Model framework

This model was developed to simulate the lifetime risk of developing diabetes-related complications rates (microvascular and macrovascular) in a cohort of patients diagnosed with type 2 diabetes [6, 7] (Figure 1). In this updated version of the model, both the level of HbA_1c (glycosylated haemoglobin) and two-hour postprandial glucose (PPG) define the degree of glycemic control [8, 9]. Each year of remaining life is simulated for all the patients in the cohort and during each cycle, the patient is exposed to the risks of developing each type of complication. These risks are determined from the degree of glycemic control, as well as other known risk factors, such as duration of diabetes.

The microvascular complications (nephropathy, retinopathy, and neuropathy) have several stages through which each patient can progress. The most severe stages for the microvascular complications are end stage renal disease, blindness or amputations. The stages of a complication are assumed irreversible – only progression to more severe stages is possible. Complications such as hypoglycaemia and foot ulcer were assumed to resolve in the course of each cycle of one year. For the purpose of this model, macrovascular complications (stroke and myocardial infarction) were considered as finite events, rather than progressive conditions.

Each simulated patient had clinical characteristics that were determined by the input distributions specified. Using a Monte Carlo technique, each patient in the cohort was assigned gender, race and age. The assignment of cholesterol level, smoking status, body mass index and systolic blood pressure was then determined using the distributions and associations observed amongst patients with type 2 diabetes [10–12].

For thirty annual cycles, the model checks each patient who has survived to that point, and updates the age, duration of disease and HbA_1c level. Over each cycle, the estimated risks of developing a new complication or progressing to the next stage of an established one are assigned to each simulated patient in the cohort. During a pre-model period of seven years, the patients were allowed to accumulate complications but costs from managing these complications are not considered in the comparisons.

The model was assessed for face validity by clinical experts and health authorities. Previous analyses using the model have been evaluated by peer review [6–9]. Source data and other independently obtained results were used as comparisons to determine predictive validity [2, 13]. Model results for relative risk over 10 years for all-cause mortality and for microvascular disease and retinopathy at 12 years were consistent with UKPDS patients in intensive and conventional treatment groups.

Risk estimates

The risk of death in this updated model was linked to both PPG and HbA_1c levels. Weibull functions were derived from the Diabetes Epidemiology: Collaborative Analysis of Diagnostic Criteria in Europe (DECODE) study [14, 15] – and estimates were based on the patients' age, gender, systolic blood pressure, total cholesterol, body mass index, smoking status, and PPG level. As in the original model, the risk of death was also assessed from the age- and gender-dependent mortality for patients diagnosed with type 2 diabetes [16], with an adjustment if nephropathy develops [17, 18]. The higher of these three death risk estimates in each model cycle was applied.

The estimates for microvascular complications (nephropathy, retinopathy, and neuropathy) were determined from the available epidemiological studies [19–21] and the risk gradients observed in the Diabetes Control and Complications Trial (DCCT) were assumed to apply to type 2 diabetes [22], an accepted assumption [23–25] confirmed by the UKPDS [2]. The risks of each microvascular complication are estimated by adjusting each according to the patient's HbA_1c level at a specific point in time (risk = 1 - e ^-λ-t, where λ = λ_bH_r ^β, and H_r is the HbA_1c value relative to a standard and β is a complication-specific coefficient) [16, 26]. The base hazard for a complication depends on factors such as duration of diabetes, race and for the retinopathy module, for example, also the probability of detection and treatment.

Evidence has recently been published that indicates PPG is an independent predictor of the occurrence of macrovascular complications, as well as of mortality [14, 27, 28]. In this updated model, the risk of stroke or myocardial infarction was estimated using Weibull functions derived from the DECODE study [15]. The risk equations derived from the DECODE study include established risk factors for macrovascular disease such as age, gender, systolic blood pressure, total cholesterol, body mass index, smoking status, as well as PPG level.

Costs

For each complication, the direct medical costs were estimated for the immediate impact of the event (costs arising in the year the event occurs) and the subsequent impact of the complication (costs accrued in years subsequent to the year of the event). Clarke et al combined resource use data collected from the UKPDS with cost estimates for these services, and published regression equations for estimating the cost of major complications [29]. The annual hospital in-patient costs, and non-hospital costs (general practioners, nurses, podiatrists, opticians, dieticians, hospital outpatient clinics) were estimated using these regression equations for the event year and subsequent years. As the inpatient costs were estimated for myocardial infarction, stroke, blindness, or an amputation. The inpatient costs of less severe stages of these complications were not included in these estimates the cost estimates are quite conservative. All complication costs are expressed in 1999 Great Britain Pounds (£1 GBP = $1.7 USD = €1.4 Euros). It should be noted that the cost of end stage renal disease was estimated based on data from 1996 [30]. We elected not to inflate this cost, however, as the applicability of general inflation rates to something as specialized as the management of end stage renal disease is fraught with inaccuracy and this was the most expensive complication (£21,456 per year).

The drug treatment cost estimates conservatively assumed full compliance with the treatment. The daily cost for metformin (1500 mg per day) was £0.07 [31], and £0.87 for the combination of nateglinide (360 mg/day = £0.80) with metformin (1500 mg per day) [31].

Analyses

After processing each cohort of 10,000 patients over thirty years, the model provides estimates of the mean survival time, the frequency of each type of complication, and the mean accumulated complication and treatment costs per patient. Survival time is also weighted by the quality of life; the utility assigned depending on the complications present. The utilities assigned were as follows; amputation 0.50, stroke 0.62, blindness 0.71 and myocardial infarction 0.73 [32], end stage renal disease 0.59 [33]. The cost per life year gained (LYG) and cost per quality adjusted life year (QALY) was determined. Consistent with NICE recommendations, costs were discounted at 6% and benefits at 1.5% [34]. Sensitivity analyses were conducted on model parameters and uncertainty in the base case estimates was examined using the bootstrap technique with 250 model replications, and 1000 re-samples from the results of these simulations.

Base case

The improvement in glycemic control, in terms of both the HbA_1c and the PPG, expected with the combination nateglinide with metformin is estimated to increase survival on average 0.39 years per patient (0.32 discounted years) or 0.46 (0.37 discounted) QALY (Table 3). Moreover, complications were expected to occur less frequently, or at least progress more slowly (Table 2).

Combination therapy is expected to reduce the frequency of complications and prolong survival, but also increase the average costs by an average of £2,066 per patient. To determine the impact of the nateglinide-metformin combination on the cost of managing complications, the difference in mean cost between metformin alone and the combination group was determined (Table 3). Thus, savings of £464 were estimated regarding the lifetime cost of managing complications. These arise mainly from a reduction in the costs of treating end stage renal disease (72%) and neuropathy (19%). The increase in the treatment costs due to combination therapy are therefore predicted to be partially offset by this reduction in the cost of managing complications, leaving an increment of £2,066 in the lifetime costs per patient (Table 3). This translates into a cost-effectiveness ratio of £6,772 (95%CI: £6,134 to 7,464) per additional discounted year of life, and £5,609 per discounted QALY.

Sensitivity analyses

Varying the efficacy of the combination of nateglinide and metformin on PPG values had a minor effect, a 50% reduction in efficacy led to a 3% increase in macrovascular disease related costs. Varying the impact of the combination of nateglinide and metformin treatment on HbA_1c values had a larger impact on the total cost predicted. Decreasing the efficacy by 10%, or 25% led to total cost increases of 3%, and 9%, respectively. Also a 10% increase in efficacy led to a 4% decrease in costs.