Incidence and progression of diabetic retinopathy in Sub-Saharan Africa: A five year cohort study

Aims To describe the incidence and progression of retinopathy in people with diabetes in Southern Malawi over 5 years. To document visual loss in a setting where laser treatment is not available. Methods Subjects from a cohort sampled from a hospital-based, primary-care diabetes clinic in 2007 were traced in 2012. Laser treatment was not available. Modified Wisconsin grading of retinopathy was performed using slit lamp biomicroscopy by a single ophthalmologist in 2007 and using four-field mydriatic fundus photographs at an accredited reading centre in 2012. Visual acuity was measured by Snellen chart in 2007 and by ‘Early Treatment of Diabetic Retinopathy Study’ chart in 2012. HbA1c, blood pressure, HIV status, urine albumin–creatinine ratio, haemoglobin and lipids were measured. Results Of 281 subjects recruited in 2007, 135 (48%) were traced and assessed, 15 were confirmed dead. At follow-up (median 5.3 years) ≥2 step retinopathy progression was observed in 48 subjects (36.4%; 95% CI 28.2–44.6). Incidence of sight threatening diabetic retinopathy for those with level 10 (no retinopathy) and level 20 (background) retinopathy at baseline, was 19.4% (11.3–27.4) and 81.3% (62.1–100), respectively. In multivariate analysis 2 step progression was associated with HbA1c (OR 1.2495%CI 1.04–1.48), and haemoglobin level (0.77, 0.62–0.98). 25 subjects (18.8%) lost ≥5 letters, 7 (5.3%) lost ≥15 letters. Conclusions Progression to sight threatening diabetic retinopathy from no retinopathy and background retinopathy was approximately 5 and 3 times that reported in recent European studies, respectively. Incidence of visual loss was high in a location where treatment was not available.

Introduction Sub-Saharan Africa faces an epidemic of diabetes [1]. In contrast to high income countries [2][3][4][5][6][7][8] very few cohort studies have investigated prevalence, incidence and determinants of severity and progression of diabetic retinopathy (DR) in this region [9]. Little is known about the effects of infectious disease (including Human Immunodeficiency Virus (HIV) and malaria) and anemia on the microvascular complications of diabetes. Malawi has a population of 16.4 million. Annual per capita healthcare expenditure is extremely low at US$77 [10]. The best available population based survey (World Health Organisation (WHO) STEPwise methods) reported a prevalence of diabetes of 5.6% in Malawian adults in 2009 [11]. In 2007, a cross sectional study of diabetes complications was performed at the diabetes clinic at Queen Elizabeth Central Hospital (QECH), Blantyre [12]. As part of this study our group performed slit lamp bio-microscopy to document grades of retinopathy. We found a high prevalence of both sightthreatening and proliferative DR: 19.6 and 5.7%, respectively [13]. At the time of this survey, and until 5 years later, laser treatment was not available in the public sector in Blantyre. In 2012 we recalled subjects from this cross sectional study in order to report retinopathy progression at 5 years. Additionally, we undertook a prospective, 2 year cohort study of patients attending two hospital-based diabetes clinics. Data from this study has been published elsewhere [14,15].

Materials and methods
Setting QECH is a large teaching hospital. QECH provides primary and secondary care to the people of greater Blantyre (approximately 1 million). Tertiary care is provided to the Southern Region of Malawi. At the time of this study (2012) the QECH diabetes clinic was the only public sector diabetes clinic in Blantyre. In the period 2007 to 2012 the clinic underwent a number of changes. The number of registered patients increased from approximately 800 to 2000. A vibrant nurse-led patient education programme supported by the World Diabetes Foundation commenced in 2008. Its aims were improving compliance with diet and medications and educating patients on the complications of diabetes. An electronic records system (Diabetes and Hypertension System, Baobab Health Trust, Malawi) was installed in early 2010. In 2007 medications regularly available free of charge were glibenclamide and insulin (lente and soluble). Metformin was available from private pharmacies but rarely from the hospital pharmacy. By 2012 metformin was more frequently available free of charge. However, supplies of all drugs remained intermittent. Tests available at the clinic were the same in 2007 as 2012: glycaemic control measured by fasting blood glucose (FBG), blood pressure (BP), height and weight. Measurement of lipids, glycosylated haemoglobin (HbA1c) and urine test sticks for microalbuminuria were not available routinely.

