This study was done following the methodology of umbrella review of SRM studies [16, 17]. Umbrella review is a systematic synthesis of systematic reviews or meta-analysis reports on a specific research topic.
Data source and literature search
Five databases (PubMed, Embase, Scopus, Web of sciences, and Google scholars) were searched for SRM studies on child malnutrition and IYCF practices in Ethiopia, published from January 2015 to August 15, 2019. The search for malnutrition studies was focused on the four more prevalent undernutrition conditions of public health priority in Ethiopia; i.e., anemia, stunting, underweight, wasting, and underweight. The search for IYCF practice studies was focused on the child feeding indicators recommended by WHO/UNICEF. They were (a) early initiation of breastfeeding, (b) exclusive breastfeeding during the first 6 months, (c) continued breastfeeding up to 2 years and beyond, (d) dietary diversity, and (e) meal frequency. Thus, we specifically searched for SRM studies that reported on the magnitudes and determinants the 4 malnutrition conditions and the IYCF practice indicators mentioned above. For each condition, key search terms were identified and used to develop search strategies. The key terms and phrases used for searching studies on malnutrition were ‘anemia’, ‘stunting’, ‘wasting’, ‘underweight’, ‘risk factor’, ‘predictor’, determinant’, ‘meta-analysis’, ‘systematic review’, and ‘review’. The key terms and phrases used for searching studies on IYCF practice were ‘early initiation of breastfeeding’, ‘within one-hour breastfeeding’, ‘exclusive breastfeeding’, ‘duration of breastfeeding’, ‘complementary feeding’, ‘timely initiation of complementary feeding’, ‘feeding practices’, ‘dietary diversity’, ‘dietary quality’, ‘dietary frequency’, ‘meal frequency’, ‘minimum acceptable diet’, and ‘IYCF practices’. The literature search was done by two reviewers independently, with discrepancy resolved by consensus. In addition to the systematic database searching, article searching was done using the reference list of the included studies and the ‘cited by’ and ‘related articles’ function of PubMed.
Study screening and selection
The search was restricted by language and period of publication. Only English language publications, done in the period 2015–2019, were eligible for inclusion. The time restriction was aimed to ensure the findings better reflect or relate to the current nutritional situation of the country. It was also for the magnitude and determinants of malnutrition might vary from time to time. For a study to be considered as systematic review or meta-analysis, it should have to meet the following predefined criteria: (a) presented a defined literature search strategy, (b) appraised included studies using a relevant tool, and (c) followed a standard approach in pooling studies and providing summary estimates. Studies were excluded due to any of the following reasons: (a) no report on the measures of interest for this study, (b) language other than English, and (c) narrative reviews, editorials, correspondence, abstracts, and methodological studies. When a study reported on more than one malnutrition conditions or IYCF practice indicators, all reports were extracted as long as they were reported following appropriate methods. The screening and selection of studies was conducted in two stages. First, title and abstract reviewing was done. Then, full-text reviewing was done for studies included after title and abstract reviewing.
Data extraction
Data from the included studies were extracted using a standardized data abstraction form, developed in excel sheet. For each study, the following data were extracted: (a) identification data (first author’s last name and publication year), (b) type of malnutrition condition or IYCF practice indicator assessed, (c) measure of magnitude (prevalence for malnutrition, coverage or level of practice for IYCF indicators) or measure of association (odds ratio or relative risk) with 95% confidence intervals, (d) number of studies included, (e) total number of samples included, (f) risk factors (determinant or predictor reported) for the main outcome variable(s) in the study, (g) publication bias assessment methods and scores, (h) quality assessment methods and scores, (i) data synthesis methods (random or fixed-effects model), and (j) the main conclusion of the study. When a study provided two different estimates (i.e., one based on random-effects model and the other based on fixed-effects model) on the same outcome, we extracted the estimate from random-effects model if the associated between-studies heterogeneity (Higgin’s I2) [18] was > 50% and estimate from fixed-effects model if the associated heterogeneity was < 50%.
Study quality and reliability assessment
The methodological quality of the included SRM studies was assessed using the Assessment of Multiple Systematic Reviews (AMSTAR) tool [19]. It consists of 11 questions that measure the quality of the approaches used for pooling the empirical studies included in the review and summarizing their estimates. The tool has been validated and frequently used for appraisal of the quality of SRM works. The quality scoring was done out of 11, with scores 8–11, 4–7, and < 3 indicating high, medium, and low qualities, respectively. The grading was done by two reviewers, with discrepancies resolved by discussion and consensus.
Data synthesis
Both quantitative and qualitative approaches were used to summarize the estimates of the included studies. When two or more estimates were provided on exactly the same topic, we presented the range of the estimates and also calculated a summary (pooled) estimate. The choice of the meta-analysis model was guided by the between-studies heterogeneity, which was assessed by Higgin’s I2-Statistics [18]. According to Higgins et al. I2 < 49%, 50–75%, and > 75% represents low, moderate, and high levels of heterogeneity, respectively. We intended to pool the estimates with fixed-effects models if the level of heterogeneity was < 50%. However, there was a high level of between-studies heterogeneity. Thus, the pooled (summary) prevalence estimates were calculated with the DerSimonian-Laird random-effects model, which accounts for both within-study and between-studies variations [20]. We intended to assess publication bias by visual inspection of funnel plots, Begg’s rank or Egger’s regression tests, as appropriate. However, it was not possible to assess publication bias as there were inadequate numbers of studies, which under-power any of these methods. A minimum of 10 studies is needed to evaluate publication bias [21]. Stata version 15.0 software (StataCorp, TX USA) was used for the quantitative analyses. A summary list of determinants of malnutrition and poor IYCF practices was prepared.
Ethical consideration
This study was done using data extracted from published studies. Thus, no study participants’ consent or ethical approval was needed.