This systematic review with meta-analysis was conducted and reported according to the Preferred reporting items for systematic reviews and meta-analysis (PRISMA) guidelines [8].

Eligibility criteria and search strategy: Randomized controlled trials (RCTs) were included if they met the following criteria: (i) the study compared any NSAID with placebo or another NSAID in the treatment of JIA; (ii) the study provided endpoints for the efficacy or adverse events; and (iii) the study included patients diagnosed with JIA. The details of eligibility criteria are provided in Web Table 1. For this network meta-analysis, MEDLINE (via PubMed), EMBASE, and the Cochrane Library were searched for RCTs published from January 1, 1965, to January 1, 2019, comparing the efficacy and (or) safety of NSAIDs in the treatment of JIA. The following search terms were used: "#1 "Juvenile Idiopathic Arthritis" OR "Arthritis" OR "Still Disease" OR "Rheumatoid" AND #2 "NSAIDs" OR "Agents" OR "Non-Steroidal Anti-Inflam-matory Drugs" OR "Analgesics" OR "indomethacin" OR "naproxen" OR "Naprosyn" OR "aspirin" OR "acetylsalicylic acid" OR "celecoxib" OR "Celebrex" OR "rofecoxib" OR "piroxicam" OR "ibuprofen" OR "meloxicam" OR "tolmetin" OR "diclofenac" OR "Voltaren" OR "voltarol" AND #3 "randomized controlled trial" OR "controlled clinical trial" OR "placebo" OR "drug therapy" OR "groups". In order to ensure the authenticity of data, we also searched the list of references included in the articles.

Data extraction: The following information was extracted from each study: i) study characteristics (author, year, study design); ii) patient characteristics (sample size, race, age, sex, duration of JIA, subtype); iii) interventions: any NSAID, dosage, concomitant therapy, follow-up time when outcomes were evaluated; and, iv) outcomes (efficacy, and adverse events). Data were extracted from original studies by two independent investigators, and any discrepancy between the investigators was resolved by after discussions along with a third investigator.

The primary outcomes were efficacy (the number of patients who had improvement in symptoms of arthritis) and safety (the number of patients who experienced adverse events).

Quality assessment: The risk of bias was assessed using Cochrane risk of bias tool [9], evaluating for random sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assess-ment, incomplete outcome data, selective reporting, and other bias like sample size, [10] (labeling them as low risk of bias with 200 participants or more per treatment group, unclear risk with 50 to 199 participants per treatment group, or high risk with fewer than 50 participants per treatment group), multiple publications, financial declara-tions, and participants with conflicts of interest.

Statistical analyses: First, we divided the three-arm or more-than-three-arm tests into combinations of any two arms, and performed the evidence network diagram for the comparison of each treatment. We assessed the inconsis-tency or extent of disagreement between direct and indirect evidence. For closed loops, we tested the transitivity assumption by examining loop-specific consistency between direct and indirect effects using network side splits and global consistency by comparing a model assuming consistency with an inconsistent model. When the global Wald test indicated no significant differences between the consistency and inconsistency models [11] and no significant differences in estimates based on side splits, we presented consistency model estimates. The Markov chain Monte Carlo (MCMC) method was used to obtain pooled effect sizes [6]. We calculated the risk ratio (RR) with 95% credible interval (CI; or Bayesian confidence interval). The efficacy and safety of NSAIDs in different arms were ordered according to the probability of being ranked as the best performing regimen. We did a network meta-analysis within a Bayesian framework with WinBUGS (version 1.4.3) and further analysis with Stata (version 15.1). Information on relative effects and safety was converted to a probability that a treatment is the best, second best, etc., or to the ranking of each treatment, called the surface under the cumulative ranking curve (SUCRA) [12]. The SUCRA value was 100% when a treatment is certain to be the best for efficacy but the worst for safety.

The selection of studies is shown in the flowchart (Fig. 1). Overall, 8 studies provided data of 1112 individual JIA patients receiving the following NSAIDs: celecoxib, rofecoxib, meloxicam, diclofenac, ibuprofen, naproxen, piroxicam, and tolmetin. The study sample size ranged from 26 to 310. The duration of treatments was from 2 weeks to 24 weeks. It was reported that there was no significant difference in age, sex, course of disease between the groups. The subtype of JIA contained poly-articular JIA, oligoarticular JIA, and systemic JIA. The details of each included trial were listed in Table I. Risk of bias within individual studies was assessed (Fig. 2).

Fig. 1 PRISMA 2009 flow diagram.

Fig. 2 Risk of bias assessment.

