- Email: [email protected]
Estimating the Danish Populations' Preferences for Pharmacogenetic Testing Using a Discrete Choice Experiment. The Case of Treating Depression
Volume 12 • Number 4 • 2009 VA L U E I N H E A LT H
Estimating the Danish Populations’ Preferences for Pharmacogenetic Testing Using a Discrete Choice Experiment. The Case of Treating Depression Louise Herbild, MSc,1 Mickael Bech, PhD,2 Dorte Gyrd-Hansen, PhD2 1
Danish Institute of Health Services Research, Copenhagen, Denmark and Institute of Public Health, University of Southern Denmark, Odense, Denmark; 2Institute of Public Health, University of Southern Denmark, Odense, Denmark and Odense University Hospital, Odense, Denmark [Correction added after online publication 20-November-2008: Author Affiliations have been updated]
vhe_465
560..567
A B S T R AC T Objective: The objective is to estimate willingness-to-pay (WTP) for pharmacogenetic testing in the treatment of depression. Methods: In a web-based discrete choice questionnaire, four attributes were included: 1) number of changes in antidepressants before symptom relief; 2) time with dosage adjustments due to adverse side effects and/or lack of effects; 3) cost of pharmacogenetic testing; 4) probability of benefits from pharmacogenetic testing. Respondents were asked to choose between two scenarios; 1) pharmacogenetic testing; and 2) an opt-out option reflecting a scenario without pharmacogenetic testing. The indirect utility model was assumed to be multiplicative in probability of benefits and reduced time with dosage adjustments as well as reduced number of antidepressant changes. Results: Most coefficients had the expected signs and were statistically significant. WTP for avoidance of one change in antidepressant medication
is 1571 Danish Krone (DKK), whereas WTP for reducing the period with dosage-adjustments by 1 month is DKK604. Both were statistically significantly different from zero. Conclusion: If diagnosed with depression, peoples’ WTP for pharmacogenetic testing appears to exceed its price as long as there is a reasonable probability for improvements in treatment (in the present case 10%). Utility is associated with outcomes only. Hence, other modes of provision of similar improvements in treatment may be valued equally highly. WTP estimates and the associated policy implications appear to be robust because they were unaffected by estimation model. Keywords: depression, discrete choice experiment, pharmacogenetics, preferences.
Introduction
The lack of robust evidence with regards to the optimal choice of drug, as well as starting and maintenance dosages has led to clinicians using an iterative approach to prescribing, whereby therapy is started at low dosages which are then gradually increased to decrease the risk of severe life-threatening toxic effects [4,11]. Consequently, this often leads to waiting time for alleviation of symptoms in addition to the initial minimum of 2 to 3 weeks with pharmacotherapy before symptom relief. Frequently subsequent dose-increases ends with a change in antidepressant medication and a new waiting process because treatment was either ineffective or associated with intolerable ADRs [4,7,12,13]. During this phase of treatment, patients may continue to suffer and experience severe and uncontrolled symptoms and/or ADRs [2]. The costs of depression are immense [14–17] and in the Swedish population, which has some resemblance with the Danish, annual total costs were found to equal €11,000, of which direct costs constituted €3800 [17]. Substantial costs are associated with treatment failure [18–20] as well as personal cost for patients in terms of suffering and lost quality of life.
