Incidence of sight- threatening diabetic retinopathy in an established urban screening programme: An 11- year cohort study

Aims: Systematic annual screening to detect sight- threatening diabetic retinopathy (STDR) is established in the United Kingdom. We designed an observational cohort study to provide up- to- date data for policy makers and clinical researchers on incidence of key screening endpoints in people with diabetes attending one screening programme running for over 30 years. Methods: All people with diabetes aged ≥12 years registered with general practices in the Liverpool health district were offered inclusion. Data sources comprised: primary care (demographics, systemic risk factors), Liverpool Diabetes Eye Screening Programme (retinopathy grading), Hospital Eye Services (slit lamp biomicroscopy assessment of screen positives). Results: 133,366 screening episodes occurred in 28,384 people over 11 years. Overall incidences were: screen positive 6.7% (95% CI 6.5– 6.8), screen positive for retinopathy 3.1% (3.0– 3.1), unassessable images 2.6% (2.5– 2.7), other significant eye diseases 1.0% (1.0– 1.1). 1.6% (1.6– 1.7) had sight- threatening retinopathy confirmed by slit lamp biomicroscopy. The annual incidence of screen positive and screen positive for retinopathy showed consistent declines from 8.8%– 10.6% and 4.4%– 4.6% in 2007/09 to 4.4%– 6.8% and 2.3%– 2.9% in 2013/17, respectively. Rates of STDR (true positive) were consistently below 2% after 2008/09. Screen positive rates were higher in first time attenders (9.9% [9.4– 10.2] vs. 6.1% [6.0– 6.2]) in part due to ungradeable images (4.1% vs. 2.3%) and other eye disease (2.4% vs. 0.8%).


| INTRODUCTION
By 2045, an estimated 693 million people (9.9% of the global adult population) will have diabetes 1 with around 10% having sight-threatening diabetic retinopathy (STDR, DR). 2 Early detection and treatment of STDR is key to avoiding visual impairment in people living with diabetes (PWD). [3][4][5][6] Laser photocoagulation remains the mainstay of treatment for proliferative disease, while intravitreal therapies are indicated in most patients requiring treatment for maculopathy. Systematic annual screening to detect STDR is established in several countries including the United Kingdom and has greatly improved the detection of treatable disease.
Landmark epidemiological studies on DR are around 30 years old, 7,8 and while providing important underpinning, data on incidence and progression are not directly applicable to screening programmes. As programmes become established, people with longstanding disease are detected reducing the prevlance in DR in the screening population. Data from screening programmes are limited and reporting is variable. [9][10][11] Screening populations are stable and provide large cohorts for longitudinal epidemiological studies and for interventional clinical studies of early disease, but their characteristics are changing as systematic screening becomes established and evolves. True positive cases are retained under hospital eye services (HES). Disease management has changed including improved glucose and blood pressure (BP) control, less smoking and longer life expectancy.
We conducted an observational cohort study of people attending during an 11-year period for DR screening in one established urban programme in England. Our study was conducted within the Individualised Screening for Diabetic Retinopathy (ISDR) programme of applied research in DR screening 12,13 (www.isdrp roject.co.uk). Here, we present data on incidence of all key stages in the screening pathway, screen positive, screen positive due to DR and STDR. We aimed to provide up-to-date data to inform screening programmes, policy makers and clinical services and to support future clinical research in early retinopathy. We also investigated retrospective data collected since the beginning of screening in Liverpool to gain additional insights into long-term changes in incidence.
interpretation, or writing of the report. The corresponding author had full access to all data in the study and has final responsibility for the decision to submit. positive, fewer than 50% are screen positive for diabetic retinopathy. Most are due to sight threatening maculopathy. The annual incidence of STDR is under 2% suggesting future work on redefining screen positive and supporting extended intervals for people at low risk. Higher rates of screen positive and STDR are seen in first time attenders. Those who have never attended for screening should be specifically targeted.

