HEALTH ECONOMICS, VOL. 1: 39-51

(1992)

COST EFFECTIVENESS ANALYSIS

SCREENING FOR DIABETIC RETINOPATHY: A RELATIVE COST-EFFECTIVENESS ANALYSIS OF ALTERNATIVE MODALITIES AND STRATEGIES M. J. SCULPHER MSc', M. J. BUXTON BA', B. A. FERGUSON MSc', D. J. SPIEGELHALTER PhD3 and A. J. KIRBY BSc3 ' Health Economics Research Group, Brunel University, Uxbridge, MiddIesex, UK; Health Economics Consortium, University of York, UK; MRC Biostatistics Unit, University of Cambridge, UK

'

SUMMARY Diabetic retinopathy is the most common cause of blindness among adults of working age in the UK. If the disease is detected early effective treatment can be provided and this has resulted in calls for a systematic national screening programme. Using data on the screening of 3423 diabetics collected as part of an experimental programme in three UK centres, the relative cost-effectiveness of various screening options is assessed. The paper utilises direct evidence on a number of single modality screening options, including ophthalmoscopy undertaken by general practitioners or ophthalmic opticians, and non-mydriatic photography. With the objective of increasing the sensitivity of screening and using data collected in the study, options based on two further potential screening strategies are modelled and evaluated: combined screening using both ophthalmoscopy and non-mydriatic photography; and selective screening where high-risk diabetics are directly referred to an ophthalmologist and lowrisk cases are either left unscreened or are screened by one of the single or combined modality screening options. Given the objective of early detection, effectiveness is assessed in terms of the sensitivity and specificity of the referral decisions of screening options. Both health service and private resource costs of the various screening options are estimated, the latter in terms of travel and the opportunity cost of time. Cost effectiveness is evaluated in terms of the expected cost per true positive case of diabetic retinopathy referred by the screening options. To narrow the choice between the options, those subject to three-way domination with respect to the three choice variables of sensitivity, specificity and expected cost per true positive are excluded. Amongst the remaining options, the choice is dependent on the trade-off between the higher specifics of unselective single modality screening options and the higher sensitivities and lower expected costs per true positive case detected of combined modality and selective screening options. KEY

WORDS-screening, effectiveness, diabetic retinopathy, expected cost.

INTRODUCTION Diabetic retinopathy is the most significant cause of blindness among adults of working age in the UK. This remains the case despite the existence of treatment, using laser photocoagulation, which has been proven to be effective in reducing the risk of severe visual loss, if carried out sufficiently early. *-* Evidence indicates, however, that a high

proportion of patients are being referred too late for effective treatment,' despite the possibility of early detection using ophthalmoscopy or fundus photography. A national screening programme has been proposed, lo and a number of screening modalities has been considered for routine use in the UK. Controversy has surrounded the choice of screening approach, with both support for ophtha1moscopy"- l 3 and various types of fundus

Address for correspondence: Professor Martin Buxton, Health Economics Research Group, Brunel University, Uxbridge, Middlesex, UB8 3PH, UK.

1057-9230/92/010039- 13$06.50 0 1992 by John Wiley & Sons, Ltd.

Received 14 August 1991 Accepted 21 November 1991

40

M. J. SCULPHER E T A L .

photography. 14-” Amongst those supporting ophthalmoscopy, debate has centred on whether community-based screeners, such as general practitioners and ophthalmic opticians, should undertake screening, or whether only hospitalbased ophthalmological staff are sufficiently skilled. 18-22 Given limited health care resources, it is crucial to evaluate proposed approaches to screening in terms of both their effectiveness and cost-effectiveness. In 1985, a comparative study was set up by the Department of Health in the UK, to consider the relative effectiveness and cost-effectiveness of various screening modalities, using either ophthalmoscopy or fundus photography, in appropriately referring cases of serious diabetic retinopathy to a consultant ophthalmologist. Within this study, a total of 3423 diabetics cared for largely by a general practitioner (i.e. community-based diabetics) were screened between 1985 and 1988 in three English centres-Exeter, Oxford and Sheffield. The effectiveness of screening based on ophthalmoscopical examination by general practitioners and opticians was assessed, together with that of nonmydriatic photography, by comparing the referral decision proposed by these screening modalities with that of a hospital-based ophthalmologist who screened each patient as a reference standard. Cost-effectiveness was evaluated in terms of expected cost per true positive case detected, including those met by the health service and those met privately. A detailed description of the study design, together with the effectiveness and costeffectiveness results for these directly observed screening options, has been published elsewhere. 23*24 The low sensitivities of these options casts doubt on the feasibility of their role in routine clinical practice.23 In an attempt to identify more sensitive options, this paper extends the previous analysis by using the data collected on observed single modality screening options to model the results of potential options based on more complex strategies: the combination of ophthalmoscopy and non-mydriatic photography; and selective screening, where high-risk diabetics are directly referred. These latter strategies are compared with single modality screening options in terms of their sensitivity, specificity and expected cost per true positive case of diabetic retinopathy detected. The choice of the many

resulting options is narrowed by identifying options that are dominated by others in terms of these choice variables. METHODS

