***Eligibility criteria*** **Participants** Studies of children (≤18 years old) with mild-to-moderate chronic or persistent asthma will be included. We will include studies exclusively comprising of paediatric patients, and those involving adult and paediatric patients if data of paediatric age groups are accessible and can be extracted. This meta-analysis will also include studies of ‘children with recurrent wheeze’ or ‘preschool wheezers’. Currently, the diagnosis of asthma in young children is challenging because there are no universally accepted diagnostic criteria.[28][1] Only a subset of young children with recurrent wheezing episodes later develops physician-diagnosed asthma.[29][2] ,[30][3] In addition, there is a wide range of differential diagnoses in recurrent wheeze that mimic paediatric asthma, such as cystic fibrosis, congenital malformation of the airways and foreign body aspiration.[31][4] Therefore, children with recurrent wheeze may or may not have asthma. The likelihood of asthma in such patients depends on the presence/absence of risk factors (eg, family history).[32][5] Despite these problematic issues, we have decided to enrol children with recurrent wheeze for the following three reasons. First, recurrent wheezing is a major risk factor of asthma. As shown in studies of the Asthma Predictive Index, the combination of wheezing episodes (≥3 episodes/year) and other criteria is strongly associated with the risk of asthma (up to 77% chance of active asthma).[30][6] Second, in addition to symptoms and risk factors, the therapeutic response is often important for diagnosis of paediatric asthma,[32][7] ,[33][8] and empirical evidence indicates that children with recurrent wheeze may benefit from regular ICS use.[31][9] Finally, previous systematic reviews/meta-analyses have not often distinguished children with asthma from those with recurrent wheeze.[34][10] Owing to these reasons, we consider that children with asthma and those with recurrent wheeze share similar (although not identical) clinical characteristics and responses to ICS therapy. We will include only data of physician-diagnosed wheezing (≥3 times, separately) to ensure the consistency of patients’ symptoms. In trials of children with recurrent wheeze, we carefully review (1) whether the risk factors of asthma (eg, atopic status or family history) are described, and (2) whether differential diagnoses of wheeze are investigated. If these issues are insufficiently examined or documented, the authors will discuss whether such reports will be eligible for inclusion into the meta-analysis. **Interventions** We will include RCTs to examine the effectiveness of ICS in asthmatic children for ≥4 weeks. We will only include studies using ICS without co-interventions because the effectiveness of ICS is difficult to assess separately in trials with co-intervention (eg, ICS/LABA combination therapy). We will limit studies evaluating the effectiveness of ICS in current use (ie, studies of ICS that are no longer used, such as HFA-chlorofluorocarbon, will be excluded). Therefore, this study will include the BDP HFA-metered dose inhaler (MDI), BUD (dry powder inhaler (DPI) and nebules), ciclesonide (HFA-MDI), flunisolide (HFA-MDI), FP (HFA-MDI and DPI) and mometasone furoate (MDI and DPI). **Comparisons** This study will include clinical trials comparing one ICS with other active or inactive intervention(s), such as other types of ICS, other classes of drugs (eg, antileukotrienes) or placebo. The comparator should also be a single intervention because of the reason aforementioned. **Outcomes** In meta-analyses, researchers often declare the primary end point of the study.[35][11] However, this practice is difficult in asthma studies.[36][12] There are several domains in asthma control, such as a pulmonary function test or symptoms (eg, exacerbation), and, according to expert opinion, no single primary end point is recommended for assessment of responses to asthma.[37][13] Therefore, our planned study will not define a single primary end point but, instead, it will examine different end points to determine a more complete understanding of asthma control by ICS ([table 1][14]).[36][15] Study outcomes should be clinically relevant, and, ideally, they should be patient centred.[38][16] Additionally, outcomes of a sufficiently large number of trials should be pooled in the analysis. Summarising a large sample size would lead to more precise and confident estimation and, in NMA, combining small sample size studies could result in biased estimates.