Comparative efficacy and safety of different drugs in patients with systemic juvenile idiopathic arthritis: A systematic review and network meta-analysis

Abstract

Background:

The goal of this study was to estimate the relative efficacy and safety of different biological agents (infliximab, canakinumab, baricitinib, anakinra, adalimumab, tofacitinib, tocilizumab, and rilonacept) compared with placebo for systemic juvenile idiopathic arthritis (JIA) patients, through a network meta-analysis.

Methods:

Pubmed, Embase, and Cochrane Library were searched from database inception to July 2023 for randomized controlled trials comparing different biological agents (infliximab, canakinumab, baricitinib, anakinra, adalimumab, tofacitinib, tocilizumab, and rilonacept) or placebo directly or indirectly in JIA. Bayesian network meta-analyses were conducted. Data was extracted and analyzed by R with gemtc package. The treatment options were ranked using the surface under the cumulative ranking curve (SUCRA) value.

Results:

We identified 10 randomized controlled trials and analyzed 898 participants. Canakinumab (odds ratio 55.0, 95% credible intervals 2.4–67.0) was more effective than the placebo, and the difference was statistically significant. However, there was no statistical significance between other drugs versus placebo in terms of the modified ACRpedi30 (P > .05). The SUCRA shows that canakinumab ranked first (SUCRA, 86.9%), anakinra ranked second (SUCRA, 77.7%), adalimumab ranked third (SUCRA, 61.9%), and placebo ranked the last (SUCRA, 6.3%). Nevertheless, there were no notable discrepancies in the occurrence of adverse events, hepatic-related adverse events, infectious adverse event, serious adverse events, and serious infection following treatment with canakinumab, anakinra, tocilizumab, rilonacept, or the placebo. Based on the clustergram of modified ACRpedi30 and adverse events, canakinumab is suggested for JIA according to the surface under SUCRAs considering the symptom and adverse events simultaneously.

Conclusions:

Among patients with JIA, canakinumab exhibited the highest likelihood of being the optimal treatment for achieving the modified ACRpedi30 response rate, and neither of the tested biological agents carried a significant risk of serious adverse events.

1. Introduction

Systemic juvenile idiopathic arthritis (sJIA) is a juvenile idiopathic arthritis (JIA) subtype uniquely characterized by a combination of systemic features and arthritis.^[1,2] Currently available therapies for sJIA include nonsteroidal anti-inflammatory drugs, glucocorticoids, synthetic disease-modifying antirheumatic drugs, and biologic synthetic disease-modifying antirheumatic drugs.^[3,4] Improved comprehension of the biology behind sJIA has resulted in better patient outcomes through the creation of therapies that target cytokines.^[5] In the case of active JIA, incorporating a biological agent into the initial treatment plan may be necessary.^[6] The role of interleukin-1 (IL-1) is critical in the development of sJIA, including its role in fever induction, activation of endothelial receptors resulting in rashes, and elevation of IL-6 levels in circulation.

Anakinra is an IL-1 receptor antagonist that is non-glycosylated and created through genetic recombination.^[7] Tocilizumab is a monoclonal antibody that humanizes the anti-human IL-6 receptor, which prevents IL-6 signaling.^[8] Canakinumab, on the other hand, is a monoclonal antibody that is fully human and anti-IL-1β, selectively binding to IL-1β and preventing its signaling.^[9] Lastly, rilonacept is a fusion protein that combines the cytokine receptor extracellular domains of both receptor components necessary for IL-1 signaling and is of human origin.^[10]

Several randomized controlled trials (RCTs) have aimed to assess the effectiveness and safety of anakinra (an IL-1 receptor antagonist), canakinumab (an IL-1 antibody), rilonacept (another IL-1 receptor antagonist), and tocilizumab (an IL-6 receptor antibody).^[10–12] However, due to a lack of direct comparison trials, the relative efficacy and safety of these biological agents remain unclear. As there is a lack of direct comparison trials, network meta-analysis (NMA) is a method that can be employed to merge evidence from RCTs of various treatments to determine the impact of 1 treatment in comparison to another.^[13]

The objective of the current study is to utilize NMA to analyze the relative effectiveness and safety of different biological agents (infliximab, canakinumab, baricitinib, anakinra, adalimumab, tofacitinib, tocilizumab, and rilonacept) in patients with JIA.