Study design and screening arrangements Within the study there was a total of 3423 community-based diabetics. These patients represent a sample of diabetics who are under the sole care of their general practitioner, who have not received any systematic hospital-based eye care, and who would represent the key target of a regular organised screening programme. A total of 2891 patients were screened, using ophthalmoscopy through dilated pupils, by a primary screener: by a general practitioner in Exeter (n = 1207), Oxford (n = 628) and Sheffield ( n = 641) or by an ophthalmic optician (for one group in Oxford (n = 415)). These patients, together with 532 other diabetics in Sheffield who were without a grimary screener, were also screened using a 45 non-mydriatic Canon camera using Polaroid prints, which was either based at a hospital ophthalmic clinic or taken to the general practice; photographs were assessed by a consultant ophthalmologist. Each diabetic also received an examination by an ophthalmologically-trained clinical assistant using ophthalmoscopy. Diabetics either visited hospital for their clinical assistant screen (Oxford and Sheffield) or the latter was undertaken at the diabetic’s general practice (Exeter). The clinical assistants in the three centres undertook identical screens in terms of technology and data collection; they represent the reference standard against which the effectiveness of the primary screeners and of the camera was assessed, as well as an indication of the prevalence of diabetic retinopathy amongst the screened population (the prior probability of the disease’s presence). Screeners using ophthalmoscopy, together with ophthalmologists when assessing the photographs, were required to indicate whether they were proposing to refer a given patient to hospital for suspected sight-threatening diabetic retinopathy.

Alternative approaches to screening With the aim o f , identifying more sensitive

41

DIABETIC RETINOPATHY

screening options and using the data collected in the study, options based on two further potential approaches t o screening are evaluated in this paper.

Combined screening options. Given the limitld field of view of the camera, it is feasible that manifestations of sight-threatening diabetic retinopathy on the peripheral fundus will not show up in the photographs. By matching for each patient the referral decision proposed by the GP or optician with that proposed by the ophthalmologist, on inspection of the photographs, the notional referral decision of a combined ophthalmoscopy and photography modality is generated. Given the low sensitivities of the single modality screening options directly observed in the study, 23 the logical decision rule used is that a patient is considered referred by the combined modality options (i.e. a positive screen)

if either of the single modality options confer a positive screen.

Selective screening options. The onset of serious diabetic retinopathy is known t o be associated with both the type and duration of diabetes,25326both of which should be known t o the diabetic’s general practitioner and do not require additional clinical examination. Selective screening options are modelled based on the direct referral t o an ophthalmologist of a high-risk subgroup of diabetics-defined, in the base-case, as insulin-taking (Type I) diabetics with a duration of the disease of 10 years or more-without any prior screening. Low-risk cases may then be left unscreened or screened using one of the single or combined modality options. A total of thirteen screening options are, therefore, considered in this paper; these are listed in Table 1.

Table 1. Details of screening to options evaluated Option GP Optician Hospital-based camera GP-visiting camera

Basis of screening Single modality screening using ophthalmoscopy Single modality screening using ophthalmoscopy Single modality screening using non-mydriatic photography with photos assessed by ophthalmologist Single modality screening using non-mydriatic photography with photos assessed by ophthalmologist Combination modality

Arrangement Patient examined at general practice. Special visit or part of general assessment. Patient examined at optician’s premises. Special visit. Patient’s fundi photographed by clinical assistant at hospital. Special visit. Patient’s fundi photographed by clinical assistant at general practice. Special visit or part of general assessment.

Patient examined and fundi photographed by clinical assistant at general practice. Special visit or part of assessment. Patient examined at optician’s premises Combination modality Combined optician during special visit; fundi photographed at and GP-visiting general practice by clinical assistant on camera special visit or as part of general assessment. 7- 13 Selective screening High-risk diabetics directly referred High-risk cases not screened, but sent for with low-risk by GP to ophthalmologist; immediate diagnosis. Low-risk cases diabetics not low-risk cases either not screened screened as for 1 to 6 above screened or screened or screened by one of the options by 1 to 6 above 1 to 6 above Combined GP and GP-visiting camera

42

M. J . SCULPHER E T A L .

The effectiveness of screening modalities and strategies The effectiveness of screening options is assessed in terms of their sensitivity and specificity. By comparing, for each patient, the referral decision proposed by the screening option being evaluated with that of the clinical assistant’s reference screen, the former’s sensitivity and specificity are estimated which are independent of the prior probability of the disease’s presence. Sensitivity is defined as the proportion of positive cases correctly referred by the option; specificity is defined as the proportion of negative cases correctly not referred by the option.

The resource costs of screening: health service costs Data were collected on the health service resources used during all types of screening as part of the study. 24 Information was obtained on the staff involved (grade and time present), the travel details of staff where applicable, the consumables used (drugs, film) and the location of screening for estimation of overheads. The base cost analysis assumes that diabetics have to make a special visit to the screening clinic in order to receive their screening test@). Resource use is valued using 1989-90 UK unit costs and, as far as possible, the approach is one of an analysis of short-run marginal costs. The fixed capital cost of the camera is translated into an annual equivalent cost based on an expected life of the camera of 10 years and a discount rate of 6 per cent; this cost, together with the camera’s annual maintenance cost, is allocated on a per patient basis assuming an annual throughput of 2000 patients, to which the cost of ophthalmologist assessment is added. The cost of an ophthalmic optician screen is based on the 1989-90 rate of reimbursement of opticians by the Family Practitioner Committee for health service eye tests. This is considered to represent an estimate of the opportunity cost of an optician screening a given diabetic: such a screen would involve the optician forgoing other health service-funded eye tests. The extent to which this assumption is crucial to the results of the analysis is assessed by way of sensitivity analysis.

