Table 7.

Treatment predictors (Use of number of drugs of each WHO group) of treatment success versus death and treatment failure among children and adolescents treated for rifampicin-resistant tuberculosis stratified by age

Use of number of drugs of each WHO group	Treatment success versus unfavourable outcomes (death/treatment failure)
	Age < 5 years				Age 5 to <10 years				Age 10 to <15 years				Age 15–19 years
	Univariable logistic regression models		Multivariable logistic regression modela		Univariable logistic regression models		Multivariable logistic regression modela		Univariable logistic regression models		Multivariable logistic regression modela		Univariable logistic regression models		Multivariable logistic regression modela
	Crude OR (95% CI)	p-value	aOR (95% CI)	p-value	Crude OR (95% CI)	p-value	aOR (95% CI)	p-value	Crude OR (95% CI)	p-value	aOR (95% CI)	p-value	Crude OR (95% CI)	p-value	aOR (95% CI)	p-value
Number of drugs group A b
1	1.33 (0.66–2.68)	0.42	0.93 (0.43–2.00)	0.85	0.89 (0.51–1.56)	0.69	0.89 (0.48–1.62)	0.70	1.70 (1.2–2.40)	0.0027	1.32 (0.92–1.89)	0.13	1.3 (1.08–1.56)	0.0066	0.95 (0.77–1.16)	0.62
2	1.08 (0.4–2.9)	0.88	0.73 (0.25–2.10)	0.55	1.62 (0.65–4.1)	0.30	1.03 (0.37–2.87)	0.96	1.11 (0.67–1.83)	0.69	0.94 (0.57–1.67)	0.92	1.73 (1.29–2.32)	0.00027	1.66 (1.21–2.28)	0.0016
3	0.28 (0.07–1.14)	0.076	0.42 (0.09–2.02)	0.28	3.12 (0.40–24.4)	0.28	2.23 (0.26–18.71)	0.46	3.03 (1.55–5.94)	0.0013	2.81 (1.39–5.70)	0.0040	2.21 (1.65–2.96)	<0.0001	2.36 (1.72–3.23)	<0.0001
Number of drugs group B c
1	1.68 (0.78–3.62)	0.19	1.78 (0.78–4.04)	0.17	1.11 (0.51–2.43)	0.80	1.26 (0.57–2.80)	0.56	1.34 (0.74–2.45)	0.34	1.3 (0.69–2.45)	0.41	1.55 (1.08–2.22)	0.016	1.18 (0.80–1.75)	0.41
2	1.83 (0.67–4.97)	0.24	1.47 (0.50–4.29)	0.48	2.06 (0.73–5.84)	1.74	2.17 (0.71–6.66)	0.17	0.61 (0.33–1.16)	0.13	0.64 (0.32–1.26)	0.19	0.67 (0.47–0.98)	0.040	0.48 (0.32–0.73)	0.00052
Number of drugs group C d
2	0.95 (0.2–4.5)	0.95	0.95 (0.18–4.98)	0.95	0.64 (0.14–2.92)	0.57	0.73 (0.15–3.56)	0.70	1.2 (0.65–2.25)	0.555	1.27 (0.64–2.51)	0.49	0.96 (0.7–1.31)	0.80	1.09 (0.77–1.54)	0.62
3	0.96 (0.21–4.3)	0.96	0.78 (0.15–3.95)	0.76	0.82 (0.19–3.63)	0.79	1.21 (0.25–5.79)	0.81	1.73 (0.94–3.17)	0.078	1.47 (0.75–2.88)	0.26	1.32 (0.97–1.79)	0.075	1.08 (0.77–1.50)	0.67
4 or more	1.1 (0.24–5.11)	0.90	0.86 (0.16–4.58)	0.86	0.79 (0.18–3.51)	0.75	1.15 (0.24–5.66)	0.86	1.79 (0.96–3.32)	0.065	1.39 (0.69–2.76)	0.36	1.73 (1.27–2.36)	0.00055	1.24 (0.88–1.74)	0.22

^aMultivariable model, including all predictors listed (Number of drugs group A, Number of drugs group B, Number of drugs group C), and adjusted for the following baseline clinical characteristics: age, sex, HIV status, history of any past tuberculosis treatment, having pulmonary disease and smear +.

^bUse of no drugs from group A, reference for predictor Number of drugs group A.

^cUse of no drugs from group B, reference for predictor Number of drugs group B.

