High burden of undernutrition among primary school-aged children and its determinant factors in Ethiopia; a systematic review and meta-analysis

Systematic review and meta-analysis from computerized databases and gray literatures form the evidentiary bases of this study.

Preferred reporting items for systematic review and meta-analysis (PRISMA) guidelines were used to prepare and present this systematic reviews and meta-analysis [18]. To identify relevant articles, a comprehensive search between 2006 and September 3, 2019 was performed in the databases of PubMed, Science Direct, Google Scholar, and Cochrane Library. All searches were limited to articles written in English and grey literature was searched through the review of reference lists and hand searches. Moreover, to find unpublished papers, select research centers, including Addis Ababa Digital Library and Saint Mary’s University repositories were searched. Studies identified by our search strategy were retrieved and managed using Endnote X7 (Thomson Reuters, Philadelphia, PA, USA) software. The search was conducted from September 1 to November 25, 2019. During the search process the authors used the following keywords and MeSH terms: prevalence AND undernutrition OR stunting OR wasting OR underweight OR thinness AND associated factor OR determinant factor AND primary school-aged children AND Ethiopia.

All relevant articles were screened and extracted by two authors (AA and MM) independently using a standardized data extraction format in Microsoft™ Excel tool adapted from the Joanna Briggs Institute (JBI) [19]. Name of first authors, publication year, study area, study province, study design, mean age, response rate, sample size, and 95% CI prevalence of undernutrition as (stunting, underweight and thinness/wasting) were extracted from all selected full papers.

In this meta-analysis, the qualities of the included article were assessed using a modified version of the Newcastle-Ottawa Scale for cross-sectional studies adopted from Madhavan et al. [20]. This tool has three main sections. The first section, scored on the basis of one to five stars, focuses on methodological quality of each study (i.e., sample size, response rate, and sampling technique). The second section of the tool considers the comparability of the study cases with a possibility of two stars to be gained. The last section is concerned with the outcomes and statistical analysis of the original study with a possibility of three stars to be gained. In addition, quality appraisal of included studies was evaluated by two authors (MD and MM) independently and any discrepancy was resolved by a third author (MA). Articles attaining a Newcastle-Ottawa Scale (NOS) score of ≥7 stars out of 10 were considered as high quality and eligible for inclusion. In this study, all eligible studies were determined to have high quality scores.

Two authors (DB and MA) assessed the risk of bias for included articles independently by using the Hoy et al.’s internal and external validity tool for prevalence studies based on a review of the literature [21]. The tool consists of ten items addressing four domains of bias plus a summary risk of bias assessment [21]. Any scoring difference was resolved by discussion mediated by a third author (AM).

All included studies used a structured pre-tested questionnaire for face to face interviews and anthropometric measurement for weight was measured with minimal clothing and no shoes using an electronic beam balance in kilogram (kg) to the nearest of 0.1 kg and length was measured with standing position in centimeters (cm) to the nearest of 1 cm. Finally the anthropometric measures of weight, height, and age values were converted into Z-scores of the indices according to WHO standards [22]. Since the tools used by primary studies are pre-tested and standard measurements were used at various settings, it permits meaningful comparisons.

There were two outcomes considered for this systematic review and meta-analysis. The first outcome was to determine the pooled prevalence of undernutrition in terms of stunting, underweight and thinning/wasting among primary school-aged children in Ethiopia. Underweight is reflected when the weight for age is more than 2 standard deviations (SD) below the WHO Child Growth Standards median; stunting reported when height for age is more than 2 SD below of the WHO Child Growth Standards median; and wasting is weight for height is more than 2 SD below the WHO Child Growth Standards median [23].

