Note: Where available, the PDF/Word icon below is provided to view the complete and fully formatted document
Diabetes prevention and care: a cost-benefit model for ranking intervention strategies. Paper prepared for the 7th Annual National Health Outcomes Conference, Canberra, 27-28 June 2001



Download PDFDownload PDF

Diabetes Prevention and Care:

A Cost-Benefit Model for Ranking Intervention Strategies

Agnes Walker

National Centre for Social and Economic Modelling (NATSEM) University of Canberra 2601

Paper prepared for the 7th Annual National Health Outcomes Conference, Canberra.

27-28 June 2001

iii

Abstract

The paper first describes the Diabetes Model which NATSEM developed in collaboration with the Diabetes Centre at the Prince of Wales Hospital, Sydney, with support from the Commonwealth Department of Health and Aged Care.

The aim was to build a model capable of assisting policy analysts with assessments of the health and cost impacts of a range of possible future diabetes care and prevention interventions. The model is of the dynamic cohort group type, covering the 1995 to 2050 period. Health outcomes are measured through the Disability Adjusted Life Years (DALY) statistic, and expenditures are estimated on a ‘per person cost of treatment’ basis.

The paper illustrates the model’s capabilities through simulation of two different types of intervention scenarios. The advantages and disadvantages of these are compared both in terms of costs and health outcomes. Finally, methods for using the model results to assist with the ranking of the scenarios are described.

iv

Author note

Agnes Walker is Principal Research Fellow at the National Centre for Social and Economic Modelling, University of Canberra.

Acknowledgments

The author is grateful to Professor Stephen Colagiuri of the Sydney Diabetes Centre for initiating and guiding the development of the model, to Michele McLennan of the same Centre for preparing and documenting the input data to the Model, to members of the Steering Committee set up under the Commonwealth Department of Health and Aged Care contract for valuable comments on earlier reports, and to Colin Mathers of the Australian Institute of Health and Aged Care for assistance with the Disability Adjusted Life Years (DALY) computations.

Abbreviations and explanations

Some of the commonly used general abbreviations in this paper are: ABS Australian Bureau of Statistics

Type 1 diabetes Insulin dependent diabetes mellitus

Type 2 diabetes Non-insulin dependent diabetes mellitus

QALYs Quality Adjusted Life Years

DALYs Disability Adjusted Life Years

Some of the commonly used explanations are:

Base case or 'do nothing' case The simulation under which current disease patterns and policies remain unchanged over the 1995 to 2050 period

NPV Net Present Value of costs (or benefits). With a discount rate of

zero, the NPV for costs is simply the sum of the year-by-year costs over the study period.

v

Contents

Abstract iii

Author note iv

Acknowledgments iv

Abbreviations and explanations iv

1 Background 1

2 Introduction 1

3 Overview of the Diabetes Model 2

3.1 Model structure 2

3.2 Elements of Diabetes Model 3

4 The Illustrative Scenarios Simulated 5

4.1 Scenario 1: Additional screening and promotion 6

4.2 Scenario 2: Improved treatment 9

5 Simulation results 10

5.1 Scenario 1: Additional screening and promotion 10

5.2 Scenario 2: Improved treatment 13

5.3 Sensitivity of results to changes in parameters 15

6 Ranking of Scenarios 15

Bibliography 18

1

1 Background

The diabetes model, the prototype of which was developed in 1997 and 1998 by NATSEM in collaboration with the Prince of Wales Hospital Diabetes Centre, was further developed and validated during 1999 under a Department of Health and Aged Care contract.

The related report was submitted to the Department in December 1999- see Walker, Colagiuri and McLennan, 1999. That report described the up-graded diabetes model, simulated a realistic intervention scenario as specified by the Steering Committee set up by the Department, and reported on tests concerning the sensitivity of model results to changes in assumptions and parameter values. At their meeting of 14 July 2000, Steering Committee members unanimously endorsed the above report.

This paper makes use of the up-graded model with a view to illustrate its capabilities in policy relevant applications.

2 Introduction

Epidemiological studies of common non-communicable diseases reported in the literature often concern the tracking of a particular aspect of a disease in already diagnosed patients - see for example Eastman et al 1997(a and b), Institute for Medical Informatics and Biostatistics 1997(a and b). However, in Australia - as in some other developed countries - a key resource allocation issue is the appropriate mix of prevention and treatment strategies, rather than the treatment of already diagnosed patients only.

