Effectiveness and safety of the available preventive tuberculosis treatment regimens for children and adolescents: systematic review and network meta-analysis

Abstract

Background

Although children are at a higher risk of progressing to tuberculosis (TB) disease compared with adults, research and publications addressing TB management have mostly emphasised adult disease.

Methods

MEDLINE, Embase, CENTRAL and grey literature were searched from inception to 3 May 2021, and updated on 12 December 2024 for the systematic review and network meta-analysis (NMA). Randomised controlled trials of children and/or adolescents under 18 years with TB infection, contacts of drug-susceptible TB, without HIV, evaluating the effectiveness and safety of available regimens for the treatment of TB were included. Meta-analysis was conducted using a random effects model. Risk of bias was assessed using the Cochrane RoB tool 2. Effect estimates (RR and OR) are presented with 95% CIs. Certainty of evidence was evaluated using the Grading of Recommendations, Assessment, Development and Evaluation (GRADE) approach.

Results

Eleven studies were obtained. All network geometries had open and disconnected loops, except adverse reactions for which we conducted a full NMA. We found that 4 months of isoniazid and rifampicin (4HR) reduces the number of active TB cases at 1 year, 2 years and 5 years of follow-up (RR 0.49 (95% CI 0.32 to 0.76)) compared with isoniazid for 9 months (9H). The isoniazid + rifapentine for 3 months (3HP) regimen (RR 1.09 (95% CI 1.03 to 1.15)), 4HR (RR 1.07 (95% CI 1.01 to 1.14)) and rifampicin for 4 months (4R) (RR 1.12 (95% CI 1.05 to 1.20)) regimens have higher treatment adherence compared with 9H. High variability was found in the reporting of adverse reactions, which were more likely to occur with 3HP compared with 4HR (OR 4.56 (95% CI 1.22 to 16.96)) and 4R (OR 6.37 (95% CI 2.11 to 19.19)). Adverse reactions occur less frequently with 4R (OR 0.34 (95% CI 0.20 to 0.58)) compared with 9H.

Conclusions

This study synthesised available data on paediatric short and long-course regimens. Short course regimens may have higher adherence (3HP, 4HR, 4R) and may reduce active TB incidence (isoniazid + rifampicin for 3 months (3HR) and 4HR) compared with 9H.

PROSPERO registration number

CRD42021271512.

WHAT IS ALREADY KNOWN ON THIS TOPIC

• Children with tuberculosis infection are at high risk of progressing to tuberculosis disease, but evidence on tuberculosis preventive treatment regimens has largely been extrapolated from adult populations.

WHAT THIS STUDY ADDS:

• Before this review, no systematic review had synthesised evidence exclusively from randomised trials in children and adolescents comparing all available preventive treatment regimens.

• Short-course regimens may have a higher adherence (isoniazid + rifapentine for 3 months (3HP), isoniazid + rifampicin for 4 months (4HR) and rifampicin for 4 months (4R)), may reduce the incidence of active tuberculosis at 2 years (isoniazid + rifampicin for 3 months (3HR) and 4HR) and may reduce the risk of adverse reactions (4R), compared with long-course regimens (isoniazid for 9 months (9H)).

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY:

• These findings support the use of short-course regimens for children and adolescents with tuberculosis infection and highlight the need for additional high-quality paediatric trials, particularly in children under 5 years of age, to strengthen future guideline recommendations.

Introduction

Tuberculosis infection (TBI) is a state of persistent immune response to stimulation by Mycobacterium tuberculosis (M. tuberculosis) antigens with no evidence of clinically manifest tuberculosis (TB) disease.¹ It is estimated that 7.5 million children are infected with M. tuberculosis each year, and 5%–10% may develop TB disease without TB preventive treatment (TPT).² In addition, more than 1.5 million children and adolescents up to 19 years of age fall ill with TB every year, with the highest proportions observed in children less than 5 years of age.³

Although children are at a higher risk of progressing to TB disease compared with adults, research and publications addressing essential aspects of management and control of TB have mostly emphasised adult disease.⁴ In contrast, paediatric TB has received less attention, mainly due to greater challenges in diagnosis and the lower priority given within TB control programmes.⁴ However, paediatric TB remains a public health problem due to the higher risk of severe disease and mortality among young children compared with adults.⁴ Preventing the progression of TBI to active TB through medication is a critical component of global TB control efforts.⁵ For this reason, safe and effective TPT strategies, as part of global TB control efforts,⁵ should be targeted at population groups at higher risk of progressing to TB disease.

Isoniazid preventive therapy (IPT) has traditionally been used due to efficacy and TB risk reduction in children.⁶ However, the prolonged treatment duration, non-adherence and side effects present significant challenges.^{5 7} Rifamycin-based regimens, including rifampicin and rifapentine, are alternatives to IPT. Short-term rifampicin regimens likely have greater adherence and lower adverse events compared with prolonged IPT regimens.⁵

To date, no systematic review has synthesised all available evidence from randomised controlled trials (RCTs) comparing TPT regimens in children. While a previous review and network meta-analysis (NMA) combined data from adults and children,⁸ another review included both non-randomised studies and RCTs.⁹ A third review, which included a meta-analysis (MA), compared IPT to placebo or no prophylaxis in children,⁶ and a fourth review, also with a MA, included HIV-infected children with and without known TB exposure.¹⁰

Due to the absence of an evidence synthesis on the topic focused only in children, we conducted a systematic review and NMA to determine the effectiveness and safety of all available regimens for the treatment of TBI in children and adolescents under 18 years of age not infected with HIV who are in contact with drug-sensitive TB cases.

Methods

Search strategy and eligibility criteria

The protocol for this review was registered with the PROSPERO database (CRD42021271512). A more detailed description of the methods is provided in the protocol published elsewhere.¹¹ The study followed the recommendations by the Cochrane Handbook for Systematic Reviews of Interventions,¹² and this report was prepared following the Preferred Reporting Items for Systematic Review and Meta-Analysis (PRISMA) statement for NMA.¹³ Using the Ovid platform without language restrictions, we searched MEDLINE, Embase and the Cochrane Central Register of Controlled Trials (CENTRAL). The original search was conducted on 3 May 2021 and updated on 9 August 2022, 24 June 2024 and 12 December 2024. The search strategies are described in online supplemental appendices 1–3.

We included all RCTs of children and/or adolescents under 18 years of age, with TBI, in contact with individuals with drug-susceptible TB, regardless of the definition the authors used for TBI. RCTs that compared at least one drug regimen (rifampicin for 3 to 4 months (3R, 4R), isoniazid with rifampicin for 3 to 4 months (3HR, 4HR), isoniazid with rifapentine for 3 months (3HP), isoniazid for 6 months (6H), isoniazid for 9 months (9H), isoniazid for 12 months (12H)) with another drug regimen and/or placebo and/or no treatment were included. Primary outcomes were the incidence of active TB at 2 years of follow-up and treatment adherence. The incidence of active TB at 1 and 5 years of follow-up, bacteriological confirmation of TB within the first 2 years after exposure, adverse reactions (excluding hepatotoxicity), hepatotoxicity, discontinuation of treatment due to adverse events and mortality at 5 years of follow-up were included as secondary outcomes.

