OBJECTIVE

To compare four screening strategies for diabetic macular edema (DME).

RESEARCH DESIGN AND METHODS

Patients attending diabetic retinopathy screening were recruited and received macular optical coherence tomography (OCT), in addition to visual acuity (VA) and fundus photography (FP) assessments, as part of the standard protocol. Two retina specialists provided the reference grading by independently assessing each subject’s screened data for DME. The current standard protocol (strategy A) was compared for sensitivity, specificity, quality-adjusted life-year (QALY) gained, and incremental cost-effectiveness ratio (ICER) with three alternative candidate protocols using a simulation model with the same subjects. In strategy B, macular hemorrhage or microaneurysm on FP were removed as surrogate markers for possible DME. Strategy C used best-corrected instead of habitual/pinhole VA and added central subfield thickness (CST) >290 μm on OCT in suspected cases as a confirmation marker for possible DME. Strategy D used CST >290 μm OCT in all subjects as a surrogate marker for suspected DME.

RESULTS

We recruited 2,277 subjects (mean age 62.80 ± 11.75 years, 43.7% male). The sensitivities and specificities were 40.95% and 86.60%, 22.86% and 95.63%, 32.38% and 100%, and 74.47% and 98.34% for strategies A, B, C, and D, respectively. The costs (in U.S. dollars) of each QALY gained for strategies A, B, C, and D were $7,447.50,$8,428.70, $5,992.30, and$4,113.50, respectively.

CONCLUSIONS

The high false-positive rate of the current protocol generates unnecessary referrals, which are inconvenient for patients and costly for society. Incorporating universal OCT for screening DME can reduce false-positive results by eightfold, while improving sensitivity and long-term cost-effectiveness.

Diabetic retinopathy (DR) is a common cause of irreversible blindness, and its incidence increases with the duration of diabetes (1). Diabetic macular edema (DME) and proliferative diabetic retinopathy (PDR) are the two major causes of visual loss in DR (2). Timely treatment for DME, such as intravitreal injections of anti-vascular endothelial growth factor (anti-VEGF), can help prevent visual loss (3,4) and be cost-effective, because early cases with less affected vision are associated with lower treatment costs (5). For the community, detecting early DR is more cost-effective than managing its late complications (6). Screening of patients with diabetes allows for early detection and treatment of DR (including DME), when patients are often asymptomatic or only mildly symptomatic, and can help reduce the financial burden associated with managing advanced diseases (7,8). DR screening based on fundus photography (FP) is widely used, including U.K. and Hong Kong (6,9). Currently, all public patients with diabetes managed by the General Out-Patient Clinics in Hong Kong receive regular (6–18 months, depending on risk factors) FP, and the photographs are graded according to a standard protocol adopted from the U.K. National Health Service (10). Patients with sight-threatening DR, defined as having their worse eye graded as preproliferative DR (PPDR) or PDR (R2, R2.5, R3), maculopathy (M1), or ungradable (U) on FP, are referred to the ophthalmology Specialist Out-Patient Clinic (Eye SOPC) for further management (6).

We previously reported a DR prevalence of 39% (n = 68,067) and maculopathy (M1) prevalence of 8.6% (n = 15,010) among 174,532 patients with diabetes screened over a period of 3.5 years in our local screening program (6). However, detecting DME reliably with FP remains challenging, because DME-associated macular thickening may not be visually apparent, even with stereo FP (6). As a result of this limitation of FP, standardized surrogate markers (macular exudates, hemorrhages, and/or aneurysms) are used to indicate possible DME (11). Figure 1A shows how these markers were identified in a FP to give an M1 grading, although DME would only be confirmed or excluded after referral to ophthalmologists. In addition to FP, impaired visual acuity (VA) is another criterion used for an M1 grading. However, because some patients with refractive errors may not be using corrective prescription glasses during DR screening with the current standard protocol, some unnecessary referrals may occur from underestimation of VA in these patients.

Figure 1

A typical case of DME. A: Typical color FP centered on the macula in a patient with diabetes demonstrating retinal hemorrhage (red triangles) and exudates (white triangles). B: Typical cross-sectional image of the macula showing the center foveal depression as well as the details of the different layers of the retina. C: Macula of a patient with diabetes demonstrating typical features of DME, including loss of the foveal contour, thickening, and intraretinal cystoid edema.

Figure 1

A typical case of DME. A: Typical color FP centered on the macula in a patient with diabetes demonstrating retinal hemorrhage (red triangles) and exudates (white triangles). B: Typical cross-sectional image of the macula showing the center foveal depression as well as the details of the different layers of the retina. C: Macula of a patient with diabetes demonstrating typical features of DME, including loss of the foveal contour, thickening, and intraretinal cystoid edema.

Close modal

While the inclusiveness of the current grading protocol for maculopathy (M1) may help improve the sensitivity of screening for DME, it can also decrease the specificity and result in many false positives or unnecessary referrals. These referrals carry downstream costs in specialist consultations and specialized investigations such as fluorescein angiography and optical coherence tomography (OCT). The resulting total cost incurred on the health care system can be substantial (12).