The resource costs of screening: private travel and time costs A 10% sample of patients screened during the study was asked to provide data to facilitate an estimate of the private costs incurred in attending for screening. 24 Details were sought on patients’ (and, where applicable, on their companions’) journey and travel arrangements to and from the clinic and, where some form of payment was made at the point of use, on the cost of transportation. When the journey was made by car, appropriate guidelines have been used to cost the journey which incorporate relevant fixed and variable costs.27 The time patients and any companions spent visiting the screening clinics is also valued based on data collected on their activities forgone and occupation. When their time would have been spent at work, the time is valued at the gross cost of employment using average hourly manual or non-manual rates of pay for the area. 28 When patients were forgoing non-working time this is valued using hourly cost guidelines generated from survey work used in transport studies. 29 Screening based at the general practice The base-case analysis assumes that patients are obliged to make a special visit to a given clinic for screening. It may be the case, however, that, for some screening options, diabetics will already be attending the clinic for a general check-up. In the case of screening methods involving the general practice, patients can be screened (by their GP, a camera or their GP and a camera) during the same visit as they have a general check-up. This represents the alternative-case analysis in which onIy the additional resource costs incurred as a direct result of eye testing are allocated to the screening test. The cost and cost-eflectiveness models A given diabetic has a probability of being placed in one of four decision groups by the screening process: true positive, false negative, true negative and false positive; the probabilities are determined by the sensitivity and specificity of the screening options and by the prior probability of disease. For selective screening without screening of low-risk cases, only two decision ’

43

DIABETIC RETINOPATHY

groups exist-high-risk positive and high-risk negative-the probabilities of which are determined by the probability of being high-risk and the prior probability of the disease within that sub-group. Selective screening with screening of low-risk patients has six decision groups: true and false positive, and true and false negative for lowrisk cases; high-risk positive and negative for high-risk patients. Each of these decision groups for the different types of screening will have costs associated with them; these will be screening costs with or without diagnostic costs. For example, the cost of a false positive is the cost of screening and diagnosis. In the case of the high-risk positive and negative decision groups for selective screening, diagnostic costs only are incurred due to direct referral. For each type of screening the expected cost of a screening episode is calculated, representing the sum of the cost of being in each of the relevant decision groups multiplied by the associated probability of a given patient being placed in each decision group. Given the primary objective of detecting cases with diabetic retinopathy, the cost-effectiveness ratio used in this evaluation is the expected cost per true positive case detected, which is derived by dividing the expected cost per screening episode by the probability of being a true positive. Estimates of the expected cost per true positive case detected will be slight over-estimates because a proportion of true positive cases would have presented symptomatically in a world without screening. Given that no data exist on the proportion which would present in this way or on the stage and cost of presentation, no estimate has been made. However, the absolute effect on expected costs per true positive case detected would be identical for all options: the product of the proportion presenting symptomatically and the present value of the cost of that future presentation. The paper’s primary interest is in the relative cost-effectiveness of the screening options, so this overestimation will not weaken the analysis. It is often considered that false negatives in screening programmes may be deterred from symptomatic presentation because of the false reassurance of their screen, resulting in possible higher treatment costs due to late detection. As regards screening diabetics for retinopathy, the estimation of these costs is not considered neces-

sary because such deterrence is unlikely: on failing eye sight, diabetics are not likely to delay presentation to an optician because other factors unrelated to diabetic retinopathy can be associated with sight loss. The relative cost-effectiveness analysis is based on a prevalence (i.e. one-off) screen of the community-based diabetic population. For unselective screening, and overall prior probability (prevalence) of retinopathy of 5.8% is used.23 For selective screening relevant epidemiological parameters are taken from the study: the prior probability amongst high- and low-risk diabetics is taken as 17.3% and 4.7% respectively, with the probability of being high-risk as 8.3%. The identification of preferred screening options is based on three choice variables: expected cost per true positive case detected, sensitivity and specificity. Those options subject to three-way domination a.re identified and excluded; that is, options which have a higher expected cost per true positive case detected as well as a lower sensitivity and specificity than one or more other screening options. RESULTS Single modality options As detailed in Table 2, for the base-case analysis the average (health service, private travel and time) cost per screen ranges from €14.87, for the GP-visiting camera, to f25.46, for hospital-based fundus photography. The figures in parentheses relate to the costs for the alternative-case analysis of screening taking place ,as part of a general check-up at the general practice. For such screening involving the GP, it has been assumed that the additional time taken for retinopathy screening is 8 minutes (i.e. allocating no staff or patient waiting time) and that no additional patient travel cost is incurred; hence the average total screening cost per patient falls from €20.66 to f5.02. For screening using photography as part of a general check-up at the general practice, only private travel and time cost is avoided due to the fact that, in the study, visiting clinical assistants took the photographs: the cost incorporates staff and patient waiting times and these would justifiably remain, even if the patient is already ._ present at the clinic. Hence the average total

44

M. J. SCULPHER E T A L .

Table 2. The average health service, and patient travel and time cost per patient of single screening options Modality

Ophthalmoscopy (1) General Practitioner t (2) Ophthalmic Optician Fundus Photography (3) Hospital-based t (4) GP-Visiting Diagnosis$ Hospital-based clinical assistant

Health service cost* E

Patient travel cost* E

Patient time cost*

15.07 (4.59) 10.40

1.33 (0) 2.54

4.26 (0.43) 6.37

20.66 (5.02) 19.31

16.55 12.12 (12.12)

4.43 1.12 (0)

4.48 1.63 (0.65)

25.46 14.87 (12.77)

18.23

4.43

6.35

29.01

E

Total* E

Notes: t = Based on data pooled across centres. X = Assumes cost of diagnosis equal to that of hospital based clinical assistant.

