PLATO7 was a multicenter, double-blind, randomized trial comparing ticagrelor (180-mg loading dose, 90 mg twice daily thereafter) and clopidogrel (300–600-mg loading dose, 75 mg daily thereafter) for the prevention of cardiovascular events in 18624 patients hospitalized with ACS with or without ST-segment elevation.

The primary endpoint was a composite of death from vascular causes, MI, or stroke, measured as the number of events. At 12 months, the primary endpoint had occurred in 9.8% of patients receiving ticagrelor compared with 11.7% of those receiving clopidogrel (hazard ratio [HR], 0.84; 95% confidence interval, 0.77–0.92; p<0.001).

The model used to evaluate ticagrelor was adapted for the Portuguese situation from a European model (Nikolic et al.8). This model, which has been the subject of several publications, consists of two parts: a short-term decision tree for the first year based on data from the PLATO trial, and a Markov model for long-term extrapolation, in order to cover all the major events associated with resource use and clinical outcomes during a patient's lifetime. The model is similar to other decision analysis models on ACS.9–11

Figure 1 illustrates the model structure.

ACS: acute coronary syndrome; MI: myocardial infarction.'> ACS: acute coronary syndrome; MI: myocardial infarction.' title='Model structure (adapted from Nicolic et al.8). ACS: acute coronary syndrome; MI: myocardial infarction.'/>

Figure 1.

Model structure (adapted from Nicolic et al.8). ACS: acute coronary syndrome; MI: myocardial infarction.

(0.34MB).

In the first year, patients are allocated to nodes in the decision tree according to the estimated probability of suffering each event, which differs between the two treatment strategies according to the treatment effect observed in the PLATO trial. Each node is assigned estimates of health care use and QALYs.

After the first year, it is assumed that patients are no longer on treatment with ticagrelor or clopidogrel, without direct repercussions (treatment effects or post-treatment relapse) for the cycles of the Markov model.

The Markov cycles have a duration of one year. Each health state is associated with an estimate of health-related quality of life expressed in QALYs and of health resource use. The patient cohort progresses through the Markov model according to the estimated probability of transition based on the PLATO data.

In each year patients without events have a risk of suffering non-fatal MI or non-fatal stroke (transitions 1 and 2); in the case of a non-fatal event, patients progress to the “Non-fatal MI” or “Non-fatal stroke” state. Patients at the “No event” state are also at risk of death (transition 3) every year, and if this occurs, they progress to the “Death” state, an absorbing state following which no further transitions are possible.

The states “Non-fatal MI” and “Non-fatal stroke” represent the first-year outcome in terms of survival, costs and quality of life of patients who suffered a non-fatal event, but who are still at risk of dying and thus moving to the “Dead post event” state (transitions 4 and 5); the “Non-fatal MI” and “Non-fatal stroke” states are tunnel states, meaning that patients can only remain in them for one cycle. Patients who are still alive after a year pass to the “Post MI” or “Post stroke” states, which like the “Non-fatal MI” and “Non-fatal stroke” states represent patients’ outcomes in terms of survival, costs and quality of life in the second and subsequent years after a non-fatal event following entry to the PLATO trial.

The “Post MI” and “Post stroke” states are associated with risk of death (transitions 6 and 7) as well as with costs and QALYs.

For short-term calibration of the model, the probabilities were based on data from the PLATO trial (Table 2). Four transition probabilities were obtained to calibrate the decision tree nodes:

• 1.

  The probability of non-fatal MI;

• 2.

  The probability of non-fatal stroke;

• 3.

  The probability of death from any cause;

• 4.

  The probability of having no further event, defined as the difference between 1 and the sum of the probabilities of the other three events.

At the end of the first year, there are differences between patients treated with ticagrelor and clopidogrel, reflected in the distribution between the different health states. In the long-term model no patients are receiving either medication and so the transition probabilities are the same for both arms. Even so, the long-term treatment effects differ as a function of the different starting points.

The annual mortality risks in the “No event” state (transition 3) are estimated using age-specific mortality rates taken from Portuguese life tables.12 It should be borne in mind that patients in this state have been event-free for at least a year since the initial ACS, and the evidence suggests that ACS patients are at greater risk of suffering a fatal event in the following 12 months, the risk decreasing thereafter.13,14 An HR of 2 was applied to mortality rates for age and gender to incorporate the increased risk associated with ACS. A standard deviation of 0.10 was assumed for the logarithm of the HR for the probabilistic sensitivity analysis and the ratio was assumed to follow a normal distribution on the logarithmic scale.

