Routine health information use among healthcare providers in Ethiopia: a systematic review and meta-analysis

The rationale for the review

In Ethiopia, the magnitude of using routine health information for evidence-based decision-making was evaluated by different scholars. However, the review result indicated that there were inconsistent findings ranging from the lowest 22.5%12 to the highest 78.5%30 both in Amhara National Regional State, Ethiopia. Moreover, the lowest and the highest effect sizes were obtained from relatively low (395) and high (720) sample sizes, respectively. Because of this, it would be challenging for managers, planners and health practitioners to decide which evidence to employ for making decisions. Thus, the primary goal of this review was to aggregate the findings of many studies to derive a single impact estimate.

Review questions

1. What proportion of routinely generated health information is used by healthcare providers in Ethiopia?

2. What are the factors that are associated with routine health information use among healthcare providers in Ethiopia?

Protocol and registration

The Preferred Reporting Items for Systematic Reviews and Meta-Analyses 2020 checklist was used to develop the review protocol.32 This review has been registered at PROSPERO with registration identification number CRD42022352647.

Eligibility criteria

The primary research included in the review should focus on the routine health information use, with or without its associated factors. The papers may be published or unpublished but must be written in English. All of the papers reviewed should have been produced and published at any time prior to August 2022. However, publications without a full text and abstract as well as editorial reports, letters, reviews and commentaries were excluded from the study.

Search strategy

Between 20 August 2022 and 26 August 2022, a thorough and methodical search of the literature was conducted using electronic databases such as PubMed, Global Health, Scopus, EMBASE, African journal online, Advanced Google Search and Google Scholar.

Study selection and data extraction

The Mendeley software, author names, location and setting, participant counts, study dates and study duration were all used to remove duplicate papers found in various databases. The inclusion and exclusion criteria of the review were also used for the study selection process. To ensure the homogeneity of our search, each reviewer independently selected the appropriate papers for the review, which were then gathered. This method was in line with the Cochrane review handbook’s fundamental guidelines for choosing studies and extracting data, which note that data may be presented in a number of formats but are commonly translated into a format appropriate for meta-analysis. Additionally, multiple reports of the same study need to be linked together, and data should be extracted from study reports by at least two people, independently.33 So that, this review employed two individuals (MMT and TMY) for both study selection and data extraction.

Both experts received a set of agreed-upon inclusion and exclusion criteria (MMT is the lead researcher and TMY volunteered). Each of them evaluated the articles and decided which ones to include or leave out of the analysis. In light of this, MMT rated 23 articles to be included but rated to exclude 867 articles out of the total articles accessed (890), while TMY rated 20 items to include but rated 870 articles to omit. However, both raters agreed to combine their exclusion of 865 items and inclusion of 18 articles. It was discovered that the computed kappa statistics was 0.80. According to the interpretation of Cohen’s kappa,34 there was a high level of agreement between raters because the kappa value was 0.80, which was between 0.80 and 0.90 and meant that 64% to 81% of the data were considered credible. Because both raters used dialogue to resolve their differences, a third rater was not required. Data about study region, study setting, study year, publication status, study design, sample size, sampling technique and response rate was collected using a checklist.

Outcome measurement and quality assessment

Data extraction was made after a careful review of outcome measurement. The outcome of interest for this review was routine health information use which was defined as the use of routine health data to track daily health service activities, create weekly plans, improve service delivery, display updated data, purchase drugs, mobilise resources, assist community mobilisation, identify the root of a health issue in the community, predict outbreaks and prioritise diseases.11 In order to reduce garbage in, garbage out problem of meta-analysis and Newcastle-Ottawa Quality Assessment Scale (adapted for cross-sectional studies)35 was used as indicated in table 1 below. MMT graded the articles’ quality using an evaluation tool that has three parts: selection (five stars), comparability (two stars) and outcome (three stars) (table 1).