Network meta-analysis for efficacy: Network meta-analysis was only performed when studies were sufficiently homogeneous regarding outcome criteria. Thus, for efficacy, three studies [14-16] with the efficacy criteria of achieving an American College of Rheumatology Pediatric-30 (ACR Pedi 30) response [22-23] were eligible. Four NSAIDs (celecoxib, rofecoxib, meloxicam, and naproxen) were compared with at least one other active drug directly and indirectly. There were no significant differences between any two NSAIDs regarding efficacy (Fig. 3). The ranking of treatments based on cumulative probability plots and SUCRAs is shown in Web Fig. 1. In terms of efficacy, celecoxib (6 mg/kg bid) had the highest probability of being most effective (SUCRA = 76.4%), while two doses of meloxicam ranked last.

Fig. 3 Network meta-analysis of efficacy of non-steroidal anti-inflammatory drugs for juvenile idiopathic arthritis.

Network meta-analysis of the safety: Nine NSAIDs (celecoxib, rofecoxib, meloxicam, naproxen, ibuprofen, aspirin, diclofenac, piroxicam, and tolmetin) were directly compared with at least one other active drug. There were no significant differences between any two NSAIDs regarding safety (Fig. 4). Ranking probability based on SUCRA values indicated that rofecoxib (0.3 mg/kg/d) had the highest probability of being the safest treatment (SUCRA=33.0%), followed by piroxicam (SUCRA =35.5%). Tolmetin and aspirin appeared to have the worst safety probability (SUCRA=82.3% and 82.0%, respectively) (Web Fig. 2).

Fig. 4 Network meta-analysis of safety of non-steroidal anti-inflammatory drugs for juvenile idiopathic arthritis.

Inconsistency plots assessing network inconsis-tencies between direct and indirect estimates showed a low possibility of inconsistencies that might significantly affect the results. In addition, the results of the random and fixed-effects models yielded the same interpretation, indicating that the results were robust.

We conducted a network meta-analysis of currently available literature regarding NSAIDs for children and adolescents with JIA. Unlike previous meta-analyses, we were able to generate a ranking order for the relative efficacy and safety of NSAIDs in patients with JIA. We found that the rate of efficacy observed in all treatment groups in our study were above the pooled composite placebo response rate (28.9%) reported in a meta-analysis of six placebo-controlled trials [25]. However, no statistically significant differences were observed between NSAIDs in terms of efficacy or safety. The findings are similar to the previous meta-analysis on NSAIDs for osteoarthritis in adults [26,27]. The SUCRA ranking suggests that celecoxib had better efficacy, while piroxicam and rofecoxib have higher safety probabilities, compared to other NSAIDs.

The most common adverse effects across all treatment groups were gastrointestinal side effects, rash, headache, and pyrexia. These side effects occurred more frequently within the aspirin, tolmetin, and ibuprofen groups, resulting in more non-compliance. Estimates of NSAIDs-associated gastropathy range from 0.7-75%, depending on different study designs [28-32]. Most of the gastrointestinal disorders were mild, while serious gastropathy such as gastrointestinal perforation and massive gastrointestinal hemorrhage was lower than adults. The combination of glucocorticoid, leflunomide, and methotrexate can aggravate gastrointestinal adverse reactions. While children have a very low risk of cardiovascular thromboembolic and serious gastro-intestinal events, prolonged use of NSAIDS into adulthood could make them vulnerable to such risks, especially when associated with other risk factors such as obesity or smoking [36].

We used the GRADE (Grading of Recommendations Assessment, Development, and Evaluation) approach to assess the quality of the evidence related to our outcomes. The eight included studies themselves were of moderate quality; however, there may be circumstances where the overall rating for a particular outcome would need to be adjusted per GRADE guidelines [37]. The sample size for some comparisons was assessed as a high bias of risk, which largely restricts the quality of meta-analysis. Among the included studies, there were no two studies that investigated the same type of NSAID compared with another type of NSAID, which might overestimate the efficacy and safety of treatments. Additionally, the inconsistent criteria of efficacy make it impossible to make a comprehensive comparison of some NSAIDs. Also, there was no data on the stratification of subtypes and concomitant therapy, thus it is unlikely to eliminate the impact of these factors on efficacy and safety. The follow-up time points were limited from only 2 to 24 weeks. The quality of the evidence (GRADE rating) for the efficacy and safety of NSAIDs is very low, meaning there is no evidence to support or refute the findings.