In recent years, retinal OCT has been developed to produce accurate and quantitative information on the cross-sectional thickness of the macula. It uses laser interferometry to collect data along the thickness of the macula, which are then used to reconstruct a cross-sectional image of the macula with resolution in the order of microns (13,14). Pharmacologic dilation of the pupil for the OCT examination is unnecessary, and the entire process is painless, fast (within minutes), safe, noninvasive, noncontact, and without radiation exposure (13,14). OCT is reliable and repeatable in identifying macular thickening in patients with various macular diseases (15) and has become the gold standard in diagnosing DME (16,17), where it has been reported as having higher accuracy than FP (18). Figure 1 shows a typical cross-sectional image of a normal macula (Fig. 1B) and that of a patient with DME (Fig. 1C).

In a previous report on cases graded as M1 on FP, up to 42.1% did not have significant macular edema on subsequent OCT (19). This highlights the fact that the surrogate markers of the current grading protocol (exudates, hemorrhages, aneurysms) may have poor correlation with the presence of DME. We believe that a better screening strategy for detecting DME can be found by comparing strategies using different grading protocols, including some that incorporate the use of OCT as an additional screening tool.

### Aims and Objectives

The aim of this study was to compare the performance and cost-effectiveness of different DME screening strategies with varying modifications to the current standard screening protocol, including: removal of macular hemorrhages or aneurysms as surrogate markers; ensuring all refractive errors are corrected for VA assessment; and incorporating OCT scans on selected, or all cases.

### Subjects

Ethical approval was obtained from the University of Hong Kong/Hospital Authority Hong Kong West Cluster Institutional Review Board for this cross-sectional, observational study. All subjects were patients with diabetes recruited from the Diabetic Complications Screening Program of the Hospital Authority’s Hong Kong West Cluster, from 1 February 2014 to 31 January 2016. At each screening session (conducted once per working day), screening-naive patients were offered the opportunity to join the study as subjects and receive an OCT scan, in addition to the standard VA testing and FP. To avoid selection bias, we only recruited the first five consecutive, consenting subjects for each session. Subjects with ungradable FPs or OCT scans, due to poor image quality, were excluded from further analysis.

### Four Screening Strategies

All subjects underwent the same set of screening procedures, but only a preset portion of the obtained clinical data were used by each of the four strategies for determining the maculopathy grading (maculopathy not present = M0, maculopathy present = M1). Strategy A is the current standard screening protocol, which is based on the protocol used by the U.K. National Health Service (10), with grading based on habitual/pinhole VA and the presence of surrogate markers (macular exudate, hemorrhage, or microaneurysm) on FP. Strategy A was used as the control and base model for the other strategies, which incorporated various modifications; strategy B removed macular hemorrhage or microaneurysm as surrogate markers for M1; strategy C used the best-corrected VA instead of habitual/pinhole VA, where any preexisting refractive errors were corrected before VA testing, and also used macular OCT for suspected M1 on FP for final confirmation; whereas strategy D applied universal macular OCT as the sole determinant of M1 in all cases.

### Screening Procedures

The screening procedures have been reported previously (6). Toward the end of the screening visit, a macular volume scan using OCT (Cirrus HD OCT 4000; Carl Zeiss Meditec, Dublin, CA) was performed on all subjects.

### Definition of Maculopathy (M1) on OCT

In DME, the foveal thickness (also known as central subfield thickness) is increased due to the presence of intraretinal edema, which can be readily detected with macular OCT. We used central subfield thickness (CST) ≥290 µm on OCT as a grading criteria for M1 in strategy C and strategy D, based on published data and the model of OCT machine used (4,20).

### Definitive Diagnosis of DME

To determine the reference standard for comparing the performance of each strategy, the definitive diagnosis of DME was made by the independent assessment of all screening data for each subject by two of the investigators, who are experienced retina specialists (I.Y.H.W., R.L.M.W.). For any diagnostic discrepancy, the specific case would be jointly reviewed by the two investigators again, for a final consensus.

### Model Structure

We used the current screening protocol (strategy A) as a benchmark for comparison with the other three screening strategies. A model was formulated to simulate the current screening practice regarding diabetic maculopathy: M0 gradings are not referred, while M1 gradings are referred to the Eye SOPC. Figure 2 illustrates the workflow. Patients with DME confirmed by ophthalmologists at the Eye SOPC would be offered treatment, which includes laser therapy (macular focal or grid laser photocoagulation), or intravitreal anti-VEGF injections. The model would estimate the costs for each subject in the first 12 months after the initial Eye SOPC assessment. Because local data regarding the probabilities of subsequent treatment or procedures are not available, we used best estimates based on our private communications with other local retina experts, medical record review of past DR patients managed at our Eye SOPCs, and using references from local and overseas prevalence studies on DME (4,6,21). On average, we estimated subjects would require an Eye SOPC assessment every 2 months during the first 12 months, with an estimated 10% default rate. An estimated 50% of subjects would receive laser therapy, and 25% would receive initial anti-VEGF treatment with three injections (as usual practice). We estimated that laser therapy would be repeated in 75% of subjects every 4 months for persistent DME; whereas for anti-VEGF treatment, 50% would receive repeated injection after the first three injections, until a total of six injections, on average, have been given in the first 12 months.