Depression was rated as the leading cause of worldwide nonfatal disability by the World Health Organization in 2003 [1]. In Denmark, the prevalence of people needing treatment for depression is estimated to be 3.5%, although the risk of developing the disease in a lifetime perspective is 15% [2]. Depression is difficult to treat, and only 50% to 70% of patients respond to initial pharmacotherapy [2–4]. Recommended antidepressant dosages is based on comprehensive plasma concentration levels, which is sought by taking account of factors such as age, sex, other concomitant pharmaceuticals, as well as alcohol consumption, coffee, cigarette, or drug abuse, etc. For a large part of drugs, these recommended plasmaconcentration levels encompass rather wide intervals and concentration–response curves are not well-established [5,6]. Also, the consequences of choosing one drug as first treatment as opposed to another is hard to assess [7–9]. Different antidepressants are therefore tested successively to find a potent drug with a minimum of associated adverse drug reactions (ADRs)—a well-known phenomena resulting in severe problems with nonadherence to pharmacotherapy for as many as 30% to 60% of patients [2,10]. Examples of common ADRs are dry mouth, constipation, blurred vision, nausea, tremor, insomnia, and cardiac disturbance [5]. Address correspondence to: Louise Herbild, Danish Institute of Health Services Research, PO Box 2595, Dampfaergevej 27-29, DK-2100, Koebenhavn OE, Denmark. E-mail: [email protected] 10.1111/j.1524-4733.2008.00465.x
560
Taking Genetics into Account Some studies have identified that genetic variation in the drug metabolizing enzyme, CYP2D6, may be associated with individuals’ response to certain antidepressants (as well as some antipsychotics), in terms of both risk of treatment failure as well as prevalence of adverse drug reactions and observed plasma concentration levels [21–23]. Knowledge about patients’ genetic status with respect to this enzyme might reduce the length as well
© 2008, International Society for Pharmacoeconomics and Outcomes Research (ISPOR)
1098-3015/09/560
560–567
561
Pharmacogenetics and Population Preferences Table 1
Examples of antidepressants primarily metabolized by CYP2D6
Fluoxetine, Fluvoxamine, Paroxetine, Citalopram, Doxepin, Amitriptyline, Imipramine, Trimipramin, Nortryptyline, Maprotilin, Clomipramine, Venlafaxin, Mirtazapine, Mianserin [25].
as number of periods with dosage adjustments as patients’ metabolic capacity can be taken into consideration along with age, sex, lifestyle, and concomitant pharmaceuticals from the very beginning when the first dosage of an antidepressant agent is prescribed. This kind of personalized medicine—testing for genetic differences between individuals when prescribing medicines—is termed pharmacogenetic testing. The benefits of pharmacogenetic testing for the CYP2D6 gene when prescribing antidepressants is compromised by the fact that far from all antidepressants are metabolized this way [21,24]. Table 1 provides a list of substrates primarily metabolized by the CYP2D6 [25]. Aside from that, the partly unknown and complex nature of psychiatric diseases complicates the prediction of response to pharmaceuticals, as this is also affected by individual differences in pathways, receptors, disease pathology, as well as lifestyle, age, sex, and concomitant pharmaceuticals [21,26–29]. Hence, in spite of a relatively high prevalence among Caucasians where 20% to 30% have either no functioning CYP2D6 enzymes, reduced number of enzymes, or varying levels of duplications of the enzyme relative to the dominant part of Caucasians [21,30–35], no evidence on the benefits of pharmacogenetic testing is available [4,36,37]. Chou et al. [38] found trends toward greater number of ADRs and prolonged hospital stays as well as increased costs among psychiatric patients with genetically increased or decreased enzymatic activity. A large population-based cohort study [4], which examined the influence of genetically decreased enzymatic activity on intolerability of antidepressants by studying dosages, frequency of switching to another antidepressant, or discontinuation of pharmacotherapy, found an increased risk of switching between antidepressants among patients with decreased metabolic capacity as well as lower mean antidepressant maintenance dosages in these patients [4]. Studies indicate that pharmacogenetic testing for metabolizing enzymes in general might benefit between 15% and 40% of drug treatments, but numbers vary between studies [30,31,39]. A conservative assumption based on expert opinion is that only 2% to 5% of depressed patients tested would benefit from pharmacogenetic testing of the CYP2D6 [40]. Ultimately, the benefits of testing will depend on the most frequently used antidepressant (and the way this is metabolized) as well as the population under investigation as the frequency of the “non-normal” deviant metabolizers differs across ethnic groups [31]. Possible benefits of pharmacogenetic testing are reductions in the time-to-target-dose as well as reductions in the number of changes in antidepressant medication as patients can be characterized by whether they are expected to be able to tolerate ordinary dosages or whether dosages should either be decreased or increased, or if a pharmaceutical not primarily dependent on the CYP2D6 enzyme should be chosen as first choice. Pharmacogenetic testing for changes in the CYP2D6 gene has been introduced at one psychiatric hospital in Denmark and is available at the price of DKK1630 (€219) [41]. A single blood sample (or mouth swab) is required; and because our genes do not change, the test will never need to be conducted again. Other psychiatric hospitals are likely also to consider implementing the pharmacogenetic test in the future. We therefore wanted to see if pharmacogenetic testing for CYP2D6 and the
potential consequences for the treatment of depression would be of any value to potential patients and if this could be modeled by means of a discrete choice experiment. The purpose of this study was to elicit the Danish populations’ preferences for the introduction of routine pharmacogenetic testing among newly diagnosed depressed patients about to be treated with antidepressants. The aim was to estimate the monetary use value of the pharmacogenetic test.