Active TB was defined as the disease caused by being infected with M. tuberculosis,¹⁴ confirmed bacteriologically or diagnosed clinically based on the TB diagnostic criteria of the American Thoracic Society/Infectious Diseases Society of America/Centers for Disease Control and Prevention clinical practice guidelines.¹⁵

Data extraction and risk of bias assessment

Two researchers (VS-J, YHM) independently screened the titles/abstracts and full-text articles. Data extraction from the studies meeting all inclusion criteria was carried out independently by the researchers (VS-J, YHM, AFE-B), using a standardised and pre-established format in Microsoft Excel. Disagreements in screening, full-text and/or data extraction were resolved by consensus discussion or by seeking adjudication by a third reviewer (JA-R). Authors of included studies were contacted to request clarification and/or additional data, if missing. The Cochrane Risk of Bias tool, V.2 (RoB2),¹⁶ was used to assess the risk of bias (RoB) for each outcome of the included RCTs. Two of the reviewers (VS-J, YHM, AFE-B) independently performed the RoB2 assessment for each study. Discrepancies between reviewers regarding RoB assessment were resolved by consensus or by seeking adjudication by a third reviewer (IDF).

Statistical analysis

The network geometries were plotted for each primary and secondary outcome in R V.4.2. Each geometry node indicates the available interventions, with the size proportional to the number of patients randomised to this intervention. The lines connecting the nodes represent the number of direct comparisons between two interventions, with the thickness of the line proportional to the number of direct comparisons that exist between the interventions included in the RCTs. Eight nodes (6H, 9H, 12H, 3HP, 3HR, 4HR, 4R and NT/PB) comprised the NMAs. No treatment and placebo were combined into one node, given that different treatment options have been implemented for children with TBI since the 1950s.

Pairwise meta-analyses were performed on the available direct comparisons for each outcome. Effect estimates, along with 95% CIs, were calculated using the risk ratio (RR) for dichotomous outcomes. We assessed heterogeneity by calculating the I² statistic to estimate the percentage of variability not explained by sampling error.¹⁷ Publication bias was evaluated only for outcomes with at least 10 studies per direct comparison. We created network plots for each outcome based on the available direct comparisons. We conducted Bayesian NMAs using a random effects model with a Markov-chain Monte Carlo method based on the expected heterogeneity because we considered that the transitivity and consistency assumptions were met.¹⁸ We evaluated transitivity visually and based on previously defined effect modifiers, as well as consistency, using global and local tests in closed loops.¹⁹ We estimated network OR with their 95% CI. Finally, we calculated the surface under the cumulative classification curve (SUCRA) values for each intervention, by outcome. A detailed description of the statistical methods has been provided in our published protocol. We used R (RStudio, V.4.2; R Core Team, 2024) and Stata (V.17.0; StataCorp LP, College Station, Texas) for statistical analyses regarding NMA, and Review Manager V.5.4 for pairwise meta-analyses.

Certainty of the evidence

The confidence in the estimates (also called quality or certainty of the evidence) for each reported outcome was assessed independently by two reviewers (VS-J and IDF), using the Grading of Recommendations, Assessment, Development, and Evaluation (GRADE) approach for direct evidence²⁰ and for the NMA.^{21 22}

Results

Overview of studies

We identified 1887 references from databases and 10 through hand citation searching (figure 1). After title and abstract screening, 108 studies were considered potentially eligible and were reviewed in full text. One citation could not be retrieved.²³ Of these, 98 studies were excluded (online supplemental appendix 4) and 10 publications were included (table 1). One RCT included two study periods with different numbers of participants for each intervention in each period,²⁴ which we analysed as two separate studies. Therefore, we included 11 studies.^24–33

Figure 1. PRISMA 2020 (54) flowchart. PRISMA, Preferred Reporting Items for Systematic Review and Meta-Analysis.

Table 1. Characteristics of included studies (ordered by study identification number).

