Evaluation of the Costs and Outcomes From Changes in Risk Factors in Type 2 Diabetes Using the Cardiff Stochastic Simulation Cost- Utility Model (DiabForecaster)

By McEwan, Phil; Peters, John R; Bergenheim, Klas; Currie, Craig J

Key words: Cost-effectiveness – Cost-utility – Economics – QALY – Risk factors – Type 2 diabetes

ABSTRACT

Aims: The aim of this study was to determine the mean costs and outcomes associated with modifiable risk factors in patients with type 2 diabetes and to determine equivalent changes to these risk factors in terms of financial costs and health outcomes.

Methods: The Cardiff Stochastic Simulation Cost-Utility Model (DiabForecastei), which evolved from the Eastman model, was used to follow a cohort of 10 000 patients over 20 years.

Results: Costs were affected most significantly by changes in the total cholesterol to HDL cholesterol (Total-C:HDL-C) ratio and in HbA^sub 1c^. Unit increases in Total-C:HDL-C increased discounted costs by 200 per patient; for ratios > 8 units, unit increases led to cost increases of 300 per patient. Unit increases in HbA^sub 1c^ increased per patient discounted costs from 200 (5-6%) up to 2900 (10-11%). Similar patterns were observed for QALYs. Estimates of equivalence showed that a 1% reduction in HbA^sub 1c^ was equivalent to an 0.4 increment in QALYs, which was equivalent to a reduction of 44 mmHg in SBP, 18.2 mg/dL in HDL, 100 mg/dL in total cholesterol or 1.8 units of Total-C:HDL-C ratio. A 1% reduction in HbA^sub 1c^ was also equivalent to 108 less cost, which was equivalent to a 13.0 mmHg decrease in SBP or a 0.57 unit decrease in the Total-C:HDL-C ratio.

Conclusions: This model provides reliable utility estimates for diabetic complications and may eliminate uncertainty in cost- effectiveness analyses of treatment. These data also provide a novel way of comparing the value of treatments that have multiple effects.

Introduction

The evaluation of alternative treatments for diabetes requires long-term forecasting of costs and outcomes using modelling techniques, and a number of groups have developed such models1-4. Many of these models have been compared and contrasted at the ‘Mount Hood Challenge’5, which provides a forum for comparing the costs and clinical events predicted by different models using the same input profiles. This process has identified areas where models disagree and factors that lead to disagreement. Largely because of the complexity of diabetes and its manifestations, there are still refinements to be made to these models; for example, costs prior to mortality6, costs and outcomes of multiple complications7, and importantly for economic evaluation, inclusion of reliable estimates of health utility in various complication states. Although several refinements are yet to be made, existing models have been validated and can answer important questions associated with the care and treatment of diabetes.

At present, modifiable risk factors that are characteristic of the insulin resistance syndrome (IRS), including glucose control, blood pressure and lipid profiles8, are used to predict long term outcomes but are likely to evolve over time8,9. Models incorporating these risk factors are driven by various risk functions including those developed by the Framingham investigators10 and those from the United Kingdom Prospective Diabetes Study (UKPDS)11,12. The cost of achieving therapeutic targets, such as blood pressure or lipid lowering, can be specifically associated with single or multiple interventions. Whether these targets are achieved through pharmacotherapy or lifestyle changes, the various modifiable risk factors are associated with characteristic cost and outcome consequences.

The purpose of this study was to determine the mean cost and outcome consequences of particular biochemical states in subjects with type 2 diabetes and to determine how these factors interact in terms of cost and outcomes. These data should give a clear indication of the most cost-effective population treatment strategy. An additional objective was to address a key deficiency in existing economic models; i.e. the use of unreliable utility estimates, particularly with respect to multiple complication states.