Participants
Patient selection in the 2007 cross sectional study has been described elsewhere [12,13]. Briefly, consecutive subjects attending for routine out-patient review between March and June 2007 were invited to participate. Of 620 subjects included in the study 281 were examined by an ophthalmologist. Sampling was ad hoc (i.e. not consecutive): subjects had slit lamp examination if the ophthalmologist was present at the particular clinic at which they were recruited. At this time laser treatment was not available in the public sector. The 2007 study was not planned as a cohort study therefore no contact details were recorded. Tracing of subjects between December 2011 and November 2012 was systematic. The QECH diabetes clinic electronic patient record system was searched by subject name by a research nurse. Identified persons were then contacted by phone or home visit. The study team attended the diabetes clinic weekly between May and November 2012 to approach patients in the clinic waiting room. The majority of deaths in Malawi are not registered. The relatives of deceased subjects were visited at home by a study nurse in order to confirm the death. Death was recorded if confirmed by a first degree relative or 'Traditional Authority' (village leader in rural districts).

Procedures
Clinical assessment in the 2007 study has been described elsewhere [12,13]. Briefly, visual acuity (corrected with pin-hole) was measured using a Snellen chart. FBG, HbA1c and HIV status were tested. Slit lamp biomicroscopic retinopathy grading was performed by one ophthalmologist (SG). Retinopathy and maculopathy were classified by feature-specific grading as described in the Liverpool Diabetic Eye Study (LDES) [16] (S1 Fig).
Clinical assessment of subjects in 2012 was the same as in our 24 month cohort study described in detail elsewhere [14]. Briefly, uncorrected and pinhole visual acuity was assessed using an Early Treatment of Diabetic Retinopathy Study (ETDRS) chart. Thresholds for moderate visual impairment (50 to 59 letters; equivalent to 6/24 Snellen) and severe visual impairment or blindness (<50 letters; equivalent to 6/36 or worse) were set according to the WHO [17]. For subjects with corrected visual acuity in the better eye of less than 80 letters, the principle cause of vision loss was documented by the examining ophthalmologist (PB). Hypertension was defined according to WHO criteria [11]: systolic blood pressure !140 mmHg, or diastolic blood pressure ! 90 mmHg, or taking anti-hypertensive medication. Point-of-care testing was offered for haemoglobin level and HIV (Malawian national protocol [18]). Anemia was defined according to the WHO: male 130 g/l; female 120 g/l [19]. Laboratory testing of venous blood samples was performed for HbA1c, fasting glucose, HDL cholesterol, LDL cholesterol, triglycerides and serum creatinine. Urine albumin-creatinine ratio was measured.
As in 2007 retinopathy and maculopathy were classified by feature-specific grading using the LDES scale [16]. In contrast to 2007 dual grading of digital fundus photos of four 45˚standard fields [16] was performed at the Liverpool Reading Centre by accredited graders. Sight threatening diabetic retinopathy (STDR) was defined as any of the following: retinopathy level 40-71+ (moderate pre-proliferative DR or worse); level 3-4 maculopathy (macular exudates in a circinate pattern or within one disc diameter of the foveal centre or CSME (ETDRS definition [20])); or other retinal vascular disease related to diabetes: central or branch retinal vein occlusion, central or branch retinal artery occlusion.