* = Figures in parentheses relate to alternative-case analysis which assumes that screening takes place as part of general check-up at the general practice.

screening cost per patient falls from f14.87 to f12.77. Table 3 provides details of the effectiveness and cost-effectiveness results for single modality screening options. Sensitivities range from 0.4, for the hospital-based camera, to 0.58 for the GPvisiting camera; specificities range from 0.91, for GPs, to 0.97 for the GP-visiting camera. The expected costs per true positive case detected of these single modality options range, for the basecase analysis, from f497, for the GP-visiting

camera, to E l 178, for the hospital-based camera; for the alternative-case analysis they range from f273 for the GP up to f1178.

Combined modality options Table 3 also details the effectiveness and costeffectiveness of combined modality screening options. When fundus photography is added to GP ophthalmoscopy, sensitivity increases to 0.8 and specificity falls to 0.86. If it is assumed that

Table 3. Sensitivity, specificity and expected cost per screening episode and per true positive case detected for single and combined modality screening options Subject to 3-way Expected cost Expected cost per screening per true positive way domination? Sensitivity Specificity episode* E case detected* f Y/N*

Modality/Strategy Single Modality Options Ophthalmoscopy (1) General Practitioner t (2) Ophthalmic Optician Fundus Photography (3) Hospital-Based? (4) GP-Visiting Combined Modality Options (5) G P and GP-Visiting Camera (6) Optician and GP-Visiting Camera

t

0.53 0.48

0.91 0.94

24 (9) 22

784 (273) 784

y (N) y (-)

0.4 0.58

0.96 0.97

27 17 (15)

1178 497 (434)

N (N)

0.8 0.67

0.86 0.89

34 (19) 37 (35)

734 (419) 968 (914)

N (N) N (N)

y (-1

Notes: * = Figures in parentheses relate to alternative-case analysis which assumes screening takes place as part of general checkup at the general practice. t = Based on data pooled across centres.

45

DIABETIC RETINOPATHY

the camera is taken to the general practice and that the patient receives both screens at the same appointment, the expected cost per true positive case detected falls to &734in the base-case analysis; in the alternative-case analysis, where both types of screening take place as part of a general check-up at the general practice, this cost falls to f419. When fundus photography is added to optician ophthalmoscopy, sensitivity increases to 0.67 and specificity falls to 0.89. If it is assumed that a patients’ fundi are photographed at the general practice on a separate occasion, requiring a separate visit by the patient, the expected cost per true positive case detected of this combined strategy, for the base-case analysis, increases over that of the optician alone, to €968; for the alternative-case analysis of photography only taking place as part of a general check-up, this cost falls to f914.

Selective screening options Table 4 details the effectiveness and costeffectiveness of the various selective screening options. Sensitivities range from 0.25, when the low-risk sub-group is left unscreened, to 0.85, when it is screened by the combined GP and GPvisiting camera option. Specificities range from 0.8, when low-risk diabetics are screened by the

combined GP and GP-visiting camera option, to 0.93, when they remain unscreened. In the basecase analysis, the expected costs per true positive case detected range from f168, when low-risk diabetics are not screened, to f863, when they are screened by the hospital-based camera. In the alternative-case analysis, where some screening options for low-risk diabetics take place as part of a general check-up at the general practice, this range remains unchanged but the expected costs per true positive case detected where low-risk patients are screened by GPs, GP-visiting camera, combined GP and GP-visiting camera and combined optician and GP-visiting camera fall by 59%, 11070, 40% and 12% respectively.

Sensitivity analysis The cost of screening using the non-mydriatic camera is based on a throughput of 2000 patients per annum. The robustness of the costeffectiveness ratios for such screening to changes in annual throughput has been considered and found to be relatively unimportant because the fixed cost of the camera, when allocated on a per patient basis, is relatively unimportant compared to the variable cost.24 For example, a halving of annual patient throughput to 1000 only increases the expected cost per true positive case detected of

Table 4. Sensitivity, specipity and expected cost per screening episode and per true positive case detected for selective screening options

Strategy

High-risk patients directly referred, low-risk patients: (1) Not screened (2) Screened by GP (3) Screened by Optician (4)Screened by GP-visiting camera (5) Screened by hospital-based camera (6) Screened by GP + GPvisiting camera (7) Screened by Optician + GPvisiting camera

Expected cost per screening Sensitivity Specificity episode? f

Expected cost Subject to 3-way per true positive domination? case detected? f Y/N t

0.25 0.65 0.61 0.68

0.93 0.84 0.87 0.9

2 24 (lo), 22 18 (16)

168 650 (266) 63 1 443 (394)

N (-1 N (N)

0.55

0.89

27

863

y (-1

0.85

0.8

33 (20)

679 (407)

N (N)

0.75

0.83

37 (32)

842 (745)

N (N)

N (N) N (N)

Notes: * = Based on definition of high-risk of Type I with 10 years or more duration of diabetes. t = Figures in parentheses relate to alternative-case analysis which assumes screening takes place as part of general check-up at the general practice.

