Dabigatran versus warfarin under standard or pharmacogenetic-guided management for the prevention of stroke and systemic thromboembolism in patients with atrial fibrillation: a cost/utility analysis using an analytic decision model

A Markov state transition model was built to simulate the C/U of anticoagulation therapy for a virtual population of 10 000 individuals with a new diagnosis of AF under the perspective of the public health care system (Figure 1). The virtual population consisted of new AF patients with a mean age of 64 years (sd. = 8), who never had a previous stroke and who didn’t have a contraindication to anticoagulation therapy.

Outcomes considered were major hemorrhagic and thromboembolic events per 100 person-years, direct medical costs and quality-adjusted life-years (QALY). The model consists of daily cycles starting at the first day of anticoagulation treatment and ending 5 years later or at the death of an individual. The 5 years time horizon was chosen in order to take into account the chronicity of anticoagulation therapy in AF while avoiding to speculate about the long term effects of dabigatran.

Input parameters were retrieved from an extensive literature search of guidelines on anticoagulation in FA patients, and peer-reviewed published studies prioritized according to the following order: Quebec, other provinces of Canada, United States of America (USA), Europe and Australia.

The model begins by presenting the following three options: 1) SD-W; 2) GT-W; and 3) dabigatran 150 mg twice per day. Dabigatran was chosen among the new oral anticoagulants because it is the first one that has been approved in Canada and the first one that has been included in drug lists of the Canadian provincial public insurance schemes.

In the warfarin options (SD-W and GT-W), treatment monitoring is performed by measuring the international normalized ratio (INR). INR may be below (<2), within (between 2 and 3), or above (> 3) the therapeutic range. In the SD-W base case scenario, the time spent in each category (below, within and above) was based on the results of the RE-LY clinical trial comparing dabigatran and warfarin[16], while in the GT-W option, it was calculated using data from Anderson et al. 2007[13]. We assumed that after one year 100% of individuals, whatever the warfarin group they belong to, had reached a stable maintenance dose[27]. The proportion of patient time spent in each INR category (below, within and above) was considered to be similar for both groups[27]. In the dabigatran option, it was assumed that no laboratory monitoring was required.

According to event probabilities[16, 17, 28, 29], patients move into the Markov model through the following health states: no major event, major hemorrhagic event, major thromboembolic event, and death (Figure 2).

Major thromboembolic and hemorrhagic events were considered according to the definition of Fihn et al.[30]. Hemorrhages were classified into two categories: 1) intracranial (intracerebral and subdural hematoma) and 2) extracranial. We assumed that all extracranial hemorrhages were gastrointestinal as they constitute the majority of extracranial hemorrhages related to anticoagulant therapy regardless of the type of treatment[28, 31]. Major thromboembolic events considered were strokes, myocardial infarctions (MI), deep vein thromboses (DVT) and pulmonary embolisms (PE). Following the occurrence of a major event, the model assumed that an individual visits the emergency room and that he has a defined probability of being hospitalized for a short or a long period of time. In addition, the model considers the probability of survival and of death after each event as well as the probability to have mild or severe post-event sequelae (see Table 1). It also considers a complete cessation of anticoagulant therapy in case of intracranial hemorrhage and a one month cessation in case of gastrointestinal bleeding[26].

Costs included in the study are those of the public health care system. Only direct costs were estimated (Table 2).

Health services consumed along the course of an anticoagulation therapy relate primarily to services associated with prescription, warfarine monitoring and management of thromboembolic or hemorrhagic complications (deep vein thromboses, pulmonary embolisms, strokes, myocardial infarctions, intracranial, sub-dural and extracranial hemorrhages)[42, 49‐55] as well as the cost of follow-up in case of sequelae from a stroke or an intracranial hemorrhage.

The quantification of services consumed was based on the literature. In all cases, we considered for baseline values, recommendations by the Canadian guidelines on the management of thrombosis and anticoagulation[42, 49‐55]. For items not found in the guidelines, we used relevant economic studies published in peer-reviewed journals. Unit prices for services consumed were calculated from the administrative data of the Quebec public health care system. The lowest prices from the list of drugs covered by Quebec public healthcare insurance (Régie d’assurance maladie du Québec (RAMQ)) were used to estimate the cost of outpatient medication, to which were added 6% for wholesalers as well as the pharmacist fees paid by RAMQ. Diagnosis related-groups (DRG) data were used to calculate the average cost of hospitalization to which were added the physician fees paid by RAMQ. The ministry of health SIFO data bank was used to calculate activity center unit prices for ambulatory care. These prices were increased to reflect the contribution of support activity centers to clinical services, using the direct method[56]. The rate base used for these sources was the values of the 2010-2011 fiscal year.

Utility scores used for each health states are presented in Table 1. They were retrieved from published studies[21, 47, 48]. Utility scores were then used to weight the time spent in each health state to produce QALYs.

Simulations were carried out on the SCHNAPS platform simulator[57, 58] that runs on the CLUMEQ network super computers. In the “individual sampling model (ISM)” simulation that was performed, each virtual individual is generated and has his own path in the simulation process. Simulations were repeated 1000 times with different virtual populations of 10 000 individuals. However, each of the 1000 different virtual population generated was used for the three options compared. The principal outcome measured was the incremental cost-utility (ICUR) ratio calculated using the difference in average individual cost over the 1000 simulations divided by the difference in average QALYs. We considered the literature that proposes that an option can be accepted as cost-effective if its ICUR is CAD$ 50 000/QALY or less[59]. An option was considered dominated if it was more costly and less effective (less QALY) in comparison to its alternative (strict dominance) or if it’s incremental cost-effectiveness ratio was greater than that of the next, more effective, and more expensive alternative (extended dominance). All costs and QALY were discounted at a baseline annual rate of 3%.

Univariate and multiway probabilistic sensitivity analyses were performed using the parameters that were foreseen as having a possible impact on the outcomes (Table 1). One way sensitivity analyses were performed to evaluate the eventual impact of each single parameter on the results. We tested the minimum and the maximum values (from the 95% confidence intervals) for each of these variables. For the GT-W option, we tested several patient percentage times in therapeutic range (TTR) values in order to find the value for which GT-W would be cost-effective compared to SD-W. Subsequently, multi-way probabilistic sensitivity analyses using Monte Carlo simulations were performed using a virtual population of 5000 individuals. We assumed that event probabilities and utility scores followed a beta distribution, that costs followed a gamma distribution while relative risks were assumed to have a log-normal distribution[60]. A C/U acceptability curve[61] was then produced from 1000 Monte Carlo iterations in order to better define the joint uncertainty of the parameters on C/U ratios.

The Markov state decision model and the parameterization were validated by two experts (MA, PM) in genetics and anticoagulotherapy.

The simulation process was validated at each step to ensure that the data generated matched expected data. This consisted in verifying that the number of events (major bleedings and major thromboembolisms) corresponded to their predefined expected occurrence according to the literature[49‐54].

This project was approved by the Research Ethics Committee of Laval University.