Methods
Discrete Choice Experiment A discrete choice experiment (DCE) was used to elicit preferences for pharmacogenetic testing among a panel of Danes, 18 to 65 years old. In a DCE, respondents are faced with hypothetical choices between treatment scenarios that differ in terms of specified attributes and attribute levels. The value of the good (in this case the pharmacogenetic test) is determined by the magnitude of these attributes, and by varying the levels of these in hypothetical scenarios, respondents are forced to make trade-offs. Those incremental changes in attributes which are relevant for respondents’ choices can then be identified. On the basis of the relative weighting of the attributes, a compensation variation measure that conforms with demand theory is derived [42,43].
Survey Design Seven hundred one respondents were e-mailed a link to the study questionnaire sent out by an independent research agency (Gallup) in August 2007. The sample was stratified by age, sex, geographic location, and education. Participants in the research panel from which respondents were sampled were recruited randomly from telephone interviews and invited to participate in the Internet-based research panel. Participation in the research panel is rewarded by participation in a lottery. The questionnaire was developed on the basis of a literature review, expert opinion and three focus group interviews with depressive patients. The focus group interviews were conducted in an iterative manner from which problems related to the pharmaceutical treatment of depression were specified in more detail and related to those elements of the pharmaceutical treatment that the pharmacogenetic test could have an impact upon. A more detailed description of the development of the questionnaire and results from the focus group interviews are available in English elsewhere [44]. From the focus group interviews, it was clear that patients clearly prefer if the correct antidepressant could be administered at the right dosage from the very beginning of their pharmaceutical treatment. Two attributes found to be of importance to patients’ experience of the pharmaceutical treatment which could be affected by pharmacogenetic testing were identified: 1) the number of changes in antidepressant medication before symptom relief; and 2) the time with dosageadjustments due to lack of effect and/or unacceptable ADRs. These two attributes are hereafter called the effect-attributes and both relates to the experience of the health improvement associated with the treatment. The number of changes in antidepressant medication was important to patients because each change represents a realization that the current antidepressant will not be effective or is associated with intolerable ADRs, that time has been wasted, and that a new course of treatment with associated risks of ADRs must be initiated. Although changing antidepressants was associated with a negative utility due to frustration and the facing of new risks, the time with dosage adjustments was found to be frustrating because of the waiting time and the lack
Herbild et al.