Statistical analysis
Retinopathy grades were analysed by patient according to the worse or only gradeable eye. Visual acuity scores were analysed by patient according to the better eye. For the purposes of analysis visual acuity scores from 2007 (corrected Snellen acuities in the better eye) were converted to ETDRS letter scores using a standard conversion table [21]. Comparison was then made with 2012 ETDRS letters measurement in the better eye. Primary outcome was progression of retinopathy by !2 steps on the LDES scale (either 2 step progression in one eye or 1 step progression in both eyes). A multiple logistic regression model (backwards stepwise; probability of removal of 0.2) was constructed to determine the odds ratio and 95% confidence intervals for the primary end point at 5 years. For the majority of variables the baseline value was used in the analysis. For those variables only measured at the 2012 visit this value was used (uACR, LDL, HDL, triglycerides). An initial 11 variables were studied: HbA1c, duration of diabetes, baseline grade of DR, type of diabetes, sBP, haemoglobin level (2012), urine albumin creatinine ratio (uACR)(2012), triglycerides (2012), HIV status, age and sex. uACR did not demonstrate a linear association with probability of 2 step progression; a logarithmic transformation (base 10) was used in analysis. All tests were two-sided and a p value <0.05 was taken to indicate statistical significance. All calculations were performed using STATA version 12 (StataCorp, College Station, TX, USA). The study was approved by the University of Malawi, College of Medicine Research Ethics Committee and the University of Liverpool Research Ethics Committee. All participants gave written informed consent.  Table 1. Compared to subjects who were traced and assessed in 2012, subjects who were not seen in 2012 demonstrated higher median age and systolic BP and a higher prevalence of STDR (based on univariate statistical tests with no adjustment for multiple comparisons). Subjects not seen in 2012 demonstrated worse baseline visual acuity scores in the better eye (data not shown; p = 0.0001, Χ 2 test for trend). Table 2 shows demographic, clinical and biochemical measurements for subjects included in the 2007 study and seen again in 2012 categorised by grade of retinopathy at baseline.
Of the 135 subjects from the 2007 cohort 41 were recruited (by systematic random sampling from the QECH diabetes clinic) into our 24 month cohort study. Of the 41 subjects; 38 were seen at 12 months and 36 at 24 months providing 6 and 7 year longitudinal data (Table 4 and  S5 Table). Of 34 subjects without STDR at baseline (2007) [14,15], 5 year progression data detailed in this manuscript and 6 and 7 year data from the 41 subjects described above to give a composite graph of DR progression over 7 years.