K E Y W O R D S
incidence, prevalence, screen positive, screening, sight-threatening diabetic retinopathy

Novelty statement
• In an established urban diabetic retinopathy (DR) screening programme in England, screen positive rates show a consistent fall over time to a low level. • The annual incidence of sight-threatening diabetic retinopathy (STDR) is low at under 2%. New proliferative disease is rare. The majority of referrals from screening programmes are not due to DR. • Higher rates of screen positive and STDR occur in first time attenders, in particular previous non-attenders, and in type 1 disease who should be targeted. • Once a steady state has been reached, screening programmes should consider revising referral thresholds and extending intervals for low-risk individuals.

Precis
• What is already known about this topic? Prevalence of diabetes is increasing worldwide against a background of inadequate and overstretched resources. Screening detects DR at a stage at which vision loss can be prevented; it is effective and cost-effective in high-income settings. Current programmes are designed to address incidence and prevalence based on 30-year-old data • What is the key question? What is the incidence of DR, types of STDR and a screen positive result in an established screening programme? Can high risk groups be identified? • What are the new findings? In an urban DR screening programme established for 30 years with a stable population, rates of screen positive and screen positive due to DR show a consistent fall over time to a low level. We report a consistently low annual incidence of STDR at under 2%. New proliferative disease is rare. Most referrals to the hospital eye service are for non-DR-related findings. Higher rates of screen positive and STDR are seen in first time attenders; previous non-attenders are at particular risk. Reporting of outcomes in the literature is highly variable.  Figure 1 and apply to all English screening programmes. All PWD aged ≥12 years registered with a Liverpool general practice (GP; primary care) are considered eligible for routine screening apart from the following categories: moved out of area; seen by another programme; blind (no perception of light in both eyes). Of those individuals who are eligible, two groups are excluded: opted out of screening; medically unfit. Individuals are suspended from routine screening if they are attending the HES for any of the following: management of active diabetic eye disease; slit lamp biomicroscopy (SLB) due to ungradeable photographs; digital surveillance (monitoring of DR more frequently than annually by retinal photography in a dedicated surveillance clinic). The remaining people undergo active screening in the community by a qualified retinal screener at one of six primary care facilities across the city. Screening in the NDESP involves technician-based digital photography through dilated pupils with at least two 45° colour photographs and manual grading using the grading schema shown in Table S1. Screen positive is defined as moderate pre-proliferative retinopathy (R2 in the NDESP scheme) and above; and/or maculopathy (grade M1); and/ or other significant eye disease; or ungradeable images. 14 Screen positive individuals undergo clinical examination by an ophthalmologist trained in medical retina in dedicated HES-based clinics. For the purposes of our analysis, this examination defines 'true positive' for STDR defined as follows: moderate/severe pre-proliferative DR or proliferative DR (sight threatening retinopathy, STR) and/or sight threatening maculopathy (STM), i.e., any of the following features: multiple blot haemorrhages, venous beading, intraretinal microvascular abnormalities, new vessels, preretinal/ vitreous haemorrhage, fibrovascular proliferation, exudates within 1 disc diameter (1500 μm) of the foveal centre, group of exudates within the macula more than 0.5 disc area in size, retinal thickening within 1 disc diameter of the foveal centre, haemorrhages/microaneurysms with reduced vision.