Methods

Simulation model

This study provides an evaluation of both cost-effectiveness and cost-utility in diabetes using the Cardiff Diabetes Cost Utility Model (DiabForecaster), a discrete event stochastic simulation model written in Visual Basic and embedded within Microsoft Excel. The fundamental structure of the model was based on the Eastman DCCT model3 which, by default, forecasts costs and outcomes over a 20- year period. Table 1 details the baseline characteristics for the simulation cohort taken from the Eastman publication and a schematic representation of the model is shown in Figure 1. The major revisions to the Eastman model related to the cardiovascular module, which was updated to accommodate the UKPDS risk engines to predict coronary heart disease (CHD)11 and stroke12 events. All-cause mortality was modelled using UK 2001-2003 gender-specific interim life tables. The model retained the original structure and transition probabilities for modelling the progression of microvascular complications detailed by Eastman and colleagues. The model can be run in two modes: by default it follows 10000 newly diagnosed diabetic patients over 20 years, updating in yearly increments; however, it can also run up to 60 years with a user- specified annual incidence of type 2 diabetes. A steady state was typically achieved after 40 years of iterations.

Data sources

Health-related utility was measured using EuroQol (EQ-5D) data obtained from the Health Outcomes Data Repository (HODaR), which contains data from patients discharged from the third largest hospital trust in the UK. The methods underlying HODaR have been described elsewhere13. In brief, patients in Cardiff and the Vale of Glamorgan were sent a questionnaire six weeks following discharge from hospital between January 2002 and July 2003. Patients with diabetes were identified by a previous history of inpatient admission where diabetes was recorded as a coexisting diagnosis. Utility estimates were obtained for individuals with no complications and for patients with the following single complications: myocardial infarction (MI), stroke, peripheral vascular disease (PVD; with and without amputation), end-stage renal disease (ESRD), symptomatic retinopathy and severe vision loss. Patients were also categorised by the number of complications they developed. The utility values used in the model are shown in Table 2.

Table 1. Baseline characteristics for the simulation cohort

Figure 1. Flow diagram of the Cardiff stochastic simulation cost- utility (DiabForecaster) model. CV, cardiovascular; MI, acute myocardial infarction. Simulation endpoints: a. Cardiovascular endpoints: fatal and non-fatal myocardial infarction; fatal and non- fatal stroke. b. Retinopathy health states: no retinopathy; background retinopathy; proliferative retinopathy; macular edema; severe vision loss. c. Nephropathy health states: no nephropathy; microalbuminuria; gross proteinuria; end-stage renal disease. d. Neuropathy health states: no neuropathy; symptomatic neuropathy; lower extremity amputation

As primary complications occurred within the simulation they were assigned the relevant utility value. The more prevalent complication states had a sufficiently large number of direct observations so that direct utility estimates could be used for subjects with secondary complications (highlighted in bold in Table 2). In the cases where there were insufficient patient numbers for reliable point estimates, the impact of multiple complications on quality of life was determined via an additional decrement in utility. Subjects within HODaR were classified as having one, two, or three or more macrovascular or microvascular complications and average utility values obtained. The difference in average utility for patients in each category was determined, as detailed in Table 2. For example, the observed decrement in utility from one to two diabetic complications was 0.496 – 0.446 = 0.05. Therefore a person with an MI and also severe vision loss would have 0.05 subtracted from their initial utility value of 0.661.

Costs

All macrovascular and microvascular costs were broken down into the costs associated with the event in the year in which it occurred and the subsequent maintenance costs. The majority of costs were obtained from the UKPDS14: average costs for a representative individual from the UKPDS cohort were used. Symptomatic PVD costs were obtained using Cardiff data15 and ESRD costs were also obtained locally (see Table 3). Drug costs from the British National Formulary16 were utilised for lipid-lowering, anti-hypertensive and anti-diabetic therapies. Costs were discounted at 6% per annum and benefits at 1.5% per annum.

Table 2. Mean utility estimates for single and multiple complication states in type 2 diabetes used in the simulation model

Table 3. Unit cost of clinical events, treatment and maintenance costs used in the simulation model

Model output

The main outputs of the model were total costs, total number of clinical events and quality-adjusted life years (QALYs). The model was run using published demographic profiles and baseline modifiable risk factors detailed in Table 1. One-way sensitivity analysis was conducted by increasing each parameter in turn and re\cording the impact on expected cost, QALYs and the predicted number of clinical events.