In this context Richardson and Robertson (1999, pp350-1) note that there is a “lack of both epidemiological and economic evaluation which is a prerequisite to sensible decision making”. The population-wide broad ‘group’ model described in this paper aims to alleviate this situation in the case of diabetes. By helping researchers to rank combinations of prevention and care strategies - in terms of not only costs relative to benefits, but also health outcomes - the model has the potential to aid the decision making process in a quantitative fashion.

It has been proposed that the model described in this paper be eventually linked to a detailed individual-based (that is microsimulation) model able to track the three per cent or so Australians already diagnosed with diabetes. The broad ‘group’ model would then determine the numbers diagnosed under different ‘prevention’ scenarios, while the microsimulation model would follow the progression patterns of diabetes in greater detail.

2

3 Overview of the Diabetes Model

The Diabetes Model considers both Type 1 and Type 2 diabetes. Type 1 diabetes - that is insulin dependent diabetes mellitus - is usually diagnosed in people under 40 years of age. Such people produce little or no insulin themselves and require lifelong injections of insulin. Type 2 diabetes - that is non-insulin dependent diabetes mellitus - is usually diagnosed in people aged 40 years or over. Such people produce insufficient insulin and are resistant to the insulin which is produced. Walker, Colagiuri and McLennan (2001) provides details of the characteristics and prevalence of diabetes, a disease which is one of the Federal Government’s six priority health areas.

As part of its 2001-2002 Budget, the Federal Government announced a close to $50 million package over four years “to improve prevention, early diagnosis and management of diabetes - a disease affecting an estimated 900,000 Australians aged over 25, with half of these cases being currently undiagnosed” -Minister for Health and Aged Care (2001, p.2).

In developing the Diabetes Model, its structure was arranged so that all the above aspects of the disease - prevention, diagnosis and management - can be analysed, and that the benefits of interventions in any of these areas can be compared with their respective costs.

3.1 Model structure

The Diabetes Model, programmed in EXCEL97, is of the dynamic cohort group type, covering the period 1995 to 2050. The model takes as input ABS population projections by gender and by 5-year age groups. It considers the total Australian population and is able to track particular age/gender cohorts over time. It considers treatment costs for diabetes as well as the health outcomes arising from treatment. Health outcomes are measured by the Disability Adjusted Life Years (DALY) indicator, which is computed as the sum of Years of Life Lost and Years Lived with Disability.

The model and the methodoloies used aredescribed in detail in Walker, Colagiuri and McLennan (2001). The stages considered - in terms of categories of people with and without diabetes and per unit treatment costs - are summarised in Table 1.

3

Table 1: The 15 stages and treatment costs identified in the Diabetes Model

Stages Modelled

1 People without diabetes 2

Low risk High risk

People with diabetes

3 undiagnosed

4 diagnosed without complicat's 5

6

Poorly controlled Moderately controlled Well controlled

diagnosed with complications

7 early stage

8

9

10

11 end stage 12 13 14

eye (retinopathy) kidney (microalbuminuria, macroalbuminuria) limbs (neuropathy, peripheral vascular disease) heart (angina and TIA)

eye (vision threatening retinopathy, blindness) kidney (renal failure) limbs (foot ulcer and amputation) heart (MI, CABG, PTCA and stroke)

15 Death for the population overall and due to diabetes

(specified externally, ie in input data)

Treatment

Costs

Costs of treating diabetes per person, per year medical costs (doctor, hospital and drugs) for:

- each of the four 'early' and four 'end-stage' complications;

- each of the three ‘diagnosed without complications stages.

3.2 Elements of Diabetes Model

The elements of the Diabetes Model are summarised in Figure 1. The model takes as input a wide range of demographic and diabetes related information, such as ABS population projections; death rates (generally and due to diabetes); life expectancy data; prevalence rates by the stages of diabetes considered (Type 1 and Type 2); a

4

‘trend factor’ which allows for the expectation that past trends towards increases in the prevalence of diabetes will continue; the ratio of people with diabetes who have been diagnosed to those as yet undiagnosed; data on the duration of diabetes; and treatment costs by stages of disease progression (Type 1 and Type 2).

Figure 1: Elements of Diabetes Model

Next, year-by-year, the model selects from the total Australian population the number of persons in each of the 15 stages (see Table 1). It then computes, stage-by-stage, treatment costs and DALYs for those diagnosed. The methods of computation, validation and sensitivity testing of DALY estimates are detailed in Walker, Colagiuri and McLennan (2001, Chapter 6).

