Prevalence, Incidence and Ecological Determinants of Diabetic Retinopathy in Iran: Systematic Review and Meta-analysis

Abstract Purpose To estimate the pooled prevalence and incidence of diabetic retinopathy (DR) in Iran and to investigate their correlations with the Human Development Index (HDI), healthcare access (i.e., density of specialists and sub-specialists), and methodological issues. Methods Electronic databases such as PubMed, Embase, Scopus, Web of Science, Google Scholar, and local databases were searched for cohort and cross-sectional studies published prior to January 2018. Prevalence and incidence rates of DR were extracted from January 2000 to December 2017 and random effects models were used to estimate pooled effect sizes. The Joanna Briggs Institute critical appraisal tool was applied for quality assessment of eligible studies. Results A total of 55,445 participants across 33 studies were included. The pooled prevalence (95% CI) of DR in diabetic clinics (22 studies), eye clinics (4 studies), and general population (7 studies) was 31.8% (24.5 to 39.2), 57.8% (50.2 to 65.3), and 29.6% (22.6 to 36.5), respectively. It was 7.4% (3.9 to 10.8) for proliferative DR and 7.1% (4.9 to 9.4) for clinically significant macular edema. The heterogeneity of individual estimates of prevalence was highly significant. HDI (P < 0.001), density of specialists (P = 0.004), subspecialists (P < 0.001), and sampling site (P = 0.041) were associated with heterogeneity after the adjustment for type of DR, duration of diabetes, study year, and proportion of diabetics with controlled HbA1C. Conclusion Human development and healthcare access were correlated with the prevalence of DR. Data were scarce on the prevalence of DR in less developed provinces. Participant recruitment in eye clinics might overestimate the prevalence of DR.


INTRODUCTION
Diabetic retinopathy (DR) is the leading cause of vision loss in adults aged 20-74 years, and remains one of the foremost causes of blindness and visual impairment worldwide. [1][2][3][4] Despite significant development in the prevention and control of DR, the proportion of DR increased by 7.7% among all declining causes of blindness between 1990 and 2015. [5] The prevalence of DR strongly correlates with both the duration of diabetes and the level of glycemic control. [6] Therefore, timely management of DR stemming from screening programs, appropriate referral for treatment, and improving healthcare accessibility are important in preserving vision in diabetics. [7] Although the treatment of DR can decrease the risk of visual loss by 60%, it imposes a heavy cost to the healthcare system. [8] Despite some improvements in diagnostic assessment and treatment options, [6] the lack of qualified healthcare services along with a Westernized lifestyle have caused the burden of DR to be high and on the rise in developing countries. [7,8] DR is a pressing public health matter, probably due to suboptimal access to diabetes care services such as eye care professionals and eye care services, especially in low-to middle-income countries. [8] Low human development might be another correlate of the increasing burden of DR. To the best of our knowledge, no study has investigated the correlation between human development and DR. The only study on this subject assessed the association between the Human Development Index (HDI) and the number of studies published on DR. [9] HDI is abstracted from income, education, and life expectancy markers and ranks areas into different levels of human development. [10] Despite the high prevalence of diabetes, there are few reliable national studies on the incidence, prevalence, and correlates of DR in developing countries. [11] Of note, Iran is a country in transition, having a high variety of healthcare access options and human development as well as a huge variation in the prevalence and incidence of DR across its geographic regions. [12][13][14] Therefore, besides assessing the prevalence and incidence of DR, their adjusted correlations with HDI and healthcare access were investigated in the current observational study.

Protocol and Registration
The Meta-analysis of Observational Studies in Epidemiology (MOOSE) guidelines were followed. [15] The study protocol was approved by the Shiraz University of Medical Sciences (ethical approval code: IR.SUMS.MED.REC.1397.256).