| ISDR cohort study participants
All PWD registered with the LDESP covering a single urban health district in the north west of England were offered F I G U R E 1 Diabetic eye screening cohort management in England and Wales. Source: https://www.gov.uk/gover nment/ publi catio ns/diabe tic-eye-scree ning-cohor t-manag ement -overv iew/diabe tic-eye-scree ning-cohor t-manag ement Accessed 27 January 2020 • How might this impact on clinical practice in the foreseeable future? Low rates of STDR suggest that the purpose of screening should be reassessed once a steady state has been reached with changes in the referral threshold and extended intervals for people at low risk. People newly added to a programme, in particular those with a history of previous non-attendance, should be specifically targeted.
inclusion in the ISDR programme. The patient cohort was established in a recruitment process approved by the Preston North West NHS Ethics Committee (13/NW/0196) and the Royal Liverpool and Broadgreen University Hospitals NHS Trust (RLBUHT) governance teams. GPs in Liverpool were invited to participate commencing in June 2013 and continued until 2016 when all had agreed to participate. Recruitment occurred for all PWD in each practice. Newly diagnosed PWD and those PWD moving into area were added via the LDESP throughout the study. Consent was through an 'opt-out'. An invitation letter, information booklet and an opt-out form were sent to eligible participants. The dataset comprised data acquired both prospectively from June 2013 to September 2017 and retrospectively for PWD who did not opt out from April 2006 to 2013. This latter data exhibited a censoring effect: governance rules removed data from PWD who had died prior to a practice joining the study and who therefore had no opportunity to opt out.

| Data sources and analysis
Data on PWD in the UK NHS are collected in primary care, within screening programmes and in secondary care. A purposebuilt real-time dynamic data warehouse was developed to store data from these three previously unlinked sources: This analysis incudes only individuals attending for DR screening in the community and their first follow-up in the HES. Collection of data on people attending the HES for management of active diabetic eye disease, SLB screening due to ungradeable photographs or hospital-based digital surveillance of DR was not possible; therefore, we are unable to report whole population data. For the purpose of analysis, the values of the time-dependent clinical variables closest to the time of the screen episodes (i.e., annual screening episodes) within a time window of 1 year prior to 1 week after the screening episode were used. For positive screen events, data from the first SLB recorded within 1 year of the positive screen event were used to provide the final outcome for that event. The first recorded screening attendance was defined as the first recorded screening attendance where there were no earlier recorded screening attendances or SLB records for that individual. Any other recorded attended screening appointment was defined as any screening appointment where there had been at least one earlier attended screening appointment or SLB.
Two thousand two hundred sixty-five participants from the cohort study were recruited to an RCT within the ISDR programme and allocated to individualised interval screening. 12 For those allocated to 24-month intervals (who under normal circumstances would have been screened annually) data on annual incidence were adjusted (with an assumed attendance proportion of 0.85): 2015/16 226 individuals (266 * 0.85), 2016/17 1232 individuals (1449*0.85). Of these, 2.7% were assumed to have type 1 diabetes and 89.9% type 2 diabetes. All were assumed to be screen negative at the first appointment (which under normal circumstances would have occurred 12 months before the allocated appointment at 24 months). Additionally, 96.2% were assumed to be R0R0 and 3.8% to be R1R0.

| Dataset
The analysis dataset was extracted from the ISDR data warehouse on 25 September 2017. The size of the dataset varied throughout the 11 years studied with year-end effects and variation due to sampling time. Figure 2 shows the distribution of data in the ISDR cohort study. Of 30,771 invited to participate up to this time point, 2191 (7.1%) opted out and 196 had no LDES or HES data leaving data on 28,384 PWD available for analysis. Table S2 shows the numbers of people on the LDESP register categorised by inclusion in the ISDR study and attendance at 1 or more screening appointments in a screening year.

| Censoring effect
The number of individuals with at least one screening appointment recorded in our dataset increased each year between 2006/07 and 2014/15. Between 2006/07 and 2016/17, the median age increased from 60.3 to 64.5 years and the median disease duration from 2.3 to 7.0 years. In order to assess whether censoring altered our estimates of retinopathy, we examined rates of STDR during the time period with the most complete data: 2013/14 to 2016/2017. Of 28,580 individuals in the ISDR cohort, 2538 (8.9%) died. Rates of STDR were calculated from the last attended appointment. In individuals in whom STDR status was not missing, there was no significant difference in STDR rates between subjects who died and those who did not: 164/2311 (7.1%; 95% CI 6.1-8.2) vs. 1571/23,331 (6.7%; 6.3-7.0). We also performed exploratory modelling to investigate the effect of death in subjects pre-2013 (data not shown). Together, our analyses indicate that death appears to have had little or no confounding effect on estimated rates of STDR prior to 2013/2016.