Validation

The model was constructed using an established structure with slight modification to accommodate more appropriate risk engines for cardiovascular endpoints. Face validation of the model was conducted to ensure correct logical functioning. The model output was validated by comparing how well the model reconstructed data from the UKPDS17, Eastman3 and Cardiff18-20. For validation purposes, the baseline profiles shown in Table 1 were adjusted to match those associated with the relevant validation study.

Results

Validation

The model was run with baseline demographic profiles observed in the UKPDS cohort and glycated haemoglobin (HbA^sub 1c^) set to 6.5%. Consistent incidence rates per 1000 person years were produced for fatal and nonfatal MI (13.0 vs. 12.3), stroke (5.2 vs. 5.0) and all cause mortality (15.8 vs. 15.5) for the UKPDS observations and DiabForecaster, respectively (Figure 2). Output from DiabForecaster also compared well to findings from the Eastman model. Cumulative incidences of microvascular complications are shown alongside the output from the Eastman cohorts in Figure 3. When the model was run to steady state, the output compared well with the annual incidence of macrovascular events and mortality actually observed in Cardiff. The annual event rates (per cent) observed in Cardiff compared with forecast event rates (total events/total person years) were 1.2 vs. 1.5 for MI, 1.8 vs. 1.9 for stroke and 4.2 vs. 4.3 for mortality, respectively (Figure 4).

Figure 2. Model validation: a comparison of the incidence of cardiovascular outcomes from the Cardiff stochastic simulation cost- utility model (DiabForecaster) to observed data in the UKPDS study. UKPDS: United Kingdom Prospective Diabetes Study

Figure 3. Model validation: comparison of microvascular outcomes from the Cardiff stochastic simulation cost-utility model (DiabForecaster) to observed data in the Eastman study. BDR: background diabetic retinopathy; PDR: proliferative diabetic retinopathy; ME: macular oedema; Vision Loss: retinopathy or blindness; MA: microalbuminuria; GPR: gross proteinuna; ESRD: end- stage renal disease; LEA: lower extremity amputation

Figure 4. Model validation: a comparison of cardiovascular outcomes of the Cardiff stochastic simulation cost-utility model (DiabForecaster) to those observed in real-life in the Cardiff population

Costs and outcomes

The forecast discounted costs and QALYs are listed in Table 4 along with the number of expected clinical endpoints. Costs were affected most significantly by changes in the total cholesterol to HDL cholesterol (Total-C:HDL-C) ratio and by changes in HbA^sub 1c^. For each unit change in the Total-C:HDL-C ratio, mean discounted costs per patient increased by 200; for ratios > 8 units, a one unit increase in the ratio led to a discounted cost increase of 300 per patient. Unit (per cent) increases in HbA^sub 1c^ increased costs per patient from 200 (5% to 6%) up to 2900 (10% to 11%). Changes in systolic blood pressure (SBP) of 10 mmHg brought about a consistent 100 average increase in total costs.

Similar patterns were observed for QALYs. From a baseline QALY of 10.3 there was an average 0.2 decrement in QALYs for unit increases in the Total-C: HDL-C ratio. Unit changes in HbA^sub 1c^ brought about an average 0.4 unit decrement and a 10 mmHg increase in SBP brought about an 0.1 unit decrement in QALYs.

Fatal and non-fatal MIs were affected most notably by changes in the Total-C:HDL-C ratio. The inclusion of an HbA^sub 1c^ parameter in the CHD risk engine resulted in an increase in predicted MIs for corresponding increases in HbA^sub 1c^, but a decrease in stroke events due to increased cardiovascular mortality. Elevations in SBP predicted similar increases in MI and stroke events. Increases in the number of microvascular complications resulted almost entirely from changes in HbA^sub 1c^. Decreases in blindness and amputation observed with increasing Total-C:HDL-C ratio and SBP were due to increased mortality.