In conclusion, this Bayesian network meta-analysis involving eight RCTs with low quality of evidence showed that, in terms of efficacy, celecoxib (6 mg/kg bid) ranked best among the four NSAIDs (celecoxib, rofecoxib, meloxicam, and naproxen). In terms of safety, rofecoxib, piroxicam, and meloxicam may be better than others. However, the limitations of the study and suboptimal quality of evidence bar us from making strong conclusions about the comparative efficacy or safety of NSAIDs used to treat JIA. Further well-designed RCTs are needed to figure out the best NSAID for JIA.

Acknowledgments: Mao Song for technical assistance and Xiao-hui Tan for language editing.

Contributers: TX: conceived and designed the study, critically revised the manuscript; SC: acquired data, interpreted data, and drafted and critically revised the manuscript. ZY, ZZ, ZJ: critically revised the manuscript; SC, TX: screened and selected articles; SC, ZY, TX: assessed the quality of included trials. All the authors read and approved the final manuscript.

Competing interests: None stated. Funding: None.

1. Ross EP, Taunton RS, Prudence M, et al. International League of Associations for Rheumatology Classification of Juvenile Idiopathic Arthritis: Second Revision, Edmonton, 2001. J Rheumatol. 2004;31:390-2.

2. Sarah R, Sheila TA, Timothy B, et al.. 2019 American College of Rheumatology/Arthritis Foundation Guideline for the Treatment of Juvenile Idiopathic Arthritis: Therapeutic Approaches for Non-Systemic Polyarthritis, Sacroiliitis, and Enthesitis. Arthritis Rheumatol. 2019;71:846-63.

3. Andrew RM, Sheena D, Geoffrey TM, Henry JM. Tolerability and adverse events in clinical trials of celecoxib in osteoarthritis and rheumatoid arthritis: Systematic review and meta-analysis of information from company clinical trial reports. Arthritis Res Ther. 2005;7:644-65.

4. Hetao H, Jianke P, Weiyi Y, et al. Celecoxib vs diclofenac sodium in patients with knee osteoarthritis: A protocol for systematic review and meta-analysis. Medicine (Balti-more). 2020,99:e19680.

5. Ruijie W, Pin L, Heng J. The efficacy of celecoxib for pain management of arthroscopy: A meta-analysis of rando-mized controlled trials. Medicine (Baltimore). 2019;98: e17808.

6. Deborah MC, Ades AE, Higgins JPT. Simultaneous comparison of multiple treatments: Combining direct and indirect evidence. BMJ. 2005;331:897-900.

7. Ferrán CL, Aurelio T, Chris C, David M, Brian H. Network meta-analysis for comparing treatment effects of multiple interventions: An introduction. Rheumatol Int. 2014;34: 1489-96.

8. Moher D, Liberati A, Tetzlaff J, Altman DG. Preferred reporting items for systematic reviews and meta-analyses: The PRISMA statement. BMJ. 2009;339:b2535.

9. Higgins JPT, Thomas J, Chandler J, et al. Cochrane Handbook for Systematic Reviews of Interventions version 6.0 (updated July 2019). Cochrane, 2019. Available from: http://www.training.cochrane.org/ handbook.

10. Dechartres A, Altman DG, Trinquart L, Boutron I, Ravaud P. Association between analytic strategy and estimates of treatment outcomes in meta-analyses. JAMA. 2014;312: 623-30.

11. Shim S, Yoon BH, Shin IS, Bae JM. Network meta-analysis: Application and practice using Stata. Epidemiol Health. 2017;39:e2017047.

12. Salanti G, Ades AE, Ioannidis JPA. Graphical methods and numerical summaries for presenting results from multiple-treatment meta-analysis: An overview and tutorial. J Clin Epidemiol. 2011;64:163-71.

13. Bhettay E, Thomson AJ. Double-blind study of ketoprofen and indomethacin in juvenile chronic arthritis. S Afr Med J. 1978;54:276-8.

14. Foeldvari I, Szer IS, Zemel LS, et al. A prospective study comparing celecoxib with naproxen in children with juvenile rheumatoid arthritis. J Rheumatol. 2009;36:174-82.

15. Reiff A, Lovell DJ, Adelsberg JV, et al. Evaluation of the comparative efficacy and tolerability of rofecoxib and naproxen in children and adolescents with juvenile rheumatoid arthritis: A 12-week randomized controlled clinical trial with a 52-week open-label extension. J Rheumatol. 2006;33:985-95.

16. Ruperto N, Nikishina I, Pachanov ED, et al. A randomized, double-blind clinical trial of two doses of meloxicam compared with naproxen in children with juvenile idiopathic arthritis: Short- and long-term efficacy and safety results. Arthritis Rheum. 2005;52:563-72.

