Conclusions
Our study provides definitive IRs for sight-threatening diabetic retinopathy by demographic characteristics routinely collected in DES. There is a strong relationship in CIR increases with retinopathy severity at baseline, and a monotonic increase in rates with advancing yearly intervals, more pronounced in younger people (table 2 and online supplemental table 2). Presence of retinopathy in both eyes, age, and ethnicity are strong determinants for incident sight-threatening diabetic retinopathy. We have defined and calculated sociodemographic associations with sight-threatening diabetic retinopathy, and we provide a web calculator (https://bit.ly/NEL-diabetic-eye-screening-risk-calculator) to estimate disease trajectories of individuals with different sociodemographic profiles. Young, male, non-white ethnic groups with longer duration of diabetes show higher sight-threatening diabetic retinopathy hazards when compared with older, female, white groups, and those with shorter duration of diabetes.
Diabetic retinopathy status at baseline was the most important predictor for sight-threatening diabetic retinopathy referral to hospital eye services. When compared with people with no retinopathy at baseline, the threefold and almost eightfold increase in hazards of sight-threatening diabetic retinopathy for non-sight-threatening diabetic retinopathy in one eye and non-sight-threatening diabetic retinopathy in both eyes, respectively, is consistent with the reported HRs in the literature.5 6 The absence of diabetic retinopathy and the presence of non-sight-threatening diabetic retinopathy in one or both eyes provides valuable information for simple risk stratification in groups of people that were previously considered of low homogeneous risk.6
Younger age groups have been reported to have higher incidence of sight-threatening diabetic retinopathy20–24 and a recent analysis from observed data has shown that young people are at increased risk of experiencing significant delays in diagnosis of sight-threatening diabetic retinopathy if biennial screening intervals were to be implemented.25 Derived from our analysis, we provide evidence of the presence of a different disease trajectory with increased risk of sight-threatening diabetic retinopathy in younger people (HR per 5-year rise in age 0.92, p<0.001). The causes are likely multiple, and result from a complex interplay of different factors that could be partially explained by higher levels of non-attendance to DES,12 13 and by suboptimal control of diabetes and major modifiable risk factors26 27 in young people. Individuals younger than 45 years of age are at a critical stage of their work, or career development, and the lifetime burden and health costs of sight-threatening complications in this population is of considerable public health importance.
Males are at higher risk of sight-threatening diabetic retinopathy development than females. Similar effect sizes to what is reported in our study are available.12 21 24 An electronic medical record (EMR)-based study analysing the development of sight-threatening diabetic retinopathy in people with diabetes in the UK found an HR of 1.22 (95% CI 1.01 to 1.47) and 1.15 (95% CI 1.06 to 1.26) for males newly diagnosed, and males with known diagnosis of diabetes when compared with females, respectively.21 Similarly, Mathur et al reported an increased relative risk of diabetic retinopathy (HR 1.08, 95% CI 1.05 to 1.09), and severe diabetic retinopathy (HR 1.25, 95% CI 1.12 to 1.39), in males when compared with females.22 Lawrenson et al found a 23% increase in odds (95% CI 1.15 to 1.35) of sight-threatening diabetic retinopathy in males when compared with females in a 15-month limited follow-up DES study.12 Non-attendance to DES,12 13 and hormonal differences28 are possible underpinning factors, but the evidence remains contradictory.
South Asian and black ethnic groups have been reported to have higher prevalence of diabetes,22 29 30 and are more likely to develop both sight-threatening diabetic retinopathy30 and visual impairment31 than white people. Sivaprasad et al30 reported, in a predominantly white (66%) population, an 82% (HR 1.82, 95% CI 1.61 to 2.06) and 99% (HR 1.99, 95% CI 1.81 to 2.18) increase in risk of sight-threatening diabetic retinopathy in South Asian and black ethnic groups when compared with white people, respectively. Mathur et al22 showed a 25% (HR 1.25, 95% CI 1.00 to 1.56) increase in risk of severe retinopathy (defined as advanced, proliferative, or laser-treated diabetic retinopathy) in South Asian patients when compared with white patients, however, a third of the ethnicity data were missing. Scanlon et al5 showed a 55%, 58%, and 24% increase in hazards of sight-threatening diabetic retinopathy in African, Caribbean, and other ethnic groups when compared with white people in a small (n=1 223) dataset from South London. More recently, an EMR-based study with over 98% with usable ethnic coding, identified increased risk of sight-threatening diabetic retinopathy in African (HR 1.36, 95% CI 1.02 to 1.83), Indian (HR 1.38, 95% CI 1.17 to 1.63), Pakistani (HR 1.28, 95% CI 1.04 to 1.55), Bangladeshi (HR 1.36, 95% CI 1.19 to 1.54), Caribbean (HR 1.22, 95% CI 1.03 to 1.43), and other (HR 1.25, 95% CI 1.06 to 1.47) ethnicities with a new diagnosis of diabetes when compared with white individuals.21 And Mangelis et al32 identified a 39% increase in hazards of sight-threatening diabetic retinopathy (p=0.009) in African Caribbeans with type 1 diabetes when compared with non-African Caribbean people. These results stress the need to address health inequalities across ethnic groups to improve prevention of sight-threatening complications.
Our study has several strengths. First, a large sample size providing considerable statistical power to detect associations (or absence of association) with sight-threatening diabetic retinopathy stratified by age, sex, ethnicity, retinopathy severity at baseline, and deprivation. Second, IRs provide an important source of reference to inform power calculations for future clinical trials. Third, there is high quality in ethnicity recording with usability of 98.5%. Fourth, the prevalence of any diabetic retinopathy falls in line with previous reports (27.5% prevalence overall, 49.1% and 26.4% in people with type 1 and type 2 diabetes, respectively) and is representative of the UK.5 22 33 Fifth, retinopathy classification was performed by trained assessors following a multilevel internally and externally quality-assured grading protocol that meets UK national recommendations.11
The limitations of our study are as follows. First, HbA1c, blood pressure, blood lipids, medication history, or body mass index were not available for analysis. However, estimates of our Cox model are in alignment with reports from a previous EMR-based study which controlled for the above-mentioned variables.21 Second, we cannot exclude human errors in grading of retinal fundus images despite the well-established grading protocol, but it is expected that, based on hospital eye service outcomes, findings of the study would not substantially differ.
Leveraging an unprecedented multi-ethnic and sociodemographically diverse population undergoing DES, our study delivers a contemporary analysis of sight-threatening diabetic retinopathy incidence and associated factors. IRs provided in our analysis are valuable for future research requiring estimates of transition probabilities or sample size. Our survival analysis revealed significant associations based on simple sociodemographic variables available in routine DES which offer significant information for risk stratification among people with diabetes. Understanding IR and associations of sight-threatening diabetic retinopathy in cohorts such as ours, illuminates paths for future research, identifies areas to optimise service planning, and equity in eye care. Further work to devise accurate prediction models and assess the potential contribution of clinical/metabolic, imaging, and imaging-derived data to risk prediction are essential next steps.