Search strategy and study selection: This study was performed according to recommendations of the PRISMA statement, which is available as a supplementary PRISMA checklist.24 In October 2016, 10 electronic databases were searched: PubMed, Scopus, ISI Web of Science, WHO Global Health Library (WHO GHL), Virtual Health Library (VHL), Google Scholar, POPLINE, New York Academy of Medicine Grey Literature report (NYAM), Controlled Trials (mRCT), WHO International Clinical Trials Registry Platform (ICTRP), and System for Information on Grey Literature Report in Europe (SIGLE). The following search terms were used: [honey AND (asthma OR (bronchial hypersensitivity) OR (bronchial hyperreactivity) OR (respiratory hypersensitivity) OR (airway hypersensitivity) OR (airway hyperreactivity)] with no restrictions on language or publication date.
Three reviewers independently screened titles and abstracts to select potential full-text articles for further scrutiny according to inclusion and exclusion criteria. The inclusion criterion was any article that discussed the effect of honey on bronchial asthma, including animal and human studies.
All original articles were included without restriction to study design. No restriction of language, country, socioeconomic status, or time period was applied. Our exclusion criteria were as follows: unreliably extracted data, overlapped data sets, books, conference articles, theses, case reports, nonoriginal articles (reviews and analyses), and articles without available full text (conference, editorial, and author response). Any disagreement was resolved through discussion and arbitrated with the senior authors. The study selection procedure is summarized in a systematic review flow chart.
Data extraction Data were extracted independently by three reviewers using standard data extraction forms and any disagreement was resolved through discussion and consensus between reviewers and senior researchers was achieved. If there was one study with more than one publication report, they were compared and the publication with the most complete dataset was included. The data extraction sheet included the title of the article, name of authors, year of publication, characteristics of population, and country of recruitment, as well as data about sample size, study design, follow-up, and individual characteristics of participants. Confounders, form of honey, and outcome of the study were also recorded.
Quality assessment Two authors assessed the quality of studies independently without blinding to authorship or journal. Discrepancies were resolved by discussion with the senior authors. Three different quality assessment tools were utilized according to the study design of each article. Three included randomized controlled trials (RCTs) were assessed by Cochrane Collaboration’s tool for assessing risk of bias in randomized trials. Three cross-sectional studies were assessed by Australian Cancer Network’s quality assessment tool for cross-sectional studies. Two animal studies were assessed using SYRCLE’s risk of bias tool for animal studies. The metrics used to assess the quality of RCTs included random sequence generation (selection bias), allocation concealment (selection bias), blinding, incomplete outcome data (attrition bias), selective reporting (reporting bias), and other sources of bias. The metrics used to assess the quality of cross-sectional studies included subject selection, comparability of groups analyzed on demographic characteristics, and participation rate. On the other hand, selection bias, performance bias, detection bias, and attrition bias were used to assess animal studies.
Analyses All data were analyzed using comprehensive metaanalysis software, version 3 (Biostat, NJ, USA). Whenever possible, dichotomous and continuous variables were analyzed to compute the pooled odds ratio (OR) and standardized mean difference, respectively. The corresponding 95% confidence intervals of pooled effect size were also calculated using a fixed-effects model. No heterogeneity was significant enough to use a random-effects model due to the relatively small number of articles. Heterogeneity was assessed with Q statistics and I2 test considering it significant with I 2 value >50% or P value <.01.