Introduction

Diabetes mellitus is a metabolic disease that may lead to chronic microvascular and macrovascular complications [1]. Diabetic retinopathy (DR)—the most common complication of diabetes mellitus—is one of the leading causes of preventable blindness in the adult population [2]. Vision impairment and blindness due to diabetes may be irreversible if timely treatment is not provided, affecting the individual’s functional capabilities and self-care [3]. Moreover, DR is considered a risk factor for other diabetes complications [4].

The International Diabetes Federation estimates that 537 million adults live with diabetes in 2021 [5]. In Brazil, a systematic review estimated a 6.9% prevalence of diabetes in the population based on studies published after 2010 [6]. With an aging population, coupled with growing rates of diabetes, a higher burden of DR and demand for eye care and treatment are expected [2]. The international literature on DR epidemiology has several population studies, such as the WESDR [7], UKPDS [8], DCCT [9], and ETDRS [10], and a recent systematic review by Teo et al. has concluded that—amongst individuals with diabetes—the global prevalence of DR is estimated at 22.27% [2]. However, factors such as varying levels of surveillance, different socio-economic factors, and health systems organization can prompt differences in estimated DR prevalence among countries [11,12,13,14].

Brazil is a large upper-middle-income country that hosts the world’s sixth largest population of individuals with diabetes [15]; it is also the country with the largest free public health care system [16], on which around 75% of its population relies [17]. National data on the prevalence of DR are lacking, but regional studies indicate a prevalence ranging from 7.6 to 44.4% of individuals with diabetes, with great regional and methodological variations in each survey [18,19,20,21,22,23].

The diagnosis of DR comprises the detection of ophthalmological lesions in ophthalmoscopy or color fundus photographs that are considering in classifying DR. The classification defines the prognosis and the need for treatment. More advanced degrees of DR have a worse prognosis. DR is classified as proliferative and non-proliferative, being divided into mild, moderate and severe; macular edema may or may not be present [24].

Because DR is a major public health issue, demanding thoughtful resource allocation, and since blindness is preventable with timely treatment, planning from health authorities is crucial. Since no national strategies or standardized workflows for DR screening and management in the Brazilian public health system currently exist [25], estimating the prevalence of DR and its regional variations is a crucial step for designing such policies and for an effective resource and workforce allocation. This study aims to assess the prevalence of DR in Brazil; additionally, this study aims to evaluate other aspects of DR epidemiology, such as geographic differences and risk factors.

Methods

This report describes a systematic review and meta-analysis of studies describing the DR prevalence in individuals with diabetes in Brazil. All procedures herein described were conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses and Meta-analyses Of Observational Studies in Epidemiology guidelines [26]. The protocol for this review was registered and publicly available at PROSPERO (CRD42022362777).

Search strategy

Three databases (PubMed, LILACS, and EMBASE) were systematically searched using terms related to diabetes, retinopathy, and prevalence. Papers written in English, Portuguese, or Spanish were retrieved, from inception to February 2022. The detailed search strategy can be consulted in Additional file 1: Table S1.

Eligibility criteria

Articles meeting the following criteria were included: (1) designed as cross-sectional, cohort, or case–control studies, (2) conducted in Brazil, (3) describing the frequency of adults with DR among those with type 1 diabetes or type 2 diabetes. Studies including pregnant women, patients with diabetes other than type 1 or type 2 diabetes mellitus, or conducted outside Brazil were excluded.

Study selection

A.N.G. and M.A.R. independently reviewed titles and abstracts considering eligibility criteria. Once the initial screening was completed, full-texts were reviewed by both researchers. Discrepancies in all steps were resolved by consensus. Figure 1 shows a PRISMA diagram depicting the study selection process.

Fig. 1
figure 1

Flowchart of studies

Full size image

Data extraction

Two reviewers (G.L. and M.A.R.) independently extracted relevant data from the included studies using a standardized form and following a predetermined protocol. Extracted data included: title, first author, year of publication, language, study objective, study design, year of data collection, municipality and federative unit studied, sample size, gender, type of diabetes, duration of diabetes, skin color, mean age, frequency of DR, frequency by type of DR, classification of DR used, and diagnostic method of DR.

Risk of bias assessment

Risk of Bias (RoB) was assessed in duplicate by L.P.S. and T.A.C., using a tool developed by Hoy et al. [27] for RoB estimation in prevalence studies. The tool comprises 10 items, classified as low or high risk of bias and a summary item that stratifies studies in low, moderate, or high risk of bias. Disagreements were resolved by discussion, with involvement of a third author when necessary. RoB plots were generated using the ‘robvis’ package for R (version 0.3.0) [28].