Eligibility Criteria
Observational studies (prospective or retrospective cohort and cross-sectional) were included if they provided sufficient information about the incidence and prevalence of DR and clinically significant macular edema (CSME). No restriction was applied on the year of publication or type of diabetes, and all studies published in Persian or English language were included. These two languages covered all studies published about the Iranian population and we did not find studies in other languages.
congresses that were held during the study period around the world and specifically in Iran, and we selected and hand searched the abstract books that were obtainable as much as possible. We also searched university websites for thesis and reports that were related to the subject during the study period. References of all included studies were also searched for potentially eligible studies. In cases where the full text of an article was unavailable, the corresponding author was contacted. Documents were catalogued using Endnote X4.

Study Selection
In case of repeated publications from one study, the newest publication was included. Titles, abstracts, and the full texts of retrieved articles were reviewed independently by two experts in the field and eligible articles were selected. Studies on children, pregnant women, or non-Iranian populations were excluded as well as experimental studies, secondary studies, letters, case-reports, case-series, commentaries, and editorials. Clinical trials were also excluded. Studies with a non-representative sample or irrelevant comorbidities were also excluded. Also, in the case of disagreement regarding exclusion or inclusion of a study, a consensus was reached through discussion between the authors.

Data Extraction and Risk of Bias Assessment
We performed a comprehensive literature review and designed a conceptual framework. An excel (MS Office) data sheet based on clinical principles was prepared for data extraction using our framework. The following variables were included in the data extraction form: first author, publication year, study year, study location (i.e., province/district), urbanization ratio (number of participants from the urban area/number of participants from the rural area; based on the data presented in each individual study), sampling site (diabetes clinic, eye clinic, or general population), sampling design (random, multistage, convenient, unknown), study design (cross-sectional, cohort), sample size (overall and for subgroups), age range (or mean age), duration of study, gender (or female/male ratio), diagnostic methods of DR, [16] proportion of diabetics (in population-based studies), type of diabetes (or type I/type II ratio), type of DR (i.e., non-proliferative diabetic retinopathy [NPDR]) and its stage (mild, moderate, or severe), proliferative diabetic retinopathy (PDR), CSME, mean HbA1C (overall, among DR patients, and among non-DR patients), number of DR patients (or estimated prevalence and its standard error), number of incident cases of DR (or cumulative incidence rate; only for cohort studies), number of person-years of follow-up (in cohort studies), duration of DM (overall, among DR patients, and among non-DR patients), mean age at the onset of DM, mean duration of DR, and proportion of patients with newly diagnosed DM. The aforementioned data were extracted (if available) and 25% of the extracted data were randomly cross-checked by another author. Also, in the case of disagreement regarding data extraction, a consensus was reached through discussion between the authors. Since several studies had not reported exact values for some variables, more than 30 disagreements on the most appropriate estimates for these missing values were discussed in the team.
The Joanna Briggs Institute ( JBI)'s critical appraisal tool [17] was applied for quality assessment of eligible studies. Quality assessments and critical appraisals were performed by two different authors independently. In the case of disagreement regarding the quality score, a consensus was reached through discussion between the authors (4 studies out of the included 33 studies).

Data on Human Development and Healthcare Access
The density of specialists and subspecialists (numbers of specialists and subspecialists in the healthcare system of each province to the total population in that province) was retrieved from a recent reliable report by Haghdoost et al [18] as indices of healthcare access. Density ratios were categorized into quantiles. Haghdoost et al gathered the number of physicians based on the questionnaires filled out by medical universities all around Iran. Their study was a part of a project to define the national treatment map of Iran in 2025 (Naghsh-e Rahe Darman-e Iran). In order to control the precision of completing the questionnaire, their data was crosschecked with the Medical Registry Information System and different medical insurance companies. [18]  HDI indices (less developed, moderately developed, and developed) were retrieved for each province according to the study performed by Safaeipour et al [12] and used as an independent variable in meta-regression modeling.