| Demographics
Overall demographics of the 11-year cohort are shown in Table 1. Gender proportions were relatively stable across all screening years (55%-58% male, 40%-43% female, 1%-3% unknown). The proportion of individuals with type 1 diabetes decreased over time from 6.7% to 4.7%, while the proportion with type 2 diabetes increased from 76.1% to 81.6%. The proportion with unknown diabetes type decreased proportionally from 17.2% to 13.7%. The median HbA1c was stable across all years (median: 50-52 mmol/mol; 6.7%-6.9%) although there was a high proportion of missing data prior to 2012/13.

| Incidence
Annual incidences of screen positive, STDR, STR and STM are shown in Table 2; 28,384 PWD attended at least one screening episode over the 11 years (total 133,366 screening episodes included). 6  death; moved out of area) as well as missing data which could not be recovered. There was some variation across the years of rates of screen positive with an overall reduction over the 11 years from 8.8%-10.6% in 2007/09 to 4.4%-6.8% in 2013/17. The annual incidence of screen positive due to DR also decreased over the 11 years from 4.4%-4.6% in 2007/09 to 2.3%-2.9% in 2013/17. In contrast rates of screen positive for ungradeable images were stable, fluctuating around 2%-3%, similarly other eye disease was stable at around 1%. Rates of STDR (true positive; examined by a medical retina specialist using SLB) were relatively stable between 1.3% and 2.2%. STR at 0.4%-0.9% and STM at 1.1%-1.8% were also stable. Figure 3 illustrates these trends over time. Out of those with a known retinopathy level, the proportion of subjects with no DR rose steadily from 61.8% in 2006/07 to 79.4% in 2016/17. Tables 3 and 4 show the annual incidences of screen positive, STDR, STR and STM for individuals attending their first recorded screening appointment and for those attending a screening appointment which was not their first recorded screening episode, respectively. Data are displayed from screening year 2007/08 to allow for at least a full year of data collection prior to a first recorded screening event. Compared to people already attending screening, rates in first time attenders for screen positive were ×1.6 higher (9.9% vs. 6.1%) and for STDR ×1.4 higher (2.2 vs. 1.6). Much of this screen positive effect appears to be due to unassessable images (×1.8, 4.1% vs. 2.3%) and other eye disease (×3.0, 2.4% vs. 0.8%). Figures S1 and S2 show a graphical representation of the annual incidences of screen positive, screen positive for DR, STDR and STR in these two groups.

| First time attenders
People living with diabetes attending for DR screening for the first time are a heterogeneous group comprising people newly diagnosed with diabetes and those with existing diabetes who were previous non-attenders. We investigated the annual incidences of screen positive, STDR, STR and STM for individuals attending their first recorded screening appointment where disease duration was ≤1 year (i.e., newly diagnosed) and where disease duration was greater than 1 year (Tables S3 and S4). The overall screen positive rate across all study years in the newly diagnosed group and those with diabetes >1 year was 8.4% (8.0-8.9, 95% CI) and 14.3% (13.2-15.4), respectively. Both figures are higher than the rate in people who had already attended at least one screening appointment: 6.1% (6.0-6.2). Similar but stronger effects were seen for rates of screen positive for DR (newly diagnosed diabetes 2.2% [2.0-2.5]; diabetes duration >1 year attending a first recorded screening event 7.1 [6.3-7.9]; at least one