Figure 2

Flowchart of subjects entering the screening protocol with the respective probabilities of various outcomes. PPV, pars plana vitrectomy; SOPD, Specialist Out-Patient Clinic.

Figure 2

Flowchart of subjects entering the screening protocol with the respective probabilities of various outcomes. PPV, pars plana vitrectomy; SOPD, Specialist Out-Patient Clinic.

Close modal

### Evaluation of Strategies

The performance of the four strategies was compared using various indices of effectiveness, including sensitivities, specificities, positive predictive value (PPV), and negative predictive value (NPV). The health economics of each strategy was also evaluated using quality-adjusted life-years (QALYs) gained. In the Global Burden of Diseases Study, there is currently no specific disability weight (DW) for DME (2225). We used the DW for diabetes, which was 0.015 for patients with diabetes without any related complications, including DR, and 0.552 for DR-related blindness (24). This was then calculated and multiplied by the number of true positives (confirmed DME) referred to Eye SOPC for each strategy.

### Costs

We estimated the per-person provider cost of 1) the screening program, 2) SOPC assessment by an ophthalmologist, and 3) treatment cost of DME up to 1 year after the initial SOPC assessment. We used an ingredient-costing approach, whereby the estimated costs (or exact costs, when available) were collected on the quantities of the different resources used. We included all costs, whether they were a direct financial cost to the program or an indirect cost, such as the opportunity cost associated with surgical intervention, such as pars plana vitrectomy, for example. Capital item costs were calculated on the assumption that each has a 5-year life span with annual maintenance costs included. Staff costs (optometrists and ophthalmologists) were calculated based on the average time spent on each subject and the corresponding portion of that staff’s salary. Detailed costs were estimated from private communications with program managers and local experts from various hospitals across Hong Kong, as well as published data from the web page of the Hospital Authority (26) and the Government of the Hong Kong Special Administrative Region (27). The costs of the resources used were converted into monetary terms in Hong Kong dollars (HKD) and then converted into U.S. dollars (USD), based on the rate of USD 1.00 = HKD 7.75 (prevailing rate as of August 2016).

### Cost-effectiveness Analysis

Cost-effectiveness ratios (CERs) were calculated using the QALYs gained and the estimated total program costs. Incremental cost-effectiveness ratios (ICERs) were calculated as the cost difference between the control (strategy A) and the alternative candidate strategy (strategy B, C, or D), divided by the difference in QALYs gained. The different strategies were assessed for cost-effectiveness by comparing their cost per QALY gained with a standardized threshold. Because there is no Hong Kong–specific threshold, two standards of willingness-to-pay were used to assess the outcomes:

1. We used USD 50,000/QALY gained as an arbitrary threshold, with <USD 50,000/QALY gained defined as cost-effective.

2. As recommended by the World Health Organization (28), we also used the gross domestic product per capita (GPD) of Hong Kong in 2014 (HKD 310,113, or USD 39,963) as a threshold, such that strategies costing (per QALY gained): 1) less than the GDP are considered “very cost-effective”; 2) between one and three times the GDP are considered “cost-effective”; while 3) more than three times the GDP are considered “not cost-effective” (29).

### Safety Assurance

All subjects with M1 grading were referred to Eye SOPC, as per current standard of care. Subjects initially screened as M0, but subsequently assessed to have DME by our investigators, were also referred to the Eye SOPC.

### Demographics and Performance Indices

A total of 2,277 subjects were recruited, with a mean age of 62.80 ± 11.75 years, and 996 (43.7%) were men. DME, as diagnosed by our two independent retina specialists, was present in 105 subjects, giving a prevalence of 4.61%. The prevalence of maculopathy (M1), as graded by the different strategies, varied widely from 1.49% (strategy C) to 14.67% (strategy A), depending on the sensitivity and specificity of the specific screening protocol. The mean VA and CST of eyes with M0 or M1 grading for each strategy is reported in Table 1. There was no diagnostic discrepancy between the two independent retina specialists. The sensitivities, specificities, PPV, and NPV for each strategy are reported in Table 2. Apart from strategy D, none of the other screening strategies have sensitivity >50% for the detection of DME. Strategy D had 100% sensitivity because no subjects with DME had CST <290 µm on OCT. There were 36 subjects (1.7%) who had CST >290 μm (range 292–322 μm) but no DME after independent assessment of their OCT and FP by the two reviewing investigators. The PPVs were much higher when OCT was incorporated into the screening protocol (strategies C and D), although the NPVs remained mostly unchanged due to the low overall prevalence of DME in our subjects, as expected from screening asymptomatic patients. Although strategy C had 100% specificity and PPV, it also had a lower sensitivity than the current protocol (strategy A), being <33%.