Finally, both treatment costs and DALYs are discounted with a view to arrive at a single summary statistic for the selected simulation period (1995 to 2050 as default). The discounting of future benefits is a standard practice in economic analysis. It was used in the Diabetes Model for computing the Net Present Values1 of the treatment cost and health outcomes streams associated with future benefits, and the implementation cost of the intervention program.

To estimate the effects of a proposed intervention scenario, the model is first run under the ‘Base case’ (or ‘Do nothing’ case) option - which assumes that current

1 The Net Present Value (NPV) of a cash stream over a period is the sum of the discounted values in each year of the period. If the discount rate is zero (which will provide an undiscounted estimate), the NPV is simply the sum of the dollar values for each year of the study period.

OUTPUTS

DIABETES MODEL

Type 1

Type 2

INPUT DATA default: Base Case

Most variables can be changed by

user

Numbers diagnosed with diabetes - for each age-gender group, by stages of diabetes - totals ( males, females, persons)

Treatment costs by stages of diabetes, 5-year intervals to 2050

Estimates of Net Present Values - for treatment costs and DALYs - using discounted chashflow

techniques

Estimates of Disability Adjusted Life Years (DALYs) - as sum of Years of Life Lost and Years Lived with Disability

5

disease patterns and policies remain unchanged over the study period - and the ‘Scenario’ - in which a proposed government intervention policy is modelled. The effects of the intervention are then estimated as the difference between the Base and Scenario simulations in terms of treatment costs and health outcomes.

By relying on the cost-benefit approach, the model is well suited to the ranking of proposed diabetes interventions. If cost benefit results from similar models on other diseases are available, then the approach also allows rankings between diabetes and other diseases.

4 The Illustrative Scenarios Simulated

Diabetes interventions worth of further investigation were listed by HAC and AIHW (1998, p.11) as:

• effective prevention strategies;

• improvements in early detection of diabetes;

• improvements in the quality of diabetes care, especially prevention programs for diabetes-related complications; and

• progress in patient management.

Since interventions of this kind attracted funding in the 2001-02 Budget (Section 3), we constructed from these alternatives the two illustrative scenarios for further analysis. The two scenarios, which for simplicity only concern people with Type 2 diabetes, are:

• Scenario 1: preventive action (additional screening so as to ‘capture’ more of those undiagnosed), and

• Scenario 2: improved treatment (so as to reduce the number of people with diabetes progressing toward complications).

Apart from being policy relevant, these two scenarios are also interesting in that they are able to illustrate many of the capabilities of the Diabetes Model. Simulations involving a single, but more realistic scenario are described in Walker A, Colagiuri S. and McLennan (1999).

6

4.1 Scenario 1: Additional screening and promotion

Reasons for measures such as prevention, early diagnosis and better management of diabetes have been given by HAC and AIHW (1998, p. 12) as “The earlier a person with diabetes is diagnosed, the sooner treatment can be started to control blood glucose levels and delay the onset and progression of many diabetes related complications.”

In developing Scenario 1, we considered a combination of programs that were able to:

• improve prevention (eg through managed modification of risk factors),

• earlier detection (eg through screening),

• encouragement of individuals to not only seek diagnosis (eg by raising community awareness), but also to individually take action to change their life styles. These latter actions could arise in response to specially targeted health promotion programs (HAC and AIHW (1998, pp. 11-12).

Since the purpose of Scenario 1 is illustrative, the above elements have not been worked out in detail at this stage. However, some indication can be given as to what would be involved with, for example, developing a nation-wide screening program. The relevant issues are discussed below.

In general, most patients become diagnosed with diabetes through opportunistic screening by health professionals (HAC and AIHW 1998, p. 12). One could envisage a national screening program targeting Australians aged 40 years or more who are also overweight or obese.2 To obtain an estimate of the maximum total cost of the national screening program, the number of persons aged 40 or more who were overweight or obesity could be multiplied by a ‘per-person-screened’ cost estimate. If the likely take-up of the program was available, then this maximum could be deflated accordingly. Costs could be further contained through, for example, targeting people who had relatives with diabetes.

For purposes of our illustrative Scenario 1 we assumed that the cost of the total intervention program (screening, raising community awareness and health promotion) was $200 million annually. This amount was assumed to be spent in each year of the study period (1995 to 2050).

2 . Mathers, Vos and Stevenson (1999, p.115) note that, in 1995, some 7.3 million adult Australians (56 per cent of the adult population) were overweight, with 2.4 million of these being obese (18 per cent of the adult population). The number of overweight or obese people were assessed using the Body Mass index (BMI) measure. The BMI was calculated as weight (kg) divided by height squared (m2). Being overweight or obese were indicated by a BMI of 25 or more, and 30 or more respectively.