2. Methods

This systematic review and NMA was performed in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines and the PRISMA Network Meta-Analysis Extension statement.^[14]

2.1. Search strategy

Two authors (BW and YZ) comprehensively searched Pubmed, Embase, and Cochrane Library from database inception to July 2023. A structured search was performed using the following search string: infliximab, canakinumab, baricitinib, anakinra, adalimumab, tofacitinib, tocilizumab, and rilonacept. We refrained from imposing language restrictions on the retrieved articles. The entire process was meticulously supervised by a third author (ZZ), and any discrepancies were adeptly resolved through collaborative discussions. In addition, we will manually search reference lists, related citations, and gray literature from websites. It is important to note that ethics approval was deemed unnecessary for this systematic review and meta-analysis, as it did not entail any direct patient contact.

2.2. Inclusion criteria and exclusion criteria

We considered studies that were eligible for NMA if they met the PICOS criteria (population, intervention, comparator, outcomes, study design). Population: sJIA patients. Intervention: the intervention group received infliximab, canakinumab, baricitinib, anakinra, adalimumab, tofacitinib, tocilizumab, rilonacept, or placebo. Outcome: modified ACRpedi30, adverse events, hepatic-related adverse events, infectious adverse event, serious adverse events, and serious infection. Study design: RCTs. The exclusion criteria encompassed the following categories: case-control studies, animal studies, cadaver studies, single case reports, comments, letters, editorials, protocols, guidelines, publications derived from surgical registries, and review articles. Additionally, clinical studies with fewer than 10 patients were not included.

2.3. Literature selection

All relevant studies which were collected were imported into Endnote X7 (EndNote X7, Thomson Reuters, New York, NY), and then duplicate literatures were excluded. The 2 researchers (JP and LZ) independently screened the literature by reading the titles and abstracts of the studies and excluded irrelevant articles. Studies were screened according to the PICO framework, and irrelevant studies were excluded. A senior reviewer (BW) is consulted in case of disagreement regarding which studies to include.

2.4. Data extraction

The available data was extracted independently from the included studies by 2 reviewers (JP and LZ). Information including the first author, publication year, country, characteristics of participants (such as sample size, number of patients, randomized period, and disease duration), and study design (such as intervention and control method) were independently extracted by 2 reviewers (YW and YZ), and then a cross-check was conducted for these data. The primary outcome included modified ACRpedi30, adverse events, hepatic-related adverse events, infectious adverse event, serious adverse events, and serious infection. Modified ACRpedi30 responses, included an improvement of 30% in at least 3 of the 6 core criteria for sJIA and a worsening of 30 or more in no more than one of the criteria.^[12] If the data were missing or could not be extracted directly, we contacted the corresponding authors to ensure that the information is integrated.

2.5. Quality assessment

Two authors (YW and YZ) independently assessed the methodological quality of the included studies using the Cochrane risk of bias assessment tool, as detailed in the Cochrane Handbook for Systematic Reviews of Interventions.^[15] This tool covers 7 domains: random sequence generation, allocation concealment, blinding of participants/personnel, blinding of outcome assessors, management of incomplete outcome data, selective reporting, and evaluation of other possible sources of bias. Should any disparities arise in the assessments conducted by the 2 reviewers, a third reviewer (BW) will be consulted to facilitate resolution.

2.6. Data analysis and statistical methods

The Bayesian NMA were conducted using R (version 3.5.1, R Foundation for Statistical Computing, Vienna, Austria; https://www.r-project.org/) with the gemtc package, which utilized JAGS (version 4.3.0, Plummer), and involved 20,000 iterations for each of the 4 MCMC chains, with an initial burn-in period of 5000 iterations. Mean differences were calculated using a random-effects model for heterogeneous studies and a fixed-effects model for homogeneous ones. Good homogeneity among all points was determined if their distances from each other fell within 95% of the Limits of Agreement. The node splitting results demonstrated overall consistency, as both direct and indirect comparisons for all outcomes of interest yielded P-values >.05. In cases where heterogeneity was detected, further investigation was carried out, with I² values >50% indicating heterogeneity. For the overall results, we also considered the total I2.pair and I2.cons, with values closer to 0 indicating homogeneity. Furthermore, we calculated the surface under the cumulative ranking curve (SUCRA), where a higher SUCRA value indicated a more favorable result for an individual intervention. A random-effects network meta-regression was performed within a Bayesian hierarchical framework using the gemtc package in R. To assess the presence of small-study effects, we employed a funnel plot using STATA 14.0 (Stata Corp., College Station, TX).

