Article Text
Abstract
To determine whether the recommended screening interval for diabetic retinopathy (DR) in the UK can safely be extended beyond 1 year. Systematic review of clinical and cost-effectiveness studies. Nine databases were searched with no date restrictions. Randomised controlled trials (RCTs), cohort studies, prognostic or economic modelling studies which described the incidence and progression of DR in populations with type 1 diabetes mellitus or type 2 diabetes mellitus of either sex and of any age reporting incidence and progression of DR in relation to screening interval (vs annual screening interval) and/or prognostic factors were included. Narrative synthesis was undertaken. 14 013 papers were identified, of which 11 observational studies, 5 risk stratification modelling studies and 9 economic studies were included. Data were available for 262 541 patients of whom at least 228 649 (87%) had type 2 diabetes. There were no RCTs. Studies concluded that there is little difference between clinical outcomes from screening 1 yearly or 2 yearly in low-risk patients. However there was high loss to follow-up (13–31%), heterogeneity in definitions of low risk and variation in screening and grading protocols for prior retinopathy results. Observational and economic modelling studies in low-risk patients show little difference in clinical outcomes between 1-year and 2-year screening intervals. The lack of experimental research designs and heterogeneity in definition of low risk considerably limits the reliability and validity of this conclusion. Cost-effectiveness findings were mixed. There is insufficient evidence to recommend a move to extend the screening interval beyond 1 year.
- Diabetic Retinopathy
- Mass Screening
- Systematic Review
- Diabetes Mellitus
- Screening Interval
This is an Open Access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/
Statistics from Altmetric.com
Introduction
Diabetic retinopathy (DR) is a serious complication of diabetes mellitus and a major cause of visual loss globally.1 ,2 Prevalence of DR is rising.2 Early detection and timely treatment of sight-threatening DR have reduced the incidence and progression of visual loss.3–8 Screening for DR (using ophthalmoscopy and fundus photography) is accurate, safe and cost-effective.9–11
DR screening is recommended in many countries.12–16 However there is often no complete register of patients and non-uniformity of interval, coverage, uptake, screening methods and grading.17 ,18 The generalisablity of findings from one country to another has been questioned.19 There has been considerable international debate about extending the screening interval in the UK and the USA. A previous systematic review20 advocated a longer screening interval as a cost saving measure with various caveats but the authors did not formally critically appraise the evidence base, and two studies have since been published.
This systematic review aims to investigate the effects of longer screening intervals (vs an annual screening interval) in people with diabetes in order to inform screening decisions in the UK.
Methods
Study eligibility criteria
Randomised controlled trials (RCTs), cohort studies, prognostic or economic modelling studies which described the incidence and progression of DR in populations with type 1 diabetes mellitus (T1DM) or type 2 diabetes mellitus (T2DM) of either sex and of any age investigating annual with longer intervals were eligible (articles where annual screening is compared with a shorter interval were also included, because for some patient groups this interval may have to be more frequent). Studies investigating DR but not related to screening intervals, full text non-English publications, editorials, letters or commentaries were excluded.
Search strategy
Electronic searches were performed on 1 October 2013 in nine databases including Medline, SCOPUS, EMBASE, Cochrane Database of Systematic Reviews, Cochrane Central Register of Controlled Trials, Cochrane Methodology Register, Database of Abstracts of Reviews of Effects, Health Technology Assessment Database, and the National Health Service (NHS) Economic Evaluation Database. No date or language restrictions were applied. Search terms were left deliberately broad. Reference lists of relevant systematic reviews were checked. Online supplementary appendix 1 gives details of the search strategy.
Study selection
Abstracts from the different databases were merged and duplicates removed. Two reviewers independently screened abstracts using prepiloted eligibility criteria and also independently reviewed the identified full texts. Differences were discussed and agreed with the input of a third adjudicator where necessary.
