This study employs a systematic review and meta-analysis approach to estimate the prevalence of undiagnosed diabetes in Ethiopia. The reporting of the study adheres to the PRISMA (26) (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines (27–29). To prevent duplication and uphold transparency, this review protocol was registered with the International Prospective Register of Systematic Reviews (PROSPERO) under the registration number CRD42021266676.

The inclusion criteria for this review were based on the Condition, Context, and Population (CoCoPo) framework, which considered studies focused on the prevalence of UDM and associated factors (conditions) in the Ethiopian (context) adult age group (population) (30). Both published and unpublished studies in the English language were considered. However, studies specifically addressing gestational diabetes, case reports, and qualitative designs were excluded from this review.

The search strategy was aligned with the systematic review question and inclusion criteria (CoCoPo). Two authors conducted a pretest of the PubMed database, and the actual electronic search was conducted between September 25 and October 5, 2023. Three reviewers systematically searched through PubMed, Web of Science, APA PsycInfo (EBSCO), CINAHL, MEDLINE, and Hinari, focusing on articles published in English between 2012 and 2023. Hinari, a World Health Organization (WHO) portal for low and middle-income countries, granted access to databases such as Web of Science, SCOPUS, AIM, CINAHL, IRIS, and African Journals Online. Although we did not find a study, grey literature was explored through institutional repositories, specifically Addis Ababa University and the WHO library. The search process was meticulously documented in a PRISMA flow chart, ensuring a systematic and thorough approach to literature identification (27).

For each key concept, relevant free-text words and MeSH terms were employed, and combinations were established using Boolean operators like “AND” and “OR.” This approach facilitated the retrieval of pertinent articles, accounting for different synonyms used in the literature. A search strategy was developed using fundamental concepts in the review questions: “prevalence,” “magnitude,” “burden,” “undiagnosed diabet*,” “diabet*,” “population,” “factor*,” “determinant*,” “influenc*,” “risk factor*,” “predictor*,” “Ethiopia” and English language (S1).

Following the search, all identified studies were imported into Endnote Desktop v.20 for efficient reference management and citation generation. Duplicate sources and publications that do not directly relate to the research question were removed. After removing duplication, the selection process followed the following process. Firstly, an initial pilot test of sample evidence sources was conducted to ensure clarity and consistency in applying inclusion and exclusion criteria. Following the pilot test, two independent reviewers (TWA and EA) screened the title and abstract against the inclusion criteria. Finally, the full text of the selected articles in the initial phase underwent a detailed assessment against the inclusion criteria by two additional independent reviewers (HG and AG). At any stage, disagreements between two reviewers were resolved through discussion and involving a third reviewer (BFS).

The reviewer team employed a standardized data extraction form to extract the screening articles. The data extraction form was pilot-tested to assess the reliability and consistency of the extraction, as well as the appropriateness and usability of the form. Following pilot testing, two reviewers (TWA and EA) independently extracted data from papers included in the review using a finalized data extraction form (S2). The data extraction form included (but was not limited to) the first author, region, sample size, study period, study setting, study method and design, study purpose, outcome measurement tool, key findings as relevant to the review question (comorbid hypertension, overweight, smoking, family history of diabetes, sedentary behavior, and waist circumference), and participant demographic data (age). Disagreements in data abstraction between the first two reviewers were resolved by involving a third reviewer (BFS).

Given that the Newcastle–Ottawa Scale (NOS) (31) is a commonly used tool for appraising non-randomized studies (32), it was employed to assessed the quality of the included studies. The NOS comprises three categorical criteria (low, fair, and good) with a maximum score of ten points. This tool evaluates representatives of the sample, sample size, non-respondents, ascertainment of exposure, independent blind assessment, and statistical tests. Based on the NOS, we considered a study to have fair to good quality scores. Unfortunately, all primary studies were rated as fair to good quality (S3) (33–36).

The primary outcome of this review was to determine the pooled prevalence of UDM among the adult population in Ethiopia. To analyze the second outcome (factors), relevant variables associated with UDM were extracted from each included article. In examining factors linked to the primary outcome, Adjusted Odd Ratios (AORs) from the primary studies were utilized to ascertain the association between independent variables and the presence of UDM.

The risk of bias in the included studies underwent assessment using a 10-item rating scale designed for prevalence studies (37). The assessment tool encompasses factors such as representative sample size, data collection method, reliability and validity of study tools, case definition, and prevalence periods of the studies. Each observational study article was classified based on a low risk of bias (affirmative responses to domain questions) or a high risk of bias (negative responses to domain questions). A score of 1 (Yes) or 0 (No) was assigned to each study for each domain, and these domain scores were aggregated to derive an overall study quality score. Scores falling within the range of 8–10 were deemed to indicate a “low risk of bias,” scores of 6–7 signified a “moderate risk,” and scores ranging from 0 to 5 were indicative of a “high risk.” In instances of disagreements between reviewers regarding the least risk of bias classification, resolution occurred through consensus.