3. Results

3.1. Search results

A total of 1562 relevant studies were collected from databases (PubMed, Embase, and Cochrane Library) based on the search strategies. Additional records were identified through other sources (reference lists, n = 20). We used Endnote Software (Version X7, Thompson Reuters, CA) to remove 1029 duplicate studies. According to the title and abstract, 520 relevant studies were excluded, and then, 4 studies were removed by reading the full text. Finally, 10 studies^{[10–12,16–22]} were included in this meta-analysis in accordance with the inclusion criteria. The PRISMA flow diagram is listed in Figure 1.

Figure 1.

Flow diagram of the literature selection process.

3.2. Study characteristics

The general characteristic of the included studies can be seen in Table 1. All of them evaluated the efficacy of infliximab, canakinumab, baricitinib, anakinra, adalimumab, tofacitinib, tocilizumab, rilonacept, and placebo for JIA patients. All of the included literatures were published between 2011 and 2023. The number of patients ranged from 24 to 225. Randomized period ranged from 2 weeks to 54 weeks. JIA subtype included SoJIA, early PA JIA, JO AS, and ERA and JIA. The age of the patients ranged from 5.8 to 12.6. Disease duration ranged from 0.3 to 15.3 weeks.

Table 1.

General characteristic of the included studies.

Author	Country	No. of patients	Intervention	Comparison	Randomized period	JIA subtype	Age (yr)	Disease duration (wk)	Study
De Benedetti 2012[11]	Italy	112	Tocilizumab	Placebo	12 weeks	SoJIA	9.6	5.4	RCT
Ilowite 2014[10]	USA	71	Rilonacept	Placebo	4 weeks	SoJIA	10	2.6	RCT
Lovell 2013[16]	USA	24	Rilonacept	Placebo	4 weeks	SoJIA	12.6	3.1	RCT
Quartier 2011[12]	France	24	Anakinra	Placebo	4 weeks	SoJIA	8.5	3.7	RCT
Ruperto 2012[17]	Italy	84	Canakinumab	Placebo	2 weeks	SoJIA	8.5	2.2	RCT
Ruperto 2021[18]	Italy	225	Tofacitinib	Placebo	4 weeks	SoJIA	6.9	3.4	RCT
Ramanan 2023[19]	USA	220	Baricitinib	Placebo	4 weeks	SoJIA	5.8	2	RCT
Tynjälä 2011[20]	Finland	60	Infliximab	Placebo	54 weeks	Early PA JIA	10.3	0.3	RCT
Horneff 2012[21]	Germany	32	Adalimumab	Placebo	12 weeks	Jo AS	32	15.3	RCT
Burgos-Vargas 2015[22]	Germany	46	Adalimumab	Placebo	12 weeks	ERA JIA	46	12.9	RCT

ERA = enthesitis-related arthritis, JIA = juvenile idiopathic arthritis, PA = polyarticular, RCT = randomized controlled trials, SoJIA = systemic-onset JIA.

3.3. Risk of bias

Risk of bias summary and risk of bias graph can be seen in Figure 2 and Figure 3, respectively. Of the 5 RCTs, 3 studies did not report the random sequence generation and thus should be listed as unclear of risk of bias. One study did not report the allocation concealment and should be listed as unclear risk of bias. One study did not report the blinding of participants and personnel and should be listed as unclear risk of bias.

Figure 2.

Risk of bias graph of the included studies.

Figure 3.

Risk of bias summary of the included studies.

3.4. Results

3.4.1. Modified ACRpedi30

A total of 10 studies, including 9 treatments (infliximab, canakinumab, baricitinib, anakinra, adalimumab, tofacitinib, tocilizumab, rilonacept, and placebo) contributed to the clinical outcome of the modified ACRpedi30.

As displayed in Figure 4A, the network structure diagrams detailed the direct comparisons between different drugs in the modified ACRpedi30. Network meta-analysis showed considerable heterogeneity with global I² = 0%.

Figure 4.

(A) Network structure diagrams of modified ACRpedi30. (B) Forest plot of the modified ACRpedi30 as compared with placebo. (C) Surface under the cumulative ranking curve (SUCRA) probabilities of different treatments for modified ACRpedi30. (D) Funnel plot of the different treatments for modified ACRpedi30.

In head-to-head comparison, canakinumab (odds ratio 55.0, 95% credible intervals 2.4–67.0) was more effective than the placebo, and the difference was statistically significant. However, there was no statistically significant between other drugs versus placebo in terms of the modified ACRpedi30 (P > .05, Fig. 4B).

