Subsequent vascular events

UK-specific data are used to ensure that event rates match the likely distribution in the UK. The probabilities of further MIs, strokes and vascular deaths for individuals with a history of MI are derived from patients on the Nottingham Heart Attack Register, whereas the probabilities of subsequent strokes and vascular deaths for patients with a history of a stroke are derived from patients on the South London Stroke Register.¹⁷⁶

Logistic and multivariate regression analyses were used to estimate the probability of experiencing secondary events within 1 year of a qualifying primary event. First, a logistic regression was used to estimate the probability of experiencing a secondary event of any type, that is, the combined rate of non-fatal MI, non-fatal stroke and vascular death. Multivariate regression analysis was then used to determine the distribution of secondary events between each type, should an event occur. The results confirm the importance of accounting for age in the model. For patients experiencing an MI, the probability of a secondary event within 1 year is strongly correlated with age (mean probability of 14.7% at age 45 years and 29.5% at age 85 years). Similarly, for patients experiencing a stroke, the probability of a secondary event within 1 year increases with age (mean probability of 5.4% at age 45 years and 29.8% at age 85 years), while patients with unstable angina have a mean probability of a secondary event of 8.7% at age 45 years compared with 31.3% at age 85 years.

Similar analyses were performed to estimate the probabilities of subsequent events in subsequent years. In the absence of data from individuals with a history of multiple events, these results are used to inform all subsequent events. This is a conservative approach as the application of these data implies that there is no additive effect on fatal or non-fatal event rates from previous events. Uncertainty in these event rates is explored using multivariate distributions.

List of assumptions used in the economic model

• For individuals in the event-free health state, the Weibull curves derived from the GPRD are used to predict the time to ACM. These curves are valid for up to a maximum of 15 years, after which standard life tables are used.
• Individuals enter the model with the mean characteristics of the patients in the MTC; thus, they have an average age of 45.5 years and a mean BMI of 34.92 kg/m², 25.7% are male and 33.2% are diabetic.
• At the end of the active treatment period, BMI reverts to the baseline value in a linear fashion over a 3-year period.³⁸
• For rimonabant, as changes in BMI at 6 and 12 months were not available for inclusion in our MTC, we use the average of 1.76 kg/m² (relative to placebo change) as reported in a previous economic evaluation.¹⁶⁸
• For the comparator arm (no active treatment), we assume just one visit with the practice nurse at baseline and no additional monitoring.

TABLE 32. Regressions used for subsequent events (Nottingham Heart Attack Register data)

TABLE 33. Regressions used for subsequent events (South London Stroke Register data)

TABLE 34. Results of ACMM regression

TABLE 35. Actual and predicted EQ-5D scores and mean errors in predicted values

TABLE 36. Ratio of fatal CHD to stroke

TABLE 37. Monitoring costs

Cohort size

The number of individuals required to capture the individual patient variation in a typical cohort was determined by examining the average costs and QALYs derived from cohorts of increasing numbers of patients. With a sample size of 200,000, there is still a small amount of variation in the estimated average costs (Figure 18) and QALYs (Figure 19). These variations have stabilised when using sample size of 400,000.

FIGURE 18. Stability analyses for cohort size, average discounted cost per patient

FIGURE 19. Stability analyses for cohort size, average discounted QALYs per patient

To ensure that our results represent those of an average cohort, we use a sample size of 1,000,000 for the deterministic analyses. However, because of computational limitations, we use a sample size of 400,000 and 200 Monte Carlo simulations in the stochastic analyses.

TABLE 38. Percentage of lives needed to be lost for the cost per QALY compared with placebo to be > £20,000: using deterministic results

TABLE 39. Sensitivity analyses