562 of symptom relief and/or intolerable ADRs, as well as the uncertainty regarding the response of an increased or decreased dosage. In the questionnaire, it was clearly explained to respondents that the number of changes and the total time with dosage adjustments (across all medication attempts) could vary independently. This is reality in practice, where a prescribing doctor may choose to switch medications early in the process; or alternatively opt for longer periods with dosage adjustment in an attempt to avoid switching. Hence, a reduced number of changes may involve longer or shorter total time with dosage adjustments, and a longer dosage-adjustment period may involve fewer or more medicine changes. Clinically, the relevance of the two attributes is supported by evidence on increased number of antidepressant changes as well as difference in maintenance dosages among patients with genetically altered enzymatic capacity relative to patients with normal capacity [4]. The pharmacogenetic tests’ ability to affect the two effectattributes depends upon whether the “first choice” prescribed antidepressant is metabolized by the CYP2D6 and whether the patient has a non-normal metabolic capacity. Other factors such as specific pathways, lifestyle, concomitant pharmaceuticals, etc., may also affect the effects of the pharmacogenetic test. Hence, a significant proportion of patients will not gain from pharmacogenetic testing, but at the time the test is performed, one cannot determine who these individuals are. Consequently, we included a third attribute representing the probability of getting any benefit (i.e., any reduction in the time with dosage adjustment and any reductions in the number of antidepressant changes) from the test. Finally, a price of the test was included, allowing for the calculation of marginal willingness-to-pay (WTP) estimates. Each respondent was presented with eight choice-sets, each containing two scenarios; one scenario represented treatment outcomes with the pharmacogenetic test and the other represented a status quo scenario with treatment outcomes as standard practice is today without the pharmacogenetic test. This status quo scenario represents an opt-out option [45]. In this case, the opt-out option is specified in accordance with the expected outcomes of current practice, and the respondent is thereby provided with the necessary information required to make an informed choice. Respondents were given a thorough description of depression and the symptoms characterizing the disease as well as the pharmacological treatment and the related problems. The attributes were carefully explained and respondents were told to think of the test as an insurance-policy “you do not know whether you will benefit, but you will have to pay with certainty. If you do not benefit your treatment outcome will be equivalent to the scenario without the test.” Respondents were faced with the following question: “Imagine that you have just been diagnosed with depression for the first time. Which of the two treatment possibilities would you choose?” An example of a choice set is shown in Table 2. The treatment scenario without the pharmacogenetic test was constructed as a fixed status quo scenario and represented a treatment profile for a group of moderately to severely depressed patients’ first episode of depression. Levels for the two effectattributes characterizing this scenario were based on median first episode values obtained in a previous study among moderately to severely depressed patients treated without the test in accordance with current practice [46]. The levels were four changes in antidepressants and 5 months with dosage adjustments. A positive price-attribute was associated with the pharmacogenetic test scenario. Respondents were informed that on an average, having the test would not affect out-of-pocket pharmaceutical expenses.
Table 2 Example of a choice set presented for the respondents in the discrete choice experiment Imagine you have just been diagnosed with depression for the first time. Which of the two treatment possibilities would you choose? Likelihood of getting improvements from the test: Price of the test (expenses for medicine remains constant): Number of changes in antidepressant medicine: Time with dosage adjustments due to lack of effect and/or unacceptable adverse side effects:
Genetic test
No genetic test
50% DKK18.000
DKK0
2 changes
4 changes
3 months
5 months
DKK, Danish Krone.
In the pharmacogenetic treatment scenario, attribute levels were conservatively reduced by one or two changes in antidepressants, and the period with dosage adjustments was reduced by 2 or 4 months. These values were entered into the regression analyses by way of difference coding as reductions of one and two, and two and four, respectively. To secure that our priceattribute covered both a lower as well as an upper bound of WTP [47], eight levels were chosen for the price attribute in the pharmacogenetic treatment scenario, ranging from DKK200 (€27) to DKK18,000 (€2420). The price of the test was stated as an out-of-pocket, once in a lifetime payment. The prevalence of genetically non-normal metabolizers in Denmark is almost 20% [35]. In the present study, we assumed that half of these could benefit from pharmacogenetic testing. Hence, the base-case probability of benefiting from the test was set at 10%. In other stated preference studies, it has been found that respondents are frequently insensitive to scope, and this problem is enhanced when respondents face probabilities [48,49]. We therefore chose to present respondents with another probability level to test if individuals are indeed sensitive to the level of probability. This level was set at 50% to distinguish it from the base-level. The probability attribute was thus coded as a dummy variable (which equaled 1 if the probability was 50%; otherwise zero) and the associated coefficient was used to test for sensitivity to scope. Attributes and attribute levels as well as coding are shown in Table 3.
