The protocol for this review was registered in PROSPERO (https://www.crd.york.ac.uk/prospero/display_record.php?ID=CRD42022355933), and amendments were being made during the review process. The “Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 statement: an updated guideline for reporting systematic reviews” was used as a framework for all sections of the review (12) (Supplementary material 1).

Analytical, prevalent, and retrospective cross-sectional studies, as well as mixed study designs, were included. All published studies in English, both in community and institutional settings, and on the WTP for SHI among the formal sector in Ethiopia were considered. The following study parameters were also used to decide which studies to include: outcome variables, population (study units), year of the study, context (regions), sample size, and response rate.

All other studies with incomplete data and a high risk of bias were excluded. If a study had both published and unpublished copies with identical reports, the unpublished copy was excluded. Furthermore, if a study was published in more than one journal, it was considered a duplicate, and the most recently published one was chosen to be included in the review.

On September 1, 2022, database searches were conducted on Scopus, Hinari, PubMed, Google Scholar, and Semantic Scholar (Supplementary material 2). Resources from PubMed and Hinari were searched manually. However, Scopus, Google Scholar, and Semantic Scholar have been searched using the “Perish or Publish” database searching tool version 8 (13). Registries like the Ethiopian Health Insurance Service (EHIS) and the general web were also searched for additional information. The databases were searched using text words and indexed terms, which included “willingness to pay,” “social health insurance,” “factors,” and “Ethiopia. “Additional filters were also employed: year of study, publication year, content type, discipline, and language. To find more relevant studies, the reference lists of studies that met the inclusion criteria were searched.

After duplicates and irrelevant studies were excluded using “Zotero” reference manager version 6, two reviewers, EMB and HNT, screened the included studies independently. The selection of studies was carefully screened by these two researchers. First, the articles were refined by their title and abstract; second, by full-text revision by these authors, independently and finally together, until reaching consensus. When disagreements arose, a third reviewer was contacted to resolve the difference. Then, all studies that fulfilled the eligibility criteria and had a score of “low” or “medium” risk of bias were included.

A Microsoft Excel spreadsheet was used for data extraction. Two reviewers, EMB and HNT, independently extracted the data, compared their conclusions, and came to an agreement. If not, a third reviewer was invited to help these two reviewers reach consensus. Moreover, the authors of the studies were contacted to collect the missing information.

The outcome variable, the population (study units), the year of study, the context, the sample size, the response rate, and funding sources were extracted by the excel spreadsheet. The main outcome of this review was WTP for SHI. The additional outcomes were the factors affecting the WTP for SHI.

The risk of bias for the included studies was assessed independently by two reviewers, EMB and HNT, using the Joanna Briggs Institute (JBI) critical appraisal checklists. The bias was assessed on: the criteria for inclusion in the sample; the description of study participants and setting; the validity and reliability of measurement; confounding and strategies to deal with it; and the appropriateness of the outcome measure. Accordingly, studies with a score of 7 or higher were labeled as low risk, 5–6 medium risk, and 4 or lower high risk. Then, those studies with low and medium risk were included in the review. Any inconsistencies were resolved by discussion and involving a third reviewer, as necessary.

Prevalence, proportion, inverse variance, and odds ratios were calculated for each study. For the summary effect, the X², z-value, p-value with a 95% CI, and odds ratio were computed.

We used thematic analysis for the qualitative synthesis. Coming to the quantitative part, first, data (such as events, non-events, participants, and sample sizes) were extracted from each study using a Microsoft Excel spreadsheet. Then, preliminary effect measures, like the prevalent rate and proportion, as well as odds ratios of WTP for SHI, were computed in the spreadsheet. Finally, a generic inverse variance analysis was employed to estimate the overall effect sizes using Revman-5. The summary odds ratio with 95% CI was computed based on the random effect model. Sub-group analyses were conducted to compare the effect estimates across studies based on region (context) and profession. The level of overall statistical significance was determined using a p-value < 0.05 with a 95% CI.

Reporting bias was assessed by considering whether the studies were published or not. It was also examined by the year of studies and the publication years of them. For those studies with incomplete or missing data, the study authors were contacted. The studies with incomplete data were excluded.

The I² statistic was used to evaluate between-study heterogeneity. The influence of each study on the overall meta-analysis was measured using inverse variance (percentage of weight). The funnel plot was used to examine the possibility of bias between studies (publication bias). Sensitivity analysis was performed by unchecking outlier studies.