Author and year	Country	Age	Population type	Immunological tests used	Interventions	Randomised patients included	Follow-up/active TB incidence	Adverse events
Villarino et al25	USA, Canada, Brazil, China, Spain.	Age range 2–17 years.	Not migrants	TST	RPT+INH 12 weekly doses administered through directly observed therapy RPT doses: 10–14 kg: 300 mg. 14.1–25 kg: 450 mg. 25.1–32 kg: 600 mg. 32.1–50 kg: 750 mg.>50kg: 900 mg. INH doses: >12 years: 15 mg/kg. 2–11 years: 25 mg/kg. Daily INH for 9 months (270 doses) dispensed in 30-day batches, self-administered or under direct observation. >12 years: 5 mg/kg. 2–11 years: 10 mg/kg.	RPT+INH 12 weekly doses: 552 INH 9 months: 506	Follow-up 33 months after enrolment. Incidence 1 year: 3HP: 0/471. 9H: 1/434 2 years: 3HP: 0/471. 9H: 1/434	3HP: Attributed to Tx: 11 grade 1–2 events; 3 grade 3 events; no grade 4 events; no grade 5 (death) or serious adverse events. Not attributed to Tx: 25 grade 1–2 events, 3 grade 3 events, 1 grade 4 event, no grade 5 (death) or serious adverse events. 9H: Attributed to the Tx: 5 grade 1–2 events; 1 grade 3 event; no grade 4, grade 5 (death) or serious adverse events. Not attributed to Tx: 35 grade 1–2 events, 5 grade 3 events, 2 grade 4 events, 2 grade 5 (death) events and 7 serious adverse events.
Spyridis et al24 (2007) Period 1	Greece	Age range: 0–15 years.	Migrants	TST	INH+RIF, 10 mg/kg/day (max. dose: 300 mg and 600 mg, respectively) administered once a day before lunch for 4 months INH 10 mg/kg once a day before lunch (max dose: 300 mg) for 9 months	INH+RIF 4 mo: 238 INH 9 months: 232	Follow-up 7–11 years. Incidence 1 year: 4HR: 26/220. 9H: 48/200 2 years: 4HR: 26/220. 9H: 48/200 5 years: 4HR: 26/220. 9H: 48/200	Attributed to Tx. Serious drug-related adverse events were not detected. Nausea and epigastric pain (grade 1–2) were reported by 7 adherent patients in 4HR and 13 adherent patients in 9H.
Spyridis et al24 (2007) Period 2	Greece	Age range 0–15 years.	Migrants	TST	INH+RIF, 10 mg/kg/day (max. dose: 300 mg and 600 mg, respectively) administered once a day before lunch for 4 months INH+RIF, 10 mg/kg/day (max. dose: 300 mg and 600 mg, respectively) administered once a day before lunch for 3 months	INH+RIF 4 mo: 236 INH+RIF 3 mo: 220	Follow-up 3–7 years. Incidence 1 year: 4HR: 30/221. 3HR: 23/209 2 years: 4HR: 30/221. 3HR: 23/209 5 years: 4HR: 30/221. 3HR: 23/209	Attributed to Tx. Serious drug-related adverse events were not detected. Nausea and epigastric pain (grade 1–2) were reported by two adherent patients in 4HR and two adherent patients in 3HR.
Magdorf et al26 (1994)	Germany	Age range: 0–17 years.	Not migrants	TST	RIF 350 mg/m2 of body surface area for 4 months. RIF 350 mg/m2 of body surface area and PZA 30 mg/kg for 2 months. INH 200 mg/m2 of body surface area for 6 months.	RIF 4 months: 50 INH 6 months: 50	Follow-up 2 years. Incidence 1 year: 4R: 0/50. 6H: 0/50 2 years: 4R: 0/50. 6H: 0/50	6H events: hair loss (therapy break for 7 days, without the symptoms reappearing) and slight reversible behavioural disturbances (therapy continued. Symptoms disappeared spontaneously).
Egsmose et al27 (1965)*	Kenya	Age range: <6–14 years.	Not migrants	TST	INH 5–10 mg/kg (daily average dose) for 12 months. <10 kg = 50 mg/day. 10–19 kg=100 mg/day. 20–39 kg=200 mg/day. 40–59kg=300 mg/day. 60–69 kg=400 mg/day. >70 kg = 500 mg/day The placebo tablets and INH tablets were identical for researchers and patients. Placebo was distributed according to the same criteria as INH	INH 12 months: 208 Placebo: 198	Follow-up 2–4 years. Incidence 1 year: 12H: 7/208. Placebo: 10/198 2 years: 12H: 7/208. Placebo: 12/198	14 deaths occurred (7 in INH and 7 in control). In the control group, 4 contacts died before any follow-up was done; the cause of death was not ascribed to TB and these 4 are excluded from the analysis. None of the remaining 10 contacts had pulmonary lesions or were excreting tubercle bacilli at previous examinations, and it was not possible in any case to assign the cause of death to TB. Other contacts left the study through migration, death, ceased to cooperate, but results are mixed from children and adults.
Diallo et al28 (2018)	Australia, Benin, Brazil, Canada, Ghana, Guinea, Indonesia	Age range 0–17 years.	Not migrants	TST or IGRA (QFT or T-Spot)	RIF 10–20 mg/kg/day for 4 months (administered by participant/caretaker) <3 kg: 75 mg. 3–4.9 kg: 75 mg. 5–7.4 kg: 75 mg. 7.5–9.9 kg: 150 mg. 10–14.9 kg: 225 mg. 15–19.9 kg: 300 mg. 20–24.9 kg: 375 mg. 25–29.9 kg: 450 mg. >30: 600 mg. INH 10–15 mg/kg/day for 9 months <3kg: 50 mg. 3–4.9 kg: 50 mg. 5–7.4 kg: 75 mg. 7.5–9.9 kg: 100 mg. 10–14.9 kg: 150 mg. 15–19.9 kg: 200 mg. 20–24.9 kg: 250 mg. 25–29.9 kg: 300 mg. >30: 300 mg.	RIF 4 months: 428 INH 9 months: 416	Follow-up 16 months after randomisation.	Authors assessed whether the event was attributed to Tx if a trial drug was discontinued. No adverse events resulting in permanent discontinuation of a trial drug occurred in either group. We report minor adverse events (grade 1 according to supplemental material; did not result in the discontinuation of a trial drug): fever or night sweats, weight loss, sputum, cough, skin problems, gastrointestinal problems and neurologic problems.
Comstock et al29 *	USA (Alaska)	Age range of population: 0–14 years.	Not migrants	TST	INH 4–8 mg/kg/day (average dose range) for 12 months (approximately 5 mg/kg/day of INH). Participants were advised to take the total daily dose at one time, usually with the morning meal. Placebo tablets were similar in size, shape, hardness, gloss and taste. Participants were advised to take the total daily dose at one time, usually with the morning meal.	INH 12 months: 1544 Placebo: 1529	Follow-up 43–76 months (69.3 median). Incidence 1 year: 12H: 2/1544. Placebo: 6/1529	12H: 2 deaths among small children, who had access to the medication at home, taking a fatal dose of isoniazid. 2 children went to the ER with severe convulsions but recovered after treatment. 1 child had an excessive dose of isoniazid, no convulsions occurred, but was excessively sleepy the next day. Other data is mixed for children and adults.
Biraro et al30 (2015)*	Uganda	Age range of population: 6–18 years.	Not migrants	TST or IGRA (QFT).	INH 5 mg/kg, up to a max. of 300 mg, plus pyridoxine 25 mg daily for 6 months, self-administered. No treatment.	INH 6 months 7 No treatment: 6	Follow-up 6 months.	There were no documented side effects of isoniazid.
Bush et al31 (1965)*	Japan	Age range of population: 2 months to15 years.	Not migrants	TST	INH 12 months, 100 mg tablets. 2–18 months: ½ tablet. 19 months to 6 years: 1 tablet. 7–15 years: 1½ tablets. >16 years: 2½ tablets. Placebo: tablets identical in appearance to those of INH. Placebo was distributed according to the same criteria as INH, for 12 months.	INH 12 months: 479 Placebo: 450	Follow-up 1 year. Incidence 1 year: 12H: 4/479. Placebo: 4/450
Ferebee and Mount32 (1962)*	USA (39 communities)	Age range of population: 2 months to 20 years.	Not migrants	TST	INH 12 months, 100 mg tablets, 4–7 mg/kg dose range (majority received 5 mg/kg) 2–18 months: ½ tablet. 19 months to 6 years: 1 tablet. 7–15 years: 2 tablets.>16 years: 3 tablets. Placebo: tablets identical in appearance to those of INH. Placebo was distributed according to the same criteria as INH, for 12 months.	INH 12 months: 8076 Placebo: 8222	Follow-up 1 year. Incidence 1 year: 12H: 26/8076. Placebo: 61/8222	During medication year persons assigned to both groups died of nontuberculous causes (data mixed for children and adults). 1 death is directly attributable to isoniazid overdose and died from the acute toxic effects. Among the reasons for individuals discontinuing the pills are adverse events, but data are mixed for children and adults.
Martínez Alfaro et al33 (1998)*	Spain (Albacete)	Age range of population: < 20 years.	Not migrants	TST	INH+RIF for 3 months, 10 mg/kg/day (max. dose 300 mg/day of INH and 600 mg/day of RIF). Doses were adjusted by weight in children. INH for 9 months (5 mg/kg/day) Max. 300 mg/day of INH. Doses were adjusted by weight in children.	INH+RIF 3 months: 15 INH 9 months: 13	Follow-up INH+RIF 3 months: 19±11 months INH 9 months: 16±10 months Incidence 1 year: 3HR: 1/15. 9H: 0/13	No serious adverse events were reported (grade 4 and 5). No adverse events were observed in 23 patients under 20 years of age. Mild adverse events: 2 in the 3HP group; 3 in the 9H group. Including elevated GPT transaminases (<5 times the normal value), gastrointestinal symptoms, headache, neuralgia, photosensitivity and thrombocytopenia.

^*From the mixed studies (children and adult populations), only the paediatric population data that could be extracted, was included in the analysis.

AE, adverse events; AR, adverse reactions; BCG, Bacillus Calmette-Guérin; CDC, Centers for Disease Control and Prevention; 6H, isoniazid for 6 months; 9H, isoniazid for 9 months; 12H, isoniazid for 12 months; 3HP, isoniazid + rifapentine for 3 months; 3HR, isoniazid + rifampicin for 3 months; 4HR, isoniazid + rifampicin for 4 months; IGRA, interferon-gamma release assay; INH, Isoniazid; Max, maximum; NA, not applicable; ND, no data; PZA, pyrazinamide; QFT, QuantiFERON-TB Gold; 4R, rifampicin for 4 months; RIF, rifampicin; RPT, rifapentine; TB, tuberculosis; TBI, tuberculosis infection; TST, tuberculin skin testing; TU, tuberculin units; Tx, treatment.

Study characteristics

Of the 11 studies, 5 included both child and adult populations, with data that could be extracted on children.^{27 29 31–33} Four studies evaluated 12H versus placebo,^{27 29 31 32} four studies evaluated 9H,^{24 25 28 33} one study evaluated 3HP²⁵ and no treatment,³⁰ and two studies evaluated 4R,^{26 28} 3HR,^{24 33} 4HR²⁴ and 6H,^{26 30} each. The doses of simultaneously administered TPT (3HP and 3–4 hour) and rifampicin monotherapies were within the established range of interest.¹¹ For isoniazid monotherapy, the approximate dose in most included studies was 5 mg/kg, which is 2 mg/kg lower than the lower limit of the range considered adequate.¹¹ For 3HP, doses were administered weekly through directly observed therapy.²⁵ On the other hand, self-administration of anti-TB drugs was the predominant form in the included studies. One study included participants up to 18 years of age,³⁰ while two studies included participants up to³² or less than 20 years of age.³³ Only two studies included the paediatric migrant population.²⁴ Likewise, three studies did not report the percentage of contacts with intrafamilial infection or household contact origin.^{25 26 33}

Risk of bias

Six studies had a high overall RoB,^{24 25 27 28 33} including three outcomes from the study by Diallo et al,²⁸ while three studies raised some concerns^{26 30 31} and three studies had a low overall RoB.^{28 29 32} This last category included two outcomes from one study.²⁸ A high RoB in the randomisation process and measurement of the outcome domains was found in 43.4% of the studies. However, most of the study outcomes presented a low RoB in the domains of deviations from intended interventions, missing outcome data and selection of reported outcomes, respectively (online supplemental appendix 5). Among the five studies included in the outcome of incidence of active TB at 2 years of follow-up, four had a high overall RoB, and only one raised some concerns (figure 2A). For the seven studies assessing treatment adherence, five had a high RoB, while two raised some concerns (figure 2B). The RoB assessment of the secondary outcomes is detailed in figure 2C–I.