Statistical Design A fractional factorial experimental design optimizing Defficiency was generated in SAS [50] using the MktEx macro, resulting in a design with 32 hypothetical alternatives. These 32 Table 3 Attributes and attribute levels used in the discrete choice experiment
Attribute Likelihood of getting improvements from the test: Price of the test
Number of changes in antidepressant medicine: Time with dosage adjustments due to lack of effect and/or unacceptable adverse side effects:
Attribute levels (coding) in the scenario representing pharmacogenetic testing 10% (0), 50% (1) DKK200; DKK600; DKK1000; DKK1500; DKK3000; DKK6000; DKK9000; DKK18,000 2 changes (2), 3 changes (1) 1 months (4), 3 months (2)
563
Pharmacogenetics and Population Preferences alternatives were distributed across four blocks of eight using the MktBlock macro in SAS that creates a blocking factor, which is uncorrelated with every attribute of every alternative while ensuring the maximum degree of design efficiency. The 32 alternatives were each paired with the fixed status quo alternative such that a total of 32 choice sets were constructed. The design is available from the authors on request. In the design, we allowed for two-way interaction effects to be estimated between the probability of getting an effect and the attributes on the effect i.e., reductions in the number of antidepressant changes and time with dosage adjustments.
Theory of Expected Utility Using the two attributes on “number of changes in medicine” and “time with dosage adjustments before symptom relief” to describe the pharmacogenetic test (the effect-attributes), we implicitly assume that reductions in these represent the total observable value of the pharmacogenetic test, i.e., its utility-generating properties. As the underlying assumption of the DCE is that individuals are utility-maximizing, we expect respondents to maximize utility with respect to the two effect-attributes subject to the price of the pharmacogenetic test. The utility gain depends upon the probability by which the effects will be provided. According to the theory of expected utility (EUT), we assume that the expected utility is a multiplicative function of probability and effect. Respondents are thus expected to trade among treatment scenarios in a way that maximizes their utility from the effects of the test subject to the probability of getting these and the price. Hence, we assume that the utility function can be represented by the following specification:
ΔU test = αˆ test ASC + βˆ price Δx1 + βˆ change Δx2 + βˆ time Δx3 + βˆ change ∗ prob Δx2 Δd50 + βˆ time ∗ prob Δx3Δd50
(1)
The coefficient βˆ price represents the marginal disutility of the price of the test (because this has to be paid in addition to existing costs of treatment without pharmacogenetic testing), and βˆ change and βˆ time represents the utility of marginal changes in any of the two effect-attributes when the probability of effect is 10% (reference level). Dx2Dd50 and Dx3Dd50 constitute added utility associated with changes in the effect-attributes when the probability level is 50% (d50 equals one if the probability is 50%; else, zero). The coefficient αˆ test for the alternative specific constant (ASC) captures any utility associated with the test per se.
version 9.2 (Stata Corp., College Station, TX). WTP-estimates were calculated as marginal rates of substitution (MRS) between the coefficients of the effect-attributes and the coefficient on price, e.g., MRS = −Δβˆ change Δβˆ price. Confidence intervals (CI) for WTP-estimates were calculated using the parametric bootstrapping procedure specified by Krinsky-Robb with 10,000 replications [51–53]. To test for preference heterogeneity across relevant groups without introducing problems of scale [54], we compared WTPestimates in a number of stratified analyses or included interaction terms between the attributes and dummy variables representing sub-groups. We tested 1) if WTP differed across income groups with a household income above or below DKK350,000 (€47,000); 2) if current or former pharmaceutical treatment for depression among respondents affected WTP; 3) if prior diagnosis of depression and/or having friends or relatives with present or past diagnosis of depression and experience with antidepressants affected WTP; 4) if there was a difference in WTP across sex, age, and education; and 5) if perceived difficulty in answering the DCE-questions was associated with differences in WTP-estimates.
Results A total of 701 questionnaires were sent out. Four respondents were noneligible to participate, 74 did not wish to participate, 39 did not complete any of the DCE questions, and 261 did not reply at all. A total of 323 completed questionnaires were returned, providing a response rate of 46%.