[39][17] From these perspectives, we will not include studies that exclusively examined biomarkers, QOL, or severity scores for the following reasons. First, how these outcomes correlate with the clinical benefit has yet to be established, and the magnitude of benefit of these outcomes is difficult to interpret for patients and even for healthcare professionals.[37][18] Second, a previous systematic review identified a few studies that examined these outcomes in paediatric patients.[40][19] Finally, for QOL and severity scores, different formulations are available and they are not interchangeable with each other.[41][20] **Exclusion criteria** We will exclude the following literature: abstracts only (eg, conference paper), studies that are not on asthma (eg, viral bronchiolitis), studies examining the dose–response relationship of ICS (because of technical difficulties in incorporating data into the meta-analysis), safety assessment studies of ICS and short term or intermittent use of ICS. ***Information sources*** The primary literature search will rely on PubMed and the Cochrane Central Register of Controlled Trials (CENTRAL). We will enrol all RCTs, including those of cross-over or quasirandomised design, that are published in full-text articles in the English language. We will use medical subject headings and text words related to ‘child’, ‘asthma’ and ‘ICS’ for the literature search.[40][21] To ensure literature saturation, we will scan the reference lists of included studies or relevant reviews that are identified through the search. ***Search strategy*** The search strategy is provided [here][22]. ***Study selection process*** One of the authors (MT) will scan the titles and abstracts of all the literature retrieved by the initial search and select eligible articles for review of the full text. Two other authors (HK and KT) will independently review full-text articles to assess eligibility and select citations to be meta-analysed. Studies that reported at least one core outcome will be selected (shown in [table 1][23]). **Extraction of data** The authors will also extract data independently using a prestandardised data abstraction form. Any disagreements will be resolved by discussion among all the authors. The process of literature selection will be published (eg, web-appendix style). **Risk of bias in individual studies** We will assess the quality and risk of bias of eligible studies, such as the method of randomisation, treatment allocation concealment, blinding the outcome assessor and dropouts. For this purpose, the Cochrane risk assessment tool will be used.[42][24] ,[43][25] **Confidence in cumulative evidence** We will also rely on the Grading of Recommendations Assessment, Development and Evaluation approach for quality assessment in cumulative estimates. ***Data synthesis*** **Statistical methods** [Figure 1][26]a illustrates the scheme of the proposed pairwise meta-analysis. A pairwise meta-analysis can compare head-to-head trials ([figure 1][27]a, A vs B and A vs C), but cannot compare indirect arms ([figure 1][28]a, B vs C). In contrast, NMA can compare indirect arms ([figure 1][29]a, B vs C). On the basis of a ‘consistency assumption’, the indirect effect B-C represents the difference between effect A-B and effect A-C (in this case, intervention A is referred to as a common comparator).[23][30] ,[44][31] Moreover, when there are head-to-head trials between B and C ([figure 1][32]b), NMA can combine the direct effect B-C and indirect effect B-C (ie, effect A-B—effect A-C).[45][33] In this way, NMA combines all available evidence of direct and indirect comparisons. There is an additional strength in NMA. A pairwise meta-analysis can compare only two interventions at a time.[23][34] In the situation shown in [figure 1][35]b, comparison of ‘A vs B vs C’ is not feasible, even when direct comparisons exist. In contrast, NMA can compare ≥3 interventions and determine which treatment works best. Further, NMA can compare more complex network loops ([figure 2][36]). [Figure 2][37] shows that comparative effectiveness among the ICS’ X, Y and Z can be estimated by combining direct evidence (effect B-C) and indirect effects using drug A and placebo as common comparators. On the basis of these strengths in NMA, we will evaluate the comparative effectiveness of ICS by pooling the results from head-to-head trials of ICS and from indirect comparisons among different ICS using placebo or other classes of medications (eg, antileukotriene drugs) as a common comparator. Statistical analyses will be conducted in a Bayesian hierarchical framework using a random-effects model.[46][38] We will use the gemtc package in R statistical software.[47][39] ,[48][40] This package uses a method developed by Lu and Ades.