46

M. J. SCULPHER E T A L .

the GP-visiting camera screening option by 6% to f529; a similar reduction in throughput for the combined GP and GP-visiting camera screening option increases the excepted cost per true positive by 3% to f757. The NHS reimbursement rate to opticians has been used to represent an estimate of the opportunity cost of an optician screen. It is clear that, in relative terms, the latter is not a ‘competitive’option and any higher screening cost would make it even less so; for example, a $35.00 screening cost would result in an expected cost per true positive case detected of f1028, which would be significantly higher than all other options bar the hospital-based camera. Optician screening costs would have to fall to as low as f2.45 before it became a cheaper screening option, per true positive case detected, than the GP-visiting camera. For selective screening, the base-case definition of high-risk is taken as an insulin-taking (Type I) diabetic with a duration of diabetes of 10 years or more. If reasonable alternative definitions of high-risk are used, there is no significant impact on the effectiveness or cost-effectiveness results. If

a more conservative definition is used (Type I with duration of 15 years or more), sensitivity falls by between 0.02 and 0.08, specificity increases by between 0.02 and 0.03 and expected cost per true positive case detected changes by between -6% (no screening for low-risk cases) and 7% (hospitalvisiting camera screening for low-risk cases). If a less restrictive definition of high-risk is employed (Type I with a duration of 5 years or more) then sensitivity increases, relative to the base definition, by between 0 and 0.02, specificity falls by between 0.03 and 0.04 and expected cost per true positive case detected changes by between -2% (hospital-visiting camera screening of low-risk cases) and 35% (no screening of low-risk cases). The use of combined modality or selective screening to increase sensitivity results in a small fall in specificity, increasing the risk of swamping hospital ophthalmic departments with healthy patients. Hence the health service cost of diagnosis, which is assumed to be equal to the shortrun marginal cost of screening using a hospital-based clinical assistant, might be an underestimate of the true diagnostic cost: a flood of healthy patients may necessitate new buildings,

Expected cost per true positive case

(e) I

1200 I

1000 -

600 -

10.00

70.00 Health service cost of diagnosis

30.00

50.00

90.00

110.00

(€1

Note: LR refers to screening of low-risk group only as part of a selective screening strategy.

Figure 1. The effect of changing health service diagnostic cost on expected cost per true positive case detected: selective versus unselective screening options.

DIABETIC RETINOPATHY

equipment etc. Figure 1 shows the effect of varying health service diagnostic cost on the expected cost per true positive case detected of two of the most sensitive and cost-effective screening options: the GP-visiting camera and the combined GP and GP-visiting camera options when they screen all diabetics and when they screen only low-risk cases. When health service diagnostic cost reaches about E55, it becomes cheaper, per true positive case detected, to screen all diabetics with the GP-visiting camera rather than just the low-risk cases. When diagnostic cost reaches about f75 it is cheaper, per true positive case detected, for the combined GP and GPvisiting camera option to screen all diabetics rather than low-risk cases alone. A key variable as regards the absolute cost of detecting a true positive case is the prior probability of retinopathy in a given diabetic population. The analysis is based on an initial ‘prevalence’ screen; follow-up (or ‘incidence’) screening is likely to have a much smaller pool of positive cases to detect. The second round screening undertaken in the study found an annual incidence of about 1.5% which, if incorporated into the cost-effectiveness model, would increase the expected cost of detecting a true positive case for all options by approximately 400V0, assuming sensitivity and specificity remain unchanged.

Identifying dominated screening options The final columns of Tables 3 and 4 identify ‘three-way dominated’ options in both the basecase analysis and (in parentheses) the alternativecase analysis where screening takes place as part of a general check-up at the general practice. In the base-case analysis, therefore, four screening options are subject to three-way domination: (a) single modality GP; (b) single modality optician; (c) single modality hospital-based camera; (d) selective strategy with hospital based camera screening of low-risk cases. In the alternative-case analysis single modality G P screening is removed from this list. DISCUSSION AND CONCLUSIONS Epidemiological estimates indicate that up to 98% of insulin-treated patients diagnosed before 30years of age and 50% of non-insulin-treated

