A simple algorithm to estimate the time to development of sight-threatening diabetic retinopathy

Abstract

BackgroundMore than 3 million people with diabetes in the UK are at risk of vision loss because of diabetic eye disease. Present UK recommendations specify annual diabetic eye screening in those aged 12 years or older with diabetes. Patients identified with potentially sight-threatening diabetic retinopathy are referred to ophthalmology services. Those with no retinopathy and those with non-referable retinal microvascular damage continue to be invited for annual screening irrespective of risk. Models for risk stratification require clinical information, including a measure of glycaemic control that may not be available to screening programmes. With use of data only from retinal photographs we aimed to identify patients at low, intermediate, and high risk of those offered annual appointments, to inform discussion about potential changes to recommended screening intervals. MethodsWe reviewed all patients referred from primary care to one local screening programme with assessable images of both eyes from at least three screening episodes between Jan 1, 2005, and Dec 31, 2010. Digital images were captured after dilation and were graded by a central quality assured team with NHS diabetic eye screening programme protocols. We included patients who at the first and second (index) screenings had no retinopathy or microaneuryms in one or both eyes. They were followed up until any sight-threatening diabetic retinopathy was detected or until the last available photograph with no sight-threatening diabetic retinopathy. Patients were categorised by presence of microaneuryms in neither, one, or both eyes at each screening. We excluded those with any missing data. By design, no confounding variables were included in the model. We derived life table survival plots with Kaplan-Meier estimates and calculated hazard ratios (HRs) with Cox proportional hazards models. We used log logistic models to estimate the proportion of patients with microaneuryms every year after baseline. Data were analysed with SAS (version 9.1). FindingsFrom 2005, 18–254 patients with non-referable diabetic retinopathy at two consecutive screenings were available, of whom 3700 had no further assessable images. Hence 14–554 patients were followed up for further median 2·8 years (IQR 1·3–3·3). Their median age was 65 years (IQR 56–73), and 8188 (55%) were men. Of 7246 with no evidence of retinal microvascular damage at both screenings, 120 progressed to sight-threatening diabetic retinopathy, 0·7% by 1 year. Of 1778 with microaneuryms in neither eye at first screening and in one eye at second screening, 80 progressed to sight-threatening diabetic retinopathy, 1·9% by 1 year (HR 2·9, 95% CI 2·2–3·8). Of 1159 with microaneuryms in both eyes at both screenings, 299 progressed to sight-threatening diabetic retinopathy, 11% by 1 year (HR 18·2, 95% CI 14·7–22·5). Those in higher risk groups were more likely to progress to more serious diabetic retinopathy than were those in lower risk groups. InterpretationThis screening programme of a mainly northern European population has robust quality controlled imaging and grading. Hence it needs to be validated in other populations and programmes to be generalisable. Two sequential image sets can differentiate levels of risk and could modify the interval for screening. The addition of these data to clinical indices could improve risk stratification compared with either method alone. FundingNone