Figure 2. Risk of bias (ROB2) per outcome of included studies. Outcomes: (A) Incidence of active TB at 2 years of follow-up; (B) treatment adherence; (C) incidence of active TB at 1 year of follow-up; (D) incidence of active TB at 5 years of follow-up; (E) bacteriological confirmation of TB within the first 2 years of exposure; (F) adverse reactions (other than hepatotoxicity); (G) hepatotoxicity; (H) discontinuation of treatment due to adverse events and (I) mortality at 5 years of follow-up. Risk of bias domains: D1, randomisation process; D2, deviation from the intended interventions; D3, missing outcome data; D4, measurement of the outcome; D5, selection of the reported result. 6H, isoniazid for 6 months; 9H, isoniazid for 9 months; 12H, isoniazid for 12 months; 3HP, isoniazid + rifapentine for 3 months; 3HR, isoniazid + rifampicin for 3 months; 4HR, isoniazid + rifampicin for 4 months; NT/PB, no treatment/placebo; 4R, rifampicin for 4 months; TB, tuberculosis.

Network geometry

For the primary outcomes, the networks were disconnected and open, while displaying eight intervention nodes (figure 3A,B). For the incidence of active TB at 2 years of follow-up, intervention nodes were displayed in three separate comparisons, while for treatment adherence, they were displayed in two. For secondary outcomes of incidence of active TB at 1 year of follow-up, bacteriological confirmation of TB within the first 2 years after exposure and mortality at 5 years of follow-up, network geometries were disconnected and open (figure 3C,E,I). The network geometries for incidence of active TB at 5 years of follow-up, adverse reactions, hepatotoxicity and discontinuation of treatment were connected, yet open (figure 3D,F–H). NMA was only performed for adverse reaction outcome.

Figure 3. Network geometries per outcomes. Outcomes: (A) Incidence of active TB at 2 years of follow-up; (B) treatment adherence; (C) incidence of active TB at 1 year of follow-up; (D) incidence of active TB at 5 years of follow-up; (E) bacteriological confirmation of TB within the first 2 years of exposure; (F) adverse reactions (other than hepatotoxicity); (G) hepatotoxicity; (H) discontinuation of treatment due to adverse events and (I) mortality at 5 years of follow-up. 6H, isoniazid for 6 months; 9H, isoniazid for 9 months; 12H, isoniazid for 12 months; 3HP, isoniazid + rifapentine for 3 months; 3HR, isoniazid + rifampicin for 3 months; 4HR, isoniazid + rifampicin for 4 months; NT/PB, no treatment/placebo; 4R, rifampicin for 4 months; TB, tuberculosis.

Primary outcomes

Incidence of active TB at 2 years of follow-up

Five studies were included in this outcome.^24–27 The comparisons of 12H versus NT/PB and 4R versus 6H did not connect with the rest of the geometry (figure 3A). As a result, the NMA was not conducted for this outcome. Paired forest plots based on the available direct evidence are presented (online supplemental appendix 6). TPT with the combination therapy 4HR may reduce the number of TB-sensitive child contacts without HIV who develop active TB at 2 years of follow-up by 51% (RR 0.49 (95% CI 0.32 to 0.76)) compared with 9H (low certainty, online supplemental appendix 7). For the comparison of 3HP versus 9H, we found they may not be different (RR 0.31 (95% CI 0.01 to 7.52)) although the evidence is uncertain (very low certainty, online supplemental appendix 8). Similarly, regimens 4HR and 3HR (RR 1.23 (95% CI 0.74 to 2.05)) (very low certainty, online supplemental appendix 9) and 12H and NT/PB (RR 0.56 (95% CI 0.22 to 1.38)) (low certainty, online supplemental appendix 10) may not be different in terms of developing active TB at 2 years of follow-up. The effect for the comparison 4R versus 6H could not be estimated, as no events were reported in either group.

Treatment adherence

Seven studies were included in this outcome.^{24–26 28 31 33} Despite including all eight treatment regimens for the outcome, the NMA was not conducted due to disconnected and open network geometry (figure 3B). Direct evidence was graphically represented using paired forest plots (online supplemental appendix 11). Adherence to treatment in children who received 3HP may be slightly higher (RR 1.09 (95% CI 1.03 to 1.15)) compared with 9H (low certainty, online supplemental appendix 8). Likewise, the 4HR (RR 1.07 (95% CI 1.01 to 1.14)) and 4R (RR 1.12 (95% CI 1.05 to 1.20)) regimens likely increase the adherence compared with 9H, respectively (moderate certainty, online supplemental appendices 7 and 12, respectively). Finally, the evidence suggests that there might be little to no difference in terms of treatment adherence between 3HR and 9H (RR 0.94 (95% CI 0.78 to 1.13)) (low certainty, online supplemental appendix 13), 4HR and 3HR (RR 0.99 (95% CI 0.94 to 1.03)) (low certainty, online supplemental appendix 9), 4R and 6H (RR 1.09 (95% CI 0.96 to 1.25)) (low certainty, online supplemental appendix 14), and between 12H and NT/PB (RR 1.02 (95% CI 0.99 to 1.04)) (low certainty, online supplemental appendix 10).

Incidence of active TB at 1 year of follow-up

Nine studies were included in this outcome,^{24–27 29 31–33} four of which evaluated 12H versus NT/PB. NMA was not conducted due to disconnected and open network geometry. Forest plots of the direct comparisons are presented in online supplemental appendix 15. Two studies were excluded from this outcome analysis due to only a 6-month follow-up³⁰ and the unspecified time of diagnosis of two TB cases in the 9H group within the 16-month follow-up period.²⁸ One case occurred in a patient who completed the 9H regimen, while another developed isoniazid-resistant TB (without completing treatment) more than 9 months after discontinuation.²⁸

TPT with 12H likely reduces the number of HIV-negative children in contact with drug-sensitive TB who develop active TB after 1 year of follow-up by 51% (RR 0.49 (95% CI 0.33 to 0.71)) compared with NT/PB (moderate certainty, online supplemental appendix 10). The same occurred with 4HR, which may reduce the number of children who develop active TB after 1 year of follow-up by 51% (RR 0.49 (95% CI 0.32 to 0.76)) compared with 9H (low certainty, online supplemental appendix 7). Finally, regarding the comparisons of 3HP versus 9H, 3HR versus 9H and 4HR versus 3HR, the evidence suggests that there might be little to no differences between them, in terms of active TB at 1 year (online supplemental appendices 8, 9 and 13).