Data extraction and assessment of methodological quality
Two independent reviewers extracted data including: study characteristics, population characteristics, screening data, postscreening data, outcome measures and conclusions. Economic appraisals data were extracted and assessed using Consolidated Health Economic Evaluation Reporting Standards statement21 and an adapted checklist for economic models was also used.22 Observational studies were assessed using the Critical Appraisal Skills Programme tool.23 Risk prediction studies were evaluated, but focused on external validation and impact on outcomes.24 Disagreements were resolved through discussion with a third reviewer.
Data synthesis and analysis
Evidence was synthesised narratively, because of heterogeneity in populations, screening intervals, measurement methods, outcomes and uptake. Evidence on incidence and/or progression of DR was synthesised separately in relation to (A) screening intervals and (B) risk factors.
Results
Fourteen thousand and thirteen records were identified after duplicates were removed and 142 were screened at full text level. Of these 113 were excluded, leaving 29 eligible records, reporting on 26 unique studies (figure 1).1 ,6 ,9 ,25–47 One study was reported in two publications: one reporting prevalence of DR in relation to screening intervals25 and the other incidence and progression of DR in relation to prognostic factors.26 A study by Mehlsen et al was reported in two publications.27 ,28 Analyses from two other publications were based on the same study population.29 ,30 To avoid a potential for double-counting, the abovementioned six publications were considered as three unique studies. One additional study was identified from searching of reference lists.31
Incidence and progression of DR
Incidence and/or progression of DR was reported in 15 studies.1 ,25–40 There were no randomised or non-randomised comparative studies. Ten observational studies followed progression of retinopathy in single populations with defined screening intervals.1 ,25 ,26 ,29–36 Five studies described risk stratification algorithms. None were externally validated.27 ,28 ,37–40
Most observational studies used annual screening protocols25 ,26 ,34–36 although actual screening intervals varied. Two studies described screening every 3 years31 ,33 and one every 2 years.30 Populations studied were from UK,1 ,25 ,26 ,33 ,35 ,36 Iceland,29 ,30 Australia34 and Sweden.32 Different strategies for grading retinopathy were employed (see table 1).
The Critical Appraisal Skills Programme tool (see online supplementary appendix 2) indicated that observational studies reported well. Completeness of follow-up varied, with attrition reaching up to 31% in two studies.25 ,36 Three studies had lower losses to follow-up (under 25%32 ,29 ,35), and reported that baseline characteristics did not differ between responders/completers and those lost to follow-up. In one study1 for which the rate of non-attendance for a repeat screen was relatively low (13%), the authors reported that non-attendees tended to be older with a longer duration of diabetes in comparison with attendees. Completeness of follow-up data was not reported for four studies.26 ,30 ,33 ,34
Incidence and/or progression of DR in relation to screening intervals was reported in six observational studies (see online supplementary appendix 3).25 ,26 ,28 ,34–37 ,40 All six studies suggested that the annual screening interval for patients with T1DM/T2DM at low risk (ie, no visible DR at baseline, adequate diabetes control) could be safely extended to 2 years or beyond. In three studies the incidence and progression of DR, either actual or predicted, were similar between annual and 2-yearly screening.25 ,28 ,34 Younis et al,35 ,36 found that about 95% of patients with T1DM-T2DM with no DR at initial screening remained free of sight threatening diabetic retinopathy (STDR) for the mean screening interval of 5.4 years. In a large single arm study in more than 20 000 patients, no significant association between a longer screening interval (18–24 months vs 12–18 months) and prevalence of referable DR (OR=0.93, 95% CI 0.82 to 1.05) or STDR/maculopathy (OR=1.05, 95% CI 0.72 to 1.52) was found.25 In the largest study, of 155 000 patients,40 predicted risk of progression in patients with T2DM with no DR at two consecutive screens for the 2-year screening interval was 0.15% or less compared with 0.03% or less for the annual interval. Studies which investigated 2-year screening intervals for patient subgroups at increased risk of progression (eg, visible DR of any grade, poorly controlled diabetes, duration of DM >10 years), suggested that these subgroups may need to be screened more frequently.