For the “Non-fatal MI” state (transition 4), the mortality risk observed in the PLATO trial was compared to the risk in the life tables and an HR of 6 was applied. A standard deviation of 0.3 was assumed for the logarithm of the HR for the probabilistic sensitivity analysis and this ratio was assumed to follow a normal distribution on the logarithmic scale.

For the “Non-fatal stroke” state (transition 5), the relative mortality risk in the first year after ACS compared with age- and gender-matched individuals in the general population was based on the results of three studies.15–17 The study by Dennis et al.,15 with a risk of 7.43, was considered the most robust since it had the longest follow-up and the largest number of patients. As there were more cases of ACS in the ticagrelor group, it was decided to use 7.43 as a conservative figure in the analysis of this treatment. A standard deviation of 0.35 on the logarithmic scale was assumed.

For the transition from “Post MI” to “Dead” (transition 6), it was assumed that patients in the “Post MI” state, having suffered reinfarction, are at higher risk than patients in the “No event” state. Patients in the “Post MI” state have survived a year after the recurrent event and have a lower mortality risk than those in the “Non-fatal MI” state. It is therefore likely that the risk of patients in the “Post MI” state is between that of the “No event” and “Non-fatal MI” states. An HR of 3 was accordingly applied to maintain the logical consistency of the values in the model's parameters. A standard deviation of 0.15 on the logarithmic scale was assumed for the probabilistic sensitivity analysis.

For the “Post stroke” state (transition 7), the studies used are highly consistent in their attribution of mortality risk in the second and subsequent years after the initial stroke.18 The figure of 2.07, from Dennis et al.,15 was taken as the relative mortality risk in the first year, while for the second and subsequent years it was taken as the mean of the other years of follow-up, which was consistent from the second year onwards. A standard deviation of 0.10 on the logarithmic scale was assumed for the probabilistic sensitivity analysis.

The transition parameters of the Markov model in the base case are summarized in Table 3.

The estimates of utility used in each node of the decision tree were based on EQ-5D questionnaire data11 collected within the PLATO study.

The mean QALYs for each node were estimated using ordinary least squares regression with the nodes and treatment groups introduced as dummy variables. The equations include age and gender as explanatory variables to allow estimates of QALYs for different ages and for men and women separately.

The mean estimated QALYs for each node and for each treatment group are presented in Table 4.

For the “No event” state the estimate of QALYs for event-free patients in the PLATO trial was used. The mean estimate of ticagrelor- and clopidogrel-treated patients (0.875) was applied for patients aged <70 years. When men and women were analyzed separately, the estimated QALYs for men were 0.05 higher for men than for women; the estimate also varies with age at trial entry, and as patients grow older in the model, a proportional decrement due to age was applied.

The QALY estimates for the Markov model are summarized in Table 5.

The model's time horizon was set at a patient's lifetime and an annual discount rate of 5% was applied, in accordance with the guidelines for economic evaluation studies in Portugal.6

Probabilistic sensitivity analysis

A second-order Monte Carlo simulation was used to account for the uncertainty arising from single model inputs, so that the uncertainty of the cost-effectiveness analysis reflects the uncertainty in the decision to implement a given treatment rather than the uncertainty of the model inputs.19,20 The distribution of incremental costs and QALYs is represented graphically in Figure 2, which is based on 10000 simulations of the model.

QALYs. QALY: quality-adjusted life year.'> QALYs. QALY: quality-adjusted life year.' title='Distribution of incremental costs and QALYs. QALY: quality-adjusted life year.'/>

Figure 2.

Distribution of incremental costs and QALYs. QALY: quality-adjusted life year.

(0.22MB).

The probabilities of ticagrelor being cost-effective at different levels of willingness to pay for one QALY (acceptability curve) were also estimated (Figure 3).

Figure 3.

Cost-effectiveness acceptability curve for ticagrelor.

(0.1MB).

The mean ICUR in the simulation was €5517/QALY; for a willingness to pay of €5541/QALY there is a 50% probability of acceptance. The 95% confidence intervals are €2314/QALY for 2.5 percentile and €11790/QALY for 97.5 percentile, well below the usual acceptability threshold of €20000/QALY.

The authors declare that no experiments were performed on humans or animals for this study.

The authors declare that no patient data appear in this article.

The authors declare that no patient data appear in this article.