Respondent Descriptives Males represented 47% of respondents. Compared with the general Danish population, respondents aged 18 to 24 years were slightly under-represented and respondents with a high level of education were over-represented [55]. Nineteen respondents did not like the idea of pharmacogenetic testing or/and would never want to have the test. Some of these were among the 43 nontraders always choosing the no-test scenario. As we wanted to estimate the use value of the test, these respondents were excluded because we interpret these as being nonusers within the given spectrum of attribute levels. Three respondents had already had the pharmacogenetic test and were also excluded. This left us with a total sample of 277 individuals each answering eight choice sets.
Preferences for the Pharmacogenetic Test Econometric Analysis The data were analyzed using the conditional logistic regression procedure, clogit, in the statistical software package of STATA Table 4
Results of the conditional logistic regression analysis are presented in Table 4. As expected, the coefficient on the price attribute is negative, indicative of decreasing utility as the price of
Results from the DCE assuming a multiplicative utility function in accordance with expected utility theory
ASC Price Changes (reductions in medicine changes) Changes*d50 (added utility if probability of effect increases to 50%) Time (reduction in the time with dosage adjustments) Time*d50 (added utility if probability of effect increases to 50%) N (observations/respondents) AIC* LR chi-square Pseudo R2
Coefficient
t-value
0.3999 -0.0003 0.3999 -0.022 0.1528 0.1655
0.47 -19.29 3.42 -0.15 2.62 2.22
Prob > |t| 0.637 0.000 0.001 0.884 0.009 0.027 4432/277 2354 730.12 0.2377
WTP (DKK) 430 — 1571 -70 604 645
*AIC = 2 k - 2 ln (L), where k is the number of parameters in the model and L is the maximized value of the likelihood function for the estimated model. AIC, Akaike information criterion; CI, confidence interval; DKK, Danish Krone; WTP, willingness-to-pay.
90% CI (DKK) -1007 — 809 -1055 230 162
1860 — 2331 903 986 1135
Herbild et al.
564 the test increases. The coefficients of the effect-attributes are positive, signifying a utility increase associated with decreases in the number of antidepressant changes or reductions in time with dosage adjustments. This is all in accordance with a priori expectations and the coefficients are statistically significant (P > |t| less than 0.01). The coefficient of the interaction term between changes in antidepressant medicine and probability is unexpectedly negative, but not statistically significantly different from zero, although the interaction between the time-attribute and probability is positive and significantly different from zero (P = 0.027), though smaller than expected (according to EUT). Preferences for reductions in the number of changes in antidepressant products do not seem to be affected by the probability level at which improvements are given. The constant represents the value of the pharmacogenetic test per se, i.e., irrespective of its effects on treatment outcome. It is statistically insignificant, which implies that the value of the test is associated with outcomes only.
Marginal WTP Estimates of WTP for a marginal change in attributes are also shown in Table 4, including 90% CIs. The WTP estimate for a 10% probability of a reduction of one in the number of antidepressant changes is DKK1571 (90% CI DKK809–2331) and the WTP for a reduction of 1 month in the time with dosage adjustments with 10% probability is DKK604 (90% CI DKK230– 986). WTP for reductions in a number of changes is not statistically higher when the probability of obtaining this gain is 50%, although WTP for an increase in the chance of obtaining a month of reduction in the time with dosage adjustment to 50% is DKK645 (90% CI DKK162–1135). The total value of 1 month reduction in the time with dosage-adjustments with 50% probability is thus DKK1249. The CI for the WTP for the ASC is wide and not statistically significantly different from zero, signifying that on average people only value the pharmacogenetic test by the reductions in the attributes, although some might value the more intangible benefits of the test as well because it is slightly positive. Overall, the results imply that if the pharmacogenetic test with a chance between 10% and 50% can reduce the number of antidepressant changes by one and shorten the dosage adjustment period by 1 month, the WTP for the pharmacogenetic test exceeds the price of the test (current price is DKK1630).