17. Giannini EH, Brewer EJ, Miller ML, et al. Ibuprofen suspension in the treatment of juvenile rheumatoid arthritis. Pediatric Rheumatology Collaborative Study Group. J Pediatr. 1990;117:645-52.

18. Haapasaari J, Wuolijoki E, Ylijoki H. Treatment of juvenile rheumatoid arthritis with diclofenac sodium. Scand J Rheumatol. 1983;12:325-30.

19. García MO, Maldonado CJA, Cuttica R, Garay SM. Piroxicam in juvenile rheumatoid arthritis. Eur J Rheumatol Inflamm. 1987;8:49-53.

20. Kvien TK, Høyeraal HM, Sandstad B. Naproxen and acetylsalicylic acid in the treatment of pauciarticular and polyarticular juvenile rheumatoid arthritis. Assessment of tolerance and efficacy in a single-centre 24-week double-blind parallel study. Scand J Rheumatol. 1984;13:342-50.

21. Levinson JE, Baum J, Brewer E, et al. Comparison of tolmetin sodium and aspirin in the treatment of juvenile rheumatoid arthritis. J Pediatr. 1977;91:99-804.

22. Giannini EH, Ruperto N, Ravelli A, Lovell DJ, Felson DT, Martini A. Preliminary definition of improvement in juvenile arthritis. Arthritis Rheum. 1997;40:1202-9.

23. Burnett HF, Regier DA, Feldman BM, Miller FA, Ungar WJ. Parents’ preferences for drug treatments in juvenile idiopathic arthritis: A discrete choice experiment. Arthritis Care Res (Hoboken). 2012;64:1382-91.

24. Demirkaya E, Ruperto N , Galasso R. A meta-analysis to estimate the "real" placebo effect in juvenile idiopathic arthritis (JIA) trials. Pediatr Rheumatol. 2011;9:192-92.

25. Song GG, Seo YH, Kim JH, Choi SJ, Lee YH. Relative efficacy and tolerability of etoricoxib, celecoxib, and naproxen in the treatment of osteoarthritis: A Bayesian network meta-analysis of randomized controlled trials based on patient withdrawal. Z Rheumatol. 2016;75:08-16.

26. Castellsague J, Riera GN, Calingaert B, Lorenzo CV. Individual NSAIDs and upper gastrointestinal compli-cations: A systematic review and meta-analysis of obser-vational studies (the SOS project). Drug Saf. 2012;35: 1127-46.

27. DeWitt EM, Sherry DD, Cron RQ. Pediatric rheumatology for the adult rheumatologist I: Therapy and dosing for pediatric rheumatic disorders. J Clin Rheumatol. 2005;11: 21-33.

28. Keenan GF, Giannini EH, Athreya BH. Clinically significant gastropathy associated with nonsteroidal antiinflammatory drug use in children with juvenile rheumatoid arthritis. J Rheumatol. 1995;22:1149-51.

29. Mulberg AE, Linz C, Bern E, Tucker L, Verhave M, Grand RJ. Identification of nonsteroidal antiinflammatory drug-induced gastroduodenal injury in children with juvenile rheumatoid arthritis. J Pediatr. 1993;122:647-9.

30. Hermaszewski R, Hayllar J, Woo P. Gastro-duodenal damage due to non-steroidal anti-inflammatory drugs in children. Br J Rheumatol. 1993;32:69-72.

31. Dowd JE, Cimaz R, Fink CW. Nonsteroidal anti-inflammatory drug-induced gastroduodenal injury in children. Arthritis Rheum. 1995;38:1225-31.

32. Bertagnolli MM, Eagle CJ, Zauber AG, et al. Five-year efficacy and safety analysis of the adenoma prevention with celecoxib trial. Cancer Prev Res (Phila). 200;2:310-21.

33. Bresalier RS, Sandler RS, Quan H, Baron JA. Cardio-vascular events associated with rofecoxib in a colorectal adenoma chemoprevention trial. N Engl J Med. 2005;352: 1092-102.

34. Muntner P, He J, Cutler JA. Trends in blood pressure among children and adolescents. JAMA. 2004;291:2107-13.

35. Juonala M, Järvisalo MJ, Mäki TN, Kahoen M, Viikara JS, Raitakari OT. Risk factors identified in childhood and decreased carotid artery elasticity in adulthood: The cardiovascular risk in young finns study. Circulation. 2005; 112:1486-93.