Population. School-age children (6–14 years old).

Exposure. Predictors or determinants of stunting. The determinants are characteristics or exposures that increase the likelihood of stunting among school-age children. Those factors include residence site, educational status of mothers or caregivers, sex of the child, age of the child, and household food security status.

Comparison. The reported reference group for each determinant or associated factors in each study (e.g. stunting in children residing in an urban area versus rural area, children from food secured versus food in secured households).

Outcome. We will include studies that assess the prevalence of stunting and its determinants among school-age children in Ethiopia.

Preferred Reporting Items for Systematic Reviews and Meta-Analysis will be used for the preparation and presentation of this meta-analysis. Online databases (Medline, PubMed, Scopus, and Science direct), Google, Google Scholar, and other grey literature will be used to search articles until June 2020. The searching strategy for PubMed online database is added as a supplementary file with this protocol (Additional file 2 in S1 File). Snowballing will be used to screen the references of identified articles for possibly relevant studies. Studies identified by our database searching strategy will be retrieved and managed using Endnote X8 software [22].

Medical Subject Heading (Mesh), keywords, and free text search terms will be used. As the search terms, we will include alternative terms for stunting and will combine them using Boolean operators

Search terms. Search (((“Stunting” OR “Stunt” OR “Growth and Development” OR “Growth” OR “Body Size” OR “Body Height” AND “school-age children” OR “6–14 years old children”) AND (Prevalence OR burden OR magnitude OR incidence) AND (predictors OR “risk factors” OR determinants OR associated factors OR causes”) AND (Ethiopia OR Etiopia OR Ethio)))

According to the WHO growth reference standard, Stunting is defined as height for age Z (HAZ) scores below -2 Z-score value [2].

Two authors (SB and AW) will review the studies, based on inclusion and exclusion criteria. At the first stage, relevant articles will be considered based on their title.

Abstracts of these selected titles will be incorporated at the second stage. In the third stage, the full-text screening will be conducted. In case, if articles are not open access, we will contact the corresponding author at least three times. If the authors are not willing to provide the full text, that specific article will be excluded from the study.

Studies that are approved by both authors in the review processes will be included. Any differences in ideas will be resolved through discussion until an agreement is reached. A third reviewer (NW) may be consulted if disagreement happens between the two reviewers. Lastly, we will organize a final list of articles for data extraction

After all eligible articles are identified, two independent reviewers (SB and DA) will extract the relevant data using an organized format on Microsoft Excel Spreadsheet.

Discrepancies between data extractors will be discussed to reach a consensus. If a consensus cannot be reached, the authors will consult a third reviewer (NW). For each included article, we will record the first author’s last name, year of publication, the setting where the study was conducted, study design, study period, sample size, the response rate, the population, outcome definition, comparison groups, and pertinent associated factors, and the effect estimate.

For prevalence studies, prevalence, the logarithm of prevalence, and SE of the logarithm of prevalence will be calculated. Similarly, for determinants, OR, logarithms of OR, and SE of the logarithms of OR will be calculated. For any difficulties that might be encountered during data extraction, the corresponding author(s) will be communicated.

The quality assessment will be conducted by five authors (SB, AW, NW, KU, and DA) using the checklist of the JBI appraisal tool for cross-sectional, cohort, and case-control studies (Additional file 3 in S1 File). This tool will include different questions based on the study designs. For studies with cohort, case-control, and analytical cross-sectional study designs, question items 10, 8, and 11 will be employed, respectively. For studies with simple cross-sectional studies, a tool with nine-question items will be used. The tools have ‘Yes’ and ‘No’ types of questions and scores will be given 1 for ‘Yes’ and 0 for ‘No’ responses. Scores will be summed and converted into a percentage.

Only studies that scored ≥50% will be considered for both systematic review and meta-analysis of prevalence. For any scoring disagreements, which might happen between the abstractors, the sources of discrepancy will be investigated through revision. If disagreements persist, after the detailed review, the average scores of the reviewers will be calculated. In the same way, for determinants, each determinant with the outcome variable will be critically evaluated. A similar cut-off point that we will be using for prevalence studies will be applied to determinants of stunting.

The extracted data will be imported into STATA version 16 (Stata Corp LLC, Texas, USA) software. A Narrative description of the study population will be performed. Tables and figures will be used to summarize the selected studies and results.

The pooled prevalence of stunting among school-age children in Ethiopia will be demonstrated using the random effect model [23]. The Freeman Tuckey variant of the arcsine square root transformation of proportions will be used to avoid variance variability when controlling proportions close to one [24]. We will assess heterogeneity by using the chi-squared test on Cochran’s Q statistic with a 5% level of statistical significance [25] and I² statistic test [26], assuming that I² value of 25%, 50%, and 75% is representative of low, moderate, and high heterogeneity, respectively [26]. If the heterogeneity is significant (I² > 75%) and p-value < 0.05 will be declared as the presence of heterogeneity. Thus, subgroup analyses and meta-regression will be performed to investigate sources of heterogeneity.

To identify influential studies, sensitivity analysis will be conducted [25]. Publication bias will be examined by the visual inspection of funnel plots [27] and Egger’s test [28]. A p-value < 0.10 will be considered indicative of statistically significant publication bias, if evidence of publication bias present, thus the trim-and-fill (Duval and Tweedie’s) method will be performed [29].

The existence of an association between the determinants and stunting will be estimated based on the effect size. Then, the statistical significance level will be declared at a p-value of less than 0.05.