Study design and protocol registration
The protocol of the current systematic review and meta-analysis was designed following the “Preferred Reporting Items for Systematic Reviews and Meta-Analysis Protocols” (PRISMA-P 2015) guidelines [19] and prospectively registered in the PROSPERO database with the protocol registration number of CRD42020187022.
Study setting
This review was conducted in Ethiopia, which is a highly populated country in eastern Africa. Currently, the population is estimated to be more than a hundred million. Because of the rapid growth of the population, the number of healthcare facilities is significantly increasing [20, 21]. Currently, the healthcare facilities are grouped into three major categories, including primary, secondary, and tertiary levels. Earlier in 2011, a total of 22,792 health facilities were registered in the country to provide different health services for the population in their catchment area. From this, about 125, 2999, 15,668, and 4000 contributed by hospitals, health centers, health posts, and private clinics, respectively [22]. The health centers and health posts provide primary healthcare services, and approximately 40000 and 3000–5000 population, respectively, is allocated for them. Likewise, primary hospitals were established to serve about 60000–10000 population. General and specialized hospitals cover a wide area, and they mainly provide specialized and referral services for 1 to 5 million population [23]. Currently, with the rapid increment of health facilities, the ratio of the healthcare worker to the health facilities is still inadequate [24].
Article searching strategy
Before starting the actual work, the PROSPERO database was searched to check the presence of similar projects related to this topic. Literature searching strategy, selection of eligible articles, data extraction, data analysis, and result reporting has done according to the PRISMA guidelines [25]. Articles searched from Science Direct, HINARI, Medline through PubMed, African Journals Online (AJOL), and TRIP database databases using a combination of keywords and Boolean functions the PubMed search string is attached as a supplementary file (Supplementary materials 1). All the mentioned databases searched in English without publication year restriction. The database-specific search strings were developed according to the database requirements. Besides, to include as many articles as possible, manual hand searching on google and Google Scholar was done. Moreover, reference lists of both included and excluded studies were screened. The most recent database search was done on April 1, 2020.
Article selection, eligibility, and data extraction
The searched studies imported into EndNote X9 software and duplicate articles were removed. Both authors screened the articles independently by title, abstract, and full-text to identify eligible studies. Studies were considered as eligible if they were primary studies and accessed in full-text format, conducted in Ethiopian settings, and published in English from peer-reviewed journals. Besides, studies with prevalence data clearly stated or if missed the presence of adequate data to calculate the prevalence (known sample size and number of satisfied customers) considered. The data abstraction form prepared in the Microsoft Excel Spreadsheet which includes; first author's name, year of study, publication year, region, type of health facility, study group, study design, sample size, sampling technique, and the number of study participants satisfied with the laboratory services. Both authors extracted the data independently, and any disagreement (inconsistency) was resolved by discussion.
Quality assessment
The quality assessment was done independently by authors using the Joanna Briggs Institute (JIB) quality assessment tool for prevalence studies [26]. The tool has nine quality domains with yes, no, unclear, and not applicable response options including; 1) appropriate sampling frame, 2) proper sampling technique, 3) sufficient sample size, 4) description of the study subject and setting, 5) appropriate data analysis, 6) use of valid methods, 7) use of valid measurement, 8) appropriate statistical analysis, and 9) adequate response rate [26]. Operationally, 1 and 0 values provided for yes and (no and unclear) responses, respectively. Finally, the composite and mean scores are computed. Studies with quality scores below the mean value and "mean score and above" were categorized as having a high and minimal risk of bias, respectively. The quality of data abstraction (inter-rater agreement) assessed using Cohen's Kappa, and the inter-reliability coefficient was found to be 0.784 (p < 0.001) that indicates excellent agreement.
Data synthesis and analysis
Data were analyzed using the malaprop program of STATA software, and the Freeman Tukey double arcsine transformation (ft) was enabled to include proportions close to 0 and 1 [27]; otherwise, those studies with 1 and 0 proportions could be omitted and lead to a biased estimate. This program computes the weighted pooled estimate and then perform back-transformation on the pooled estimate. The time transformed prevalence is weighted very slightly towards 50%, which enables 0 prevalence studies included in the analysis [28]. When there is evidence of across study heterogeneity, the random-effects model is recommended for analysis [29]. In this case, the Dersimonian and Laird method is most used [30]. The presence of heterogeneity among studies checked using 𝐼2 test statistics, which estimates the presence of observed differences between-studies due to heterogeneity. The 𝐼2 value can range from 0 to 100%, and 0% indicates the absence of heterogeneity; whereas, 100% is a definitive indicator of significant heterogeneity. The 25%, 50%, and 75% values represent low, medium, and high heterogeneity between studies, respectively [31]. In addition, a p-value of <0.05 is used to declare heterogeneity [32]. In this meta-analysis, the I2 value was high (97.77%), which an indication of significant heterogeneity. Due to this reason, the analysis conducted using a random-effects model at 95% CI as opposed to the fixed effects model to adjust the observed variability among studies. The possible sources of heterogeneity are investigated through stratified analysis, sensitivity analysis, and meta-regression. Visual inspection of funnel plots and results of Egger's weighted statistics were used to investigate the presence of publication bias and small-study effects. All data management and statistical analysis performed using STATA software version 16.0 (Stator LLC College Station TX 77845, USA for windows version).