7

We also assumed that individuals modified their life styles of their own accord, but as a result of greater awareness. Because of this no additional costs were associated in this Scenario with modification of risk factors.

Scenario 1 was modelled through the parameter changes shown in Table 2, with ‘optimistic’ and ‘pessimistic’ assumptions for the trend rate.

Table 2: Settings for the Base Case and Scenario 1 - Type 2

Base case Scenario

Settings of Input Dataset

Change in Prevalence rates - Males 0 +5 per cent

Change in Prevalence rates - Females 0 +1 per cent

Ratio of Undiagnosed to Diagnosed - Males 1 0.90

Ratio of Undiagnosed to Diagnosed - Females 1 0.82

Ratio of those Without Complications in All Diagnosed 0.4 0.5

Trend factor (per cent per annum) ‘optimistic’

‘pessimistic’

2.0

-

1.5

2.0

Diabetes as share of all deaths 0.02 0.01

Prevalence - early complications 20% lower

Prevalence - late complications 30% lower

Costs

Cost of intervention per year ($ million, 1995 prices) - 200

Table 2 shows that, to model the effects of Scenario 1, estimates3 needed to be made regarding the impact of the additional screening strategy on:

3 In developing actual Scenarios, such estimates will in general be based on the findings of earlier research (eg clinical trials, existing screening programs, etc).

8

• prevalence rates for diabetes (both in general and for complications);

• the ratio of undiagnosed to diagnosed persons with diabetes;

• the ratio of persons without complications in all diagnosed;

• the share of deaths from diabetes in all deaths - this only affecting the Years of Life Lost part of DALYs; and

• the trend factor which accounts for the expectation that the past trend of increasing prevalence for diabetes will continue (albeit at a lower rate under the Scenario than in the Base case).4

Nationwide organised screening is expected to result in earlier diagnosis, leading to higher prevalence rates (10 per cent higher for women and 5 per cent for men, as men have been shown internationally to be less responsive to government health initiatives).

Widespread screening of the targeted population would result in a decline in the number undiagnosed (by 10 per cent for men and 18 per cent for women). Also, more Australians ‘at risk’ would be diagnosed earlier. As more people with diabetes are diagnosed - and thus come under medical supervision - their treatment commences earlier. This leads to a reduction of the frequency and extent of complications - assumed to be 20 and 30 per cent lower for early and end-stage complications respectively (Table 2).

Taking an optimistic outlook, the ‘trend factor’ is likely to be lower under Scenario 1 than in the Base case (declining from 2 to 1.5 per cent a year). Behind these numbers lies the assumption that individuals will modify their life styles of their own accord - eg as a result of their now greater their health awareness. Because of this no additional costs are associated with the modification of risk factors under that assumption. However, at the pessimistic end of the projections it could be assumed that implementation of the Scenario would not have any impact on behaviour as far as risk factors were concerned. In that case the ‘trend rate’ was assumed in the simulations to remain unchanged from that of the Base case (that is remain set at 2 per cent per annum) - Table 2.

The overall effect is that, although under Scenario 1 more people are diagnosed (placing an upward pressure on treatment costs), the number of patients with complications are significantly lower (placing a downward pressure on treatment

4 This trend, which in the Base case arose mainly from increased prevalence of risk factors such as obesity and inactivity - see McCarty et al (1996, p.23) - will now be lower as obesity and inactivity are less of a problem under Scenario 1.

9

costs). Thus, the model quantifies the net effect of these two opposing trends in terms of direct medical costs (ie more people being diagnosed, but less people with complications). In addition the model estimates differences in health outcomes between the Scenario and the Base case through the DALY indicator.

Better community awareness would result in more people adopting an improved diet and lose (or maintain) weight. Encouraged by the specially targeted health promotion programs, more Australians at risk would exercise regularly.

The overall effect would be that, although under Scenario 1 more people would be diagnosed (placing an upward pressure on treatment costs), the number of patients with complications would be significantly lower than without the intervention (placing a downward pressure on treatment costs).

In carrying out the simulation for Scenario 1, the model quantifies the net effect of these two opposing trends in terms of direct medical costs (ie more people being diagnosed, but less people with complications) and compares such costs between the Base case and Scenario 1. In addition the Model estimates differences in health outcomes between the Scenario and the Base case through the DALY indicator.