Incidence of active TB at 5 years of follow-up

Two studies were included for this outcome.²⁴ NMA was not conducted due to disconnected and open network geometry. Forest plots of the direct evidence are presented in online supplemental appendix 16. TPT with 4HR may reduce the number of children who develop active TB after 5 years of follow-up by 51% (RR 0.49 (95% CI 0.32 to 0.76)) compared with 9H (low certainty, online supplemental appendix 7). The evidence suggests that there might be little to no differences in terms of active TB at 5 years between 4HR and 3HR (RR 1.23 (95% CI 0.74 to 2.05)) (low certainty, online supplemental appendix 9). Another study²⁷ was not included in the outcome analysis due to its follow-up period of 2 to 4 years.

Bacteriological confirmation of TB within the first 2 years after exposure

Five studies^{25 27–29 32} and 3 comparisons reported this secondary outcome. NMA was not conducted due to disconnected and open network geometry. Forest plots of the available direct evidence are presented (online supplemental appendix 17). TPT with the 12H regimen is likely to reduce the number of children who develop bacteriologically confirmed TB within the first 2 years after exposure (RR 0.41 (95% CI 0.27 to 0.63)) compared with NT/PB (moderate certainty, online supplemental appendix 10). No differences in this outcome are apparent between shorter regimens of 3HP (online supplemental appendix 8) or 4R (online supplemental appendix 12) and 9H.

Adverse reactions

Ten studies were included^{24–30 32 33} and due to the connected comparisons, this was the only outcome in which a full NMA was conducted. Direct evidence is presented in online supplemental appendix 18, while NMA is shown in tables 2 and 3. According to direct evidence, TPT with the 4R regimen is likely to reduce the number of children who experience adverse reactions (RR 0.37 (95% CI 0.23 to 0.61)) compared with 9H (moderate certainty, online supplemental appendix 12). Six comparisons may have little to no differences in adverse reactions (online supplemental appendices 7–10, 13 and 14). The comparison of 6H vs NT/PB did not yield an effect estimate due to the absence of events (online supplemental appendix 19).

Table 2. NMA and GRADE certainty of evidence on adverse reactions.

Treatment comparisons	Direct effect estimate	Number of studies	GRADE direct evidence*	Indirect effect estimate†	GRADE indirect evidence	NMA effect estimate	GRADE NMA without inconsistency	Final GRADE NMA
12H vs 3HP	–	0	NA	3.86 (0.015–993.94)7 12	Low	3.86 (0.015–993.94)	Low	Low
12H vs 3HR	–	0	NA	16.50 (0.05–4725.46)7 14 17	Very low	16.50 (0.05–4725.46)	Very low	Very low
12H vs 4HR	–	0	NA	17.65 (0.069–4483.38)7 14 21 21	Low	17.65 (0.069–4483.38)	Low	Low
12H vs 4R	–	0	NA	24.63 (0.107–5670.35)7 23 24	Low	24.63 (0.107–5670.35)	Low	Low
12H vs 6H	–	0	NA	4.73 (0.05–423.86)7 23 27	Low	4.73 (0.05–423.86)	Low	Low
12H vs 9H	–	0	NA	8.36 (0.035–1977.16)7 12 17 21 24	Low	8.36 (0.035–1977.16)	Low	Low
12H vs NT/PB	4.00 (0.61–26.35)	3	Lowa,e	–	NA	4.00 (0.61–26.35)	Low	Low
3HP vs 3HR	–	0	NA	4.26 (0.74–24.59)12 14 18	Low	4.26 (0.74–24.59)	Low	Low
3HP vs 4HR	–	0	NA	4.56 (1.22–16.96)12 14 21	Very low	4.56 (1.22–16.96)	Very low	Very lowf
3HP vs 4R	–	0	NA	6.37 (2.11–19.19)12 23 24	Very low	6.37 (2.11–19.19)	Very low	Very lowf
3HP vs 6H	–	0	NA	1.22 (0.04–31.70)12 23 27	Low	1.22 (0.04–31.70)	Low	Low
3HP vs 9H	2.16 (0.83–5.68)	1	Very lowa,e	–	NA	2.16 (0.83–5.68)	Very low	Very low
3HP vs NT/PB	–	0	NA	1.03 (0.005–191.50)7 27	Very low	1.03 (0.005–191.50)	Very low	Very low
3HR vs 4HR	1.06 (0.15–7.58)	1	Very lowa,e	1.084 (0.12–9.62)14 17 21	Very low	1.07 (0.25–4.62)	Very low‡	Very low
3HR vs 4R	–	0	NA	1.49 (0.31–7.08)14 17 23 24	Very low	1.49 (0.31–7.08)	Very low	Very low
3HR vs 6H	–	0	NA	0.28 (0.009–8.89)14 17 23 27	Very low	0.28 (0.009–8.89)	Very low	Very low
3HR vs 9H	0.51 (0.07–3.68)	1	Lowa,e	0.50 (0.05–4.43)12 14 17 21 24	Low	0.51 (0.12–2.19)	Low‡	Low
3HR vs NT/PB	–	0	NA	0.24 (0.001–50.30)7 14 17 27	Low	0.24 (0.001–50.30)	Low	Low
4HR vs 4R	–	0	NA	1.39 (0.49–3.94)14 17 23 24	Very low	1.39 (0.49–3.94)	Very low	Very low
4HR vs 6H	–	0	NA	0.26 (0.01–6.79)14 20 21 23 27	Very low	0.26 (0.01–6.79)	Very low	Very low
4HR vs 9H	0.47 (0.18–1.21)	1	Very lowa,e	0.48 (0.02–7.85)12 14 17 21 24	Very low	0.47 (0.19–1.15)	Very low‡	Very low
4HR vs NT/PB	–	0	NA	0.22 (0.001–41.38)7 14 21 27	Low	0.22 (0.001–41.38)	Low	Low
4R vs 6H	0.19 (0.01–4.10)	1	Lowa,e	–	NA	0.19 (0.01–4.10)	Low	Low
4R vs 9H	0.34 (0.20–0.58)	1	Moderatea	–	NA	0.34 (0.20–0.58)	Moderate	Moderate
4R vs NT/PB	–	0	NA	0.16 (0.00–26.71)7 23 24 27	Low	0.16 (0.00–26.71)	Low	Low
6H vs 9H	–	0	NA	1.76 (0.07–39.57)12 17 21 23 24 27	Very low	1.76 (0.07–39.57)	Very low	Very low
6H vs NT/PB	0.85 (0.01–50.10)	1	Moderatea	–	NA	0.85 (0.01–50.10)	Moderate	Moderate
NT/PB vs 9H	–	0	NA	2.089 (0.012–353.18)7 12 17 21 24 27	Low	2.089 (0.012–353.18)	Low	Low

^*Reasons for downgrading certainty of evidence in superscript: a. risk of bias, b. inconsistency, c. indirect evidence, d. publication bias, e. imprecision, f. intransitivity, g. incoherence.

^†Superscripts describe the direct comparisons that inform the indirect result.

^‡Inconsistency p value=0.9867.

GRADE, Grading of Recommendations, Assessment, Development, and Evaluation; 6H, isoniazid for 6 months; 9H, isoniazid for 9 months; 12H, isoniazid for 12 months; 3HP, isoniazid + rifapentine for 3 months; 3HR, isoniazid + rifampicin for 3 months; 4HR, isoniazid + rifampicin for 4 months; NA, not applicable; NMA, network meta-analysis; NT/PB, no treatment/placebo; 4R, rifampicin for 4 months.

Table 3. League table of adverse reactions.