Table 1

Level of VA and CST on OCT scan of the macula of subjects when screened with the various strategies

Mean logMAR VAMean CST (µm)
Normal*DME positive*Normal*DME positive*
Strategy A
M0 0.1937 (n = 1,881) 0.1883 (n = 62) 247.31 (n = 1,881) 300.36 (n = 62)
M1 0.2249 (n = 291) 0.3189 (n = 43) 247.25 (n = 291) 346.88 (n = 43)
Strategy B
M0 0.1983 (n = 2,077) 0.2530 (n = 81) 246.83 (n = 2,077) 323.62 (n = 81)
M1 0.2262 (n = 95) 0.3252 (n = 24) 250.90 (n = 95) 348.78 (n = 24)
Strategy C
M0 0.2015 (n = 2,172) 0.1901 (n = 71) 247.30 (n = 2,172) 299.56 (n = 71)
M1 N/A (n = 0) 0.3189 (n = 34) N/A (n = 0) 346.88 (n = 34)
Strategy D
M0 0.2015 (n = 2,172) N/A (n = 0) 247.30 (n = 2,172) N/A (n = 0)
M1 N/A (n = 0) 0.2813 (n = 105) N/A (n = 0) 333.48 (n = 105)
Mean logMAR VAMean CST (µm)
Normal*DME positive*Normal*DME positive*
Strategy A
M0 0.1937 (n = 1,881) 0.1883 (n = 62) 247.31 (n = 1,881) 300.36 (n = 62)
M1 0.2249 (n = 291) 0.3189 (n = 43) 247.25 (n = 291) 346.88 (n = 43)
Strategy B
M0 0.1983 (n = 2,077) 0.2530 (n = 81) 246.83 (n = 2,077) 323.62 (n = 81)
M1 0.2262 (n = 95) 0.3252 (n = 24) 250.90 (n = 95) 348.78 (n = 24)
Strategy C
M0 0.2015 (n = 2,172) 0.1901 (n = 71) 247.30 (n = 2,172) 299.56 (n = 71)
M1 N/A (n = 0) 0.3189 (n = 34) N/A (n = 0) 346.88 (n = 34)
Strategy D
M0 0.2015 (n = 2,172) N/A (n = 0) 247.30 (n = 2,172) N/A (n = 0)
M1 N/A (n = 0) 0.2813 (n = 105) N/A (n = 0) 333.48 (n = 105)

MAR, minimum angle of resolution; N/A, not applicable.

*

Based on result from reference standard as determined by two independent qualified ophthalmologists.

Table 2

Screening result using the four strategies

Strategy AStrategy BStrategy CStrategy D^
Normal# (disease not present)DME# (disease present)TotalNormal# (disease not present)DME# (disease present)TotalNormal# (disease not present)DME# (disease present)TotalNormal# (disease not present)DME# (disease present)Total
M0 (test negative), n 1,881 62 1,943 2,077 81 2,158 2,172 71 2,243 2,136 2,136
M1 (test positive), n 291 43 334 95 24 119 34 34 36 105 141
Total, N 2,172 105 2,277 2,172 105 2,277 2,172 105 2,277 2,172 105 2,277
Sensitivity, % 40.95 22.86 32.38 100
Specificity, % 86.60 95.63 100.00 98.34
PPV, % 12.87 20.17 100.00 74.47
NPV, % 96.81 96.25 96.83 100
Strategy AStrategy BStrategy CStrategy D^
Normal# (disease not present)DME# (disease present)TotalNormal# (disease not present)DME# (disease present)TotalNormal# (disease not present)DME# (disease present)TotalNormal# (disease not present)DME# (disease present)Total
M0 (test negative), n 1,881 62 1,943 2,077 81 2,158 2,172 71 2,243 2,136 2,136
M1 (test positive), n 291 43 334 95 24 119 34 34 36 105 141
Total, N 2,172 105 2,277 2,172 105 2,277 2,172 105 2,277 2,172 105 2,277
Sensitivity, % 40.95 22.86 32.38 100
Specificity, % 86.60 95.63 100.00 98.34
PPV, % 12.87 20.17 100.00 74.47
NPV, % 96.81 96.25 96.83 100

Column: Grading by the different screening strategies. Row: Maculopathy status by OCT.

#

Reference standard taken from independent reviewers’ result. ^Results in strategy D were made based purely on the foveal thickness as measured by OCT.