After conducting a search in electronic databases and other records, a total of 7546 observational studies were identified, and 3218 were removed using Endnote V.20. Subsequently, out of the remaining 4328 studies, 4035 were excluded based on title and abstract screening. Out of the remaining 293 articles, 268 were excluded due to not meeting the inclusion criteria, such as differences in population, settings, and outcomes. Ultimately, 25 articles were included in the systematic review and meta-analysis ( Figure 1 ).

In this systematic review and meta-analysis, we included 25 observational studies, encompassing 22,193 participants aged 15 years and older. The individual study cohorts ranged in size from 237 (44) to 4,371 (45) participants. The majority of the studies retrieved (n = 9) were conducted in the Amhara region (44, 46–53) followed by Southern Nations, Nationalities, and Peoples’ (SNNP) region of Ethiopia (n=7) (54–60), Oromia region (n=6) (45, 61–65), Addis Ababa (n=2) (66, 67), and Dire Dawa City Administration (n=1) (68).

All primary studies have utilized a lab-based cross-sectional design to estimate undiagnosed diabetes mellitus. Nearly all studies employed standardized blood glucose measurement tools and defined the outcome variable based on the diagnostic criteria established by the WHO, the American Diabetes Association (ADA), and the international diabetes federation (IDF). Six different measurement tools were used to assess blood glucose levels, with the glucose oxidase method being one of them. Additionally, five studies utilized the OneTouch Ultra Easy blood glucose meter ( Table 1 ).

The quality score of the included study ranged from 5 to 8. Out of 25 studies, 17 (68%) received a low bias risk. Six studies (46, 47, 50, 55, 66, 68) had a high risk of case definition bias, and four studies (47, 51, 61, 66) had a high risk of representation bias (S4).

Both funnel plots of precision asymmetry and Egger’s intercept test revealed no publication bias in the primary studies. A visual examination of the funnel plot depicted asymmetry ( Figure 2 ), and Egger’s test of the intercept (B0) yielded a value of 0.99 (95% CI: 0.94–1.04, p < 0.05). As judged by Egger’s test, no evidence of publication bias was presented at the 5% significance level.

The sub-group analysis revealed a widespread heterogeneity across the studies ( Figure 3 ). A meta-regression analysis were conducted to identify the source of this heterogeneity Subsequently, a meta-regression analysis was performed, incorporating the study year, sample size, and region as covariates. The results indicated that the listed covariates did not exhibit significance in the presence of heterogeneity across the studies ( Table 2 ).

Adjusted odds ratios obtained from primary studies were organized into three main categories: socio-demographic factors, anthropometric and clinical aspects, and behavioral perspectives. These categories were then integrated to identify the predominant risk factors associated with UDM within the adult population of Ethiopia. Commonly extracted adjusted odds ratios from the primary studies included variables such as age, waist circumference, overweight, family history of diabetes, history of hypertension, sedentary behaviors (physically inactive), and current smokers ( Table 3 ).

The pooled effect reveals that individuals aged over 29 years are more than twice as likely to be exposed to UDM in the adult Ethiopian population (Pooled Odds Ratio (POR): 2.07; 95% CI: 1.04-3.10; I2 = 32.1%). This comprehensive review, encompassing eleven articles, consistently underscores the pivotal role of advancing age a significant determinant of UDM. The collective findings emphasize that as individuals age, they exhibit a heightened susceptibility to UDM, positioning age as a critical factor in understanding the prevalence and identification of this metabolic disorder.

In this review, the pooled effect of family history was pronounced threefold for UDM (POR: 2.98; 95% CI: 1.75-4.22; I² = 0.0%). Family history emerges as a pivotal determinant of undiagnosed diabetes in Ethiopian adults, with 11 articles consistently highlighting its substantial predictive role. Demonstrating a hereditary component, individuals with a familial history of diabetes face elevated risks. The literature explores familial influence on disease dynamics, symptomatology, and healthcare-seeking behavior. These findings underscore the importance of integrating family history assessments into diabetes risk evaluations for tailored interventions in Ethiopia. Consistent evidence across multiple studies reinforces the necessity for healthcare practitioners to consider familial predisposition in enhancing preventive measures and early interventions for UDM (44, 46, 47, 49, 51, 53, 59, 63, 65–67).