47

ones will have evidence of diabetic retinopathy after 20 years. 25*26 A number of large multicentre clinical trials have now demonstrated that treatment for diabetic retinopathy using laser photocoagulation is effective as long as it is undertaken prior to the development of symptoms associated with significant visual loss. 2 - 8 Even with these advances, which indicate that between 50% and 70% of blindness from diabetic retinopathy would be preventable by early detection and treatment, 22 the incidence of new blindness registrations has showed little sign of declining; 30,31 there is evidence that this is because diabetics with treatable retinopathy are not being referred to hospital sufficiently early. To date no prospective study has been undertaken to quantify directly the benefits of screening diabetics for retinopathy. Modelling exercises have been undertaken in the US, but these have either concentrated exclusively on Type I cases (less prevalent amongst community-based diabetics in the UK)32933 or have looked solely at the early detection and treatment of the proliferative form of retinopathy. 34 Furthermore, the relevant end-point used in these studies has beexi sightyears gained which does not facilitate crossprogramme comparison, in which case these studies are effectively relative cost-effectiveness analyses. Given the clinical evidence supporting the general case for screening for diabetic retinopathy, the study on which this paper is based was designed to address the more specific question of how to screen. The appropriate design of the study was, therefore, a relative costeffectiveness analysis, where the latter is estimated in terms of cost per true positive case detected. Those single modality screening options directly observed in the UK study were found to have disappointingly low sensitivities. 23 The purpose of the further analysis reported here is to identify whether alternative potential screening options exist which can secure the detection of higher rates of true positive cases. Conclusions of this work have important implications for other forms of screening. The combination of single screening modalities based on different technologies (in this case, ophthalmoscopy and non-mydriatic photography) can improve detection rates because, as in this study, they may detect different true positive cases. Furthermore, the identification of a ‘highrisk’ group of patients which is defined using readily accessible data but which has a significantly higher prior probability of the target

48

M. J . SCULPHER E T A L ,

disease, offers an opportunity for selective screening. Depending on the size of the high-risk group, and the prior probability of the disease within it relative to that within the low-risk group, it might be feasible to refer directly high-risk cases and screen low-risk patients in a conventional manner; alternatively, expensive but highly accurate screening technologies may be used on high-risk cases in the anticipation of a sufficiently high yield of positive cases to justify cost-effective utilisation. This study has identified three variables upon which the choice of screening option should be based: sensitivity, specificity and expected cost per true positive case detected. The results of the analysis indicate that a clear trade-off exists when considering alternative screening options for diabetic retinopathy: by combining single modality screening options, and developing selective screening strategies based on the identification of high- and low-risk sub-groups, sensitivity increases but specificity declines. Ideally, this trade-off could be resolved by the use of an allembracing outcome measure, which would embody any impact of screening on the stress and anxiety of all patients screened, the longer term health improvements for true positive cases contingent on early treatment and the particular anxiety and morbidity experienced by false positive cases. No economic evaluation of a screening programme has successfully developed such a measure. It is doubtful whether generic quality of life instruments, for example the utility

measures incorporated into estimates of qualityadjusted life years, would be suifficiently sensitive to register all these outcomes. 3 5 In the absence of an all-embracing outcome measure, the identification of screening options subject to domination as regards the three choice variables represents a means of narrowing the number of options to consider. A way of narrowing the field still further would be to reduce the number of choice variables. To de-select specificity from these variables would be to remove a surrogate indicator of the disbenefit of a false positive. However, in the case of diabetic retinopathy, such disbenefit is likely to be limited: unlike screening for diseases such as cancer, the diagnostic tests used by ophthalmologists are simple out-patient procedures, usually further fundus photography; anxiety at being referred to hospital should be minimised by most diabetics’ knowledge of the chance of complications of their disease, and of existence of effective out-patient treatment if required. If specificity is excluded as a choice variable it is possible to identify options that are subject to ‘two-way domination’, that is those with lower sensitivity and higher expected cost per true positive case detected than one or more alternative options. Table 5 lists, for both the base-case and alternative analyses, the three options not subject to two-way domination in ascending order of sensitivity; the table details sensitivity, expected cost per true positive case detected and incremental cost per additional true positive case detected by

Table 5. Sensitivity, expected cost per true positive case detected and incremental cost per additional true positive case detected of screening options not subject to three- or two-way domination

Option Selective: no screening of low risk cases Selective: GP-visiting camera screening of low risk cases Selective: combined GP and GP-visiting camera screening of low risk cases Note:

Sensitivity

Expected cost per true positive case detected* f

0.25

168

0.68

443 (394)

0.85

679 (407)

Incremental cost per additional true positive* c

601 (524) 1663 (461)

* = Figures in parentheses relate to alternative-case analysis which assumes screening takes place as part of general check-up at the general practice.

DIABETIC RETINOPATHY

the adoption of the next most sensitive screening option. One other potential means of avoiding the trade-off between high sensitivity and lower specificity is to use hospital-based ophthalmological staff to screen, either with the latter travelling to the general practice or patients coming to hospital. This study estimated the cost of hospitalbased clinical assistant screening either at the general practice or at hospital. The role of these screeners as the reference standard against which to evaluate the effectiveness of the screening options would imply a sensitivity and specificity of 1.O for the clinical assistants; in routine clinical practice, however, these may drop to, say, 0.8 and 0 . 9 respectively. With these levels of accuracy, the expected cost per true positive case detected would be &716 for the hospital-based version and &5 18 for the version visiting the general practice with the patient making a special visit; on the basis of these assumptions, such screening would not be subject to two-way domination. If clinical assistant screening took place at the general practice as part of a general check-up, the cost per true positive would fall to E472; in these circumstances the option would be subject to two-way domination. Few economic evaluations consider the private costs imposed by screening programmes. Omitting such costs is a major weakness for two reasons. Firstly, the impact of any screening is a function of take-up, and the utilisation of health services has been found to be related to the costs patients incur. 36-42 Secondly, evidence on private costs permits a wider societal viewpoint for the analysis. Private costs are a key element of the analysis reported here, ranging from 18% to 46% of total screening costs per patient. Although the identification of non-dominated screening options has focused on the expected total cost of detecting true positive cases, the link between private cost and service utilisation would emphasise that it is important to assess the relative incidence of such costs between the options. This paper has shown that the shorter distance that patients would usually have to travel to general practice-based screening and the fact that such clinics are often held in the evening results in lower travel and time costs; these costs are even lower if, as in the alternative-case analysis in this paper, patients receive their screening as part of a general checkup. The selective screening options which dominate (Table 5 ) involve low-risk diabetics