### Cost-effectiveness Analysis

In calculating the QALYs gained for each subject with DME who was correctly identified (as M1) at screening and referred to Eye SOPC, we assumed that the condition would be controlled by 12 months and that vision would return to premorbid level (4). Using methodologies as previously described (3032), we calculated that the QALYs gained per subject were 0.45. This was then multiplied by the number of correctly referred subjects (true positives) for each of the four strategies to obtain their respective total QALYs gained. The total QALYs gained, estimated monetary costs, and ICERs for each strategy are given in Table 3.

Table 3

Cost-effectiveness analysis of the four screening strategies

QALY gainedTotal cost (HKD $)Unit cost (HKD$)Total cost (USD $)Cost (USD$)/QALY gainedUnit cost (USD $)Incremental QALYIncremental cost (USD$)ICER (USD $)* 19.4983 1,125,408 494.25 145,213.90 7,447.50 63.78 — — — 10.8828 710,893 312.21 91,728.10 8,428.70 40.29 −8.62 −53,485.80 6,208.10 15.4173 715,986 314.44 92,385.30 5,992.30 40.57 −4.08 −52,828.70 12,944.90 47.6123 1,517,867 666.61 195,853.80 4,113.50 86.01 28.11 50,639.90 1,801.20 QALY gainedTotal cost (HKD$)Unit cost (HKD $)Total cost (USD$)Cost (USD $)/QALY gainedUnit cost (USD$)Incremental QALYIncremental cost (USD $)ICER (USD$)*
19.4983 1,125,408 494.25 145,213.90 7,447.50 63.78 — — —
10.8828 710,893 312.21 91,728.10 8,428.70 40.29 −8.62 −53,485.80 6,208.10
15.4173 715,986 314.44 92,385.30 5,992.30 40.57 −4.08 −52,828.70 12,944.90
47.6123 1,517,867 666.61 195,853.80 4,113.50 86.01 28.11 50,639.90 1,801.20

Costs estimates (per subject per visit) in HKDs: best-corrected VA assessment, $83.30; FP,$65.50; OCT scan of the macula, $208.30; ophthalmologist consultation at the Eye SOPC,$740; laser procedure, $1,500; anti-VEGF injection,$660; vitrectomy, $30,000; in-patient stay in the hospital authority,$3,290.

*

ICER was calculated using A as reference to each strategy.

The total and unit costs were highest for strategy D, but it also had the highest QALYs gained. Strategy B was the least costly, but QALYs gained with this strategy were also the lowest. Because strategy A was used as the control for comparison with the other three strategies, the order of ICER calculation was A, B, C, and D, which was ranked according to the estimated total cost required by each strategy, along with the cost per QALY gained (Table 3). The cost per QALY gained for strategy D was the lowest of the four strategies, and compared with the control (strategy A), its ICER was also the lowest.

### Estimation of Benefits of Screening

According to data published by the Government of the Hong Kong Special Administrative Region, GDP per capita of Hong Kong in 2014 was HKD 310,113 (USD 39,963). The costs per QALY gained were USD 7,447.50, USD 8,428.70, USD 5,992.30, and USD 4,113.50 for strategies A, B, C, and D, respectively. Qualitatively, all four strategies would be categorically considered as “very cost-effective,” whether using cost less than Hong Kong’s GDP or <USD 50,000 per QALY gained as criteria.

This study compared the current standard screening protocol (strategy A) with three alternative candidate strategies for DME screening performance, using a simulated model with the same pool of screening-naive patients with diabetes.

As we had suspected, the current protocol (strategy A) had the lowest specificity of the four strategies, with a high false-positive rate (13.4%), which would result in a substantial number of unnecessary referrals that are costly to both the health care system (specialist consultations, further investigations) and patients (travel cost, taking time off work, anxiety). Strategy A also had low sensitivity (40.95%) for a screening test, with 62 subjects with DME being falsely graded as M0 (false-negative rate 59.05%). Extrapolating this to our patients, this could mean that more than half of our patients with DME are not being detected (and appropriately referred) with the current screening protocol.

Strategy B was formulated after our pilot data had shown that macular hemorrhage/aneurysm was commonly responsible for false-positive results on FP. Although the specificity improved (95.6%) after removing this surrogate marker for DME, the sensitivity worsened and was the lowest of the four strategies (22.86%). It appears that some DME presents with only macular hemorrhage/aneurysm (without exudate), and removing this marker will result in a higher number of undetected cases, making this strategy unsuitable for DME screening.

The cost of OCT has reduced over the years and has become an indispensable part of standard retinal assessment (17). Although OCT machines may not be currently available at most DR screening sites, its commercial availability is high and has become an essential piece of equipment in most ophthalmology centers (17). In fact, in many developed countries, over-the-counter macular OCT scanning is already commercially available at many optometry practices. Also, macular OCT is the current gold standard in diagnosing DME (16,17), being able to quantify the amount of thickening as well as monitoring progress and response to treatment. However, if universal OCT scanning is incorporated into the current DR screening program, the initial capital costs and skill set requirements would invariably increase in the short term. However, this would enhance the long-term effectiveness of the program, as shown by the results of strategy D, which despite requiring the highest cost, also gained the highest amount of QALYs and was the most cost-effective in terms of cost per QALY gained among the four strategies we examined.

