This systematic review and meta-analysis were undertaken to estimate the prevalence of OIs in HIV-infected adults receiving ART in Ethiopia and identify the associated factors. The study protocol for this review has been registered in an international database, the Prospective Register of Systematic Reviews, by the University of York Centre for Reviews and Dissemination with identification number CRD42022296126 (URL).¹

Furthermore, modifications to the title and expected date of review finalization were made on October 17, 2022. A 17-item Preferred Reporting Items for Systematic Reviews and Meta-Analyses Protocols 2015 checklist was used to guide protocol development (29). Besides, the Preferred Reporting Items for Systematic Reviews and Meta-Analyses 2020 Checklist was used to report the review’s findings (30) (Additional file 1).

We have undertaken this comprehensive systematic review and meta-analysis to seek an answer to the following questions: (1) What is the national burden of OIs in adult following the initiation of ART? (2) What are the potential risk factors that affect the occurrence of OIs in adults following the initiation of ART?

The inclusion criteria for this review were based on the study characteristics and report characteristics determined by using the CoCoPop (Condition, Context, and Population) mnemonic (31). Thus, we included all observational studies (cross-sectional studies, case–control studies, and cohort studies) in this systematic review and meta-analysis.

Participants/Population: The individuals aged 15 and above years who participated in the studies that assessed the prevalence of OIs and/or associated factors were considered. According to Ethiopian national guidelines for HIV/AIDS treatment programs, adults include individuals aged greater or equal to 15 years (20).

Context: Limited to primary studies conducted in Ethiopia.

Language of publication: Articles reported in English were included.

Years of publication: Articles published between 2010 and 2022 were included.

Studies reporting the outcome of interest in pre-ART PLHIV, without full-text access; articles that contained insufficient information; findings from personal opinions; articles reported outside the scope of the outcome of interest; qualitative study design; case reports; case series; letters; unpublished data; and previous systematic reviews were filtered out.

Literature search strategies were developed using medical subject headings (MeSH) and text words related to the outcomes of the study. The search typically included the following electronic bibliographic databases: Excerpta Medica database (EMBASE), PubMed, Web of Science, African Journal of Online (AJOL), Google Scholar, and Cochrane Library to ensure complete coverage of the topic by accounting for variability between the indexing in each database. The literature search was limited to studies published in the English language between 1st January 2010 and February 30th 2022 which explored prevalence, and/or factors associated with OIs among adults after initiation of ART. The reference lists of included studies identified through the search were scanned to ensure literature saturation. Where necessary, we also searched the authors’ files to ensure that all relevant materials had been captured.

We reviewed grey literature via Google. In addition, unpublished studies were sought from the official repositories of Ethiopian universities in particular Addis Ababa University and Jimma University. For the advanced search in PubMed, the following steps comprised the search process: Initially, the search statement was divided into main concepts: prevalence, Incidence, Opportunistic infections, associated factors, adults, antiretroviral therapy, and Ethiopia. Subsequently, we gathered keywords from Google scholar, Wikipedia, and Google for each concept, which was then searched independently in PubMed to find MeSH terms in the MeSH hierarchy tree and then combined in an advanced search.

Boolean operators (AND and OR) were used to combine these concepts as follows: (Prevalence [Text Word]) OR {“Prevalence”[Mesh] AND (opportunistic infections [Text Word])} OR (“AIDS-Related Opportunistic Infections”[Mesh]) AND (HIV [Text Word]) OR (“HIV”[MeSH Terms]) AND Predictors [Text Word] OR (Risk Factors [Text Word]) OR {“Associated factors”[Text Word] OR (“Risk Factors”[Mesh])} AND (adults [Text Word]) OR (“adult”[MeSH Terms]) “anti-retroviral agents”[MeSH Terms] OR antiretroviral [Text Word] AND [Ethiopia (text word)] OR (“Ethiopia*”[Mesh]). Finally, we filtered the results to include just the most relevant ones. The search was double-blinded and conducted from February 30th to April 10, 2022, by three authors (BZ.W, AA. K, and MS.O). A separate file with the search details was supplied (Additional file 2).

The articles that were found through the electronic database searches were exported to the reference management software, EndNote X7, where duplicate studies were then eliminated. Two authors (BZ.W and MS.O) independently screened the titles and abstracts that were obtained by the search against the inclusion criteria. To describe the extent to which assessments by multiple authors are similar, inter-rater agreement was calculated after referring to the Cochrane handbook for systematic reviews.

In this case, kappa greater or equal to 75% was considered, indicating excellent agreement. The screened articles were then subjected to a full article review by two independent authors (ZZ and BG.A). A pre-defined eligibility criterion was used to determine which records were relevant and should be included in the review. Where more information was required to answer queries regarding eligibility, the remaining authors were involved. Disagreements were resolved through discussion. Moreover, the reasons for excluding the articles were recorded at each step.