12H	.	.	.	.	.	.	4.00 (0.61–26.35)
3.87 (0.02–993.94)	3HP	.	.	.	.	2.16 (0.83–5.68)	.
16.50 (0.06–4725.47)	4.27 (0.74–24.60)	3HR	1.06 (0.15–7.58)	.	.	0.51 (0.07–3.68)	.
17.66 (0.07–4483.38)	4.57 (1.23–16.97)	1.07 (0.25–4.62)	4HR	.	.	0.47 (0.18–1.21)	.
24.64 (0.11–5670.36)	6.37 (2.12–19.20)	1.49 (0.31–7.08)	1.40 (0.49–3.94)	4R	0.19 (0.01–4.10)	0.34 (0.20–0.58)	.
4.73 (0.05–423.86)	1.22 (0.05–31.71)	0.29 (0.01–8.90)	0.27 (0.01–6.80)	0.19 (0.01–4.10)	6H	.	0.85 (0.01–50.10)
8.37 (0.04–1977.17)	2.16 (0.83–5.68)	0.51 (0.12–2.19)	0.47 (0.19–1.15)	0.34 (0.20–0.58)	1.77 (0.08–39.58)	9H	.
4.00 (0.61–26.35)	1.04 (0.01–191.50)	0.24 (0.00–50.31)	0.23 (0.00–41.38)	0.16 (0.00–26.71)	0.85 (0.01–50.10)	0.48 (0.00–80.87)	NT/PB

The estimate for effectiveness is located at the intersection of the intervention defining the column, and the intervention defining the row. Results are presented as OR with the 95% credibility interval (CI) between the intervention and the common comparator (9H) in the NMA. For adverse reactions, an OR below 1.0 favours the treatment defining the column. Statistically significant results are bolded and underlined. Cell shading represents the GRADE certainty of evidence assessment.

Colour: ■ high certainty of evidence; ■ moderate certainty of evidence; ■ low certainty of evidence; ■ very low certainty of evidence.

6H, isoniazid for 6 months; 9H, isoniazid for 9 months; 12H, isoniazid for 12 months; 3HP, isoniazid + rifapentine for 3 months; 3HR, isoniazid + rifampicin for 3 months; 4HR, isoniazid + rifampicin for 4 months; NMA, network meta-analysis; NT/PB, no treatment/placebo; 4R, rifampicin for 4 months.

According to indirect evidence, 3HP may result in more adverse reactions compared with the 4HR (OR 4.56 (95% CI 1.22 to 16.96)) (very low certainty), and to 4R (OR 6.37 (95% CI 2.11 to 19.19)) regimen (very low certainty) (tables 2 and 3). Notably, adverse reaction data for 4HR are derived from studies of patients with treatment adherence, as there is no data on potential events in patients that did not accomplish the adherence definition, which reduces the certainty of the evidence. Moreover, no events were reported from a study that directly evaluated the 4R regimen (vs 6H)²⁶ (online supplemental appendix 18). In a second study that evaluated 4R directly (vs 9H),²⁸ it was unclear if all minor adverse events that did not lead to treatment discontinuation were attributable to the treatment (online supplemental appendix 18). Due to these factors, intransitivity was introduced, affecting the certainty of evidence in this indirect comparison.

The NMA evidence suggests that adverse reactions are less likely to occur with 4R (OR 0.34 (95% CI 0.20 to 0.58)) compared with the 9H (moderate certainty, tables 2 and 3). No statistically significant differences were observed for the remaining indirect comparisons. For this outcome, the 4R regimen has the highest SUCRA value at 0.84, followed by 4HR (0.72), 3HR (0.69), 9H (0.45), NT/PB (0.44), 6H (0.39), 3HP (0.25), and 12H (0.18), respectively.

Hepatotoxicity

No events were reported in four out of the five included studies^{25 26 28 30 33} for 3HP vs 9H, 3HR vs 9H, 4R vs 9H and 6H vs NT/PB. NMA was not conducted for this outcome. Evidence suggests there might not be differences between 4R vs 6H in terms of hepatotoxicity (online supplemental appendix 20).

Discontinuation of treatment due to adverse events

Seven studies reported this outcome.^{24–26 28 30 33} NMA was not performed since no events were reported in five out of the seven included studies. Direct evidence suggests that two comparisons, 3HP vs 9H and 3HR vs 9H, may not differ in treatment discontinuation due to adverse events (online supplemental appendix 21). The certainty of the evidence is presented in online supplemental appendices 7–9, 12–14 and 19.

Mortality at 5 years of follow-up

Three studies were included with three different comparisons: 12H vs NT/PB, 4HR vs 9H, and 4HR vs 3HR (online supplemental appendix 22). No events were reported in the comparisons of 4HR vs 9H and 4HR vs 3HR, making risk estimation for these two comparisons unfeasible. Regarding the comparison of 12H vs NT/PB for mortality, there may not be differences between the interventions (RR 4.95 (95% CI 0.24 to 103.05)) (very low certainty, online supplemental appendix 10).

Sensitivity and subgroup analyses

Risk of bias

In general, there was only one study per direct comparison for each outcome of interest. Excluding studies with a high RoB would further limit the network geometries. Thus, no sensitivity analysis for RoB was performed.

Children under 5 years

NMA was not performed due to the limited number of comparisons and the low or absent number of reported events; therefore, direct evidence is presented in online supplemental appendices 23–29.

Study year ranges (year the recruitment began)

For the incidence of active TB at 2 years of follow-up, three studies with initial recruitment years between 1995 and 2001 were included.^{24 25} Direct evidence is presented in online supplemental appendix 30. Regarding the NMA, the network geometry was connected with an open loop (online supplemental appendix 31). TPT with 3HR may reduce the number of children who develop active TB at 2 years of follow-up (OR 0.33 (95% CI 0.15 to 0.73)), compared with 9H (low certainty, tables 4 and 5). Similarly, the 4HR regimen reduces the incidence of active TB at 2 years of follow-up by 58% (OR 0.42 (95% CI 0.25 to 0.72)), compared with 9H (low certainty, tables 4 and 5). The 3HR regimen had a SUCRA of 0.75, followed by 3HP with 0.62, 4HR with 0.54 and 9H with 0.07, respectively.

Table 4. NMA and GRADE certainty of evidence on incidence of active TB at 2 years of follow-up by study year (sensitivity analysis).

Treatment comparisons	Direct effect estimate	Number of studies	GRADE direct evidence*	Indirect effect estimate†	GRADE indirect evidence	NMA effect estimate	GRADE NMA without inconsistency‡	Final GRADE NMA
3HP vs 3HR	–	0	NA	0.917 (0.03–24.79)3 4	Low	0.917 (0.03–24.79)	Low	Low
3HP vs 4HR	–	0	NA	0.72 (0.028–18.54)3 4 6	Low	0.72 (0.028–18.54)	Low	Low
3HP vs 9H	0.31 (0.01–7.54)	1	Very lowa,e	–	NA	0.31 (0.01–7.54)	Very low	Very low
3HR vs 4HR	0.79 (0.44–1.41)	1	Very lowa,e	–	NA	0.79 (0.44–1.41)	Very low	Very low
3HR vs 9H	–	0	NA	0.33 (0.15–0.73)3 4 6	Low	0.33 (0.15–0.73)	Low	Low
4HR vs 9 hour	0.42 (0.25–0.72)	1	Lowa	–	NA	0.42 (0.25–0.72)	Low	Low

^*Reasons for downgrading certainty of evidence in superscript: a. risk of bias, b. inconsistency, c. indirect evidence, d. publication bias, e. imprecision, f. intransitivity, g. incoherence.

^†Superscripts describe the direct comparisons that inform the indirect result.

^‡Inconsistency p value: all loops were open therefore, inconsistency analysis were not applicable.

GRADE, Grading of Recommendations, Assessment, Development, and Evaluation; 9H, isoniazid for 9 months; 3HP, isoniazid + rifapentine for 3 months; 3HR, isoniazid + rifampicin for 3 months; 4HR, isoniazid + rifampicin for 4 months; NA, not applicable; NMA, network meta-analysis; TB, tuberculosis.