Discussion
We report the progression of grades of retinopathy and visual impairment over 5 years in a cohort of people with diabetes from Southern Malawi who had no access to laser treatment.
Five year incidence of any DR in those without evidence of retinopathy at baseline was 48.4% (38.2-58.5). The five year incidence of STDR for those with level 10 and level 20 retinopathy at baseline was 19.4% (11.3-27.4) and 81.3% (62.1-100), respectively. The five year incidence of PDR for those with level 10, level 20, level 30 and level 40 retinopathy at baseline was 0%, 4.5%, 22% and 40%, respectively. Higher glycosylated haemoglobin (HbA1c) and lower haemoglobin level were risk factors for progression of retinopathy in multivariate analysis. Over the followup period 25 subjects (18.8%) lost 5 or more ETDRS letters of which 7 subjects (5.3%) lost 15 or more ETDRS letters. 2 subjects (1.5%) progressed to moderate visual impairment (50-59  letters) and 3 (2.3%) became 'severely visually impaired or blind' (<50 letters). In 56% of cases DR was the sole or equal contributing cause of visual loss. High quality cohort studies of DR from Africa or Asia are scarce. In our group's separate prospective cohort study performed in Malawi between 2012 and 2014, 2 year incidence of STDR for subjects with level 10 and level 20 retinopathy at baseline was 2.7% (95% CI 0.1-5.3) and 27.3% (16.4-38.2), respectively [15]. A subset of this cohort are taking part in further follow-up to investigate the efficacy of laser treatment for DR in this population. In Mauritius subjects examined in a population based study in 1992 [23] were re-examined for complications of diabetes in 1998 [24]. At 6 years incidence of DR and PDR in subjects with diabetes but no DR in the first study was 23.8% and 0.4%, respectively. The 5 year incidence of any DR in our study was higher (48.4%). For subjects with mild non-proliferative DR (equating to LDES level 20) and moderate non-proliferative DR (LDES L30 or L40) the incidence of PDR at 6 years was 5.2% and 29.4%, respectively. These figures are similar to the five year incidences of PDR in our study (level 20 4.5%; level 30 22%; level 40 40%). Compared to recent European studies progression to STDR from no DR (level 10) and background DR (level 20) was approximately 5 times higher (19.4% vs estimates between 3.9% [2,25] and 4.0% [26]) and 3 times higher (81.3% vs estimates between 26.8 [25] and 28.9 [2]), respectively. These disparities may be explained by differences in access to health services, standards of care for diabetes, presence of comorbidities and genetic factors. Comparisons between studies must be made with caution in view of different study designs and different follow-up rates. Of particular Five year progression of diabetic retinopathy in Sub-Saharan Africa note mortality rates are likely to differ greatly between populations and are an important cause of data censoring.
DR progression at 5 years was associated with lower haemoglobin level. In our group's prospective, 24 month cohort study lower haemoglobin was associated with presence of STDR at baseline (reported elsewhere [14]) but not with progression of DR [15]. An association between anaemia and presence of DR has been reported in Indian [27][28][29] and Chinese [30] cross-sectional studies. Potential confounders of this relationship are decreased erythropoietin production due to diabetic nephropathy and nutritional and socioeconomic status. Impaired oxygen delivery to retinal tissue is one possible mechanism to explain this association. No interventional studies have been conducted to test the effect of treatment of anaemia on the complications of diabetes. Both high iron level and iron supplementation have been associated with gestational diabetes [31,32] and therefore pose risks. In the majority of reports duration is a strong determinant of DR progression [3,4,6]. In this study, and in our group's 24 mouth cohort study of DR in Southern Malawi [15], duration of diabetes was not associated with DR progression in multivariate logistic regression. Information bias may explain why no relationship was demonstrated. It is expected that time from onset of type 2 diabetes until diagnosis will be on average longer and more variable in low resource settings than in high resource settings.
To our knowledge, ours is the first study to report longitudinal visual acuity data in African subjects with diabetes. Procedures for assessing vision differed between the baseline and final assessments. ETDRS measurements yield better VA than Snellen measurements and differences are more marked in persons with low vision [33][34][35]. The effect of these differences in this study would be to reduce the degree of visual loss recorded. A potential bias is that subjects who became visually impaired may have been less likely to attend the follow-up visit in 2012. Visual impairment may increase the chance of mortality in a society where loss of vision entails loss of economic productivity. While laser treatment was not available to the vast majority of subjects in our study, other interventions including cataract surgery were obtainable. Together these factors may explain why a greater degree of visual impairment was not seen in this cohort study.
This clinic-based study has limitations. Barriers to attendance include lack of knowledge of available health services, transportation costs, and competing tasks such as planting and harvesting staple crops. Patients living in rural locations endure long journeys to clinic and may represent a selected sub-group of the rural diabetes population. Few subjects with diet controlled diabetes attend the QECH diabetes clinic [12,13] and therefore only a small number of these patients appeared in our study. Loss to follow-up is a source of bias in this study. Although relatively few deaths were confirmed, high mortality is likely to be an important cause of data censoring. As diabetes care improves in sub-Saharan Africa the prevalence of retinopathy may (paradoxically) go up due to case survival. The epidemic of diabetes in Sub-Saharan Africa necessitates provision of services. This study provides data which is essential in order to design and test locally appropriate and sustainable interventions for DR prevention, early detection and management in the region.