Data Preparation and Statistical Analysis
Point estimates of prevalence were extracted or calculated as the number of patients with DR divided by the number of patients with DM. Additionally, prevalence estimates of DR in the population, regardless of diabetic status, were estimated as the number of participants with DR divided by the number of total participants in populationbased studies. A 95% confidence interval (CI) for all individual point estimates of prevalence was estimated where it was not mentioned. [19] Cumulative incidence proportions (per 100 person-years) were calculated as the number of new DR cases divided by the number of at-risk person-years (i.e., number of study subjects multiplied by number of follow-up years).
To ensure the independence of point estimates in primary studies as well as to prevent repeated counting of participants in primary studies, only one of the overall or subgroup point estimates of each primary study was included in the meta-analysis and meta-regression models. The heterogeneity of individual estimates of prevalence was assessed according to the I-square statistic above 50%. [20] In cases of high heterogeneity, correlates were conceptualized. Then, the significance and magnitude of their correlation were investigated using the random-effects meta-regression technique.
Subgroup analyses were performed according to the most important correlates of heterogeneity if applicable (i.e., sampling site, geographical location, HDI and healthcare access). Due to persistent heterogeneity (even after subgroup analysis), individual estimates of prevalence and incidence were pooled using the DerSimonian and Laird randomeffects modeling method. Publication bias was investigated using Begg's and Egger's tests. Data analysis and calculations were performed using Stata software, version 11.2 (Stata Corporation, College Station, TX, USA). A two-sided < 0.05 was considered as statistically significant. We also provided appropriate tables and graphs for showing our results (i.e., included studies, shortage in studies, methods applied for the diagnosis of DR by primary studies, study flowchart, and study forest plot by HDI categories). The study protocol was registered in the International Prospective Register of Systematic Reviews (PROSPERO), under the registration number CRD42018104626.

RESULTS
The initial search resulted in 2,153 records. Of these, 426 records were included in the fulltext review process, and finally, 33 studies were included in the meta-analysis.

Overview of Included Studies
Thirty cross-sectional and three cohort studies were included in the analysis, representing an overall number of 55,445 diabetic patients including 17,155 patients with DR [ Table 1]. Among cross-sectional studies, seven had a populationbased sampling design representing 24,623 participants including 5,657 diabetic patients and 2,049 patients with DR. Included cohort studies represented 1,174 diabetic patients (equivalent to 5,400 person-years) and 613 incident cases of DR.

Assessment of Heterogeneity in Individual Estimates for Prevalence of Diabetic Retinopathy
In cross-sectional studies conducted in diabetes clinics, individual estimates for the prevalence of DR were significantly heterogeneous for overall diabetics (I-square, 99.5; < 0.001; the heterogeneity was similar between genders [I-square, 99.4; < 0.001]). Also, individual estimates for the prevalence of DR were highly heterogeneous in studies conducted in eye clinics (I-square, 98.0; < 0.001) and population-based studies (I-square, 95.7; < 0.001). Additionally, in three cohort studies, the estimates of individual cumulative incidence of DR were statistically heterogeneous with a highly significant I-square of 97.9 ( < 0.001).

Determinants of DR Prevalence (Correlates of Heterogeneity of Individual Estimates)
According to the random-effects meta-regression model, HDI (as an ordinal variable of tertile of HDI; adjusted OR: 0.12, 95% CI: 0.05 to 0.34, < 0.001), density of specialists (as an ordinal variable including quintile of density ratio; Risk of bias was not associated with heterogeneity ( = 0.683). Type of diabetes was not a significant determinant of heterogeneity for individual estimates of DR prevalence ( = 0.10).
Due to the lack of enough data to achieve individual estimates of cumulative incidence, metaregression modeling for cumulative incidence was not applicable. Due to significant heterogeneity among studies, the analysis was performed through subgroup analysis based on the most important correlates of heterogeneity.

Prevalence of DR among Patients Referred to Diabetic Clinics
The overall pooled prevalence rate of DR among diabetic patients was 31

Prevalence of DR among Patients Referred to Eye Clinics
The overall pooled prevalence of DR among diabetic patients was 57

Prevalence of DR among Patients in Population-based Studies
The overall pooled prevalence of DR was 29

More Details on the Association of HDI and the Prevalence of DR
The estimation of linear correlation coefficient between HDI and the prevalence of DR among diabetics was -0.18 with a of 0.029 (a relatively low but significant linear correlation) [ Figure 4].

