Visual acuity outcome of stable proliferative diabetic retinopathy following initial complete panretinal photocoagulation

Introduction

Proliferative diabetic retinopathy (PDR) is characterised by the growth of abnormal new vessels on the retina and/or optic disc. These clinical features alone are not usually associated with visual impairment unless there is coexistent diabetic macular oedema (DMO).1 Anti-vascular endothelial growth factor (anti-VEGF) therapy is the treatment of choice for DMO. The Diabetic Retinopathy Clinical Research (DRCR) Protocol T trial reported that less than 5% of patients with DMO treated with anti-VEGF therapy lost 10 or more Early Treatment Diabetic Retinopathy Study (ETDRS) letters by the end of 2 years.2 In clinical practice, treatment naïve eyes with low or high-risk PDR that does not require vitreoretinal surgery but have coexistent DMO are treated with a combination of anti-VEGF therapy and panretinal photocoagulation (PRP).3 4 There is limited literature on the probability of visual loss in patients presenting with PDR with and without DMO over time in real life.5–7

The main cause of severe visual loss in PDR is due to complications of untreated or suboptimally treated PDR such as vitreous haemorrhage and tractional retinal detachment. The landmark clinical trials, Diabetic Retinopathy Study8 and the ETDRS5 demonstrated that PRP in patients with PDR reduces the risk of severe visual loss by 50%. Eyes with non-clearing vitreous haemorrhage or tractional retinal detachment threatening or involving the macula may require vitrectomy.9 Another rarer complication of PDR that can cause severe visual loss is neovascular glaucoma.10 In the UK, where diabetic retinopathy screening is well-established, most patients who present with PDR have low-risk characteristics and a small proportion present as high-risk PDR and do not usually require prompt vitrectomy. Very few patients with PDR present with neovascular glaucoma.11 Over the last 40 years, PRP has prevented severe visual loss in most patients who present early with PDR. However, there are recent reports that show the proportion lost to follow-up is high in patients with PDR and cost of treatment may be a factor.5–7 Contemporary data on long-term visual outcome of treatment naïve patients who present with low or high-risk PDR treated with initial complete PRP in a fully public-funded health system is lacking.

Recent studies such as Protocol S and the CLARITY trial, designed to examine the role of anti-VEGF therapy as an alternative therapy to PRP for active PDR not requiring vitreoretinal surgery at baseline, showed that eyes treated with PRP maintain presenting visual acuity (VA) over 1–2 years.12 13 Approximately 45% required further fill-in PRP over 2 years.1 Moreover, in the PRP arm of Protocol S, 34% developed vitreous haemorrhage, 10% had retinal detachment and 15% underwent vitreoretinal surgery by 2 years.1 The 5-year follow-up of Protocol S showed stable visual outcomes with PRP in approximately 60% of the initial study cohort who were retained in the study.14 Factors that determine visual outcomes in Protocol S have been explored but they explain only one-third of the variations.15 The predictors of visual outcome of patients who achieve first stabilisation of PDR in clinical practice is unclear. These findings are consistent with retrospective studies. Indeed, a large Indian multicentre study described that about 55% of patients with stable treated PDR had a VA measure of ≥6/12 Snellen in a 10 years follow-up.16 Moreover, Kaiser and colleagues similarly found in their cohort study that of the 76% of patients with PDR presenting with 6/12 or more Snellen VA could maintain this level of vision 1 year after PRP.17

The aim of this study was to assess the probability of visual loss over time of patients from first documentation of stable PDR after PRP in routine clinical practice and to determine the predictors of visual loss. We considered outcomes in eyes with and without DMO at presentation.

Methods

Results

Probability of losing one Snellen line of vision (ETDRS ≥5 letters)

Kaplan-Meier modelling demonstrates that the median survival time to loss of five ETDRS letters or more was 3.32 (95% CI, 2.94 to 3.78 years, irrespective of the DMO status at baseline (data not shown). Those with active DMO (M1A) at baseline are more likely to experience this event (median event time=2.36 years (95% CI, 1.75 to 3.32)) than those with stable DMO (M1S; 3.52 years (95 CI, 2.47 to 4.05)) and without DMO (M0; 3.61 years (95% CI, 3.16 to 3.99)) (figure 1A,B). A sensitivity analysis removing those who had vitrectomy, repeat PRP or anti-VEGF therapy in the 6 months prior to vision loss did not change these probabilities (data not shown). Multivariable Cox proportional hazards models were used to identify covariates predictive of patients losing five or more ETDRS letters (figure 1C). Increasing age, better baseline VA and coexistent active or stable DMO were predictors of loss of five or more ETDRS letters.