Incidence and/or progression of DR in relation to prognostic factors were reported in 13 studies (see online supplementary appendix 3).1 ,25–,27 ,29–,36 ,38–,40 Most studies showed that patients with baseline DR (vs no DR) were at higher risk of progressing to referable DR or STDR.26 ,33 ,35–37 ,39 ,40 In one study of patients with T2DM,26 the presence of baseline background DR (vs no background DR) was associated with an increased risk of pre-DR or proliferative DR (HR=5.00, 95% CI 4.40 to 5.60). Mehlsen et al,27 also found the number of retinal haemorrhages to be a significant risk factor for progression to STDR in patients with T1DM (OR=2.68, 95% CI 1.83 to 3.91) and T2DM (OR=2.37, 95% CI 1.84 to 3.06). Grade of baseline DR was also a significant predictor of progression to referable DR or STDR.33 ,35–37 Clinical risk factors for progression to STDR were found to be longer with duration of diabetes,1 ,26 ,27 ,34 ,35 ,37 ,38 ,40 insulin use,1 ,26 ,36 ,38 higher level of HbA1c,27 ,30 ,34 ,37 ,38 raised systolic blood pressure37 ,38 and hypertension treatment.26
Six studies27 ,28 ,37–40 describe risk stratification models to determine the most effective screening interval based on identified risk factors. These were developed using either Cox proportional hazards models, non-parametric survival models or logistic regression with data from different screening programmes.27 ,28 ,37–40 None of the models were externally validated on a UK data set, although one model37 was externally validated on a Danish database. All other models were internally validated tending to overestimate performance in comparison with external validation, due to statistical overfitting and potential lack of generalisability.48 The Danish screening programme model27 ,28 indicated that screening could be prolonged 2.9 times for patients with T1DM and 1.2 times for patients with T2DM without increasing STDR. The Aspelund risk prediction algorithm37 when externally validated on the same Danish data set could reduce the number of visits by 59% without increasing overall risk using the predictors of disease type (1 or 2), HbA1c, systolic blood pressure, gender and presence of non-proliferative DR. The current Danish screening system has screening intervals dependent on risk factors, so generalisability to the UK is problematic. Risk stratification based on two consecutive negative screens had the potential to reduce the number of screening visits in the Scottish screening service by up to 40%.40 However, probability of referable retinopathy would be higher with a 2 year interval, increasing from 0.7% to 1.2% in the English model (where screening includes mydriasis and two fields per eye)39 and from 0.05% to 0.25% in men with 10 years duration of diabetes in the Scottish model (where screening includes one field per eye).40 These rates were higher when any referable disease was included.
Economic evaluation of screening intervals
Nine papers assessed cost-effectiveness of differing screening intervals for identification of DR/prevention of DR complications.6 ,9 ,41–47 Studies used various types of simulation models conducted on hypothetical cohorts of patients with diabetes using data from existing data sets and literature.
Table 2 highlights key characteristics, methods and findings. Not all studies reported a viewpoint. The most common outcomes were sight years saved/gained, and two studies used quality-adjusted life years.46 ,47 Two studies did not discount.42 ,44 Four studies did not provide an incremental cost-effectiveness ratio.
Most items in the Consolidated Health Economic Evaluation Reporting Standards checklist21 were well reported (see online supplementary appendix 4). Using the Phillips24 checklist, publications reported a clear statement of the decision problem, objectives, data sources, model, methods and assumptions. They also reported costs including sources, and most also adequately compared results with previous models (see online supplementary appendix 5).