^dUse of one or no drugs from group C, reference for predictor Number of drugs group C.

Abbreviations: aOR, adjusted odds ratio. CI, confidence interval.

Supplemental Tables 11–17 show results of this last analysis stratified separately by age, HIV status and country (South Africa vs. India), as well as univariate analyses of associations of individual drugs with treatment success. In addition, Supplementary Tables 18–21 show complete case analysis for the main results reported in the manuscript. Some differences were observed in some predictors in magnitude and significance, but directionally the effect was the same.

The characteristics of the 2,974 participants in the overall cohort with an outcome of LTFU are shown in Supplemental Table 4. When classifying LTFU participants as having an unfavourable outcome along with treatment failure and death, the clinical predictors remained consistent; however, the strength of the adjusted odds ratios (aORs) was reduced (Supplementary Figure S1a). Conversely, when categorising LTFU participants as having a favourable outcome, there were no changes observed in terms of the selection or strength of clinical predictors (Supplementary Figure S1b). In a further sensitivity analyses, when classifying LTFU as having a successful outcome or as an unfavourable outcome, treatment predictors of successful outcome remained consistent (Supplemental Figure S2a and S2b).

INTERPRETATION

This IPD meta-analysis represents the largest dataset of children and adolescents with MDR/RR tuberculosis globally to date, providing a unique opportunity to better understand key aspects of MDR/RR tuberculosis in this vulnerable population to inform policy and practice. A key finding is the continued underrepresentation of young children and those with a clinical diagnosis among children treated for MDR/RR tuberculosis. This is also the first time that adolescents 15–19 years, frequently excluded from adult trials and from receiving newer medications, were specifically targeted for inclusion in such a review. Additionally, data are included on the use of the key repurposed drugs linezolid and clofazimine, later generation fluoroquinolones and the novel drugs bedaquiline and delamanid in children and adolescents, for which there were very limited previous data. One important finding is the association of Group A drugs with better treatment outcomes in this population.

When looking at only those <15 years of age, 80% were successfully treated, similar to the 78% reported for the same age group in the 2016 paediatric MDR/RR tuberculosis IPD meta-analysis.⁵ These similar outcomes are with shorter treatment durations and less use of injectables in the updated data. In the overall cohort which included those LTFU, 72% of children and adolescents were successfully treated. The lower proportion of successful treatment in the overall cohort may reflect the inclusion of a large number of older adolescents who tend to have more severe tuberculosis but were not in the previous IPD, and overall increased disease severity with a higher proportion of participants with a confirmed diagnosis. There is scope to improve MDR/RR tuberculosis care for children and adolescents to reach the WHO’s End TB Strategy target of ≥90% treatment success for MDR/RR tuberculosis.¹² However, outcomes are still considerably better, including among adolescents, than those reported in a 2018 IPD meta-analysis of adults treated for MDR/RR tuberculosis (cured or completed treatment 61%, failure 8%, death 14%, LTFU or other 16%).¹¹

A large number of children and adolescents were treated with later-generation fluoroquinolones levofloxacin and moxifloxacin, the newer repurposed drugs linezolid, clofazimine, and the novel drug bedaquiline, including some children <5 years treated with bedaquiline. A notable exception was the small number of children and adolescents treated with delamanid. Bedaquiline and delamanid had not yet been recommended in routine care in younger children at the time this work was undertaken. The optimal role of delamanid in MDR/RR tuberculosis treatment is still to be defined. It is used more frequently in children compared to adults due to the lack of data from clinical trials of pretomanid in children and no current recommendation for pretomanid’s use in those below 15 years of age. Emerging adult and other trial data may support a stronger future role for delamanid. The optimal duration of linezolid use in children that balances its well-known dose- and duration-dependent adverse effects with its benefit is not known. Due to limited data on adverse effects and its duration of use in our review we were unable to comment on this.

Diverse treatment regimens were used, reflecting the broad time frame and evolving guidelines and available medications over this period. Our analysis demonstrated, for the first time, that the use of regimens with two or three Group A drugs resulted in a significantly improved adjusted odds of treatment success vs. death or treatment failure in children and adolescents overall, compared to use of no Group A drug. In age stratified analyses shown in Table 7, similar results were found for children aged 5 to <15 years versus in those 15 to 19 years. Among children <5 years of age there was not a statistically significant association between the number of drugs used of each WHO group and treatment success. The number of children in this age group overall was substantially lower compared to the other groups. This is due to delays in pediatric trials, the lack of guidelines on the use of new drugs like bedaquiline in young children, and limited access to newer drugs like bedaquiline reduced the number of young children receiving these drugs. The children in this age group treated with bedaquiline would have been receiving this drug off-label and without a WHO recommendation, thus would likely have had particularly limited treatment options, more extensive resistance or other complicating factors beyond those captured by our clinical predictors. This may contribute to the point estimate for the odds of treatment success versus death/treatment failure being less than 1 for young children receiving more Group A drugs.