In total, 79 resources were identified (Figure 1). Sixty-six of them were identified from databases: Scopus (n = 7), Hinari (n = 21), PubMed (n = 11), Google Scholar (n = 17), and Semantic Scholar (n = 10). The remaining 13 sources were obtained from registries (n = 2) and websites (n = 11). Thirty-seven studies were identified after duplicates (n = 42) were removed. Following the exclusion of eight studies based on relevance, 29 studies were found to be eligible for title and abstract evaluation. Through title and abstract review, 22 studies were chosen to be eligible for full text evaluation. After one and two studies were removed due to incomplete data (14) and a high risk of bias (15, 16), 19 studies were included for the qualitative synthesis. Eighteen studies were then included for the quantitative synthesis after one was excluded due to its report of the WTP for SHI being below the set premium (3%) (17).

Most of the included studies were conducted in the Amhara region (n = 6), followed by Addis Ababa (n = 6), Tigray (n = 3), SNNPR (n = 2), Oromia (n = 1), and Harar (n = 1). In total, the sample population of the included studies was 9,325, of which 9,084 (97.42%) were found to be actual participants. The summary result of the individual study characteristics is shown in Table 1.

After the risk of bias for the included studies was assessed using JBI's critical appraisal tools, those studies with a low or medium risk were included in the study (Figure 2).

The WTP for SHI was found to be influenced by sociodemographic factors like education level (20–22, 25, 31, 32), income (17, 19–21, 27, 28, 31, 34), age (26, 28, 31), marital status (28, 33), occupation or job description (28), family size (29, 30, 34), and job experience (16); health and illness status such as self-rated healthiness (16), the presence of acute (20, 21, 31), and chronic illnesses (34); health service related factors like previous medical bills (16, 18, 21, 22, 24, 27, 28, 30, 35), referral system (26), regular medical checkup (26), and health service quality (16, 19, 24–27); awareness or knowledge (16–18, 20–23, 27, 29, 30, 32, 34, 35); perception (24, 31); attitude (17, 21, 27, 33, 34); and factors related to the scheme such as trust (17, 23), premium amount (19), and the scope of the benefit packages (19, 25).

A total of 9,084 participants were found from all 19 included studies, of which 38.46%, 26.45%, 17.83%, 11.23%, 3.03%, and 3.00% were in Amhara, Addis Ababa, Tigray, SNNPR, Oromia, and Harar, respectively (Table 1). However, for the quantitative synthesis, 18 studies with 8,386 participants were included, of whom 3,543 were willing to pay for the scheme, which provided a pooled WTP of 42.25% for SHI (Table 2).

As indicated in Table 2 and Figure 3, the pooled result showed that the WTP for SHI was found to be 16% less likely and was not found to be significant (OR = 0.84; 95% CI: 0.52–1.36). The WTP was not also found to be significant for the following sub-groups: Amhara (OR = 0.72; 95% CI: 0.33–1.58), Tigray (OR = 1.40; 95% CI: 0.43–4.58), Addis Ababa (OR = 0.72; 95% CI: 0.30–1.73), SNNPR (OR = 0.43; 95% CI: 0.01–18.06), but for Oromia (OR = 0.38; 95% CI: 0.29–0.50) and Harar (OR = 8.38; 95% CI: 5.71–12.31). However, when the outlier was unchecked (32), it became significant (OR = 0.58; 95% CI: 0.37–0.91) (Figure 4). The lowest WTP for the scheme was in Oromia, while the highest was in Harar.

We found that seven studies were conducted on homogeneous populations: teachers and health professionals. Thus, as shown in Figure 5, regarding the sub-group analysis by profession, teachers were 7.67 times more likely to pay for SHI (OR = 3.22; 95% CI: 1.80–5.76) than health professionals (OR = 0.42; 95% CI: 0.19–0.93).

The reports of the included studies for the factors affecting the WTP for SHI were not consistent. So, it was difficult to determine the direction of association for both the qualitative and quantitative synthesis.

The I²-values of the sub-group analyses were 93% to 100%, which are indicators of substantial heterogeneity (36). Thus, since the I²-value was >50%, a random-effect model was used to pool the WTP for SHI with a 95% CI (37). Through sensitivity analysis, a study (32) was found to be an outlier, though the heterogeneity was not changed much (Figure 6).

Because they must pay for health bills out of their own pockets, about 100 million individuals are forced into extreme poverty every year (57, 58), which mandates governments to design and implement effective strategies to secure their citizens from such catastrophes. The scenario is not different for Ethiopia. However, catastrophic health spending represents a sufficient but not a necessary condition for financial hardship to occur. This is because financial hardship monitoring also relies on household budget surveys, household income and expenditure surveys, household living standard surveys, or socioeconomic surveys (59). Moreover, because the Bismarck model associates the right to healthcare with employment through mandatory payroll deductions, committed compulsory implementation might be equally important (4).

The direction of association between the dependent variable (WTP for SHI) and the independent variables was not estimated due to the variability of the reports of the included studies. Thus, the variables or factors were reported using qualitative synthesis. Studies other than English were not included. The data was pooled despite high heterogeneity.