The SUCRA shows that canakinumab ranked first (SUCRA, 86.9%), anakinra ranked second (SUCRA, 77.7%), adalimumab ranked third (SUCRA, 61.9%), and placebo ranked the last (SUCRA, 6.3%, Fig. 4C). A funnel plot is drawn using the effect size centered on the comparison-specific pooled effect as an abscissa, and the standard error of the effect quantity as the ordinate. Figure 1 shows that the inverted funnel plot is essentially symmetrical; it is possible that this study has a small sample effect or a small publication bias (Fig. 4D).

3.4.2. Adverse events

A total of 9 studies, including 9 treatments (infliximab, canakinumab, baricitinib, anakinra, adalimumab, tofacitinib, tocilizumab, rilonacept, and placebo) contributed to the clinical outcome of the adverse events. As displayed in Figure 5A, the network structure diagrams detailed the direct comparisons between different drugs in the adverse events. Network meta-analysis showed considerable heterogeneity with global I² = 0%. There was no significant difference in the pairwise comparison of these treatments among the 9 groups (Fig. 5B, P > .05).

Figure 5.

(A) Network structure diagrams of adverse events. (B) Forest plot of the adverse events as compared with placebo. (C) Surface under the cumulative ranking curve (SUCRA) probabilities of different treatments for adverse events. (D) Funnel plot of the different treatments for adverse events.

The SUCRA shows that baricitinib ranked first (SUCRA, 67.4%), tofacitinib ranked second (SUCRA, 63.7%), adalimumab ranked third (SUCRA, 63.4%), and tocilizumab ranked the last (SUCRA, 32.9%, Fig. 5C).

A funnel plot is drawn using the effect size centered on the comparison-specific pooled effect as an abscissa, and the standard error of the effect quantity as the ordinate. Figure 5D shows that the inverted funnel plot is essentially symmetrical; it is possible that this study has a small sample effect or a small publication bias.

3.4.3. Hepatic-related adverse events

Nine studies were included in the analysis, encompassing a range of treatments (infliximab, canakinumab, baricitinib, anakinra, adalimumab, tofacitinib, tocilizumab, rilonacept, and placebo) that contributed to the assessment of clinical outcomes related to hepatic adverse events. Figure 6A illustrates network structure diagrams, delineating direct comparisons among different drugs regarding hepatic-related adverse events. The NMA revealed significant heterogeneity, with a global I² of 0%. No significant differences were observed in the pairwise comparison of treatments across the 9 groups (Fig. 6B, P > .05).

Figure 6.

(A) Network structure diagrams of hepatic-related adverse events. (B) Forest plot of the hepatic-related adverse events as compared with placebo. (C) Surface under the cumulative ranking curve (SUCRA) probabilities of different treatments for hepatic-related adverse events. (D) Funnel plot of the different treatments for hepatic-related adverse events.

SUCRA values indicated that anakinra achieved the highest rank (SUCRA, 78.9%), followed by adalimumab (SUCRA, 64.9%), with infliximab ranking the lowest (SUCRA, 22.5%, Fig. 6C). The inverted funnel plot, depicted in Figure 6D, demonstrated essential symmetry, suggesting the possibility of a small sample effect or publication bias in the study.

3.4.4. Infectious adverse event

A total of 9 studies, encompassing various treatments (infliximab, canakinumab, baricitinib, anakinra, adalimumab, tofacitinib, tocilizumab, rilonacept, and placebo), contributed to the evaluation of clinical outcomes related to infectious adverse events. Illustrated in Figure 7A, the network structure diagrams elucidated direct comparisons among different drugs concerning infectious adverse events. The NMA revealed substantial heterogeneity, with a global I² of 0%. No significant differences were observed in the pairwise comparison of these treatments across the 9 groups (Fig. 7B, P > .05).

Figure 7.

(A) Network structure diagrams of infectious adverse events. (B) Forest plot of the infectious adverse events as compared with placebo. (C) Surface under the cumulative ranking curve (SUCRA) probabilities of different treatments for infectious adverse events. (D) Funnel plot of the different treatments for infectious adverse events.

SUCRA values indicated anakinra as the top-ranking treatment (SUCRA, 74.8%), followed by adalimumab (SUCRA, 69.1%), with tocilizumab ranking the lowest (SUCRA, 22.1%, Fig. 7C). Figure 7D displayed an essentially symmetrical inverted funnel plot, suggesting the possibility of a small sample effect or a small publication bias in this study.