In general, the cost-effectiveness studies suggested that screening every 2 years could be safely adopted for those with no background or mild retinopathy,6 ,42 ,46 ,47 without increased risk of visual loss, and this reduces screening costs by ∼25%,42 and has no detrimental effect on years of sight saved.6 When taking into account the ability to detect other eye conditions screening every 2 years remained cost-effective.46 For those who already have retinopathy a greater degree of surveillance would be required, but results were mixed, depending on models used and assumptions made. Once retinopathy was detected, the screening interval should be 6 months;44 ,45 in contrast, two studies concluded that annual screening should be used;9 ,43 and the final study found that the most cost-effective option was to carry out two screening tests at 30-month intervals.41
Discussion
Principal findings
Studies broadly supported extending screening intervals beyond 1 year for patients with T2DM at low risk of progression to DR, such as those with well controlled diabetes on dietary treatment, with low HbA1c and no background retinopathy. Inevitably the lack of comparators in the evidence base introduces significant bias including selection bias and attrition bias—with those at least risk most likely to participate and those at most risk most likely to drop out. Both of these problems bias findings, that is, screening every 2 years will appear to be more beneficial and less risky than in fact it is. Other problems in the studies identified included heterogeneity in screening methods, in grading protocols, in defining risk, and uptake. Cost-effectiveness findings were conflicting. Three studies9 ,44 ,43 concluded that annual screening remains the most cost-effective. Risk stratification models show promise in providing tailored screening intervals based on individual risk factors, but none have been externally validated.
Strengths and weaknesses
This review is systematic with a priori study eligibility criteria and rigorous methods in selection, extraction, quality appraisal and synthesis of evidence. We were unable to pool evidence due to heterogeneity, and to assess publication bias. A thorough quality assessment was undertaken using recognised checklists.
None of the included studies allowed for direct comparison of different screening intervals in relation to the incidence of retinopathy or vision loss. Economic analyses were based on hypothetical simulation models. Strengths of the evidence base include: good reporting, real-world setting,1 ,26 ,32 ,34–36 large sample size,1 ,26 ,39 ,40 adequate methods of participant recruitment and sufficient follow-up.26 ,27 ,30 ,33 ,37 ,42 There were some notable limitations: high attrition,1 ,25 ,35 ,36 systematic differences between attendees and non-attendees and substantial heterogeneity between studies, making it difficult to compare results with regard to the occurrence or progression of retinopathy. Therefore patients with ‘no existing background retinopathy’ should be interpreted as ‘patients in whom no evidence for background retinopathy has been found’.18 There were also difficulties in measuring other risk factors such as duration of diabetes. There may be limited applicability to adults with T1DM as only a small proportion of the studies covered this group. While we did not find sufficiently robust evidence to suggest that the screening interval could safely be extended beyond 1 year, it should be noted that equally we did not find persuasive evidence that it should not be extended. Only one risk prediction algorithm was externally validated and this was on a Danish data set37 where there was a large amount of missing data, and where screening intervals are already stratified by risk so not applicable to a system such as the UK where uniform screening intervals currently pertain.
Cost-effectiveness models were of considerable complexity and included various inputs such as: progression rates between disease stages, interval between screening visits, compliance and sensitivity, and specificity of testing. In general, the models assumed equal treatment success irrespective of screening interval; most also assumed the same compliance rate and uncertainty in patient behaviour and compliance were not adequately included (eg, differential compliance with different screening intervals49). Individual patient characteristics which potentially determine optimal screening interval and the practicalities of providing individualised screening intervals, were not included. Most studies did include benefits of detecting other eye disorders. The clinical outcomes and methodologies of the models were heterogeneous and precluded meaningful synthesis in meta-analysis. Some studies did not assign utility scores to differing degrees of sight loss and models used averaged progression rates obtained from studies such as the UK Prospective Diabetes Study and the Wisconsin Epidemiologic Study of Diabetic Retinopathy, in which most patients do not progress quickly. Discrete Event Simulation models use such averages which are applied to a variety of patients whose risk of progression depends on their baseline characteristics. If this is the case, it may mean that individuals who progress at the fastest rate may not be effectively captured within the model; it is these individuals who would be most likely to influence any difference in outcomes between programmes. Three studies6 ,43 ,45 are over 20 years old, and their validity is questionable in light of medical advances, changing prevalence and costs.