| DISCUSSION
In this observational cohort study of people with diabetes attending an established retinopathy screening programme, we T A B L E 2 Annual incidences of screen positive, sight-threatening diabetic retinopathy (STDR), sight threatening retinopathy (STR) and sight threatening maculopathy (STM) for people with diabetes who attended at least one screening episode in each Note: Denominator for a given year is the total number of individuals who attended at least one screening appointment during that year. For example, for screening year 2012-13, the denominator is 13,518, and the annual incidence of overall screen positive is 5.7% (774/13,518). STM and STR are not mutually exclusive categories. 95% CI are based on the Wilson score method.
Abbreviation: DR, diabetic retinopathy. a Biomicroscopy recorded within 1 year of screening appointment when positive screen result occurred.
b Adjusted for subjects recruited to the ISDR RCT and assigned to 24-month screening intervals.
report estimates of incidence of key stages in the diabetes eye care pathway. Average annual incidences across the 11 years studied were 6.7% for screen positive, 3.1% screen positive due to DR, 1.0% other significant eye disease and 2.6% ungradeable images. 1.6% had STDR confirmed by a medical retina specialist. The prevalence of any DR was 27.2%. Our latest estimates of the incidence of STDR should be seen in the context of STDR rates recorded since 1991/92 in the LDESP. From an initial 6.9%, a steady fall was observed during the phased roll-out of screening over 16 years to 1.8% in 2007/08 (screening programme data; available on request from authors). These long-term observed data are likely to represent a 'first pass effect' where existing disease is detected in a population screened for the first time.
Other factors may have influenced the observed decline in STDR rates from the early years of screening in Liverpool. Medical care for PWD has improved over this time. Diagnostic criteria for diabetes have changed in recent years, 15 and there have been concerted efforts to detect diabetes early. A similar reduction in more advanced stages of disease was identified in a systematic literature review by Liew et al., 16 who reported a two to threefold reduction in PDR and DMO over the last 30 years. Over the 11 years, there was a gradual fall in the prevalence of DR in the screened population from 38% to 21% with equivalent falls in annual rates of screen positive and screen positive for DR. In contrast STDR, rates remained relatively stable at under 2.0%. This steady decline in diagnostic categories prior to the STDR stage requires further research. They may represent improvements in grading. A lowering of the threshold for the diagnosis of diabetes 15 may have increased the proportions with no retinopathy.
The strengths of our study include the size of the dataset and the duration of data collection.
HbA1 c levels were generally well controlled in our population. This may reflect current standards of medical care but better compliance with medical care amongst those attending screening cannot be excluded. The current prevalence of diabetes in adults aged 17+ in Liverpool is 6.6% (29,993 people), slightly lower than the 7.1% in England as a whole. 17 DR screening coverage in Liverpool is low compared to other programmes: 80.4% in year April 2016 to March 2017. 18 The main limitation of our study is the mixed prospective and retrospective dataset reflecting changes in governance regulations and data collection platforms, a common issue in long-term, observational studies of real-world collected data. Around 5% of the population died or moved away each year and without consent, we were unable to access their historical data. This censoring could have affected our findings between 2006/7 and 2012/13 and is seen in the change in median age and disease duration. Subjects with worse retinopathy would be expected to have higher mortality, so our numerators may be underestimates. However, our exploratory analyses suggested that taking this effect into account would have a negligible effect on STDR rates. The ISDR RCT commenced in 2014 resulting in a proportion of patients moving to 2-yearly screening. We corrected for this effect in the last 2 years of our analysis.
In this study, we report results from PWD attending screening. Collection of data on people attending the HES with other eye disease (who undergo screening in the HES) was not possible; therefore, we cannot report whole population data. In Liverpool in 2017/ 18, 9.4% (2054/21,853) of PWD were attending the HES for management of DR or slit lamp-based screening. Nonetheless, it is possible to estimate a rate of STDR in people attending screening of 164/10,000 PWD/annum (2193/133,366 * 10,000). Comparison of our data with other studies is limited by the inconsistent and incomplete reporting of different screening outcomes. Longitudinal cohort studies from our group, 19,20 and from Norwich 21 and Wales, 22 have estimated annual incidences of preproliferative retinopathy and maculopathy. Looker et al. 9 reported data from the first 5 years of the DR Screening in Scotland. Rates of referable eye disease equivalent to our screen positive due to DR were highest in the first 2 years of screening (7.0% 2006, 6.0% 2007) before stabilising at 4.3%, slightly higher than our 3.1%.