3.4.5. Serious adverse events

Nine studies, encompassing a spectrum of treatments (infliximab, canakinumab, baricitinib, anakinra, adalimumab, tofacitinib, tocilizumab, rilonacept, and placebo), collectively contributed to the assessment of clinical outcomes related to serious adverse events. As depicted in Figure 8A, the network structure diagrams meticulously outlined direct comparisons among different drugs regarding serious adverse events. The NMA unveiled noteworthy heterogeneity, with a global I² of 0%. No significant distinctions were observed in the pairwise comparison of these treatments across the 9 groups (Fig. 8B, P > .05).

Figure 8.

(A) Network structure diagrams of serious adverse events. (B) Forest plot of the serious adverse events as compared with placebo. (C) Surface under the cumulative ranking curve (SUCRA) probabilities of different treatments for serious adverse events. (D) Funnel plot of the different treatments for serious adverse events.

SUCRA values revealed tocilizumab as the top-ranked treatment (SUCRA, 91.1%), followed by Baricitinib (SUCRA, 58.9%), with Anakinra ranking the lowest (SUCRA, 12.3%, Fig. 8C). Figure 8D illustrated an essentially symmetrical inverted funnel plot, suggesting the potential presence of a small sample effect or a minor publication bias in this study.

3.4.6. Serious infection

Nine studies, encompassing a diverse array of treatments (infliximab, canakinumab, baricitinib, anakinra, adalimumab, tofacitinib, tocilizumab, rilonacept, and placebo), collectively contributed to the evaluation of clinical outcomes associated with serious infections. As presented in Figure 9A, the network structure diagrams intricately delineated direct comparisons among different drugs in relation to serious infections. The NMA disclosed significant heterogeneity, with a global I² of 0%. No notable differences emerged in the pairwise comparison of these treatments across the 9 groups (Fig. 9B, P > .05).

Figure 9.

(A) Network structure diagrams of serious infection. (B) Forest plot of the serious infection as compared with placebo. (C) Surface under the cumulative ranking curve (SUCRA) probabilities of different treatments for serious infection. (D) Funnel plot of the different treatments for serious infection.

SUCRA values indicated that anakinra claimed the top position (SUCRA, 62.4%), followed by adalimumab (SUCRA, 60.8%), with tocilizumab ranking the lowest (SUCRA, 12.4%, Fig. 9C). Figure 9D illustrated an essentially symmetrical inverted funnel plot, suggesting the potential presence of a small sample effect or a minor publication bias in this study.

3.5. Clustergram

Based on the clustergram of modified ACRpedi30 and adverse events (shown in Fig. 10), canakinumab is suggested for JIA according to the surface under SUCRAs considering the symptom and adverse events simultaneously.

Figure 10.

Clustergram of modified ACRpedi30 and adverse events according to surface under the cumulative ranking curve (SUCRA) values.

3.6. Meta-analysis regression

We performed meta-analysis regression to further increase the reliability of our meta-analysis. The meta-regression adjustment for the publication bias, sample size, risk of bias, disease duration, and randomized period almost did not alter the outcomes regarding the modified ACRpedi30 and adverse events (Table 2).

Table 2.

Meta-regression of the modified ACRpedi30 and adverse events.

	Covariates	b	95% credible interval of b
Modified ACRpedi30	Publication bias	0.3084	0.1203	1.2157
	Sample size	0.2732	0.0670	1.5287
	Risk of bias	0.7356	0.0519	1.5798
	Disease duration	0.4409	0.0824	1.3800
	Randomized period	0.4871	0.0244	1.2070
Adverse events	Publication bias	0.2427	0.0484	1.3709
	Sample size	0.5758	0.0718	1.1859
	Risk of bias	0.2615	0.0936	1.5732
	Disease duration	0.4496	0.0422	1.2886
	Randomized period	0.3373	0.0946	1.6126

4. Discussion

This systematic review and NMA of different drugs for the treatment of JIA included data from 10 clinical trials including 898 patients who were randomized to 8 distinct treatment protocols or placebo therapy. The quality of the evidence was typically of unclear risk of bias (5 out of 10 trials; 50%).

4.1. Principal findings and comparison with other studies

Our findings provide further clarification about the efficacy of different drugs for sJIA. Only 1 meta-analysis compared different drugs for sJIA through a NMA.^[23] They revealed that all biological agents included were efficacy and safety for treatment sJIA. Although data may be somewhat limited, it seems that both canakinumab and anakinra demonstrate comparable levels of efficacy, while tocilizumab and rilonacept exhibit reduced effectiveness when compared to canakinumab in individuals with sJIA. This study suggests that it may be prudent to consider canakinumab and anakinra for sJIA patients.^[11,24] Rilonacept displayed reduced effectiveness compared to other biological agents and is not advisable as the first-line choice for a biological agent in sJIA.^[16] The NMA uncovered significant variations in efficacy, although arriving at definitive conclusions proved challenging due to the diverse trial designs and the limited number of patients included in these studies. These discrepancies could potentially introduce bias when assessing relative efficacy.