A Microsoft™ Excel spreadsheet form was used to present the extracted data from the primary study and further analysis was done by STATA/se™ Version 14 statistical software. The standard error was calculated based on binomial distribution. Heterogeneity was assessed by computing p-values of Higgins’s I² test statics and Q-statistics among reported prevalence. Higgins’s I² statistic measures the difference between sample estimates (in percentage) which is due to heterogeneity rather than to sampling error. The pooled effect was estimated with a random effects meta-analysis model. Subgroup analysis was performed based on study province, sample size, and publication year to minimize random variations between the estimates of the primary study. In addition, univariate meta-regression was computed to identify the possible sources of heterogeneity by considering sample size, quality score, and publication year. Publication bias was assessed visually by funnel plot and statistically supported with confirmatory and/or objectivity testing with Egger’s and Begg’s test at a 5% level of significance [24]. Point prevalence, as well as 95% confidence intervals, was presented in a forest plot format. In this plot, the size of each box indicated the weight of the study, while each crossed line referred to a 95% confidence interval with the mean effect at the center. The determinant factors of undernutrition were determined by a log odds ratio at 95% level of significance.

As described in Table 1, thirty cross-sectional articles published between 2006 and 2019 were including with a total sample size of 16,642 and a response rate of 97.4%. The smallest sample size was 97 in a study in Gondar, Amhara [32] while the largest sample size was 2241 in a study done at Bahir Dar and Mecha, Amhara [54].

Regarding the prevalence of undernutrition, the smallest prevalence of stunting was 4.5% [38] while the largest was 46.5% [52]. The prevalence of underweight ranged from 4.2% [39] to 34.8% [26]; whereas wasting (thinness) ranged between 5.2% [43] to 50% [26].

Thirty articles with a pooled sample size of 16,642 school-aged children and response rate of 97.4% were considered in this systematic review and meta-analysis. The pool prevalence of stunting, underweight, and wasting in Ethiopia were 21.3% (95% CI: 17.0, 25.5), 18.2% (95% CI: 14.4, 22.0), and 17.7% (95% CI: 13.5, 21.8) respectively (Figs. 2, 7 & 8).

The estimated prevalence of stunting among school-aged children in Ethiopia from 29 articles was 21.3%. There was high heterogeneity found within the included articles as revealed in forest plot (I2 = 98%, p < 0.001) (Fig. 2).

Thus, the subgroup analysis in the study province provide the highest prevalence of stunting in Amhara region 27.6% (95% CI: 20.7, 34.5)(I² = 97.8, p < 0.001) followed by Southern Nations, Nationalities and People’s (SNNP) Region 15.0% (95% CI: 26.5, 23.4)(I² = 98.2, p < 0.001) and the lowest was noted in Oromia Region 10.5% (95% CI: 8.2, 12.7)(I² = 54.8, p = 0.084) (Table 2).

The presence of publication bias was assessed with subjectivity test of bias funnel plot visually and found substantially asymmetric and a further confirmatory test of objectivity to indicate publication bias was done with Egger’s test (p-value < 0.001) (Fig. 3). Even though the funnel plot showed substantial symmetry, the Egger’s test shows presence of bias. Thus, we performed a nonparametric trim and fill analysis to handle the observed publication bias for estimating the number of missing studies that might exist and in reducing and adjusting publication bias in meta-analysis.

In this systematic review and meta-analysis of stunting in Ethiopia, the associated factors affecting stunting were frequency of meals three times per day with odd ratio of 0.45(95%CI: 0.28, 0.72) and hand washing practice with odds of 0.73(95%CI: 0.48, 1.12) were reported from a single article as a result were not able to generate the pooled estimate. Family size of below five members was reported from three studies with pooled odd ratio of 7.95(95% CI: 0.62, 15.3), which was not found statistically significant.

Sex with pooled odd ratio of 0.73 (95% CI: 0.62, 0.85), age of children with pooled odd ratio of 0.56 (95% CI: 0.44, 0.72), and maternal education with pooled odd ratio of 1.91(95% CI: 1.33, 2.73) were found statistically significant. As we have examined the association between sex and stunting among school-aged children from seven [7] studies; the findings showed that sex had an association with prevalence of stunting, with the odds of stunting 27% higher among males as compared to females (OR = 0.73, 95% CI: 0.62, 0.85). The results did not indicate presence of heterogeneity (I2 = 0.0% and p = 0.734) (Fig. 4).