49

being screened at the general practice which might be expected to maximise utilisation; however, high-risk patients are directly referred to hospital for diagnosis. Whether the incidence of higher private costs at diagnosis (nearly twice as high when making a special visit to hospital as when making a similar visit to the general practice) influences utilisation and hence the costeffectiveness of selective, relative to unselective, screening is an important area for future research In prin-ciple a cost-utility analysis which fully encompassed all the short and long term health outcome changes resulting from screening would enable a comparison to be made between the marginal benefits and costs of the screening options under consideration and of other u!jeS of health service resources. There is, however, no apriori basis to believe that such an extended analysis would change the conclusions regarding the relative ordering of the options. Given the impracticability of such a prospective study and the inadequacy of existing data accurately to estimate cost-utility retrospectively, the decision as to which non-dominated option to adopt has to remain judgemental. This study has indicated that a clear trade-off exists with the screening of community-based diabetics for retinopathy: to increase the proportion of positive cases detected (i.e. to increase sensitivity) and hence to have the greatest impact upon health outcomes will necessitate the adoption of screening strategies which have lower specificities and/or higher expected costs per true positive case detected. The overall conclusion of this study is that, if systematic screening is to be superimposed upon a well-organised system of general diabetic care at the general practice, where eye testing can be carried out during the same appointment as other routine health checks, considerable (health service and private) cost savings can result compared to screening requiring additional visits to a hospital or ophthalmic optician. This study shows, however, that, if high detection rates take primacy over the avoidance of false positive screens, then the low sensitivity of single and combined unselective screening options results in the need to refer high-risk cases directly to an ophthalmologist and either not to screen low-risk cases or to screen them using GP ophthalmoscopy or combined GP ophthalmoscopy and GP-visiting photography. If routine health care for community-based diabetics is poorly organised and patients either do

50

M. J . SCULPHER E T A L .

not receive general check-ups or receive them irregularly, the organisation of systematic eye screening around the general practice may still provide cost savings. Again, if only the two choice variables of sensitivity and expected cost per true positive case detected are used, the choice lies between three screening options, all of which are selective, with low-risk cases not screened or screened by a GP-visiting camera or combined GP and GP-visiting camera. ACKNOWLEDGEMENTS The authors would like to acknowledge the financial support received for this study from the UK Department of Health. In addition the authors would like to thank the three study Centre Directors: Hung Cheng, John Jacob and John Talbot; Nicky Gillard and J o Holland for their typing; and all other members of the study team. The usual disclaimer applies.

REFERENCES 1 . DHSS. Causes of blindness andpartial sight among adults in 1976177 and 1980181 England. London: HMSO, 1988. 2. British Multicentre Study Group. Photocoagulation in the treatment of diabetic maculopathy. Lancet, 1975, ii: 1110-1113. 3. British Multicentre Study Group. Proliferative diabetic retinopathy: treatment with xenon arc photocoagulation. British Medical Journal, 1977; 1: 739-741. 4. British Multicentre Study Group. Photocoagulation for proliferative retinopathy: a randomised controlled trial using the xenon-arc. Diabetologia, 1984; 26: 179-183. 5 . Diabetic Retinopathy Study Research Group. Preliminary report on effects of photocoagulation therapy. American Journal of Ophthalmology, 1976; 81: 383-396. 6. Diabetic Retinopathy Study Research Group. Photocoagulation treatment of proliferative diabetic retinopathy. Ophthalmology, 1981; 88: 583-600. 7. Early Treatment Diabetic Retinopathy Study Research Group. Photocoagulation for diabetic macular edema. Archives of Ophthalmology, 1985; 103: 1796-1806. 8. Early Treatment Diabetic Retinopathy Study Research Group. Early photocoagulation for diabetic retinopathy. Ophthalmology, 1991; 98: 766-785. 9. Jones, R. B., Larizgoitia, I., Casado, L., et al. How effective is the referral chain for diabetic retinopathy? Diabetic Medicine, 1989; 6: 262-266.