Comparison of findings with other reviews
Three previous systematic reviews have been published.20 ,49 ,50 The most recent of these20 did not identify six of the papers we have presented in this paper including two of the largest cohorts with 14 55439 and 155 11440 patients with diabetes. For one study43 authors have misinterpreted biannual (every 6 months) as biennial (every 24 months). Echouffo-Tcheugui20 concluded that the screening interval could safely be extended to 2 years. Jones et al49 had broad aims and examined all aspects of cost-effectiveness of DR screening. The Wessex Institute report50 also reviewed cost-effectiveness concluding that the limited evidence base suggests that more patients may lose their sight with a 2 year interval.
Recommendations for future research
Further research is needed on:
how ‘low-risk’ patients should be identified,
how different screening and grading protocols affect performance, and
how extending the screening interval might affect uptake.
An RCT randomising either individual patients or whole screening centres to a longer interval would provide robust data upon which to base policy decisions and underpin a rigorous cost-effectiveness analysis
Risk stratification algorithms identified in this review showed considerable promise for optimising services and minimising costs, although the cost-effectiveness of such a strategy would need to be carefully considered, and risk algorithms validated.
Implications for clinicians and policymakers
Based on the strength of the evidence identified in this review, we cannot reliably predict the outcome of a change in screening intervals. While the invited interval might be 2 years, in practice with lower uptake the screening interval might extend well beyond this for some patients. Additionally, there would need to be a reliable and uniform method for identifying and recording risk of progression to STDR. Previous retinopathy screening results allow for risk stratification. This is a powerful predictor of risk, but is not currently measured consistently across the UK.18 Detection and treatment for diabetes has improved in recent years, with diabetes more likely to be diagnosed and treated. This in turn will reduce the risk of progression of DR.
There is a broader question about the level of evidence required to make changes to an existing screening programme. While the best evidence would be from RCTs or meta-analysis of several trials, this is not a practical or affordable approach for every decision on a screening programme. Most often after implementation, the screening programme will be extended, either in frequency or in the eligible population, a phenomenon known as ‘mission creep’. Here, we considered the minimum level of evidence to be a two arm randomised or non-randomised trial investigating the effect of a 1-year interval vs a 2-year interval on retinopathy rates and uptake, with appropriate cost-effectiveness analysis. This was not available in the literature.
Conclusion
Observational and economic modelling studies in low-risk patients show little difference in clinical outcomes between screening intervals of 1 year or 2 years. The lack of experimental research design and heterogeneity in definition of those at low risk limits the reliability and validity of this conclusion. Cost-effectiveness literature provides mixed results. While we did not find sufficiently robust evidence to suggest that the screening interval could safely be extended beyond 1 year, it should be noted that equally we did not find persuasive evidence that it should not be extended. However we consider that current evidence does not support a move to extend the screening interval beyond 1 year.
References
Supplementary materials
Supplementary Data
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
Files in this Data Supplement:
- Data supplement 1 - Online appendix 1
- Data supplement 2 - Online appendix 2
- Data supplement 3 - Online appendix 3
- Data supplement 4 - Online appendix 4
- Data supplement 5 - Online appendix 5
Footnotes
Contributors RL performed the searches. RL and DT conducted the screening, data extraction and quality assessment for the clinical papers. HM led the data extraction and quality assessment for the cost-effectiveness papers. ST-P reviewed the risk stratification algorithms and drafted the paper. All authors were involved in reviewing, commenting, and editing drafts of the manuscript and all approved of the final manuscript and take responsibility for the integrity of the data and the accuracy of the data analysis. ST-P is guarantor.
Funding This work was supported by The Four Nations diabetic retinopathy steering group on behalf of the UK National Screening Committee, part of Public Health England. The funders provided feedback on the protocol and the draft report, but all decisions regarding study design; collection, analysis, and interpretation of data; the writing of the report; and submission for publication were taken independently by the study team. ST-P is supported by a postdoctoral fellowship from the National Institute for Health Research (NIHR). AC is supported by the NIHR Collaboration for Leadership in Applied Health Research and Care West Midlands at University Hospitals Birmingham NHS Foundation Trust. The views expressed are those of the authors and not necessarily those of the NHS, the NIHR or the Department of Health.
Competing interests None.
Provenance and peer review Not commissioned; externally peer reviewed.