Table 4. One-way sensitivity analysis of the effect of alternative mean Total-C:HDL-C ratio, SBP and HbA^sub 1c^ values on event rates in patients with type 2 diabetes (note: interpretation should take survival effects into consideration)

Contour plots for discounted costs and QALYs as a function of changes in SBP, Total-C:HDL-C ratio and HbA^sub 1c^ levels set at 7% and 10% are shown in Figure 5. These plots highlight how the rate of change in costs and health benefits is greatest as a function of changes to lipid values and HbA^sub 1c^ using these risk engines.

Estimates of equivalent changes in alternative measurements in terms of costs and QALYs for a recognised change in one of the selected parameters are listed in Table 5. For example, a 1% reduction in HbA^sub 1c^ is equivalent, in terms of QALYs, to a reduction of 44 mmHg SBP, 18.2 mg/dL HDL, 100 mg/dL total cholesterol and 1.8 units of Total-C:HDL-C ratio; each of these changes would result in a mean increase of 0.4 QALYs over the simulation period. In terms of cost, however, a 1% reduction in HbA^sub 1c^ is equivalent to 108 less in cost, which is equivalent to a 13.0 mmHg decrease in SBP and a 0.57 unit decrease in the Total- C: HDL-C ratio.

Figure 5. Mean 20-year discounted costs and quality adjusted life years as a function of lipids (total cholesterol to HDL-cholesterol ratio), systolic blood pressure (SBP) and glycaemia (HbA^sub 1c^)

Table 5. Estimation of equivalent changes in terms of costs and QALYs from changes in alternative modifiable risk factors in patients with type 2 diabetes

Discussion

This study utilised established modelling methods that were extended to include more appropriate cardiovascular risk functions and estimates of health utility associated with single and multiple complications states. The purpose was to evaluate the relative impact in terms of costs and outcome – here QALYs – from changes to the accepted modifiable risk factors associated with type 2 diabetes.

Two other groups have attempted a similar exercise: Palmer et al.21 and a group from the Center for Disease Control (CDC)22. These studies were different in their objectives to those in this study, making direct comparison of our findings difficult. The first study, the most comparable, concluded that a 10% improvement in HbA^sub 1c^ had the most beneficial effect compared with a 10% reduction in SBP or a 10% reduction in total cholesterol.

The inclusion of health-utility values in long-term economic models of type 2 diabetes has previously been problematic owing to the scarcity of data describing utility in patients with multiple complications. Ideally, point estimates of utility should be calculated for patients with each permutation of possible complications and these values used in the model. In practice, however, only utility values associated with primary complications have been available, and it is common for the effects of multiple complications to be ascertained by multiplying or adding relevant utility values. While the incidence of further complication will inevitably affect patient quality of life, there is no evidence to support this multiplicative or additive reduction. The approach taken here was to quantify the reduction in quality of life associated with the overall number of complications and use this information to adjust the patient’s initial utility score. Since this method was based on observational data, it eliminated flawed assumptions from a crucial component of this type of economic model. Although some utility estimates are available, e.g. from Bagust using the CODE-2 data23 and the UKPDS24, both study designs were unlikely to evaluate type 2 patients with complex multiple complication profiles. CODE-2 was a type 2, general-practice-based study, and the UKPDS included only type 2 patients who were newly diagnosed. Furthermore, the HODaR data that these utility estimates are based on13 have been used to measure utility (EQ5D.ndeJ, and quality of life (SF36) and other outcome measures, in an increasingly large range of published scientific studies; e.g. obesity25, hypoglycaemia26, urological problems27 and renal failure28. Thus these generic health outcome values are uniquely and directly comparable to output from these other HODaR-based studies.

The model utilised the UKPDS risk engines to model CHD and stroke. Differences in the parameters included in these equations affect the predicted number of macrovascular events. Changes to the Total-C: HDL-C ratio had a greater effect on MI than stroke, while comparable effects were recorded for similar changes in SBP. The most obvious difference between the two risk equations was the inclusion of a parameter for glycaemia (HbA^sub 1c^) in the CHD equation only. Performing one-way sensitivity analysis emphasised this and resulted in an increase in fatal and non-fatal CHD events from baseline, but a reduction in stroke events. This reduction was a consequence of fewer people left alive in the model to develop stroke.