Statistical analysis

Overall and subgroup prevalence estimates, with corresponding 95% confidence intervals, were estimated based on reported frequencies of DR in individuals with diabetes within the included studies. Pooled estimates were obtained by a random-effects inverse variance approach with arcsine transformation, assuming heterogeneity between studies due to the epidemiological nature of primary literature [29]. Confidence intervals for individual studies were estimated using the Clopper-Pearson approach. Percentage of total variability due to between-study heterogeneity was estimated by I2 statistic. Subgroup analyses were also performed to determine whether the following variables affected prevalence estimates: geographic region (South, Southeast, Midwest, North, and Northeast), diabetes duration (shorter than 10 and longer than 10 years), type of diabetes, year of the study (before or after 2000). All analyses were performed using the ‘meta’ package (version 6.0) for R (version 4.2.1). To reduce heterogeneity, potential factors that should affect it were explored, including studies that were carried out in ophthalmology services for the diagnosis of DR.

Results

The search retrieved 1400 articles from October 1950 to February 2022, of which 103 were duplicates and were excluded. In total, 975 articles were removed based on title and abstracts; 322 full-text articles were assessed for eligibility, of which 72 met all inclusion criteria. Figure 1 shows the flowchart of study selection.

Table 1 shows the characteristics of the included studies by diagnostic criteria and method of assessment.

Table 1 Characteristics of the included studies
Full size table

A meta-analysis was conducted; Fig. 2 shows prevalence rates. The prevalence rate of DR (pooled estimate) was 36.28% (95% CI 32.66–39.97, I2 98%).

Fig. 2
figure 2

Forest plot representing diabetic retinopathy prevalence rates

Full size image

We could not assess the prevalence of DR by gender because such variable was unspecified in most studies.

Trend analyses showed an increase in the prevalence of DR in patients with longer duration of diabetes [20.73% (95% CI 11.63–31.64 I2 98%) in less than 10 years of diabetes, and 37.73% (95% CI 30.80–44.93 I2 96%) in longer than 10 years] (Additional file 2: Fig. S1) and in patients with type 2 diabetes mellitus [26.84% (95% CI 16.43–38.74, I2 98%) in type 1 diabetes mellitus and 35.69% (95% CI 30.16–41.41, I2 98%) in type 2 diabetes mellitus] (Additional file 3: Fig. S2).

In the assessment of prevalence of DR according to the year of publication of the study, no difference in the prevalence of DR [36.06% (95% CI 31.59–40.65, I2 98%) was observed in articles published after 2000 and 37.84% (95% CI 22.90–54.07, I2 96%) in those published before 2000] (Additional file 4: Fig. S3).

The analysis of prevalence rates by diagnostic method showed a prevalence of 38.15% (95% CI 33.08–43.36, I2 98%) in patients diagnosed by indirect ophthalmoscopy, and 31.11% (95% CI 19.55–44.00, I2 98%) by color fundus photography (CFP, Additional file 5: Fig. S4).

We explored potential factors that would affect the heterogeneity of the analyses, including studies that were carried out in ophthalmology services for the diagnosis of DR. Table 2 shows prevalence rates of DR and their 95% CI by Brazilian regions and adjustment to studies that were not performed in ophthalmology services. The data in this table show that the prevalence of DR was higher in the Southern region. No significant change of the heterogeneity was observed when we did this type of new analyses.

Table 2 Prevalence of diabetic retinopathy by region; analyses with and without articles done in reference centers
Full size table

Quality of studies

Figure 3 summarizes data regarding quality of studies. A total of 18 studies (25%) had an intermediate risk of bias, and the rest of the studies had a high risk of bias. Additional file 6: Fig. S5 shows the risk of bias assessment for each study. Most studies were based on cross-sectional design (59 studies, 82%). The most used design was convenience sampling (68 studies, 94%). Most studies were developed only in or including data from Southeastern and Southern Brazil (30 studies, 48.4% and 24 studies, 33.3%, respectively). In 23 studies (31.9%), the main objective was to evaluate the prevalence of DR.

Fig. 3
figure 3

Quality of studies characteristics

Full size image

Discussion

Our study provides comprehensive and up-to-date evaluations of the current DR prevalence in Brazil with the largest meta-analysis to date.

This study included 72 studies carried out in Brazil and found a 36.26% prevalence of DR. It also found a higher prevalence in patients with long-term disease, type 2 diabetes, and residents of the Southern region.

According to the IDF 2021 atlas [15], 15.7 million people live with diabetes in Brazil, being the sixth country with the highest number of people with diabetes. Teló et al.—in a systematic review with meta-analysis of Brazilian observational studies from 1980 to 2015—included 50 studies and showed an increasing prevalence of diabetes in recent decades, showing that the prevalence of diabetes in Brazil can reach 6.9% of the population in studies published after 2010 [6]. Data obtained from the National Survey of Health (2014 to 2015) showed the following prevalence of diabetes according to different criteria: 6.6% (95% CI 5.9–7.2) [glycated hemoglobin (HbA1c) ≥ 6.5% (47.5 mmol/mol)]; 8.4% (95% CI 7.6–9.1) [HbA1c ≥ 6.5% (47.5 mmol/mol) or use of antidiabetic drugs]; 9.4% (95% CI 8.6–10.1) [HbA1c ≥ 6.5% (47.5 mmol/mol) or history of diabetes]; and 7.5% (95% CI 6.7–to 8.2) [history of diabetes] [95]. Extrapolating the prevalence of DR found in our review, 5.7 million people would be living with DR in Brazil. Comparing our outcomes with other systematic reviews that evaluated the prevalence of DR, our study showed a higher prevalence than China (18.45%) [96], Africa (30.2–31.6%) [97], and Europe (25.7%) [98]. In India, a systematic review performed with 8,866 diabetic patients found a 16.1% prevalence of RD [11, 98]; in Pakistan, a systematic review estimated the prevalence of DR at 28.2%, ranging from 10.6 to 91.34% [12]; in the USA, a study carried out from 2005 to 2008 with 1495 diabetic patients showed a 47% prevalence of DR [13]; in Indonesia, a 2017 study showed a DR prevalence of 43.1% in patients with type 2 diabetes mellitus [14]. Different study methodologies applied in primary studies retrieved in these different meta-analyses are important to determine these high different figures among countries.