Data synthesis and statistical analysis

A Microsoft Excel spreadsheet was used to extract the data, which was then imported into STATA V.14 for analysis. Tables, figures and forest plots were used to describe and summarise the major investigations. A random effects model with a 95% CI was used to pool the magnitude of routine health information use among healthcare providers in Ethiopia. The OR with 95% CI was used to quantify the measure of association for factors that affect routine health information use among healthcare providers. Cochran’s Q and I² statistics have been used to evaluate the heterogeneity between studies. The percentages of about 25% (I²=25), 50% (I²=50) and 75% (I²=75) would, respectively, indicate moderate, medium and high heterogeneity.36 In order to identify the apples and oranges’ problem of a meta-analysis, subgroup analysis, meta-regression and the Galbraith plot test were conducted. Egger’s regression tests and visual assessment of funnel plot asymmetry were used to determine ‘the file drawer’ problem of meta-analysis (publication bias).

Search results

A total of 890 articles were included for this particular study. From these articles, 345 removed due to duplication and 508 were removed after reviewing its title and abstract for its relevance to the topic. Finally, 23 articles were included for the review (figure 1).

Request permission

Characteristics of included studies

A total of 20 (97%) primary studies we found were published between 2006 and 2022; the other 3 (13%) primary studies included in the review were not published yet. The majority of these studies (7; 30.43%) were carried out in the Amhara,12 15 18–20 27 30 while 1 (4.3%) was from the Tigray region37 (table 2).

Meta-analysis

Routine health information use

The summary effect is an estimation of the common effect size in a fixed-effect model, which implies that the true effect size is the same across all studies (sampling error is the only reason for variability). According to a random effects model which also assumes that the true effect size varies from study to study, the studies included in the analysis constitute a random sampling of effect sizes that could have been observed in each study. Our estimation of the mean of these effects is the summary effect (variability of the effect sizes is due to systematic error).33 As can be seen from the forest plot, the existence of high heterogeneity between included studies which could be explained by I²=97.4% (interpreted as 97.4% of the variation in effect sizes is due to between-study heterogeneity not caused by sampling error) at p<0.001. Therefore, a random effects model has a mechanism to handle this variability so that this review employed a random effects model to combine the extent of routine health information use in Ethiopia.

The prevalence of routine health information use in primary studies ranged from 22.5%12 to 78.5%.30 The pooled prevalence of routine health information use in Ethiopia was found to be 53.7% with 95% CI (47.45% to 59.95%) (figure 2).

Request permission

Figure 2

Forest plot showing the pooled level of good routine health information use (Good RHIU) among healthcare providers in Ethiopia, 2022. Note: weights are from random effects analysis.

Subgroup analysis

The authors hypothesised that region and sample size might be the sources of the high heterogeneity between studies included in the review, which was confirmed at the forest plot. In order to determine the most likely reason for heterogeneity, subgroup analysis was carried out by dividing the effect sizes by study region and sample size. The effect size showed a statically significant subgroup effect for study region and sample size at p<0.001, according to the subgroup outcome.

The pooled level of routine health information use was significantly higher in southern Ethiopia (57.2% with 95% CI (48.2% to 66%)), northern Ethiopia (56% with 95% CI (39.8% to 72.3%), eastern Ethiopia (55.83% with 95% CI (51.4% to 60.3%) and western Ethiopia (49% with 95% CI (41.2% to 57%) compared with research articles from central Ethiopia (39.5% with 95% CI (35.19% to 43.81%). There is a statistically significant unexplained variation among articles from northern Ethiopia (I²=98.8%), southern Ethiopia (I²=96.2%) and western Ethiopia (I²=89.3%) at p<0.001, but there was none for eastern Ethiopia (I²=20.2%, p=0.263) and central Ethiopia (I²=39.8%, p=0.198) (figure 3). This suggests that region may be a factor in some subgroups’ heterogeneity but not in others.

Request permission

Figure 3

Subgroup analysis of good routine health information use (Good RHIU) by region of the study. Note: weights are from random effects analysis.