In terms of recommendations for future research, these findings are considered to be quite revealing. It would be useful for other groups to replicate this modelling exercise and then to compare their findings, in a similar fashion to the Mount Hood Challenge5.

Conclusion

Diabetes is a complex disease and many factors contribute to the increased morbidity and mortality seen in diabetic patients. By providing reliable utility estimates for the various diabetic complication states, the model described here may eliminate an area of uncertainty in the proper determination of the cost- effectiveness of treatment in this group of patients. Furthermore, this study gives an important insight into the relative impact on costs and outcomes of the various biochemical targets for therapy in type 2 diabetes, and provides a use\ful tool to estimate the impact of interventions that have multiple biochemical effects.

Acknowledgements

These data were presented at the 9th Annual International Society for Pharmacoeconomics and Outcomes Research Annual International Meeting, Arlington, VA, USA, 16-19 May 2004.

Declaration of interest: This study was supported by a research grant from AstraZeneca, Mlndal, Sweden.

References

1. Bagust A, Hopkinson PK, Maier W, et al. An economic model of the long-term health care burden of Type II diabetes. Diabetologia 2001;44:2140-55

2. Palmer AJ, Roze S, Valentine WJ, et al. The CORE Diabetes Model: Projecting long-term clinical outcomes, costs and costeffectiveness of interventions in diabetes mellitus (types 1 and 2) to support clinical and reimbursement decision-making. Curr Med Res Opin 2004;20(Suppl 1):S5-S26

3. Eastman RC, Javitt JC, Herman WH, et al. Model of complications of NIDDM. I. Model construction and assumptions. Diabetes Care 1997;20:725-34

4. Eastman RC, Javitt JC, Herman WH, et al. Model of complications of NIDDM. II. Analysis of the health benefits and cost- effectiveness of treating NIDDM with the goal of normoglycemia. Diabetes Care 1997;20:735-44

5. Brown JB, Palmer AJ, Bisgaard P, et al. The Mt. Hood challenge: cross-testing two diabetes simulation models. Diabetes Res Clin Pract 2000;50(Suppl 3):S57-S64

6. Currie CJ, Morgan CLl, Dixon S, et al. Comparative estimates of the financial burden to the UK health system of hospital care for people with and without diabetes in the year before death. Diabetes Res Clin Pract 2005:67;144-51

7. Currie CJ, Morgan CL1, Dixon S, et al. The financial costs of hospital care for people with diabetes who have single and multiple macrovascular complications. Diabetes Res Clin Pract 2005:67;144-51

8. Evans M, McEwan P, Peters JR, et al. Should we routinely measure a proxy for insulin resistance as well as improve our modelling techniques to better predict the likelihood of coronary heart disease in people with type 2 diabetes? For debate. Diabetes Obes Metab 2004;6:299-307

9. Soedamah-Muthu SS, Chaturvedi N, Toeller M, et al. Risk factors for coronary heart disease in type 1 diabetic patients in Europe: the EURODIAB Prospective Complications Study. Diabetes Care 2004;27:530-37

10. Anderson KM, Odell PM, Wilson PW, et al. Cardiovascular disease risk profiles. Am Heart J 1991;12:293-8

11. Stevens RJ, Kothari V, Adler AI, Stratton IM. United Kingdom Prospective Diabetes Study (UKPDS) Group. The UKPDS risk engine: a model for the risk of coronary heart disease in Type II diabetes (UKPDS 56). Clin Sci (Lond) 2000;101:671-9

12. Kothari V, Stevens RJ, Adler AI, et al. UKPDS 60: risk of stroke in type 2 diabetes estimated by the UK Prospective Diabetes Study risk engine. Stroke 2002;33:1776-81

13. Currie CJ, McEwan P, Peters JR, et al. The Routine Collation of Health Outcomes Data from Hospital Treated Subjects in the Health Outcomes Data Repository (HODaR): Descriptive Analysis from the First 20,000 Subjects. Value Health 2005;8:581-90