Reporting and Methodological Shortcomings
One study used the survey data analysis technique to consider multistage sampling design. [50] Two studies reported age-adjusted estimates. [30,40] Only 33.3% (n = 11) of the studies reported their DR results by gender [ Table 2].

Assessment of DR
According to the current results, 54.54% (n = 18) of studies performed indirect ophthalmoscopy with pupillary dilatation for the evaluation of DR. Others probably used the same method; however, this matter was not clearly stated [ Table 3].

Publication Bias
There was no significant publication bias (Begg's test = 0.824; Egger's test = 0.075) in the current study.

DISCUSSION
In the current study, the pooled prevalence and incidence of DR including PDR and NPDR among Iranian diabetic subjects referred to diabetes clinics and eye clinics and the pooled prevalence of DR among diabetics retrieved form populationbased studies were determined based on the English language studies only. Persian articles were excluded due to poor quality score in our  assessment. We found no publication bias among the included studies. The human development, access to subspecialists and specialists, and site of study sampling had a correlation with the prevalence of DR among diabetics.
Studies from other parts of the world reflect significant differences in the prevalence of DR depending on the factors such as ethnicity, demography, and access to healthcare systems. [51,52] According to the current results, an increase in HDI was independently correlated with lower DR prevalence. In order to avoid the bias of ecological inference fallacy, this finding could only be interpreted from an ecological point of view. Therefore, we cannot directly relate the DR prevalence in individuals to the HDI in each province. However, HDI may potentially be a good index of quality of diabetes care in developing countries such as Iran at a national level. This finding is consistent with previous evidence from low-income regions of developed countries. [53] The prevalence of DR and PDR/NPDR ratio were higher in eye clinics in   comparison to the DM clinics. It might be due to the higher detection rate of DR, especially PDR in eye clinics. [54][55][56] In other words, more advanced patients are usually referred to the eye clinics. The density of subspecialists was negatively correlated with the prevalence of DR. This could be due to health literacy, greater access to healthcare services in regions where subspecialists work, leading to more effective diabetes control. This finding is in accordance with previous reports. [51,52,57] Of note, many confounding factors such as social factors could be associated with the density of subspecialists and prevalence of DR. It might also be due to the interest of subspecialists to be in areas with higher socioeconomic status in which inhabitants may also have a better access to healthy food, causing more controlled DM and less prevalence of DR.
The pooled population-based prevalence of DR among diabetics was close to the maximum estimates of DR prevalence reported in developing countries such as Pakistan (29% vs 25-29%), while it was lower than the prevalence reported in developed countries. [58][59][60][61][62] Population aging and Westernized lifestyle could have been among the factors increasing the prevalence of diabetes and DR in Iran and other developing countries. [11,58,63] In addition, the higher prevalence of DR among male subjects could be attributed to lifestyle habits such as cigarette smoking, which is consistent with studies conducted in India and Nepal. [58,59] The pooled prevalence of DR among diabetics referred to eye clinics was significantly higher than that in diabetes clinics. Routine screening by an ophthalmologist is not a common practice in developing countries. [64] Thus, this higher prevalence may be due to higher rates of referral for patients presenting with visual symptoms. There is a need for improving screening programs in primary healthcare services and communication between primary healthcare providers and ophthalmologists to ensure diabetics receive timely ophthalmic examinations.
Based on the results of the current study, the estimation of DR prevalence among diabetics in eye clinics may not be a good indicator of DR prevalence among diabetics; however, a comparison of DR prevalence in diabetic and eye clinics could help determine the sensitivity and specificity of referrals to eye clinics by clinicians and over-/underutilization of eye care in developing countries.