4.2 Scenario 2: Improved treatment

In Scenario 2 we focus on improvements in the quality of diabetes care. Although HAC and AIHW (1998) note that special consideration should be given to prevention programs for diabetes-related complications, for illustrative purposes Scenario 2 will simply consider more diagnosed patients without complication being in the well and moderately controlled groups under Scenario 2 than in the poorly controlled group. While such shifts are known to lead to a lowering in the number of people with complications, we did not take account of this in this study.

Table 3 shows that the only difference between the Base case and Scenario 2 is that the shares of the ‘poorly’, ‘moderately’ and ‘well’ controlled - amongst those diagnosed without complications - had changed.

While 40 per cent of the ‘no complications’ group was ‘poorly controlled’ in the Base case, only 12.5 per cent will be ‘poorly controlled’ under Scenario 2. Similarly, while only 35 per cent of the group was ‘well controlled’ in the Base case, 47.5 per cent will be ‘well controlled’ under Scenario 2. A somewhat greater change will occur in the case of those ‘moderately controlled’, their share increasing from 25 per cent in the Base case to 40 per cent under Scenario 2.

Compared with Scenario 1, Scenario 2 will be a low cost option because programs aiming to improve the quality of care only need to concern health professionals.

10

Thus the costly effort of ‘mobilising’ a high proportion of the Australian population is avoided. We assumed for this illustrative example an intervention cost of $4.5 million per year (1995 dollars).

Table 3: Settings for the Base Case and Scenario 2 - Type 2

Base case Scenario

Settings of Input Dataset

Share of poorly controlled in all ‘diagnosed without complications’ Males and Females 0.40 0.125

Share of moderately controlled in all ‘diagnosed without complications’ Males and Females 0.25 0.400

Share of well controlled in all ‘diagnosed without complications’ Males and Females 0.35 0.475

Prevalence - early complications no change

Prevalence - late complications no change

Costs

Cost of intervention per year ($ million, 1995 prices) - 4.5

5 Simulation results

5.1 Scenario 1: Additional screening and promotion

Treatment costs

Table 4 shows the sum of the discounted stream of medical costs, or Net Present Value (NPV) over the 1995 to 2050 period for the Base case and Scenario 1, by stage of diabetes, using a discount rate of 3 per cent. The NPV for the medical costs associated with the treatment of all Australians diagnosed with diabetes was estimated at just under A$36 billion for the Base case and around A$27 billion under Scenario 1 - a saving of around A$8.5 billion over the 1995 to 2050 period (or A$154 million a year).

11

Table 4: Direct medical costs, Base case and Scenario 1, Net Present Value over 1995 to 2050 period (discounted at 3 per cent, using 1.5% pa trend rate) Base case Scenario 1 Difference

($ million) ($ million) (%)

1 2 (2-1)/1

End stage - eyes 3436 2381

End stage - kidney 2841 1968

End stage - limbs 3326 2304

End stage - heart 12736 8791

Tot end stage 22339 15445 -30.9

Early comp - eyes 2782 2203

Early comp - kidney 3111 2464

Early comp - limbs 2915 2308

Early comp - heart 1903 1502

Tot early stage 10712 8477 -20.9

No comp - poorly cont'd 1231 1523

No comp - moderately cont'd 676 837

No comp - well cont'd 819 1013

Tot without compl's 2726 3373 23.7

Tot with compl's 33050 23922 -47.5

Total persons - All Diagnosed 35776 27296 -23.7

Next, this saving in medical costs - arising from implementation of Scenario 1 - needs to be compared with the cost of the intervention itself (ie A$200 million a year summed over the study period). In NPV terms, the cost of implementing Scenario 1 amounts to A$6 billion, implying a net saving of 8.5 - 6.0 = A$2.5 billion (or A$45 million a year) - Table 5.

The ratio of savings to ‘intervention costs’ is thus 8.5/6.0 = 1.4, indicating that - under the assumptions of a favourable change in trend rate and a 3 per cent discount rate - the savings would more than pay for the costs of implementing Scenario 1. However, Table 5 shows that the results are quite sensitive to these assumptions in that the ratio of savings to ‘intervention costs’ can fall below ‘one’ (ie 0.87 per cent) under the assumptions of ‘no change in the trend rate’ and a discount rate of 5 per cent.