Two authors (BZ.W and MS.O) abstracted the relevant data independently by using a standardized Microsoft Excel spreadsheet. For data extraction, Joana Briggs Institute (JBI)-adopted formats were employed (32). The first author’s name, sample characteristics, regions of study, year of publication, study design, study area, outcome measures, timing and procedures of data collection, response rates, knowledge of and attitudes toward epilepsy were collected.

The reliability agreement among the data extractors was evaluated and verified using Cohan’s kappa coefficient after data was recovered from 30 percent of the primary studies. As a consequence, the kappa coefficient’s strength of agreement was divided into four categories: low (0.20), fair (0.21–0.40), moderate (0.41–0.60), good (0.61–0.80), and virtually perfect agreement (0.81–1) (33). A kappa statistic value of more than or equal to 0.5 was regarded as congruent and acceptable. In the case of disagreements between the two data extractors, a third author (BG.A) was involved in adjudicating unresolved disagreements through discussion and re-checking of the original articles.

The number of adult patients who had OIs was divided by the total number of adults taking ART and multiplied by one hundred to calculate the pooled prevalence of OIs among adults after the initiation of ART. The pooled effect for associated factors was investigated using the OR. Furthermore, variables identified as risk factors for OIs in at least three studies were taken into account.

Authors of primary studies classified adherence to ART as good, fair, or poor using the definition from the national HIV care follow-up form. Good adherence was defined as taking more than 95% of prescribed doses, fair adherence as taking 85–94% of prescribed doses, and poor adherence as taking less than 85% of prescribed doses as documented in patient records. Nutritional status was also based on national HIV care and treatment guidelines, and a body mass index of less than 18.5 kg/m2 was a cut-off point for undernutrition irrespective of its severity. Furthermore, for adolescents and adults, advanced HIV diseases were defined as CD4 cell counts of 200 cells/mm3 or a WHO stage III or IV event (20).

To assess the quality of the studies, the Joana Briggs Institute (JBI) critical appraisal checklists for descriptive cross-sectional studies comprised of 8 components (Additional file_ 3A), analytic cross-sectional comprised of 7 components (Additional file_3B), cohort studies comprised of 11 components (Additional file_3C), and case–control studies comprised of 10 components (Additional file_3D) were employed. Three authors (ZZ, BG. A, AA. K, and MS.O) independently assessed the methodological quality of each study. Disagreements were resolved through consultation with a third independent reviewer (BZ.W). Studies with a score of 7 or higher after being evaluated against these criteria were considered low risk and included in this systematic review and meta-analysis.

The extracted data were imported from a Microsoft Excel spreadsheet into STATA MP 16 statistical software (StataCorp LP, 4905 Lakeway Drive, College Station, TX 7845, United States) for analysis. The heterogeneity of the results was visually examined via the forest plots with pooled estimates. Thus, its presence was confirmed subjectively with a lack of overlap between the CI.

In addition, the statistical heterogeneity was explored more formally by using Cochran’s Q test (x²) at p-value <0.10 indicating significant heterogeneity. Another heterogeneity measure, Higgins and Thompson’s I² statistics, was employed to estimate the percentages of the between-study variability where,0%,25 to50%,50 to 75%, and greater than or equal to75% indicated no heterogeneity, low heterogeneity, moderate heterogeneity, and high heterogeneity, respectively, (29).

The random-effect meta-analysis model was used to estimate Der Simonian and Laird’s pooled effect due to the presence of considerable statistical heterogeneity. Subgroup meta-analysis based on the study regions as covariates, meta-regression, and sensitivity analyses were also performed to investigate the source of statistical heterogeneity. Publication or dissemination bias was examined subjectively using funnel plots and objectively using the nonparametric rank correlation test of Begg (34) and the regression-based test of Egger for small study effects (35), with p < 0.05 being taken into consideration to declare potential publication bias. Results were presented in the form of tables, texts, and figures.

Our search was restricted to articles published in the English language between 1st January 2010 and February 30th 2022 in the electronic databases of PubMed, Web of Science, and EMBASE. In addition, Google, Google Scholar, and AJOL were searched. Through systematic and manual searching, 640 primary articles were found. Due to duplication, 572 articles were removed. The remaining 68 were screened based on their title and abstract, with 38 being eliminated as unrelated to our study. Finally, 30 full-text primary articles were evaluated against eligibility criteria, and 16 were selected for quantitative analysis (Figure 1).

This systematic review and meta-analysis included 16 studies involving 6,163 participants, with a response rate ranging from 96 to 100%. More than 50% (n = 3,558) of the participants were female. Thirteen studies employed a cross-sectional study design (25, 36–47). The remaining two were cohort studies (24, 27), and one was a case–control study (48). All studies were conducted from 2013 to 2021 in different geographic locations in Ethiopia. Four of the primary studies included were based in Ethiopia’s Southern Nations Nationalities Peoples’ Regional States (SNNPRs) (25, 27, 42, 44), three in Amhara (37, 45, 48), three in Addis Ababa (36, 38, 43), three in Tigrai (24, 41, 46), one in Oromia (47), one in Sidama (39), and one in Harari (40). The highest (88.44%) (25) and the lowest (19.72%) (37) Prevalence of OIs was reported in studies conducted in Tercha Hospital, southern Ethiopia, and Gonder University Hospital, northern Ethiopia, respectively (Table 1).