Although, overall and gender-specific pooled prevalence rates of DR among diabetics referred to diabetes clinics were relatively higher than those in population-based studies, these differences were not meaningful. In addition, the female/male ratio of estimated DR prevalence in DM clinics studies was almost equal to the population-based studies (88% vs 80%). Therefore, considering the difficulties in conducting robust population-based studies, it is reasonable to estimate the prevalence of DR in diabetes clinics, especially in developing counties such as Iran.
The pooled estimate of DR incidence among diabetics in Iran was less than recent reports from developed countries such as Canada, the UK, and Spain. [61,65,66] Moreover, the estimated incidence was relatively higher than India, South Korea, and Denmark. [67][68][69] According to the current study, the available data on the incidence of DR in Iran is inadequate; further studies are needed to determine the incidence of DR in Iran.
Among the eligible studies included in the current meta-analysis, only one study reported the prevalence of DR among Type I diabetics. [27] Accordingly, the pooled estimates provided in the current study are mostly representative of the prevalence of DR among Type II diabetics. More studies on the incidence and prevalence of DR among Type I diabetics are needed.
The results of this study showed that there is a lack of research on the prevalence of DR in less developed provinces, such as Sistan and Balochistan, Bushehr, Hormozgan, Khorasan, Khozestan, Kermanshah, Kordistan, Kohkiloieye, Chaharmahal and Bakhtiyari, and Lorestan. Missing data from these provinces could have affected the results of the current study. In addition to the paucity of data in some provinces, there is a significant heterogeneity among studies, which means that making an accurate single prevalence estimate of DR is not possible. The pooled prevalence of DR was 29%. However, the prevalence rate varied from 9% to 64.1%. Thus, it might not be completely representative of the actual DR prevalence.
In addition to the aforementioned factors, the method used for detecting DR is a major factor that can influence prevalence estimates. It is crucial to know which specific diagnostic method (i.e., dilated fundus examination, direct or indirect ophthalmoscopy, digital imaging, etc.) was used in each study. [70] The ophthalmic examination methods were not adequately explained in at least 12 included studies. In the course of DR, patients might develop maculopathy with no significant change in vision and may not seek medical care. Optical coherence tomography (OCT) is an important test in the evaluation and management of diabetic macular edema and can detect subclinical macular edema. Missing OCT evaluation could lead to many diabetics remaining undiagnosed in the early stage of DR. [71] The current study has some limitations. The pooling of data from different sources introduced potential sources of heterogeneity that could impact accuracy. Various studies could have different inclusion criteria, sample selection, or study protocols. For example, the sample from a diabetes clinic differs from an eye clinic or population-based studies. The pooled estimate of DR prevalence by gender in the current study was sometimes not compatible with the total prevalence in the subgroups, which was due to the lack of reporting DR by gender in some studies. Studies in which the diagnosis of DM was based on patient selfreport, without lab test confirmation, could have led to an overestimation of DR prevalence due to the exclusion of undiagnosed diabetes from the study sample. Although it is desirable to have the HDI and density of specialists and subspecialists at the time of each study to calculate the correlations and perform the analyses, only two up-to-date and available studies were used for estimating the value of both mentioned variables. Also, the absence of studies from the eastern, western, and southern regions of Iran could have also affected the validity and generalizability of the current findings. Although our findings on the prevalence and incidence of DR may not be generalizable to all countries, it could be useful for developing countries, especially regions with similar socioeconomic, demographic, cultural, and geographic conditions.

SUMMARY
Despite the scarcity of research in less developed regions, a reasonable estimate for the prevalence of DR among Iranian diabetic subjects is around 30% (29% among female and 37% among male subjects). HDI, density of specialists and subspecialists, and sampling site were independently correlated with the prevalence of DR in Iran. The most reliable evidence on DR prevalence is likely to be retrieved from diabetes clinics in developing countries. Furthermore, providing a list of essential items for reporting the descriptive epidemiology of DR and performing studies in less developed regions could generate stronger evidence for health policy programs.