12

Table 5: Direct Costs and Benefits of Scenario 1, 1995-2050, varying discount and trend rates (1995 prices)

Scenario1, trend 1.5% pa,

NPV 3%

($ million)

Scenario 1, trend 1.5% pa,

NPV, 5%

($ million)

Scenario 1, trend 2% pa,

NPV, 5%

($ million)

Change between the Scenario and the Base case

8,480 (154 pa)

5058 (92 pa)

3816 (69 pa)

Cost of Intervention

(at $200 million per year, current prices)

6,043 (110 pa)

4372 (79 pa)

4372 (79 pa)

Ratio of savings in medical costs to intervention cost

1.40

1.16

0.87

Because the results are sensitive to the assumptions adopted, rather than using the yardstick of whether the savings are above or below the cost of implementing the Scenario, it is preferable to study the ranking of the various Scenarios. This is because in most practical instances the rankings will be considerably less sensitive to changes in assumptions than the ‘savings to implementation costs’ ratio. Relying on rankings would thus help decision makers to identify the most advantageous options, as well as the least beneficial ones.

The break-even intervention cost for Scenario 1- at which the stream of savings in medical costs over the study period equals the stream of annual intervention costs - is around A$280 million per year (discounted at 3 per cent with 1.5 per cent trend rate), A$230 million per annum (discounted at 5 per cent with 1.5 per cent trend rate) and A$180 million a year (discounted at 5 per cent with 2 per cent trend rate). These findings suggests that the less favourable the assumptions, the lower will be the annual expenditures that can be justified in terms of medical cost savings alone.

Health outcomes

The above dollar-based rankings do not account for the human side of proposed interventions - that is the actual health gains that lie behind earlier diagnosis and less disabling complications. Below an attempt is made to quantify aspects of these non-monetary effects (Table 6).

13

Table 6: Net Present Value of DALYs, 1995-2050, Base case and Scenario 1

NPV Base NPV Scenario 1

Difference (DALYs) Difference (%)

undiscounted 6,139,632 4,588,448 -1,551,184 -25.3

discounted 3% 1,943,365 1,443,008 -500,357 -25.7

Table 6 shows that the undiscounted Net Present Value of DALYs was 6,139,632 for Type 2 diabetes in the Base case (covering the 1955 to 2050 period), or 111,630 per year on average. By comparison, the estimate for 1996 by Mathers, Vos and Stephenson (1999, p. 241) was 95,917. 5

The Net Present Value of the health improvements arising from implementation of Scenario 1 amounted to 1.55 million DALYs (or 28,203 a year) if undiscounted, and 0.5 million DALYs (or 9091 a year) if discounted at 3 per cent (Table 6). In both cases there was a decline in DALYs, corresponding to an improvement in health outcomes of around 25 per cent.

According to the model’s estimates, the 28,203 a year DALY health gain was composed of 17,726 less Years of Life Lost and 10,477 less Years Lived with Disability. Had we used a discount rate of 3 per cent, these numbers would have been around two-thirds lower.

The model estimated that the reduction in the prevalence of complications has led, between 1995 and 2000, to a decline in the number of end stage eye complications from 36,218 in the Base case to 26,579 under Scenario 1. In terms of health outcomes this means that some 9,200 less people had vision threatening retinopathy, and 460 less people were blind, than there would have been without the intervention. Similar statistics can be obtained from the model for all types of diabetes complications and for all simulation periods.

5.2 Scenario 2: Improved treatment

Treatment costs

Under Scenario 2, only for patients without complications was there a cost difference between the Base case and the Intervention scenario. The model results show that the

5 Our corresponding estimate is the per year figure of total undiscounted DALYs in the model’s first time period (ie 1995 to 2000). This amounted to 96,315 DALYs for Type II diabetes, which is very close to the Mathers, Vos and Stevenson estimate of 95,917. Walker, Colagiuri and McLennan (2001) describe in greater detail how the model was validated re DALY computations.

14

NPV of the medical costs of treating those without complications declined by 3.5 per cent relative to the Base case, using a 3 per cent discount rate (Table 7). The 3.5 per cent decline translated into a 0.3 per cent lower NPV for the cost of treating all those diagnosed with diabetes.6

At A$148 million (35,776 less 35,628) - or A$2.7 million a year - the savings in medical costs are much smaller for Scenario 2 than for Scenario 1. However the NPV of the A$4.5 million a year intervention cost is also considerably smaller, amounting to A$98.4 million with a discount rate of 3 per cent. Thus, for Scenario 2, the ratio of medical cost savings to the cost of the intervention is: 148/98.4 = 1.5.