Table 5. League table of incidence of active TB at 2 years of follow-up by study year (sensitivity analysis).

3HP	.	.	0.31 (0.01–7.54)
0.92 (0.03–24.79)	3HR	0.79 (0.44–1.41)	.
0.72 (0.03–18.54)	0.79 (0.44–1.41)	4HR	0.42 (0.25–0.72)
0.31 (0.01–7.54)	0.33 (0.15–0.73)	0.42 (0.25–0.72)	9H

The estimate for effectiveness is located at the intersection of the intervention defining the column, and the intervention defining the row. Results are presented as OR with the 95% credibility interval (CI) between the intervention and the common comparator (9H) in the NMA. For incidence of active tuberculosis at 2 years of follow-up, an OR below 1.0 favours the treatment defining the column. Statistically significant results are bolded and underlined. Cell shading represents the GRADE certainty of evidence assessment.

Colour: ■ high certainty of evidence; ■ moderate certainty of evidence; ■ low certainty of evidence; ■ very low certainty of evidence.

GRADE, Grading of Recommendations, Assessment, Development, and Evaluation; 9H, isoniazid for 9 months; 3HP, isoniazid + rifapentine for 3 months; 3HR, isoniazid + rifampicin for 3 months; 4HR, isoniazid + rifampicin for 4 months; NMA, network meta-analysis; TB, tuberculosis.

For treatment adherence, six studies were included,^{24–26 28 33} with their first years of recruitment from 1989 to 2011. The network geometry was connected with an open loop (online supplemental appendix 32). Despite a trend favouring shorter regimens (3HP, 3HR, 4HR and 4R), no statistically significant differences were observed among the comparisons made for the NMA (low to very low certainty, tables 6 and 7). According to direct evidence, the adherence to 3HP, 4HR and 4R may have a small increase compared with 9H (online supplemental appendix 33). Finally, the 4R regimen had a SUCRA of 0.68, followed by 3HP with 0.62, 3HR with 0.59, 4HR with 0.58, 9H with 0.26 and 6H with 0.25.

Table 6. NMA and GRADE certainty of evidence on treatment adherence by study year (sensitivity analysis).

Treatment comparisons	Direct effect estimate	Number of studies	GRADE direct evidence*	Indirect effect estimate†	GRADE indirect evidence	NMA effect estimate	GRADE NMA without inconsistency	Final GRADE NMA
3HP vs 3HR	–	0	NA	1.02 (0.16–6.47)5 6 8	Low	1.02 (0.16–6.47)	Low	Low
3HP vs 4HR	–	0	NA	1.06 (0.23–4.82)5 6 12	Low	1.06 (0.23–4.82)	Low	Low
3HP vs 4R	–	0	NA	0.92 (0.21–4.01)5 13 14	Low	0.92 (0.21–4.01)	Low	Low
3HP vs 6H	–	0	NA	2.35 (0.24–22.55)5 13	Very Low	2.35 (0.24–22.55)	Very low	Very low
3HP vs 9H	1.75 (0.62–4.95)	1	Lowa	–	NA	1.75 (0.62–4.95)	Low	Low
3HR vs 4HR	1.29 (0.37–4.54)	1	Lowa,e	0.183 (0.004–6.78)6 9 12	Low	1.04 (0.32–3.43)	Low‡	Low
3HR vs 4R	–	0	NA	0.903 (0.14–5.71)6 9 13 14	Low	0.903 (0.14–5.71)	Low	Low
3HR vs 6H	–	0	NA	2.30 (0.18–28.61)6 9 13	Very Low	2.30 (0.18–28.61)	Very low	Very low
3HR vs 9H	0.36 (0.01–11.01)	1	Lowa,e	2.52 (0.45–13.84)6 9 12 14	Low	1.71 (0.37–7.87)	Low‡	Low
4HR vs 4R	–	0	NA	0.86 (0.19–3.90)6 12–14	Low	0.86 (0.19–3.90)	Low	Low
4HR vs 6H	–	0	NA	2.20 (0.22–21.67)6 12 13	Very Low	2.20 (0.22–21.67)	Very low	Very low
4HR vs 9H	1.96 (0.62–6.16)	1	Lowa	0.27 (0.007–10.68)5 6 9 12 14	Low	1.64 (0.55–4.90)	Low‡	Low
4R vs 6H	2.55 (0.46–14.19)	1	Lowa,e	–	NA	2.55 (0.46–14.19)	Low	Low
4R vs 9H	1.90 (0.67–5.35)	1	Lowa	–	NA	1.90 (0.67–5.35)	Low	Low
6H vs 9H	–	0	NA	0.74 (0.10–5.52)5 9 12–14	Low	0.74 (0.10–5.52)	Low	Low

^*Reasons for downgrading certainty of evidence in superscript: a. risk of bias, b. inconsistency, c. indirect evidence, d. publication bias, e. imprecision, f. intransitivity, g. incoherence.

^†Superscripts describe the direct comparisons that inform the indirect result.

^‡Inconsistency p value=0.3173.

GRADE, Grading of Recommendations, Assessment, Development, and Evaluation; 6H, isoniazid for 6 months; 9H, isoniazid for 9 months; 3HP, isoniazid + rifapentine for 3 months; 3HR, isoniazid + rifampicin for 3 months; 4HR, isoniazid + rifampicin for 4 months; NA, not applicable; NMA, network meta-analysis ; 4R, rifampicin for 4 months.

Table 7. League table of treatment adherence by study year (sensitivity analysis).

3HP	.	.	.	.	1.75 (0.62–4.95)
1.02 (0.16–6.48)	3HR	1.29 (0.37–4.54)	.	.	0.36 (0.01–11.01)
1.07 (0.24–4.83)	1.04 (0.32–3.43)	4HR	.	.	1.96 (0.62–6.16)
0.92 (0.21–4.01)	0.90 (0.14–5.71)	0.87 (0.19–3.91)	4R	2.55 (0.46–14.19)	1.90 (0.67–5.35)
2.36 (0.25–22.55)	2.30 (0.19–28.61)	2.21 (0.22–21.67)	2.55 (0.46–14.19)	6H	.
1.75 (0.62–4.95)	1.71 (0.37–7.87)	1.64 (0.55–4.90)	1.90 (0.67–5.35)	0.74 (0.10–5.52)	9H

The estimate for effectiveness is located at the intersection of the intervention defining the column and the intervention defining the row. Results are presented as OR with the 95% credibility interval (CI) between the intervention and the common comparator (9H) in the NMA. For treatment adherence, an OR below 1.0 favours the treatment defined in the column. Statistically significant results are bolded and underlined. Cell shading represents the GRADE certainty of evidence assessment.

Color: ■ high certainty of evidence; ■ moderate certainty of evidence; ■ low lcertainty of evidence; ■ very low certainty of evidence.

GRADE, Grading of Recommendations, Assessment, Development, and Evaluation; 6H, isoniazid for 6 months; 9H, isoniazid for 9 months; 3HP, isoniazid + rifapentine for 3 months; 3HR, isoniazid + rifampicin for 3 months; 4HR, isoniazid + rifampicin for 4 months; NMA, network meta-analysis; 4R, rifampicin for 4 months.

Publication bias

The adverse reaction was the only outcome that included 10 studies. A funnel plot was visually assessed for the three studies evaluating 12H and NT/PB within this outcome (online supplemental appendix 34), though interpretation is limited due to the small number of studies. For all other outcomes, publication bias could not be assessed using a funnel plot, as none had direct comparisons with at least 10 available studies (online supplemental appendix 35).