Misra et al. 10 reported data from 20,788 PWD undergoing annual retinal photography between 1990 and 2006 in one English region. Rates of referable retinopathy (NDESP grades R2 or R3 or M1) increased from 2.0% in 1990 to 6.7% in 2001, then decreased to 4.7% by 2006. Scanlon et al. 11 reported a rate of referable retinopathy/STDR of 2.3% in a   9 We report very low rates of R3 with none in the last 3 years. Compared to STR, STM is more expensive to treat and monitor. Our results are of interest to health service providers and to researchers developing new therapies who require candidates for clinical trials.
Current UK screening definitions and pathways were established over 15 years ago to support the introduction of screening and were consensus based. 23 Now that the rates of STDR are low, it may be appropriate to revisit the definition of screen positive in order to ensure optimum resource allocation. Extended screening intervals should be considered by programme managers; data on the proportions of patients with R1/R0 and R1/R1 will inform risk stratification. 13,24 The T A B L E 4 Annual incidences of screen positive, sight-threatening diabetic retinopathy (STDR), sight threatening retinopathy (STR) and sight threatening maculopathy (STM) for individuals attending a screening appointment in each screening year (1 April-31 March) which was not their first recorded screening episode between 2007 and 2017 Screening year low rates also support moves to introduce personalised riskbased variable-interval screening using clinical risk factor data. 12,25 Of those identified as screen positive in our study, only 28.1% had true positive disease. A high proportion of screen positive cases were due to ungradeable images (most likely due to cataract) and other significant eye disease, identification of which is not an objective of screening. New technologies may overcome ungradeable images (38.7% in our study); agreement is needed on approaches to managing screen positive due to other eye disease (15.7% in our study).
Despite high rates of ungradeable images, newly screened PWD in our study demonstrated rates of STDR 1.4× higher than existing screening participants (2.2% vs. 1.6%). This effect appears to be driven by previous non-attenders and is consistently seen in other epidemiological studies. In the United Kingdom Prospective Diabetes Study (type 2 disease), 4.5% of female and 7.9% of male participants had STDR at diagnosis of diabetes, 26 while the LDES reported 6.6% (9.8% for Type 1) having STDR at first screen. 19,20 In the Scottish programme, 6.9% of first-time attenders had referable disease compared to 3% for people attending their fourth or fifth screen. 9 New systematic DR screening programmes in Asia report similar findings. In Hong Kong, where screening following the English NDESP protocol was introduced for the first time in 2014, STDR was detected in 9.8% of 174,532 new screened patients in 12 months. 27 A phased introduction of screening into populations may reduce the service delivery pressures on HES. We report higher rates of screen positive and STDR in people with type 1 diabetes. Lower rates of assessable images and other eye disease in these individuals compared to those with type 2 diabetes are likely to reflect younger age. Interestingly, type of diabetes was not identified as being predictive for referable DR by the Liverpool Risk Calculation Engine. 14 However, this may reflect a low number of events in the dataset used.
We have shown that in an established screening programme with a stable population, rates of screen positive DR show a consistent fall over time to a low level. The annual incidence of STDR is under 2% highlighting the need for future work defining screen positive and supporting extended intervals for people at low risk. Higher rates of disease are seen in new screened patients and previous non-attenders who should be specifically targeted, and in people with type 1 diabetes. Our results should be generalised with caution, particularly in populations with higher rates of STDR, higher proportions of ethnic minorities and programmes in set-up. Our data represent a benchmark against which other screening programmes can be measured and will inform both redesign of screening services and future intervention studies.