Furthermore, about 13 primary studies assessed the spectrum of OIs (24, 25, 36–45, 47).

In terms of sampling techniques, systematic random sampling techniques (37, 40–42), simple random sampling techniques (24, 27, 43–45, 47, 48), combined stratified and simple sampling techniques (39), and consecutive count (25, 36, 38) were used to select the required study subjects. Furthermore, the authors of these studies employed a structured questionnaire, and or pre-tested and refined checklists to obtain necessary data. A complete summary of descriptive characteristics of all included primary studies is provided (see Additional file 4).

Given the substantial statistical heterogeneity in the fixed-effects model, the pooled estimate was determined using a random-effects model. Thus, an overall pooled prevalence of OIs among HIV-infected adult patients on ART in Ethiopia was found to be 43.97% [(95% CI, 38.59, 49.34), I² = 92.4%, p < 0.001, n = 4,195] (Figure 2).

Subgroup analysis was conducted based on Ethiopia’s administrative regions where primary studies were based, and the year of publication. Thus, we observed regional variations in the prevalence of OIs in this review. The prevalence of OIs was found to range between 31.36% [(95% CI: 30.94, 37.78), I² = 0.0%, p < 0.0681, n = 762] in random effects pooled meta-analysis in Amhara region and 60.97% (n = 497) in Ethiopia’s Oromia region with single study (Figure 3).

Considering the introduction of the WHO test and treatment recommendation, we also conducted a subgroup analysis by categorizing it before and after 2016. Thus, the pooled prevalence of OIs after the test and treatment recommendation was found to be 44.71% [(95% CI: 37.68, 51.75), I² = 93.9%, p < 0.001, n = 3,058] (Figure 4).

Random-effects meta-regression using sample size and year of publication as covariates was performed to explore the source of heterogeneity at a 5% significance level. As shown in Table 2, these covariates were not found to be the source of heterogeneity.

We conducted a leave-out-one meta-analysis to explore the effect of each study on the overall prevalence of OIs while gradually excluding each study. Results showed that the combined prevalence of OIs did not significantly change as a result of the excluded study (Table 3).

Publication bias was assessed subjectively using a funnel plot and objectively by the regression-based test of Egger and the nonparametric rank correlation test of Begg at p < 0.05. Looking at the funnel plot (Figure 5), the number of studies on the left and right sides of the plot is the same; thus, the plot is symmetrical, indicating that no publication bias was detected.

Moreover, neither Egger’s linear regression test (t = −0.67, p = 0.517) nor Begg’s rank correlation test (z = 0.16, p = 0.876) was statistically significant for a pooled prevalence of opportunistic infections, corroborating that there is no evidence of small study effects.

About 13 (81.25% of the included studies) in this review reported the patterns of OIs (24, 25, 36–45, 47). Accordingly, oral candidiasis (25.31%) was the most prevalent OI among HIV-infected adults after the initiation of ART. Pulmonary tuberculosis (22.04%), and herpes zoster (14.95%) ranked second and third, respectively (Figure 6).

About 15 primary studies included in this meta-analysis have reported various factors as determinants of OIs among HIV-infected adult patients who were taking antiretroviral therapy in Ethiopia (24, 25, 27, 36–41, 43–48). We identified five predictor variables out of seven extracted variables, each reported by at least three primary studies, that were found to be associated with the development of OIs. These include level of adherence to ART, body mass index, baseline CD4 T lymphocytes count, and WHO HIV clinical stage. The pooled OR was used to estimate the association between these variables and the occurrence of OIs, and a statistically significant association between the outcome and predictor variables was declared at two-sided p < 0.05 with 95% CI.

To begin with, the odds of developing OIs were approximately 6 times {OR, 5.90 [95% CI (3.05, 11.40), p < 0.001, I² = 76.4%]} higher in adult PLHIV who had a good level of adherence to ART compared to fair and good adherents. Additionally, the odds of developing OIs were 3.7 times higher in adult PLHIV with under nutrition than in adult PLHIV with normal body mass index {OR, 3.70 [95% CI (2.01, 6.80), p < 0.001, I² = 71.4%]}. In this meta-analysis, it was also found that the odds of OIs occurring was 3.23 times [OR, 3.23 95% CI (2.06, 5.07), p < 0.001, I2 = 79.8%] more likely in PLHIV who had baseline CD4 T lymphocytes less than 200 cells per microliter compared to their counterparts. Moreover, the odds of exhibiting OIs were approximately 5 times [OR, 4.84 95% CI (1.83, 12.82), p = 0.002, I2 = 89.5%] more likely in adult PLHIV who had presented with advanced WHO HIV clinical stage at enrollment to the program (Table 4).