[22][41] This package also allows us to check for homogeneity and consistency, which are important assumptions in NMA that combined studies should be similar in clinical and statistical context (often referred to as transitivity assumption[23][42]). The statistical results will be presented in OR (with credible interval) and probability ranking. If we observe heterogeneity among studies, subgroup analyses will be conducted (see the subsection ‘Subgroup analysis’ below). As an example of this situation, when the dosage of ICS varies considerably among studies, we will stratify studies of ‘low’, ‘medium’ and ‘high’ dose,[18][43] ,[49][44] and combine the results within each strata. The statistical analyses will be performed by one author (MT) on the basis of previous expertise.[50][45] The gemtc R package has a unique function to check local (in)consistency, which we will use for this purpose. We will use I2 statistics to check global (in)consistency. This R package also prepares a function to generate network geometry, a graphical presentation of the network of evidence, which is an essential item of NMA reporting.[27][46] **Subgroup analysis** Heterogeneity is a potential concern in meta-analysis. If heterogeneity is detected, we plan to conduct the following subgroup analyses and will report the results when necessary: - Patients with chronic asthma versus recurrent wheezers; - Age groups stratified into three categories (0–4, 5–11, ≥12 years); - Children-specific study versus ‘children and adult’ study; - Dose stratification into low, medium and high dose. **Meta-bias** Although, overall, we adhere to the PRISMA-P statement, the method of dealing with publication bias (item 16) is not specified in this protocol. This is because identification of publication bias is more complex in NMA owing to limited numbers of studies for each pairwise comparison, heterogeneity and other limitations,[27][47] and there are no formal techniques to detect or assess the extent of publication bias. NMA is, however, a rapidly evolving research area and, if standard approaches are established at the time of our final report, we will be ready to use those skills. [1]: http://bmjopen.bmj.com/content/5/10/e008501.full#ref-28 [2]: http://bmjopen.bmj.com/content/5/10/e008501.full#ref-29 [3]: http://bmjopen.bmj.com/content/5/10/e008501.full#ref-30 [4]: http://bmjopen.bmj.com/content/5/10/e008501.full#ref-31 [5]: http://bmjopen.bmj.com/content/5/10/e008501.full#ref-32 [6]: http://bmjopen.bmj.com/content/5/10/e008501.full#ref-30 [7]: http://bmjopen.bmj.com/content/5/10/e008501.full#ref-32 [8]: http://bmjopen.bmj.com/content/5/10/e008501.full#ref-33 [9]: http://bmjopen.bmj.com/content/5/10/e008501.full#ref-31 [10]: http://bmjopen.bmj.com/content/5/10/e008501.full#ref-34 [11]: http://bmjopen.bmj.com/content/5/10/e008501.full#ref-35 [12]: http://bmjopen.bmj.com/content/5/10/e008501.full#ref-36 [13]: http://bmjopen.bmj.com/content/5/10/e008501.full#ref-37 [14]: http://bmjopen.bmj.com/content/5/10/e008501/T1.expansion.html "table 1" [15]: http://bmjopen.bmj.com/content/5/10/e008501.full#ref-36 [16]: http://bmjopen.bmj.com/content/5/10/e008501.full#ref-38 [17]: http://bmjopen.bmj.com/content/5/10/e008501.full#ref-39 [18]: http://bmjopen.bmj.com/content/5/10/e008501.full#ref-37 [19]: http://bmjopen.bmj.com/content/5/10/e008501.full#ref-40 [20]: http://bmjopen.bmj.com/content/5/10/e008501.full#ref-41 [21]: http://bmjopen.bmj.com/content/5/10/e008501.full#ref-40 [22]: https://osf.io/pcsmh/ "search strategy" [23]: http://bmjopen.bmj.com/content/5/10/e008501/T1.expansion.html [24]: http://bmjopen.bmj.com/content/5/10/e008501.full#ref-42 [25]: http://bmjopen.bmj.com/content/5/10/e008501.full#ref-43 [26]: https://osf.io/ehkcx/ "figure 1" [27]: https://osf.io/ehkcx/ "figure 1" [28]: https://osf.io/ehkcx/ "figure 1" [29]: https://osf.io/ehkcx/ "figure 1" [30]: http://bmjopen.bmj.com/content/5/10/e008501.full#ref-23 [31]: http://bmjopen.bmj.com/content/5/10/e008501.full#ref-44 [32]: https://osf.io/ehkcx/ "figure 1" [33]: http://bmjopen.bmj.com/content/5/10/e008501.full#ref-45 [34]: http://bmjopen.bmj.com/content/5/10/e008501.full#ref-23 [35]: https://osf.io/ehkcx/ "figure 1" [36]: https://osf.io/uev86/ "figure 2" [37]: https://osf.io/uev86/ "figure 2" [38]: http://bmjopen.bmj.com/content/5/10/e008501.full#ref-46 [39]: http://bmjopen.bmj.com/content/5/10/e008501.full#ref-47 [40]: http://bmjopen.bmj.com/content/5/10/e008501.full#ref-48 [41]: http://bmjopen.bmj.com/content/5/10/e008501.full#ref-22 [42]: http://bmjopen.bmj.com/content/5/10/e008501.full#ref-23 [43]: http://bmjopen.bmj.com/content/5/10/e008501.full#ref-18 [44]: http://bmjopen.bmj.com/content/5/10/e008501.full#ref-49 [45]: http://bmjopen.bmj.com/content/5/10/e008501.full#ref-50 [46]: http://bmjopen.bmj.com/content/5/10/e008501.full#ref-27 [47]: http://bmjopen.bmj.com/content/5/10/e008501.full#ref-27