Sub-Saharan Africa faces an epidemic of diabetes [1]. In contrast to high income countries [2–8] 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 sight-threatening 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].

A total of 135 subjects were assessed between December 2011 and December 2012. Median time to follow up was 5.3 years (IQR 5.1–5.4). 15 subjects were confirmed dead. (Total follow-up 150 subjects: 53% of the original 281 subject cohort) (S2 Fig). Baseline characteristics of 281 subjects included in the 2007 study categorised by follow-up are shown in 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, Χ² 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 seen in the 2007 study and subsequently traced and assessed in 2012, 15 (11.1%) were HIV positive at baseline. By 2012 17 (12.6%) were reactive: 13 taking ART; 3 known HIV+ but not taking ART; and 1 new diagnosis (WHO stage 3) [22]. 21 subjects (15.6%) declined testing. In 2012 26 subjects (19.3%; 8 men, 18 women) were anaemic as defined above. The number of subjects who were hypertensive in 2007 and 2012 was 94 (69.6%) and 98 (72.6%), respectively. Only one subject received a course of scatter laser between 2007 and 2012. Of 117 subjects with type 2 diabetes, at baseline (2007) 22 (18.8%) were prescribed insulin (with or without oral hypoglycaemics), 88 (75.2%) were managed with oral hypoglycaemics and 7 (6.0%) were managed with diet alone. At the follow-up visit (2012) 40 (34.2%) were prescribed insulin, 70 (59.8%) were managed with oral hypoglycaemics and 7 (6.0%) were managed with diet alone.

Between baseline and 5 years prevalence of any retinopathy, STDR and PDR increased from 29.6% (95% CI 21.9–37.3) to 63.0% (54.8–71.1), 12.6% (7.0–18.2) to 38.5% (30.3–46.7) and from 2.2% (0–4.7) to 5.9% (1.9–9.9), respectively (n = 135). Two (or more) step progression (from baseline) was observed in 48 subjects (36.4%; 95% CI 28.2–44.6); three (or more) step progression in 30 subjects (22.7%; 15.6–29.9). Of 115 subjects without STDR at baseline (2007) 34 (29.6%; 21.3–37.9) had developed the condition at five years (Fig 1; S1–S4 Tables). Higher baseline HbA1c, longer duration of diabetes and lower haemoglobin were risk factors for 2 step progression in multivariate analysis (Table 3).

Over the follow-up period 25 subjects (18.8%) lost 5 or more ETDRS letters of which 7 subjects (5.3%) lost 15 or more letters. Of these 25 subjects 2 (1.5%) progressed to moderate visual impairment (50–59 letters) and 3 (2.3%) became ‘severely visually impaired or blind’ (<50 letters). The most common primary causes of visual loss for the 25 subjects who lost five or more letters were DR (36%) both DR and cataract (20%) and cataract alone (28%). Therefore in 56% of cases DR was the sole or equal contributing cause of visual loss. In univariate analysis loss of 5 or more ETDRS letters was associated with presence of STDR at follow up visit (OR 3.29, 95% CI 1.32–8.18, p = 0.010) but not age (OR 1.03, 0.99–1.07, p = 0.179) or duration of diabetes (OR 1.03, 0.96–1.10, p = 0.389).

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) and seen at 7 years 12 (35.3%; 19.2–51.4) had developed the condition at 7 years. Fig 2 and S3 Fig combine data from our 24 month cohort study performed between 2012 and 2014 and published elsewhere [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.

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 follow-up 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 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 [27–29] 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 [33–35]. 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.