Request permission

Figure 1

Probability of visual acuity loss of at least five Early Treatment Diabetic Retinopathy Study (ETDRS) letters after achieving stability in proliferative diabetic retinopathy. Kaplan-Meier modelling was carried out to estimate the probability of losing five or more ETDRS letters (Snellen equivalent one line) for stratified by macular status at time of diagnosis. The cohort was substratified by the macular status at baseline (first record of stable proliferative diabetic retinopathy): absence of diabetic macular oedema (DMO; M0), active DMO on treatment (M1A) or stable DMO under observation (M1S). Median time to loss of at least five ETDRS letters was (A) plotted and (B) tabulated comparing each of these subcohorts. (C) Multivariable Cox proportional hazards considering each of the baseline covariables was carried out, including: age, gender, visual acuity (VA), macular status, phakic status, previous vitrectomy, (phakic vs pseudophakic) and previous treatment for DMO.

Follow-up of stable treated proliferative diabetic retinopathy

The distribution of last hospital appointments for our patient cohort is shown in figure 2. The mean follow-up was 3.62 (SD 1.67) years and there were no differences between the groups with and without DMO at baseline. Approximately 60% of the cohort had an observation beyond 3 years. However, patients recruited included in the study in 2019 could not have 3 years follow-up and so the follow-up data is better. The mortality rate was about 13% with no differences between groups with and without DMO at baseline.

Request permission

Figure 2

Follow-up and mortality among patients proliferative diabetic retinopathy after being deemed stable (A) Histogram of duration of total follow-up for each patient within the cohort. (B) Mean, median, SD, minimum and maximum follow-up duration in years was queried for the entire cohort (Overall) and the subcohorts stratified by baseline macular status: absence of diabetic macular oedema (DMO; M0), active DMO on treatment (M1A) or stable DMO under observation (M1S). Similarly, the proportion of patients that were followed-up beyond 1 year, 2 years, 3 years, 4 years and 5 years is shown, as well as those with a recorded death within the observation period.

Discussion

The results of this retrospective study show that there is a 50% likelihood for a real-world cohort under follow-up after achieving stability post PRP for treatment naïve PDR to maintain their presenting VA for at least 3.5 years. However, in the 50% that lost at least one line of vision, the median loss was nine ETDRS letters, irrespective of the presence of DMO at baseline. A 10-letter loss is a clinically meaningful difference as these changes exceed measurement variability.23 In the study, this was corroborated by only taking forward changes present in two consecutive visits. These findings are worse than that reported in the 5-year Protocol S cohort, where 9% lost 10 or more letters from baseline.14

A five-letter loss is particularly important in a person whose baseline VA is around 70 letters, when the VA in the other eye is already below 70 letters. Although at baseline, 65% had VA of ≥70 ETDRS letters (Snellen equivalent ≥20/40) in the study eye, only 34% maintained this vision at the point when the study eye lost at least a line of vision, highlighting that stable treated PDR may continue to affect the quality of life of patients. When we consider the VA of both eyes together, this loss of at least one line of vision in the study eye disqualifies approximately 15% of the study cohort from driving based on VA criterion alone.²⁰ The baseline predictive factors of 5-letter losers included older age, higher VA and presence of any previous, active or stable DMO. Treating DMO aggressively will likely enable more patients to retain vision that meets the legal limits of driving.