Subgroup Analyses Including an interaction term between a dummy (equal to one for high-income households; else, zero) and the price attribute indicated that WTP for the test differed between respondents with different levels of income. This was expected and indicates that respondents have taken budget constraints into consideration when they completed the DCE. Almost 17% of the respondents had been diagnosed with depression by their physician. Most of these (81%; n = 35) had also been treated with antidepressants. As the choice sets were designed with two hypothetical scenarios and these respondents still constituted potential users of pharmacogenetic testing, the respondents were not excluded. We hypothesized that preferences for the pharmacogenetic test might be affected by previous experience with antidepressants. Inclusion of an interaction term between the attribute on number of antidepressant changes and a dummy variable (equal to one if respondents had been treated with antidepressants; else, zero) demonstrated no statistically significant differences between those with and without experience. We also tested for differences in preferences across groups with a presumed different level of
knowledge of depression. Sixty-five percent of the respondents knew somebody among their nearest friends or family who had been treated pharmacologically for depression, and some of these were among the 17% who had been diagnosed with depression themselves. Including an interaction term between the timeattribute and a dummy variable (equal to one if knowledge of depression; else zero), suggested no statistically significant differences in preferences because of knowledge. We also tested for sex-differences, age-related differences, and educational differences but found no impact on preferences. Finally, we tested if reported self-perceived difficulty in answering the choicequestions could be taken as a marker of preference heterogeneity among respondents. Difficulty in answering the choice questions was associated with insignificant WTP-estimates for both effectattributes and a clear tendency to choose on the basis of price alone. In sum, we found no evidence of heterogeneity in preference structures because of personal experience of antidepressant medication, knowledge of the disease, or sex, age, and education.
Discussion
Internal Validity The results of the discrete choice exercise, conducted among a random sample of the Danish population who is to imagine having been recently diagnosed with depression, suggest that the stated expected use value of pharmacogenetic testing exceeds the current price of the test, even if the test offers relatively small expected gains. If, for example, the pharmacogenetic test provides a 1-month reduction in time with dosage-adjustments as well as one change less in antidepressant products, with a probability between 10% to 15%, the mean WTP among potential users exceeds the cost of the pharmacogenetic test (current price is DKK1630). The price of the test does however not cover the entire costs of the test because blood samples have to be taken and the results interpreted. Such analyses are associated with marginal resource consumption in terms of labor costs. Consequently, pharmacogenetic testing will only be associated with a welfare gain if the marginal labor costs associated with performing and analyzing the test are small, or if the benefits of the pharmacogenetic test exceed the 10% probability of reducing the number of antidepressant changes by one and the dosage adjustment period by 1 month. The small or statistically insignificant coefficients associated with the interaction terms between the effect attributes and the probability dummy variable (equal to one if 50%; else, zero) suggests that individuals are either insensitive to the level of probability, or that respondents’ preferences in some way do not accord with EUT. There has been much criticism of EUT and its underlying assumptions [56,57], and studies have shown that when respondents face gambles, they tend to use decision heuristics that are not in accordance with EUT [49,58–64]. Some of these studies have shown that when facing gambles, respondents tend to focus either on outcomes or probabilities rather than expected value. Our study appears to suffer from similar problems. We therefore tested if a model in which outcome and probability entered the utility function as additive components, rather than in a multiplicative manner provided a better fit to the data. This was not the case (Akaike Information Criterion of 2377 compared with 2354). With the additive model specification, all coefficients except the ASC were highly statistically significant (P > |t| less than 0.001). The WTP for an increase in the probability of an effect from 10% to 50% was estimated to be DKK1984 (SD 403). Our WTP-estimates were relatively unaffected by choice of model and our results seem to be fairly robust.
565
Pharmacogenetics and Population Preferences We conducted a number of stratified analyses to compare WTP for the pharmacogenetic test across subgroups as well as test for preference homogeneity. These showed a large degree of homogeneity across strata except for that of income, which confirmed decreasing marginal utility of income and increasing price-insensitivity with rising income, as would be expected theoretically. For a subgroup of respondents reporting difficulties with the DCE questions, choices seemed to be based primarily on the level of the price attribute. This consequently led to a higher WTP for the test itself (the ASC) and lower WTP-estimates for marginal changes in the effect-attributes.