Regarding sample size, the pooled magnitude of routine health information use was 67.7% (95% CI (59.5% to 75.8%)), 61.5% (95% CI (47.4% to 75.6%)), 53.3% (95% CI (42.5% to 64%)), 52% (95% CI (37.8% to 66%)) and 48.3% (95% CI (38% to 58.5%)) among subgroups having a sample size of <201, >500, 201–300, 401–500 and 301–400, respectively. There was in fact a substantial heterogeneity within subgroups having a sample size of 201–300 (I²=91.9%), 301–400 (I²=97.3%), 401–500 (I²=97.4%) and >500 (I²=98.4%) at p<0.001, but there was no statistically significant heterogeneity among studies in a subgroup of having a sample size of <201 (I²=37.3%, p<0.206) suggesting sample size is a cause for heterogeneity for some subgroups while it was not for others (figure 4).

Request permission

Figure 4

Subgroup analysis of good routine health information use (Good RHIU) by sample size. Note: weights are from random effects analysis.

Meta-regression

A meta-analysis with a significant amount of unexplained heterogeneity across the studies included in the review can use the statistical technique of meta-regression. It seeks to determine whether there are differences in study characteristics (methodological diversity) that could account for heterogeneity. This only works for meta-analyses that use a random effects model. Hence, study year and Performance of Routine Information System Management (PRISM) framework use were regressed to explain the existence of heterogeneity. However, the results of the regression analysis showed that there was no statistically significant association between heterogeneity in routine health information use and between study year (p=0.228), and use of PRISM framework (p=0.099) which could be interpreted as sample size and use of PRISM framework were not identified as the causes of heterogeneity as well (table 3).

A Galbraith plot test

The Galbraith plot was also used to assess heterogeneity and detect potential outliers. In the absence of substantial heterogeneity, we expect around 95% of the studies to lie within the 95% CI region: hence, 17 out of the 23 studies were outside the 95% CI region, which indicates considerable heterogeneity among the effect sizes. Two studies lie further far away from the 95% CI region considered as outliers12 30 (figure 5). Even after removing these studies, 14 out of 21 studies were still outside the 95% CI region (figure 6)

Request permission

Request permission

Assessment of publication bias

A funnel plot was inspected graphically to determine whether it was symmetrical, with the horizontal axis showing the effect estimates from individual studies and the vertical axis reflecting the SE of the effect estimate. Studies with large effect sizes were dispersed at the top of the funnel plot in the diagram, whereas studies with small effect sizes were located at the bottom. The plot’s outcome resembled an inverted funnel with symmetry, showing that there was no publication bias (figure 7). Moreover, the Egger’s test for small-study effects was also performed but unable to show evidence of the existence of publication bias at p=0.396.

Request permission

Determinants of routine health information use

In this review, the pooled estimate indicated that healthcare providers who had training on routine health information use were 1.56 times more likely to use routine health information compared with those who did not have training towards routine health information use (adjusted ORs (AOR)=1.56, 95% CI (1.12 to 2.18)) (figure 8).

Request permission

Healthcare providers competency related to data management was also pooled and healthcare providers having good data management competency were 1.94 times more likely to use routine health information compared with those healthcare providers with poor data management competency (AOR=1.94, 95% CI (1.35 to 2.8)) (figure 9).

Request permission

The pooled estimate of meta-analysis also indicated that the likelihood of using routinely generated health information was 1.66 times higher among healthcare providers working in institutions with available standardised health information guideline compared with those who were working in institutions without a standardised health information guideline (AOR=1.66, 95% CI (1.38 to 1.99)) (figure 10).

Request permission

Routine health information use was 2.07 times higher among healthcare providers getting supportive supervision compared with those who did not (AOR=2.07, 95% CI (1.55 to 2.76)) (figure 11). Moreover, the probability of using routine health information was 2.2 times higher among healthcare providers who received feedback compared with those who did not (AOR=2.20, 95% CI (1.30 to 3.71)) (figure 12).

Request permission

Request permission

Limitations of the review

• Even if the review’s researchers did their best to include all pertinent data, some pieces of the literature may still have been left out by the authors.

• Though we used different methods to assess causes of heterogeneity, it remained unresolved.

• Furthermore, despite using extreme caution when searching, including and omitting publications, the pooled estimate could be affected by these practices.