Another important observation noted in the study is the visual outcomes of patients with stable treated PDR with no baseline DMO. Only 70% of these eyes had a VA of 70 letters or better. Other causes of impaired VA in patients with treated PDR include uncorrected refractive errors, visually significant cataract and DMO.16 24 Only 11.8% had previous cataract surgery in this study. These figures are similar to the baseline characteristics of the CLARITY trial cohort that was conducted across 22 sites in the UK and the Protocol S study in USA.12 13 In addition, 8.5% had previous vitrectomy indicating more advanced PDR and 11.1% had VA of 35 letters or worse (equivalent to Snellen 20/200 and defined as severe visual impairment). Factors that were associated with poor visual outcomes in Protocol S included poor glycaemic control, higher mean arterial pressure, prior DMO and higher severity levels of diabetic retinopathy. Approximately two-thirds of the variations in visual outcome in the trial could not be explained by these associations.15 This has been similarly suggested in retrospective analyses (Wykoff et al). Although we did not have access to systemic data for our patient cohort, our results also indicate that previous or coexistent DMO is the highest risk for decline in VA in real life. We could not explain why circa 50% of eyes without DMO (676/1312) also experienced a median drop of nine letters. Although 70% of the eyes with no DMO had 70 or more letters at baseline, only 30% of patients in this group had VA of 70 letters or better despite absence of DMO at baseline. New onset DMO that did not meet the central subfield thickness of 400 µm, a requirement to start anti-VEGF in the National Health Service in England and Wales, may be a reason but the database does not capture this data. Other factors such as diabetic macular ischaemia were not quantified by OCT-A in this study. With the advent of OCT-A, it is likely that we will be better informed of its contribution to vision loss in people with PDR. A proportion of patients with PDR also develop disorganisation of the inner retina and this feature is associated with poor visual outcomes.25 The inner retinal changes in retinal nerve fibre layer and ganglion cell layer post PRP may also play a role26 27 although recent reports suggest that these changes may be age related rather than laser-induced.28

Another important finding from this study was the loss to follow-up proportions. Non-English speakers, increasing age, multiple comorbidities and lack of health insurance cover, longer distance travelled for clinical appointment and failure to attend other non-eye-related appointments have all been reported as factors associated with lost to follow-up.10–12 Although the proportions of patients with no records increased over time, some patients especially those included from 2019 will have less follow-up and the COVID-19 pandemic could have affected the longer follow-up. However, the mortality rate of 13% is high given that the mean age of this cohort was 59 years. Systematic reviews on this relation have highlighted that increasing severity of diabetic retinopathy is a risk factor for all-cause mortality in both type 1 and 2 diabetes.29 30 The findings in this study highlight the need to holistically manage these patients who remain at high risk of visual loss and PDR is a surrogate for generalised vascular dysfunction in diabetes.

This study on a large multiethnic cohort of patients with stable treated PDR in a real-world setting in the National Health Service in the UK is generalisable as the cohort comprises patients with and without concomitant DMO, vitrectomised eyes, both type 1 and 2 diabetes and varying age groups. Due to variability of VA recordings in clinical settings, we mandated 5-letter drop to be confirmed in two consecutive visits. A key assumption implicit to time-event analyses is that censored patients have the same chance of experiencing an event as those still under observation. Survival analyses make use of all available data up until censorship to calculate probabilities of surviving each interval and thereby estimate event probability at any time point. For patients with missing data at a given time point (ie, unknown time to event), the calculated time-event probability can be thought of as a combination of the patient’s own available data and imputation from other patients in their cohort with more data. Another assumption is that time of study enrolment is assumed not to affect event probabilities. It is also important to note that Cox modelling can identify associations between variables.

Limitations of observational studies of real-world clinical data are well recognised.3 They inherently feature variability and heterogeneity in patient management, follow-up and outcome measurement. As the Moorfields clinical practice is based on diabetic retinopathy and DMO protocols and treatment is free to all, we believe some of these confounders and bias may be reduced. Nevertheless, other variables such as VA recording in clinical settings may be of concern despite the requirement of two consecutive VA records to confirm five or more letter loss. The proportions of patients lost to follow-up after initial treatment of PDR is also an important discussion point since the evidence of clinical effectiveness of anti-VEGF therapy for PDR were reported.5–7 Even in a clinical trial setting, only 60% of the patients were retained by 5 years in Protocol S.14 Our study also highlights the increased risk of mortality in patients with PDR. Despite these limitations, the study has enabled hypothesis generation that diabetic macular ischaemia may be a contributory cause for deterioration of VA and that treated PDR may not be a stable disease in some individuals.