10. Retinopathy Working Party. A protocol for screening for diabetic retinopathy in Europe. Diabetic Medicine, 1991; 8: 263-267. 1 1 . Foulds, W. S., McCuish, A., Barrie, T., et al. Diabetic retinopathy in the West of Scotland: its detection and prevalence, and the cost-effectiveness of a proposed screening programme. Health Bulletin, 1983; 41: 318-326. 12. Burns-Cox, J . and Dean Hart, J. C. Screening of diabetics for retinopathy by ophthalmic opticians. British Medical Journal, 1985; 290: 1052-1054. 13. Rohan, T. E., Frost, C. D., Wald, N. J . Prevention of blindness by screening for diabetic retinopathy: a quantitative assessment. British Medical Journal, 1989; 299: 1198-1201. 14. Ryder, R. E. J., Young, S., Vora, J . P. et al. Screening for diabetic retinopathy using Polaroid retinal photography through undilated pupils. Practical Diabetes, 1985; 2: 34-39. 15. Williams, R., Nussey, S., Humphry, R. el al. Assessment of non-mydriatic fundus photography in detection of diabetic retinopathy. British Medical Journal, 1986; 293: 1140-1 142. 16. Rogers, D., Bitner-Glinziczi, M., Harris, C. et a / . Non-mydriatic retinal photography as a screening service for general practitioners. Diabetic Medicine, 1990; 7 : 165-167. 17. Taylor, R., Lovelock, L. and Tunbridge, W. M. G. Comparison of non-mydriatic retinal photography with ophthalmoscopy in 2159 patients: mobile retinal camera study. British Medical Journal, 1990; 301: 1243-7. 18. Bron, A. J., Screening for diabetic retinopathy. British Medical Journal, 1985; 290: 1025- 1026. 19. Kleinstein, R. N., Roseman, J . M., Herman, W. H. et a f . Detection of diabetic retinopathy by optometrists. Journal of the American Optometric Association, 1987; 58: 879-882. 20. Harrison, R. J . , Wild, J. M. and Hobley, A. J. Referral patterns to an ophthalmic out-patient clinic by general practitioners and ophthalmic opticians and the role of these professions in screening for ocular disease. British Medical Journal, 1988, 297:1162-1167. 21. Baker, S. B., Vallbona, C., Campbell, J . V . et al. Diabetic eye disease detection by primary care physicians. Diabetes Care, 1990; 13: 908-909. 22. Owens, D. R., Dolben, J., Young, S. et al. Screening for diabetic retinopathy. Diabetic Medicine, 1991; 8: 54-510. 23. Buxton, M. J . , Sculpher, M. J., Ferguson, B. A. et al. Screening for treatable diabetic retinopathy: a comparison of different methods. Diabetic Medicine. 1991; 8: 371-377. 24. Sculpher, M. J., Buxton, M. J . , Ferguson, B. A. et al. Relative cost-effectiveness analysis of different methods of screening for diabetic retinopathy. Diabetic Medicine, 1991; 8: 644-650.

DIABETIC RETINOPATHY

25. Klein, R., Klein, B. E. K., Moss, S. E., et al. The Wisconsin epidemiologic study of diabetic retinopathy. 111. Prevalence and risk of diabetic retinopathy when age at diagnosis is 30 or more years. Archives of Ophthalmology, 1984; 102: 527-532. 26. Klein, R., Klein, B. E. K., Moss, S. E., et al. The Wisconsin epidemiologic study of diabetic retinopathy. 11. Prevalence and risk of diabetic retinopathy when age at diagnosis is less than 30 years. Archives of Ophthalmology, 1984; 102: 520-526. 21. Automobile Association. Petrol Cars: Schedule of Estimating Standing and Running Costs. London: AA Technical Services, 1989. 28. Bruzelius, N. The value of travel time, London: Croom Helm, 1979. 29. Department of Transport. Valuesf o r journey time savings and accident prevention, London: Department of Transport, 1987. 30. Grey, R. H. B., Burns-Cox, C. J . and Hughes, A. Blind and partial sight registrations in Avon. British Journal of Ophthalmology, 1989; 73: 88-94. 31. Aclimandos, W. A. and Galloway, R. N. Blindness in the city of Nottingham. (1980-1985). Eye, 1988; 2: 431-434. 32. Javitt, J. C., Canner, J . K. and Sommer, A. Cost effectiveness of current approaches to the control of retinopathy in Type I diabetics. Ophthalmology, 1989; 98: 255-264. 33. Javitt, J. C., Canner, J . K., Frank, R. G., et al.

34.

35. 36. 37.

38.

39.

40. 41. 42.

51

Detecting and treating retinopathy: a health policy model. Ophthalmology, 1991; 97: 485-495. Dasbach, E. J., Fryback, D. G., Newcomb, P. A. et al. Cost-effectiveness of strategies for detecting diabetic retinopathy. Medical Care, 1991; 29: 20-39. Donaldson, C., Atkinson, A., Bond, J. and Wright, K. Should QALYs be programme-specific? Journal of Health Economics, 1988; 7: 239-257. Acton, J. P. Nonmonetary factors in the demand for medical services: some empirical evidence. Journal of PoIitical Economy, 1975; 83: 595-614. Manning, W. G., Newhouse, J. P., Duan, N. etal. Health insurance and the demand for medicaI care: evidence from a randomised experiment. The American Economic Review, 1987; 77: 251-277. Phelps, C. E. and Newhouse, J . P. Coinsurance, the price of time and the demand for medical services. Review of Economics and Statistics, 1974; 56: 334-342. Holtman, A. G. and Olsen, E. 0. The demand for dental care: a study of consumption and household production. Journal of Human Resources, 1976; 11: 546-560. Salkever, D. S. Accessibility and the demand for preventive care. Social Science and Medicine, 1976; 10: 469-475. Coffey, R. M. The effect of time price on the demand for medical care services. Journal of Human Resources, 1983; 18: 407-424. Cauley, S.D. The time price of medical care. Review of Economics and Statistics, 1987; 69: 59-66.

Screening for diabetic retinopathy: a relative cost-effectiveness analysis of alternative modalities and strategies.

Diabetic retinopathy is the most common cause of blindness among adults of working age in the UK. If the disease is detected early effective treatment...
1MB Sizes 0 Downloads 0 Views