Our NMA concurs with a prior study that conducted direct comparisons, unveiling a statistically significant advantage of canakinumab over placebo in relation to response criteria (modified ACRpedi30).^[6,24] What sets our NMA apart from earlier research involving RCTs and a systematic review using indirect comparisons is our ability to establish a hierarchical ranking for the efficacy and safety of biological agents in individuals with active sJIA.

sJIA is accompanied by elevated levels of multiple circulating cytokines.^[25,26] Case studies and experimental evidence support the use of monoclonal antibodies or soluble receptors to block inflammatory cytokines in sJIA.^[27,28] According to the results of reviews, the most effective of these biologics are IL-1 or IL-6 inhibitors.^[29] Many pediatric rheumatologists will use one of these drugs instead of corticosteroids in patients who have failed nonsteroidal anti-inflammatory drugs monotherapy.

Patients taking biologic response modifiers are at increased risk for infections, particularly mycobacterial, viral, and fungal infections. Therefore, the American Academy of Pediatrics published guidelines for the clinical use of this class of drugs.^[30] Guidelines recommend a thorough evaluation of the medical history to determine risk of infection, and screening as appropriate based on medical history and selected biologic agent. If it is safe for the child to delay treatment, it is recommended that biological agents be started at least 2 weeks after routine vaccination with inactivated or subunit vaccines, and at least 4 weeks after vaccination with live vaccines.^[31] Unfortunately, for many children with active sJIA, it is not feasible or safe to delay treatment and wait for vaccination. Live vaccines are not recommended during treatment with biologics. If it is considered necessary for a patient to receive live vaccines during biologic therapy, an infectious disease specialist should be consulted. Inactivated and subunit vaccines can be administered during treatment, and annual inactivated influenza vaccine is recommended.

We must approach the interpretation of our findings with caution, as our study faces several limitations. First, the wide range of follow-up time points, spanning from 15 days to 1 month, with many being relatively short, renders it inadequate for assessing long-term effects. Future investigations with extended comparative follow-ups are imperative. Second, the trials included in our analysis demonstrated heterogeneity in their design and patient characteristics. Notably, there was a substantial variation in the placebo response, ranging from 8% to 37%. These discrepancies across studies may have influenced the outcomes of our NMA. Third, our study had a narrower focus, primarily addressing the efficacy and safety outcomes of biologics in sJIA. We concentrated solely on efficacy, measured by the number of patients achieving a modified ACRpedi30 response, and safety, indicated by the incidence of serious adverse events, without encompassing a broader spectrum of outcome measures. Given the limited duration and small sample size, the evaluation of safety outcomes may have been constrained by low statistical power and exceedingly low incidence rates. Fourth, a substantial constraint was the eligibility of only a few studies with a small number of participants for inclusion and analysis. Fifth, it is essential to acknowledge that studies in pediatric rheumatology are not typically conducted concurrently in the same patient population. Many of these trials employ withdrawal designs, which, although providing a high level of evidence for efficacy, cannot be seamlessly integrated into a meta-analysis. Consequently, we must exercise prudence when interpreting the results, as all the included studies compared the treatment to a placebo rather than to other treatment modalities. As a result, it remains uncertain whether one treatment stands superior to another.

5. Conclusions

In summary, our Bayesian NMA, comprising 10 RCTs, indicates that canakinumab is most likely the top-performing treatment in terms of achieving the modified ACRpedi30 response rate. Furthermore, none of the evaluated biological agents were found to be associated with a significant risk of serious adverse events in patients with JIA. To establish the relative efficacy and safety of these biological agents in JIA patients, there is a need for long-term, large-scale RCTs.

Author contributions

Data curation: Zhenbiao Zhao.

Resources: Liming Zhou.

Software: Juan Ping, Liming Zhou.

Validation: Yushan Zhang, Zhenbiao Zhao, Juan Ping, Yongzhou Zhang.

Visualization: Baoquan Wang, Yushan Zhang, Yining Wang.

Writing — original draft: Baoquan Wang, Yining Wang.

Writing — review & editing: Baoquan Wang, Yongzhou Zhang.