The pooled odds ratio for individuals with a high waist circumference showed a more than sixfold increase in the vulnerability to UDM when compared to those with a lower waist circumference (Pooled Odds Ratio (POR): 6.32; 95% Confidence Interval (CI): 4.50-8.15; I2 = 0.0%). This comprehensive review, encompassing seven studies, consistently highlighted high waist circumference as a significant determinant of undiagnosed UDM (53–57, 61, 62, 66).

In this comprehensive review, overweight emerged as a significant determinant of UDM. The analysis revealed that individuals classified as overweight were twice as likely to develop UDM compared to those with a healthy body weight (POR: 2.05; 95%CI: 1.49-2.61; I2 = 0.0%). Eleven primary studies consistently highlighted overweight as a significant risk factor for UDM. These findings underscore the crucial role of excess weight as a substantial risk factor in UDM development, emphasizing the importance of weight management strategies in diabetes prevention efforts (44, 46, 48, 51, 52, 55–57, 63, 64, 68).

The pooled effect of fifteen studies revealed that individuals with a history of hypertension face a more than fourfold increased risk of UDM compared to those without a history of hypertension (POR: 4.80; 95%: 3.49-6.11; I² = 0.0%). This significant association was consistently observed across fifteen primary studies, collectively indicating that a history of hypertension stands out as the predominant determinant in the development of UDM. The robust findings from these studies emphasize the critical role of hypertension history as a key risk factor for UDM, recommending the necessity for targeted interventions and vigilant monitoring in individuals with a hypertensive background to prevent and manage the onset of UDM (44, 47, 48, 51–59, 61, 64, 66).

Physically inactive and smoking habits were identified as significant determinants of UDM in the primary studies. According to seven primary studies, individuals with physically inactive behaviors were more likely at risk for UDM compared to those adopting physically active behaviors. The consistent findings across these studies emphasize a significant association between sedentary habits and an increased likelihood of UDM. This underscores the importance of promoting physical activity and discouraging prolonged periods of inactivity to mitigate the risk of Undiagnosed Diabetes Mellitus among the studied population (45, 47, 57–60, 65).

Three primary studies conducted a thorough investigation into the complex relationship between smoking habits and the onset of UDM. These studies not only confirmed a correlation between smoking and UDM but also emphasized the need for a detailed exploration to uncover the specific impact of smoking on the likelihood of UDM occurrence. This nuanced approach sought to achieve a comprehensive understanding of how smoking, as a distinct lifestyle factor, contributes to the intricate landscape of diabetes (45, 51, 55).

However, the pooled effect did not indicate a significant association between sedentary behaviors (POR: 2.36; 95% CI: 0.40 to 4.32, I² = 0.0%) and smoking habits (POR: 7.18; 95% CI: -0.12 to 14.48, I² = 81.8%). Notably, the absence of a significant association in the pooled effect does not necessarily imply that smoking and sedentary behavior do not pose a risk for UDM.

This systematic review and meta-analysis possess notable strengths. It is comprehensive, pooling the results of various studies in the country to offer stronger evidence regarding different factors associated with UDM among the adult population in Ethiopia. The review incorporated a relatively large number of individuals with diabetes (N = 22,193), surpassing the sample sizes of individual studies. Moreover, it systematically investigated numerous factors associated with UDM and presented insightful suggestions. Despite these strengths, the study is not without limitations. While most of the included studies demonstrated good quality, it is crucial to note that all primary studies were cross-sectional, limiting the depth of this analysis. Additionally, despite employing extensive and diverse search strategies to uncover all possible relevant literature, some grey literature, such as conference proceedings, proved challenging to locate.

This review on UDM in Ethiopian adults has far-reaching implications for public health and healthcare strategies. The findings illuminate the prevalence and determinants of UDM, prompting crucial considerations for various facets of intervention. Firstly, heightened public health awareness campaigns are urgently needed to address undiagnosed diabetes. Educational programs are essential to inform the Ethiopian population about risks, symptoms, and consequences, fostering early detection and management. Additionally, the study underscores the necessity to fortify healthcare infrastructure, specifically in terms of diabetes screening and diagnostic facilities. Investments in accessible tools and healthcare professional training are imperative for effective undiagnosed diabetes management.

Policymakers in Ethiopia are urged to prioritize early diabetes screening, suggesting the integration of routine screenings into primary healthcare services as a key strategy for prompt identification. The comprehensive nature of the review emphasizes the importance of continued research efforts, advocating for collaboration between researchers, healthcare institutions, and governmental bodies. Lastly, evidence-based public health interventions tailored to the Ethiopian context, such as community-based screenings and lifestyle modification programs, are deemed necessary to prevent and manage diabetes effectively. This holistic approach underscores the multidimensional response required to address the challenges posed by undiagnosed diabetes in Ethiopian adults.