14. Clarke P, Gray A, Legood R, et al. The impact of diabetesrelated complications on healthcare costs: results from the United Kingdom Prospective Diabetes Study (UKPDS Study No. 65). Diabetic Med 2003;20:442-50

15. Currie CJ, Morgan CL, Peters JR. The epidemiology and cost of inpatient care for peripheral vascular disease, infection, neuropathy and ulceration in diabetes. Diabetes Care 1998;21:42-8

16. British National Formulary [online]. Available from www.bnf. org/[Last Accessed 2 Feb 2004)

17. Stratton IM, Adler AI, Neil HA, et al. Association of glycaemia with macrovascular and microvascular complications of type 2 diabetes (UKPDS 35): prospective observational study. Brit Med J 2000;321:405-12

18. Morgan CL, Curie CJ, Peters J. Relationship between diabetes and mortality. Diabetes Care 2000;23:1103-7

19. Currie CJ, Morgan CL, Gill L, et al. Epidemiology and costs of acute hospital care for cerebrovascular disease in diabetic and nondiabetic populations. Stroke 1997;28:1142-6

20. Currie CJ, Morgan CL, Peters JR. Patterns and costs of hospital care for coronary heart disease related and not related to diabetes. Heart 1997;78:544-9

21. Palmer AJ, Roze S, Valentine WJ, et al. Impact of changes in HbA1c, lipids and blood pressure on long-term outcomes in type 2 diabetes patients: an analysis using the CORE Diabetes Model. Curr Med Res Opin 2004;20(Suppl 1):S53-S58

22. The CDC Diabetes Cost-Effectiveness Group. Impact of changes in HbA1c, lipids and blood pressure on long-term outcomes in type 2 diabetes patients: an analysis using the CORE Diabetes Model. JAMA 2002;287:2542-51

23. Bagust A, Beale S. Modelling EuroQol health-related utility values for diabetic complications from CODE-2 data. Health Econ2005;14:217-30

24. Clarke P, Grau A, Holman R. Estimating utility values for health states of type 2 diabetic patients using the EQ-5D (UKPDS 62]. Med Decis Making 2002;22:340-9

25. Lee AJ, Morgan CLl, Morrissey M, et al. Evaluation of the association between the EQ-5D index (health-related utility) and body mass index (obesity) in hospital-treated people with Type 1 diabetes, Type 2 diabetes and with no diagnosed diabetes. Diabet Med 2005;22:1482-6

26. Davis RE, Morrissey M, Peters JR, et al. Impact of hypoglycaemia on quality of life and productivity in type 1 and type 2 diabetes. Curr Med Res Opin 2005;21:1477-83

27. Currie CJ, McEwan M, Poole CD, et al. The impact of the overactive bladder on health-related utility and quality of life. BJU International (in press)

28. Lee AJ, Morgan CL1, Conway P, et al. Characterisation and comparison of health-related quality of life for patients with renal failure. Curr Med Res Opin 2005;21:1777-83

CrossRef links are available in the online published version of this paper: http://www.cmrojournal.com

Paper CMRO-3257_3, Accepted for publication: 10 November 2005

Published Online: 25 November 2005

doi: 10.1185/030079906X80350

Phil McEwan(a,b), John R. Peters(c), Klas Bergenheim(d) and Craig J. Currie(b,e)

a School of Mathematics, Cardiff University, Cardiff, Wales, UK

b Cardiff Research Consortium, Heath Park, Cardiff, Wales, UK

c Department of Medicine, University Hospital of Wales, Cardiff, Wales, UK

d Department of Health Economics, AstraZeneca, Mlndal, Sweden

e Department of Medicine, School of Medicine, Cardiff University, Cardiff, Wales, UK

Address for correspondence: Dr Phil McEwan, Director, Cardiff Research Consortium, Heath Park, Cardiff, Wales, CF14 4UJ, UK. Tel: +44 (0)2920-682048; Fax: +44 (0)2920-750239; email: mcewan@cf.ac.uk

Copyright Librapharm Jan 2006