The combined analysis confirms the current WHO approach for prioritisation of second-line antituberculosis drugs for MDR/RR tuberculosis treatment in children and adolescents. Critically, it demonstrates the direct negative impact of children and adolescents not being treated with these more effective drugs. Although bedaquiline received conditional approval by the U.S. Food and Drug Administration in 2012, as of 2023 paediatric trials are still ongoing and the WHO only recommended bedaquiline for children across the entire age spectrum in 2022.¹³ Our results support the inclusion of Group A drugs in all regimens for children and adolescents where the strain is susceptible. We also observed that inclusion of two Group B drugs was associated with a reduced adjusted odds of treatment success. This may be because both Group B drugs are needed when there is more severe disease or fluoroquinolone or other additional resistance, or may be due to other residual confounding. This finding should be interpreted with caution. Pyrazinamide was used nearly universally, so its impact on treatment outcomes is difficult to assess and the effect of other drugs should be considered in light of pyrazinamide use.

A substantial proportion of participants were from routine programmatic data from South Africa and India, providing key insights on who is being diagnosed, treated and reported in routine care in these two high-burden settings. Older adolescents ≥15 years of age, who tend to have adult-type disease and are easier to confirm microbiologically, represented 68.1% of all participants. Children <5 years of age were particularly underrepresented, as they would be expected to account for 50–60% of those <15 years of age based on the observed tuberculosis epidemiology in children.¹⁴ This is a critical omission, as most deaths in children from tuberculosis globally are in those <5 years who were never started on antituberculosis treatment.¹⁵ Few children had a clinical diagnosis of MDR/RR tuberculosis, while in general, at least 50–60% of children <15 years of age would be expected to have a clinical diagnosis (diagnosis of MDR/RR tuberculosis in the absence of microbiologic confirmation based on clinical grounds and risk factors for drug resistance, including exposure to a known infectious MDR/RR tuberculosis source case).¹⁶ Children with clinically diagnosed tuberculosis, underrepresented in our study, are more likely to have paucibacillary disease which is more challenging to confirm, but usually represents less severe disease, with better outcomes and ideally suited for shorter duration treatment, as recently shown in children with drug-susceptible tuberculosis (SHINE trial).¹⁷ Conversely, children in our analysis were likely to have more advanced disease with higher bacillary burden. The UN High-Level Meeting on Tuberculosis in 2018 set a target of treating 115,000 children with MDR/RR tuberculosis from 2018–2022; however, by 2021 the WHO reported that only 17,700 or 15% of this target had been reached.¹⁸ Paediatric MDR/RR tuberculosis case finding efforts should include a specific and stronger focus on diagnosing young children based on clinical diagnosis. Inclusion of more younger children and children with clinically diagnosed MDR/RR tuberculosis would be expected to improve overall estimates of treatment success as these groups tend to have less severe tuberculosis with lower bacillary burden and excellent outcomes.

Clinical predictors of successful treatment outcome were similar to those previously described.⁵ Older age was associated with a lower adjusted odds of treatment success, likely reflecting the more severe, adult-type tuberculosis usually observed in older children and adolescents. Children and adolescents living with HIV had significantly worse outcomes, highlighting their vulnerability and the need to continue to prioritise care for this high-risk group. Detection of AFB on sputum smear microscopy, a relatively crude marker for severe pulmonary tuberculosis, was also associated with worse outcome, but missing data limited the ability to assess other measures of disease severity. Future studies and programmatic data should collect information on disease severity systematically going forward. We were unable to assess the effects of fluoroquinolone resistance and nutritional status on treatment outcome due to lack of data.