Regarding the predictors evaluated in trend analysis, the duration of diabetes increases the prevalence of DR. Duration of diabetes is an established risk factor for the development of DR and other microvascular complications in patients with diabetes [99,100,101]. The same trend was described in another systematic review [14].

The study showed a higher prevalence of DR in type 2 diabetes mellitus patients, but the literature shows that prevalence is higher in type 1 diabetes mellitus [7]. One of the possibilities is that the studies on patients with type 1 diabetes mellitus have been carried out with a short duration of disease. Moreover, since type 2 diabetes mellitus is more prevalent than type 1 diabetes mellitus, a higher number of people with type 2 diabetes mellitus are possibly being studied as compared to those with type 1 diabetes mellitus, unbalancing the final outcomes.

This meta-analysis showed that the prevalence of DR in the Southern region is higher than in the Northern region. This can be explained because disease burden components have different distributions between the North and the South in Brazil, due to economic and social disparities between regions. As a country with continental territory, racial and cultural miscegenation, Brazil experiences great social and economic problems, including socioeconomic inequality. The Federation Unit with the lowest poverty rate in 2021 was Santa Catarina (10.16%) in the South and the one with the highest proportion of poor people was Maranhão with 57.90% in the Northeast. Segmenting the country into 146 spatial strata, the one with the greatest poverty in 2021 is the Coast and Baixada Maranhense with 72.59%, while the lowest is in the municipality of Florianópolis in South with 5.7% [102]. It is theorized that patients living in the Northern region do not live long enough to develop microvascular complications of diabetes [103] and, as aforementioned, the duration of the disease is one of the markers that increase the risk of developing DR. Another difference may be because in the Northern region access to public health is more limited [104].

The gold standard method for screening for DR is CFPs [105]. In this review, eight (11%) studies used CFPs as a diagnostic method and 53 (73.6%) used ophthalmoscopy. Ophthalmoscopy and color fundus photographs are both valid strategies for DR screening. Each method has advantages and potential limitations that include cost, expertise of the examiners, and equipment. The main advantages of ophthalmoscopy are its easy handling and superior performance in cases of poor patient collaboration, on the other hand, it is not sensitive enough to detect minor signs of DR and depends on the presence of a trained operator [106]. In turn, CFPs has the advantage of providing a permanent record of retinopathy, which can be used later to document retinopathy progression and allows for a more detailed grading of retinopathy. Notably, CFPs are expensive [107].

Some studies were carried out in ophthalmology services for the diagnosis of DR, which may represent a selection bias, considering that reference services will likely have a higher rate of patients with complications. Therefore, we performed an analysis excluding the studies conducted in ophthalmology services as shown in Table 2, resulting in a decrease in the prevalence of DR.

Our study has some limitations, the most important being the high heterogeneity. Migliavaca et al. show that prevalence studies have high heterogeneity [108]. The available smaller studies conducted in Brazil are often limited in scope and may uncover confounding or conflicting results due to their small sample size. The heterogeneous nature of studies (e.g., patient selection criteria, diabetes type, setting), disparity between study methods (color fundus photography vs. ophthalmoscopy), and possible differences between urban and rural settings, with variations on the following items: access to healthcare and eating habits [109, 110] may contribute to conflicting reports of prevalence and incidence, making a direct comparison of studies difficult. Most studies did not report visual acuity or the prevalence of maculopathy or proliferative DR, both considered vision-threatening DR. Although the significance of functional outcomes, most epidemiological studies do not address them. Rates of blindness are variable among countries (high-income vs. low- to middle-income countries) depending on the existence of screening programs, specialized workforce, and possibility of timely treatment [111]. Such limitations highlight the need for consistent data capture in Brazil.

Conclusion

This study shows a high heterogeneity that is expected in systematic reviews estimating prevalence rates. It is necessary to develop multicenter studies with representative samples and standardized methodology. Screening programs are effective for the identification of early DR, and epidemiological studies are essential for their success, as they collect data that allows identification of the magnitude of the problem, as well as regional differences. Further research is needed to collect such data in Brazil, with the use of standardized criteria and consistent terminology, and the inclusion of samples that are representative of the communities from which they are drawn. Robust longitudinal collection of patient data will be essential to allow identification of the true extent of diagnosed retinal complications of diabetes, in turn providing healthcare planners with essential information to aid future decision-making.