The above results correspond to the situation where a continuous injection of A$4.5 million a year is required over the 55 year long simulation period. However, the model could also be used to estimate what the benefits would be if the Scenario 2 program was only effective over a 20 year period (ie to 2015).

Table 7: Direct medical costs, Base case and Scenario 2, Net Present Value over 1995 to 2050 period (discounted at 3 per cent, using 1.5% pa trend rate) Base case Scenario 2 Difference

($ million) ($ million) (%)

1 2 (2-1)/1

End stage - eyes 3436 3436

End stage - kidney 2841 2841

End stage - limbs 3326 3326

End stage - heart 12736 12736

Tot end stage 22339 22339 0.0

Early comp - eyes 2782 2782

Early comp - kidney 3111 2464

Early comp - limbs 2915 2915

Early comp - heart 1903 1903

Tot early stage 10712 10712 0.0

No comp - poorly cont'd 1231 813

No comp - moderately cont'd 676 705

No comp - well cont'd 819 1111

Tot without compl's 2726 2630 -3.5

Tot with compl's 33050 33050

Total persons - All Diagnosed 35776 35680 -0.3

6 This should be seen as an underestimate since the additional cost savings due to a lower level of complications (arising from better ‘control’ of the group without complications), have not been simulated in Scenario 2

15

One could hypothesise that this may occur because by 2015 the improved Scenario 2 ‘control’ shares for the ‘no complications’ group would stabilise, removing the need to continue with the A$4.5 million a year intervention program. Beyond 2015 there would be no further costs and no further benefits. Under this alternative - ie Scenario 2 operational between 1995 and 2015 only - the Diabetes Model predicts a ‘medical cost savings to intervention costs’ ratio of 0.90 (3 per cent discount rate).

Health Outcomes

Because we assumed that under Scenario 2 there would be no impact on complications, the DALYs under Scenario 2 will not be different from those in the Base case. In other words, the intervention will not affect the number of people with diabetes becoming blind, losing a limb or suffering from kidney failure.

5.3 Sensitivity of results to changes in parameters

Except for changing in some instances the discount rate, the trend rate and the length of the simulation period, we did not in this paper test the sensitivity of the model results to changes in a range of key parameters. However, systematic sensitivity testing can be very time consuming because the user is able to change a large number of parameters in the model’s input dataset.

Although in one instance we illustrated the more economical alternative of developing a ‘most optimistic’ and a ‘most pessimistic’ set of parameters - and through these indicated the band within which the model results were likely to fall - when applying the model to actual policy analysis it would be preferable to carry out a broad range of sensitivity tests. This is because, without adequate sensitivity testing, it is not possible to assess the robustness of the rankings produced by the model.

6 Ranking of Scenarios

In this Section the results presented earlier are brought together, so that an example can be given as to how the relative importance of the two Scenarios could be assessed. The reader should bear in mind that, though not unrealistic, the Scenarios have been developed for illustrative purposes only.

The results for Scenario 1 were driven by the assumed reductions in the complication rates and the relative costs of treating the various complications. By contrast in Scenario 2 it was the extent of the shift across the ‘poorly’, ‘moderately’ and ‘well’

16

controlled groups that drove the results, together with the differences in the costs of treating the three ‘no complications’ groups. As seen earlier, while the cost differences between the three groups were relatively minor, the model still estimated considerable benefits under Scenario 2. This was because the specified cost of the intervention was also relatively low. Table 8 summarises the findings for both Scenarios.

Table 8: Summary of Costs and Benefits, Scenarios 1 and 2, 1995-2050 (1995 prices, 3% discount rate, using 1.5% pa trend rate)

medical costs,NPV ($ million) NPV, health outcomes (DALYs)

Scenario 1

Change between Scenario 1 and the Base case ($ million)

8,480

-500,305

Cost of Intervention (at $200 million per year)

6,043

na

Ratio of savings in medical costs to the cost of the intervention

1.4

na

Scenario 2

Change between Scenario 2 and the Base case

96

0.0*

Cost of Intervention (at $4.5 million per year)

98.4

na

Ratio of savings in medical costs to the cost of the intervention

1.5

na

na not applicable * by assumption

The results for both the big budget Scenario 1 and the small budget Scenario 2 suggested that savings in medical costs well above the cost of the intervention were possible, but that such savings could easily fall below the cost of the intervention under alternative assumptions. Using the ‘default’ parameter settings with a discount rate of 3 per cent, the ratio of savings to intervention costs for Scenario 1 was estimated at 1.4, and that of Scenario 2 at 1.5 (Table 8). On that basis - plus the fact that Scenario 2 would cost a great deal less to implement - one might be tempted to rank Scenario 2 ahead of Scenario 1. However, Scenario 1 would also result in significant improvements in health outcomes, while Scenario 2 would have no such improvements (by assumption, for illustrative purposes only).