Our study has several additional limitations. Most participants were from routine programmatic registry data in South Africa and India. Although these are reasonably representative of many high MDR/RR tuberculosis burden settings, some caution to generalising findings to other settings is warranted. Compared to some of the smaller cohort studies, programmatic data had more missing data and less nuanced information on participant characteristics, treatment regimens, and response to treatment. Inclusion of studies with small sample sizes may lead to small sample bias.¹⁹ There was good geographic representation, with the exception of little data from high-burden countries in the Western Pacific Region. Additionally, a relatively small proportion of children were treated with zero, two or three Group A drugs. There is also likely to be additional residual unmeasured confounding related to the complex clinical, treatment and programmatic factors that contribute to treatment outcomes. Data from the original review up through 2014 did not include adolescents 15 years of age or older. Although the inclusion of adolescents in this analysis addressed an important gap, it is important to consider the under-representation of younger children, who tend to have different disease spectrum and in whom the diagnosis may be more challenging. Additionally, some aspects important to adolescent care, such as how services are delivered in the community, could not be addressed by this analysis. Regimen safety could not be reported on due to very limited available data on adverse events.

Lastly, while we include LTFU among overall outcomes to provide a complete picture of true programmatic outcomes of children and adolescents with MDR/RR tuberculosis, we excluded LTFU from our primary analyses, consistent with previous similar analyses. Our data suggested that children with an outcome of LTFU were more similar to those with a favorable outcome, and sensitivity analyses in which we imputed their outcome to a favorable one yielded similar results to the primary analyses. However, an outcome of LTFU may be associated with treatment regimens that are long and poorly tolerated or that are less effective, and may also include unreported deaths. Due to this complex interplay, we are unable to discount the possibility of bias in our analyses of predictors of treatment success. Children with an outcome of LTFU are an important subgroup to target for interventions to ensure continuity of care, support adherence and ensure improved overall treatment outcomes.

Our analysis identified key areas where improved routine data collection would be very informative, including on tuberculosis disease severity, collection of weight and nutritional status, and safety data. Harmonising key data fields for childhood tuberculosis and MDR/RR tuberculosis for collection by tuberculosis programmes would greatly improve secondary analyses of routinely collected data. The WHO recently initiated a tuberculosis IPD platform for adults, children, adolescents and pregnant women that will allow these valuable datasets to be maintained and expanded in future, will align approaches to data collection and management, and will ensure available data to inform policy and for researchers.²⁰

Although important progress has been made to improve MDR/RR tuberculosis care in children and adolescents, substantial continued efforts are required to better identify tuberculosis patients across the age spectrum, ensure equitable access to the best, shortest, most-safe medications and treatment regimens, to achieve target treatment outcomes.

RESEARCH IN CONTEXT.

Evidence before the study

A systematic review and individual participant data (IPD) meta-analysis of children <15 years of age treated for multidrug-resistant/rifampicin-resistant (MDR/RR) tuberculosis was previously undertaken in 2014. For that review, we undertook a broad search for published and unpublished studies without limitations on language or study design up until October 2014 that ultimately included IPD from 1,009 children <15 years of age from 28 studies. It demonstrated that overall outcomes for children and young adolescents were better than in adults, with 78% successfully treated, and identified predictors of treatment outcome. Since then, the treatment landscape for MDR/RR tuberculosis has evolved substantially with improved access to novel drugs, but there are limited new paediatric data. Older adolescents (15 to <20 years) with MDR/RR tuberculosis may have worse outcomes than adults, but previous reports are limited.

Added value of this study

This IPD meta-analysis represents the largest dataset of children and adolescents with MDR/RR tuberculosis described globally to date, with over 20,000 included. This is also the first time that older adolescents (15 to 19 years of age) have been specifically included in such a review. Most individuals identified were adolescents and most had microbiologically confirmed disease. This updated review includes new data on key repurposed drugs, linezolid and clofazimine, and the novel drugs, bedaquiline and delamanid, in children and adolescents, as well as later generation fluoroquinolones, for which there were very limited data previously, in regimens of reduced treatment duration and with reduced use of injectable drugs. We found that the use of two or three priority (Group A) drugs was associated with improved treatment outcomes in children and adolescents.

Implications of all the available evidence

Children and adolescents with MDR/RR tuberculosis should receive treatment with Group A drugs as a priority. Treatment outcomes in children and adolescents with MDR/RR tuberculosis were better than in adults, but still below the World Health Organization’s End TB Strategy target of ≥90% successfully treated. Our findings indicate that young children and children with a clinical diagnosis are underrepresented among those treated for MDR/RR tuberculosis, indicating the urgent need for improved case finding to target these groups.

Data Sharing

The protocol is available upon request.