The association between age category and stunting was calculated from eight studies. Accordingly, the odds of stunting was 79% higher among school-aged children less than 10 years of age as compared to children 10–15 years old (OR = 0.56, 95% CI: 0.44, 0.72). The statistics showed moderate heterogeneity (I2 = 55.3% and p = 0.028) across the included studies (Fig. 5).

We observed the association between maternal education and stunting from eight studies. The prevalence of stunting was 1.91 times higher among school-aged children whose mothers had no formal education in comparison to the more educated cohort (OR = 1.91, 95% CI: 1.33, 2.73). The resulting statistics showed moderate heterogeneity (I2 = 70% and p = 0.002) across the included studies (Fig. 6).

Prevalence of underweight in school-aged children in Ethiopia from 18 studies was 18.2%. A high heterogeneity was observed across the included articles as revealed in forest plot (I2 = 96.1%, p < 0.001) (Fig. 7).

Subsequently, subgroup analysis by the study province to indicate source of heterogeneity provides the highest prevalence of underweight at 22.7% (95% CI: 19.2, 26.3) (I² = 88.1, p < 0.001) in Amhara region followed by 12.2% (95% CI: 7.1, 17.2) (I² = 94.1, p < 0.001) in SNNP region (Table 2).

As we have studied the associated factors of underweight among school-aged children in Ethiopia, we had observed protected source of water and hand washing with soap were reported form only one article as associated factor with underweight with odd ratio of 1.18 (95% CI: 0.7, 1.98) and 2.62 (95% CI: 1.02, 6.72) respectively. Two factors (sex of school-aged children and educational status of mothers) were reported consistently over three different articles but both were not statistically associated with underweight (OR = 1.17, 95% CI: 0.54, 2.56) and (OR = 1.00, 95% CI: 0.44, 2.30) respectively.

The presence of parasitic infection reported from three studies with pooled odd ratio 2.02(95% CI: 1.10, 3.73) was found significantly significant. Thus, the prevalence of underweight was two times higher among those children infected with parasitic infection as compared to non-infected children (Fig. 8).

The pooled estimates of wasting in Ethiopia among school-aged children from 18 articles was 17.7% and the heterogeneity between the studies were observed across the included articles as revealed in forest plot (I2 = 97.5%, p < 0.001) (Fig. 9).

Accordingly, the subgroup analysis by study province to show source of heterogeneity of wasting among the included articles indicated the highest prevalence of wasting was observed at 19.3% (95% CI: 5.1, 33.4) (I² = 98.9, p < 0.001) in SNNP followed by 16.5% (95% CI: 11.7, 21.3) (I² = 96.9, p < 0.001) in Amhara region (Table 2). To determine publication bias, we performed funnel plot to show bias graphically, yielding moderate asymmetry. Further confirmation of bias was assessed with objectivity Egger’s test, the presence of publication bias was noted with (p-value =0.005) (Fig. 10). A nonparametric trim and fill analysis was conducted to handle and adjusting the observed publication bias.

The association between age category and wasting was calculated from six studies. The prevalence of wasting was 2.74 times higher among school-aged children 5 to10 years age as compared to those aged 10 to15 years (OR = 2.74, 95% CI: 1.81, 4.14) from a random effects model. The result of the statistics showed moderate heterogeneity (I2 = 76.1% and p = 0.001) across the studies (Fig. 11).

In addition we have calculated the log odds ratio for educational status of mothers on wasting/thinness of school-aged children, finding that the odds of wasting is 1.66 times higher among children at the age of 5 to10 years as compared to 10 to15 years of age (OR = 0.6, 95% CI: 0.36, 0.9). Moderate heterogeneity was observed over the included studies with their respective log odds ratio (I2 = 76.1%, p = 0.001) (Fig. 12).