Because 95.4% of our subjects did not have DME, it can be argued that strategy D wasted resources by performing mostly unnecessary OCT on those without maculopathy. There are two reasons to justify this. Firstly, more than half (almost 60%) of the DME cases went undetected using the current protocol (strategy A). Performing selective OCT for only photography-graded M1 cases (strategy C) only improved unnecessary referrals by decreasing false-positive results but did not improved the sensitivity of the current protocol (strategy A) because no M0 cases will be screened with OCT, even though most DME cases in this study were missed (graded M0) when OCT was not used. When the most stringent criteria for M1 grading among all the strategies tested was used, strategy C had a higher false-negative rate (67.62%) than our control protocol (strategy A), with 71 subjects with DME being falsely graded as M0.

Secondly, using OCT selectively as in strategy C may not be practical and may delay SOPC referral for all DME cases. This is because all suspected M1 cases that have been first graded using VA and FP will then need further arrangement to be made for an additional OCT to confirm the presence of CST ≥290 µm before being finally referred to Eye SOPC.

One could also argue that VA testing and FP in strategy D could be removed to save costs. This argument is only valid if screening for DME is the sole purpose of DR screening. However, in DR screening, VA testing is useful for monitoring long-term visual stability of patients with diabetes. Many of these patients with diabetes can develop other ocular pathologies that may affect their VA, such as age-related macular degeneration, cataract, or retinal vascular occlusions. The presenting VA when these disorders occur will be useful when referring to an ophthalmologist for further management. As for FP, all DR screening strategies were designed to detect sight-threatening DR, which apart from DME, includes PPDR, which includes moderate and severe nonproliferative DR, and PDR. Although OCT is good at detecting macular pathologies, FP features of PPDR and PDR, such as intraretinal microvascular abnormalities or retinal neovascularization, cannot be reliably detected with current OCT technology. Therefore, even with universal OCT scanning, FP would still be necessary for DR screening in the foreseeable future.

### Limitations

There were several limitations. Because a specific DW for DME is not available, we have adopted the DW for DR, which includes PPDR and PDR, in addition to DME, in our analysis as the closest available estimate. This may have overestimated the benefit of OCT in terms of QALYs gained. In addition, given the lack of relevant published data, we made assumptions for cost estimates and probabilities of various treatments and outcomes, which may have introduced some unavoidable bias. However, we have attempted to minimize this bias by basing our assumptions on related published data whenever available, such as the prevalence of DR and DME (6,18,21,33), and supplemented with relevant input from local experts. Owing to worldwide differences in reimbursements and practice preferences, it is difficult to produce a universally adaptable set of probabilities. Real-world data (34) often do not fully replicate clinical trial results (35). This study was conducted in a developed economy, where prevalence of DR was similar to other developed economies (6,21,33). Therefore, the generalizability of our results is best applied to localities with a similar economic status and health care delivery system. Lastly, for a study on disease screening, the relatively small number of subjects may have limited the impact of certain outcomes. It would be ideal if the sampling size could be increased in future studies.

Our study highlighted the need for further research in areas that are likely to influence estimates of costs and effectiveness of DR screening programs. Firstly, more precise data are needed in the epidemiology of DR, and on DR and DME progression rates. Secondly, the costs and probabilities of treatment will need to be validated by, for example, conducting a survey of local experts and program managers and reviewing longitudinal patient records. Furthermore, the follow-up adherence rate by patients in such a screening program should be determined. Finally, a clearer definition of cost-effectiveness would better facilitate interpretation and comparison of economic evaluations. In this study, we had used both GDP per capita and USD 50,000 (per QALY gained) as references. Because Hong Kong ranked ∼17th globally in 2015 for GDP per capita (36), using this as reference may have overestimated the cost-effectiveness of many health-related interventions. If we use the GDP per capita of China in 2015 (USD 7,429.70) (36) as the reference, then strategies A and B would be categorized as “cost-effective,” whereas strategies C and D remain as “very cost-effective.”

### Conclusion

Four DME screening strategies were reviewed in terms of their respective performance and cost-effectiveness using a simulated model. Using a strategy incorporating universal OCT scanning in the screening protocol, we were able to achieve 100% sensitivity and >98% specificity and also superior cost-effectiveness than other strategies, including the current standard screening protocol. Further studies, when more precise data on costs are available, are warranted to confirm the cost-effectiveness of using universal OCT in DR screening.

I.Y.H.W. and R.L.M.W. are co-first authors.

Acknowledgments. The authors would like to acknowledge the Food and Health Bureau of the Government of the Hong Kong Special Administrative Region for granting support for this project.