17

Indeed, the reduction in the NPV of DALYs by around 500,000 (or 9091 per year) under Scenario 1 meant that, under that option, many Australians would have lived longer - or suffered less disability - than would have been the case if the Scenario 1 intervention did not eventuate.

So would Scenario 2 be really preferable to Scenario 1?

To resolve this issue, policy analysts may attempt to put a monetary value on DALYs (by assigning an average dollar value to each year of life lost and each year lived with disability). However, because of the difficulty of estimating such monetary values, once again a range of values may need to be analysed.

Overall, although the simulations carried out by the Diabetes Model would assist decision makers in their efforts to rank various intervention possibilities, the final decision would still require considerable judgement.

18

Bibliography

ABS (Australian Bureau of Statistics), 1997a, 1995 National Health Survey - Diabetes, Australia, ABS Cat No 4371.0, Canberra, AGPS.

—— 1997b, Deaths, Cat No 3302.0, Canberra, AGPS

—— 1996, Projections of the Populations of Australia, States and Territories: 1995-2051, ABS Cat No 3222.0, Canberra, AGPS.

Colagiuri S., Colagiuri R and Ward J., (1998), National Diabetes Strategy and Implementation Plan, Diabetes Australia, Canberra.

Eastman R, Javitt J, Herman W, Dasbach, E et al, 1997a, Model of Complications of NIDDM - Model Construction and Assumptions Diabetes Care; 20: 725-734.

Eastman R, Javitt J, Herman W, Dasbach E et al. 1997b, Model of Complications of NIDDM - Analysis of the Health Benefits and Cost-effectiveness of Treating NIDDM with the Goal of Normoglycemia, Diabetes Care; 20: 735-744.

Goodin, R., 1982, Discounting Discounting, Journal of Public Policy, 2:53-72.

HAC (Commonwealth Department of Health and Aged Care) and AIHW (Australian Institute of Health and Welfare), 1998, Diabetes Mellitus - Summary, AIHW Cat No PHE 13, Canberra.

Department of Health and Family Services (1997), First Report on National Health Priority Areas 1996, Canberra.

Institute for Medical Informatics and Biostatistics (1997a), ACCU-SIM 2 - an Overview, Switzerland.

—— (1997b), ACCU-SIM 2 - Help File Contents, Switzerland.

Krahn, M. and Gafni, A. (1993), 'Discounting in the Economic Evaluation of Health Care Interventions', Medical Care, Vol 31, No 5, pp 403-418.

Mathers, C., Vos, T. and Stevenson, C, 1999, The Burden of Disease and Injury in Australia, Australian Institute of Health and Welfare, AIHW Cat No PHE-17, Canberra, AIHW.

McCarty DJ, Zimmet P, Dalton A, Segal L and Welborn TA., (1996), The Rise & Rise of Diabetes in Australia, 1996: a Review of Statistics, Trends and Costs, Diabetes Australia National Action Plan 1996, Canberra.

Minister for Health and Aged Care 2001, Media release, 22 May.

Murray, C. and Lopez, A., 1996, The Global Burden of Disease, published by the Harward School of Public Health on behalf of the World Health Organization and the World Bank, Harvard University Press

Richardson, J. and Robertson, I. 1999, ‘Ageing and the Cost of Health Services’, Proceedings of Conference on The Policy Implications of the Ageing of Australia’s Population, Productivity Commission and Melbourne Institute, Melbourne, 18-19 March.

19

Walker, Colagiuri and McLennan 2001, Cost-Benefit Model of Diabetes Prevention and Care: Model Construction, Assumptions and Validation, Technical Paper, National Centre for Social and Economic Modelling, University of Canberra, forthcoming.

Walker A, ‘Type 2 Diabetes - Modelling Prevention and Care Options’, paper presented at the ‘Diseases of Ageing’ session of the Science of Ageing Conference convened by the National Ageing Research Institute, 29-30 November 1999, Melbourne.

Walker A, Colagiuri S. and McLennan M, Diabetes Model: Effects of an Intervention Strategy, Report to the Commonwealth Department of Health and Aged Care, Australia, 6 December 1999.

World Health Organisation (1985), Diabetes Mellitus: Report of WHO Study Group, World Health Organisation Technical Report Series 727, Geneva.