Discussion

To our knowledge, our study is the first rigorous NMA evaluating all available TPT regimens in children and adolescents in contact with drug-sensitive TB. We identified limited high-certainty evidence comparing the regimens in HIV-negative patients. Additionally, the overall certainty of the evidence was low to very low in most of the cases. Finally, due to few treatment comparisons and sparse events, NMA was only possible for adverse reactions and sensitivity analyses of primary outcomes.

Studies from the 1960s included a large number of paediatric participants, as they were classified as an at-risk population, and compared the 12H regimen with a placebo.^{27 29 31 32} From the 1990s onwards, shorter combination regimens have been compared with isoniazid monotherapies (6H and 9H)^{26 33} and continue to be studied.^{24 25 28 30} No RCTs in children have compared different durations of isoniazid monotherapy (6H, 9H and 12H) against each other, limiting the ability to establish a closed network geometry to conclude on the best treatment duration of this intervention.

The evidence suggests that 3HR and 4HR therapies may reduce the risk of active TB at 2 years of follow-up compared with 9H, a finding corroborated in the NMA through a sensitivity analysis by study year. Treating child contacts of TB is crucial even after the first year of exposure, as the incidence rate of infection in children after 1-year of follow-up remains high, indicating a persistent significant risk of developing active TB after 1 year, according to several studies including a systematic review and meta-analysis.³⁴

Likewise, the evidence suggests that short-course regimens, such as 3HP, 4HR and 4R, are more likely to be adhered to than long-course regimens, such as 9H. This finding is crucial because long treatments are recognised as barriers to treatment completion, and the lack of adherence may result in treatment failure, disease progression, complications and ongoing transmission of infection among those who do not adhere.³⁵ Furthermore, these findings informed our recent Colombian guidelines on TPT in children.³⁶ The guideline panel considered that although the evidence was of low certainty, the benefits seem to outweigh the harms and recommended short over long regimens.³⁶ Other guidelines might benefit from this synthesis to inform further recommendations in children.

For the secondary outcomes, direct evidence suggests that the 4HR may be superior to 9H in terms of active TB at 1 and 5 years of follow-up. Likewise, with moderate-certainty evidence, 12H likely reduces the number of active TB cases at 1 year of follow-up and bacteriological confirmation of TB within the first 2 years compared with NT/PB. Due to this and its impact on effectiveness, 12H monotherapy was introduced as an antituberculosis drug in the 1950s and was the first intervention recommended by the 1965 US guidelines for the treatment of TBI.³⁷

Hepatotoxicity is a highly relevant adverse outcome, considering that liver toxicity can lead to the interruption of TPT.³⁸ Likewise, the simultaneous use of medications over a prolonged period can contribute to hepatotoxicity.³⁸ Despite this, in the present study, no significant differences were observed among the available comparisons for hepatotoxicity, discontinuation of treatment due to adverse events and mortality at 5 years of follow-up, as no estimators were obtained or only a few events were reported. This may be explained by the fact that antituberculous drug-induced hepatotoxicity is considerably lower in children than in adults.³⁹ However, as monitoring of hepatotoxicity was mainly clinical, hepatic tests were performed only if a hepatic adverse event was suspected, the lack of laboratory monitoring for every participant could be considered a limitation. Our previous studies with children taking isoniazid or rifampicin and periodically tested for monitoring hepatotoxicity support the low hepatic adverse events of these medications in children.^{40 41}

Adverse reactions classification varied across studies, with a substantial difference in the number of children who received 12H compared with others, and some interventions reporting no events, leading to very low precision. In our NMA, very low-certainty indirect evidence suggests that adverse reactions are more likely to occur with 3HR compared with 4R, 6H and 12H, and with 6H compared with 9H, though these regimens have not been directly compared in children. One systematic review, including adults and children, found no differences between 3HR and 6H in adverse events that limited treatment.⁵ The higher occurrence of adverse reactions—other than hepatotoxicity—with 3HR may be due to the use of two medications instead of one. Another systematic review reported that extended isoniazid regimens commonly present hepatotoxicity as the most frequent adverse event, while rifamycin-related events tend to be of shorter duration, resolve quickly and are less severe.⁴²

Considering that TBI primarily occurs in childhood and that children are frequently infected within households, those under 5 years old are the ones with the highest risk of developing TB.⁴³ However, most studies did not include data for this subgroup or, due to the limited number of reported events, no significant differences were found. No NMA was conducted for this subgroup, highlighting the need for trials to reduce uncertainty for this specific population.

Evidence in adults with different regimens of TPT has shown that the efficacy and effectiveness of the shorter TPT regimens is comparable to the longer regimens, ranging from 36% to 90% reduction of TB incidences, and adherence can vary but can be even more than 30% higher with shorter regimens like 4R or 3HR.^{44 45} Regarding adverse events in adults, evidence has shown that they are less frequent in children than in adults, being in adults the incidence of any adverse event 3.7% and hepatotoxicity leading to drug discontinuation of 1.1%.⁴⁶

This study presents several limitations. First, this NMA assumes transitivity, which refers to clinical and methodological similarity of intervention comparisons and requires that population, intervention, comparators and outcomes of comparisons are similar enough to estimate indirect comparisons.¹⁸ Although no major differences between groups were observed, significant differences may exist regarding weight or body mass index, percentage of household infection, type of populations, cointerventions and immunological tests. Second, for each direct comparison—except for 12H vs NT/PB—only one study provided data. Additionally, the pyrazinamide regimen was not included due to its proven risk of hepatotoxicity.⁴⁷ Similarly, the incidence of active TB beyond 5 years of follow-up was not considered, as TB cases in this scenario might not be attributable to household contacts. Third, the networks were open, and some were disconnected. As a result, no NMA was conducted for the primary outcomes and six of the secondary outcomes.

The included studies span a wide range of publication years, which could have implications mainly for how TB infection is screened, due to the incorporation of IGRA into this process in recent years. However, only two studies, one from 2015 and another from 2018, incorporated this test into their TB infection identification protocols, using it as a complementary measure rather than as a sole diagnostic test. Therefore, we consider that the effect of the current existence of this diagnostic test did not significantly affect the study.

Among its strengths, this is the first study that summarises the available evidence from trials exclusively on children and adolescents. Considering that several recommendations for TPT in children have been extrapolated from adult evidence,¹⁴ such as the shift from 12H to 9H,³⁷ summarising the results of independent studies in this population is essential to understanding the origin, development and rationale behind these recommendations. Moreover, this study combines and integrates information across interventions that have not been evaluated in individual trials, establishes a ranking of treatments, and improves generalisability. Our findings also contributed to policy changes in Colombia, as this review informed the development of recommendations,³⁶ but can be used to inform recommendations in other contexts.

Regarding research opportunities, there is an urgent need to study the comparative efficacy and safety of new TPT regimens for children under 5 years exposed to TB, including dose finding for long-acting injectable and ultra short-course regimens.^{48 49}

Conclusion

TPT should focus on improving treatment adherence and reducing the incidence of active TB, discontinuation of treatment and mortality while minimising adverse reactions. This study summarised all available data on children, including both short and long-course TPT regimens, despite the limited number of studies evaluating them. Short-course TPT regimens, such as 3HP, 4HR and 4R, may have higher adherence compared with long-course regimens like 9H. Regarding adverse reactions, 4R regimen may reduce the risk of adverse reactions compared with 9H. However, adverse reactions may be more likely to occur with the 3HP regimen compared with 4HR and 4R. Finally, the incidence of active TB at 2 years may be fewer with 3HR and 4HR regimens than with long-course (9H). There is uncertainty around the effects in the subgroups of children younger than 5 years due to the scarcity of data in this population.

The funder of the study had no role in study design, data collection, data analysis, data interpretation, or in the writing of the manuscript. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication.

Data availability statement

Data are available upon reasonable request.