Funding. This project was supported by the Health and Medical Research Fund reference number 11121801, under the Department of Health, Government of the Hong Kong Special Administrative Region.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. I.Y.H.W. contributed to discussion on the study concept, researched data, and wrote the manuscript. R.L.M.W. contributed to discussion on the study concept, researched data, and wrote the manuscript. J.C.H.C. reviewed and edited the manuscript. R.K. contributed to discussion and reviewed and edited the manuscript. V.C. researched data, contributed to discussion on the study concept, and reviewed and edited the manuscript. I.Y.H.W. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

1.
Wong
TY
,
Sabanayagam
C
.
The war on diabetic retinopathy: where are we now
?
Asia Pac J Ophthalmol (Phila)
2019
;
8
:
448
456
2.
Aiello
LM
.
Perspectives on diabetic retinopathy
.
Am J Ophthalmol
2003
;
136
:
122
135
3.
Wells
JA
,
Glassman
AR
,
Ayala
AR
, et al.;
Diabetic Retinopathy Clinical Research Network
.
Aflibercept, bevacizumab, or ranibizumab for diabetic macular edema: two-year results from a comparative effectiveness randomized clinical trial
.
Ophthalmology
2016
;
123
:
1351
1359
4.
Wells
JA
,
Glassman
AR
,
Ayala
AR
, et al.;
Diabetic Retinopathy Clinical Research Network
.
Aflibercept, bevacizumab, or ranibizumab for diabetic macular edema
.
N Engl J Med
2015
;
372
:
1193
1203
5.
Ross
EL
,
Hutton
DW
,
Stein
JD
,
Bressler
NM
,
Jampol
LM
,
Glassman
AR
;
Diabetic Retinopathy Clinical Research Network
.
Cost-effectiveness of aflibercept, bevacizumab, and ranibizumab for diabetic macular edema treatment: analysis from the diabetic retinopathy clinical research network comparative effectiveness trial
.
JAMA Ophthalmol
2016
;
134
:
888
896
6.
Lian
JX
,
Gangwani
RA
,
McGhee
SM
,
Chan
CK
,
Lam
CL
,
Wong
DS
;
Primary Health Care Group
.
Systematic screening for diabetic retinopathy (DR) in Hong Kong: prevalence of DR and visual impairment among diabetic population
.
Br J Ophthalmol
2016
;
100
:
151
155
7.
Stefánsson
E
,
Bek
T
,
Porta
M
,
Larsen
N
,
Kristinsson
JK
,
Agardh
E
.
Screening and prevention of diabetic blindness
.
Acta Ophthalmol Scand
2000
;
78
:
374
385
8.
Tung
T-H
,
Shih
H-C
,
Chen
S-J
,
Chou
P
,
Liu
C-M
,
Liu
J-H
.
Economic evaluation of screening for diabetic retinopathy among Chinese type 2 diabetics: a community-based study in Kinmen, Taiwan
.
J Epidemiol
2008
;
18
:
225–233
9.
Prescott
G
,
Sharp
P
,
Goatman
K
, et al
.
Improving the cost-effectiveness of photographic screening for diabetic macular oedema: a prospective, multi-centre, UK study
.
Br J Ophthalmol
2014
;
98
:
1042
1049
10.
GOV.UK
.
Diabetic eye screening: programme overview
,
2014
. Accessed 1 August 2016. Available from https://www.gov.uk/guidance/diabetic-eye-screening-programme-overview
11.
Bresnick
GH
,
Mukamel
DB
,
Dickinson
JC
,
Cole
DR
.
A screening approach to the surveillance of patients with diabetes for the presence of vision-threatening retinopathy
.
Ophthalmology
2000
;
107
:
19
24
12.
James
M
,
Turner
DA
,
DM
,
Vora
J
,
Harding
SP
.
Cost effectiveness analysis of screening for sight threatening diabetic eye disease
.
BMJ
2000
;
320
:
1627
1631
13.
Wong
IY
,
Iu
LP
,
Koizumi
H
,
Lai
WW
.
The inner segment/outer segment junction: what have we learnt so far
?
Curr Opin Ophthalmol
2012
;
23
:
210
218
14.
Wong
IY
,
Koizumi
H
,
Lai
WW
.
Enhanced depth imaging optical coherence tomography
.
Ophthalmic Surg Lasers Imaging
2011
;
42
(
Suppl.
):
S75
S84
15.
Wong
RL
,
Lee
JW
,
Yau
GS
,
Wong
IY
.
Relationship between outer retinal layers thickness and visual acuity in diabetic macular edema
.
BioMed Res Int
2015
;
2015
:
981471
16.
Virgili
G
,
Menchini
F
,
Casazza
G
, et al
.
Optical coherence tomography (OCT) for detection of macular oedema in patients with diabetic retinopathy
.
Cochrane Database Syst Rev
2015
;
1
:
CD008081
17.
Fujimoto
J
,
Swanson
E
.
The development, commercialization, and impact of optical coherence tomography
.
Invest Ophthalmol Vis Sci
2016
;
57
:
OCT1
OCT13
18.
Wang
YT
,
M
,
Wolfson
Y
,
Bressler
SB
,
Bressler
NM
.
Comparison of prevalence of diabetic macular edema based on monocular fundus photography vs optical coherence tomography
.
JAMA Ophthalmol
2016
;
134
:
222
228
19.
Mackenzie
S
,
Schmermer
C
,
Charnley
A
, et al
.
SDOCT imaging to identify macular pathology in patients diagnosed with diabetic maculopathy by a digital photographic retinal screening programme
.
PLoS One
2011
;
6
:
e14811
20.
Bressler
SB
,
Edwards
AR
,
Chalam
KV
, et al.;
Diabetic Retinopathy Clinical Research Network Writing Committee
.
Reproducibility of spectral-domain optical coherence tomography retinal thickness measurements and conversion to equivalent time-domain metrics in diabetic macular edema
.
JAMA Ophthalmol
2014
;
132
:
1113
1122
21.
Xie
XW
,
Xu
L
,
Wang
YX
,
Jonas
JB
.
Prevalence and associated factors of diabetic retinopathy. The Beijing Eye Study 2006
.
Graefes Arch Clin Exp Ophthalmol
2008
;
246
:
1519
1526
22.
Murray
CJL
,
Lopez
;
World Health Organization
.
The Global Burden of Disease: A Comprehensive Assessment of Mortality and Disability From Diseases, Injuries, and Risk Factors in 1990 and Projected to 2020
.
Cambridge, MA
,
Harvard University Press
,
1996
23.
Lopez
,
Mathers
CD
,
Ezzati
M
,
Jamison
DT
,
Murray
CJL
.
Global Burden of Disease and Risk Factors
.
New York
,
Oxford University Press
,
2006
24.
World Health Organization
.
Disability weights, discounting and age weighting of DALYs
,
2016
. Accessed 2 August 2016. Available from https://www.who.int/healthinfo/global_burden_disease/daly_disability_weight/en/
25.
Murray
CJL
,
Vos
T
,
Lozano
R
, et al
.
Disability-adjusted life years (DALYs) for 291 diseases and injuries in 21 regions, 1990-2010: a systematic analysis for the Global Burden of Disease Study 2010
[published correction appears in Lancet 2013;381:628].
Lancet
2012
;
380
:
2197
2223
26.
Hospital Authority
.
Fees and charges
,
2016
. Accessed 2 August 2016. Available from https://www.ha.org.hk/visitor/ha_visitor_text_index.asp?Parent_ID=10044&Content_ID=10045
27.
The Food and Health Bureau, The Government of the Hong Kong Special Administrative Region
.
Appendix B Hong Kong’s Current Healthcare System
,
2016
. Accessed 2 August 2016. Available from https://www.fhb.gov.hk/beStrong/files/consultation/appendixb_eng.pdf
28.
World Health Organization
.
Cost effectiveness and strategic planning (WHO-CHOICE)
,
2016
. Accessed 2 August 2016. Available from https://www.who.int/choice/cost-effectiveness/en/
29.
The Government of the Hong Kong Special Administrative Region
.
The Facts
,
2016
30.
Sassi
F
.
Calculating QALYs, comparing QALY and DALY calculations
.
Health Policy Plan
2006
;
21
:
402
408
31.
Devleesschauwer
B
,
Havelaar
AH
,
Maertens de Noordhout
C
, et al
.
Calculating disability-adjusted life years to quantify burden of disease
.
Int J Public Health
2014
;
59
:
565
569
32.
Feng
X
,
Kim
DD
,
Cohen
JT
,
Neumann
PJ
,
Ollendorf
DA
.
Using QALYs versus DALYs to measure cost-effectiveness: how much does it matter
?
Int J Technol Assess Health Care
2020
;
36
:
96
103
33.
Klein
R
,
Klein
BE
,
Moss
SE
,
Davis
MD
,
DeMets
DL
.
The Wisconsin epidemiologic study of diabetic retinopathy. III. Prevalence and risk of diabetic retinopathy when age at diagnosis is 30 or more years
.
Arch Ophthalmol
1984
;
102
:
527
532
34.
Kawasaki
ROY
,
Shono
M
.
Anti-VEGF treatment use in a real-world health claims database. Presented at the 56th Annual Meeting of Japanese Retina and Vitreous Society, 1–3 December 2017, at the Tokyo International Forum, Tokyo, Japan
35.
Bressler
SB
,
Liu
D
,
Glassman
AR
, et al.;
Diabetic Retinopathy Clinical Research Network
.
Change in diabetic retinopathy through 2 years: secondary analysis of a randomized clinical trial comparing aflibercept, bevacizumab, and ranibizumab
.
JAMA Ophthalmol
2017
;
135
:
558
568
36.
The World Bank
.
GDP per capita (current US\$)
,
2016
. Accessed 2 August 2016. Available from https://data.worldbank.org/indicator/NY.GDP.PCAP.CD?view=map&year_high_desc=true