External Validity Respondents aged 18 to 24 were slightly underrepresented, and people with higher education and females were over-represented relative to the Danish population. This result may be due to the fact that the questionnaire was distributed electronically via email. Nevertheless, in Denmark, the majority of the population has access to the Internet in their homes, and email is a common source of communication [65]. Hence, we do not think that the distribution method in itself has introduced a severe bias in representation. The observed bias is more likely to be due to interest, ability, and availability of time. Only if the younger and/or highly educated people have markedly different preference structures in relation to pharmacogenetic testing does the small bias in representation introduce bias in the presented results. Subgroup analyses showed that sex, youth, and a high level of education had no impact on preference structures. The levels of the attributes applied in the status quo scenario were based on estimates obtained from a similar DCE conducted among patients, where respondents were asked about their experienced number of changes in antidepressants and experienced time with dosage adjustment for their first episode of depression [46]. A majority of the interviewed respondents were likely to be moderately to severely formerly or currently depressed patients. Assuming the marginal utility of reductions in the effectattributes is affected by the absolute levels of the effectsattributes presented in the status quo scenario, the results presented here are only representative of preferences for pharmacogenetic testing among moderately to severely depressed patients.
Design of the Scenarios The price of the test was presented as an out-of-pocket once-ina-lifetime payment. Treatment in the Danish health-care system is primarily free of charge and collectively financed by taxes, but there are relatively high copayments on dentist services and pharmaceuticals. Because the pharmacogenetic test is very closely linked to pharmacotherapy, the choice of an out-of-pocket payment vehicle is unlikely to severely impair the theoretical realism of the presented scenarios. We observed 43 nontraders always choosing the opt-out alternative, and it is likely that some of these have been protest-voters, unwilling to choose the test because of an aversion to out-of-pocket payment. An alternative payment vehicle such as increases in income taxes could have been applied to avoid protest bidding, but this would have led to other problems. Payment via income tax would involve paying annually over a lifetime, for what is actually a once-in-a-lifetime investment. Hence, even minute levels of annual tax-payments would have constituted large and most likely overestimated WTP values for the pharmacogenetic test. Thus, we chose the more conservative approach, acknowledging that the WTP values based on out-of-pocket payment may provide WTP-estimates that underestimate the value of the pharmacogenetic test because of protest bidding.
The pharmacogenetic test has not yet been tested in larger randomized controlled trials, and the efficacy of the test is unknown. The chosen level for the probability attribute was consequently based on assumptions with regards to the percentage of patients taking antidepressants metabolized by CYP2D6 as well as the number of patients with genetic and nongenetic characteristics affecting the benefits of pharmacogenetic testing. In clinical settings where the predominantly used antidepressants are selective serotonin reuptake inhibitors (SSRI), the likelihood of benefiting from the test is likely to be lower than the 10% assumed in this study (because the SSRIs are metabolized by CYP2D6 to a lesser extent). In contrast, settings where tricyclic antidepressants products are predominantly used would most likely attain a larger probability of benefit than the 10% (because of larger dependence on CYP2D6-metabolism among these products). Also, in populations with a larger frequency of “non-normal” metabolizers, e.g., in Africans, benefits will be larger, although the opposite is the case in populations with a lower frequency, e.g., in Asian populations. Nevertheless, clinical and genetic research within the area is expanding so fast that even if the present “true” benefit level were available it would most likely be outdated within a few years. In addition, pharmacogenetic testing for CYP2D6-polymorphisms is likely to be beneficial in relation to pharmaceutical treatments for several conditions because other pharmaceuticals than antidepressants are metabolized by the CYP2D6-enzyme. Finally the effects of pharmacogenetic testing for other genetic changes than those of the CYP2D6, i.e., in other metabolizing enzymes and/or clinical relevant pathways, is likely to also influence the potential benefits of pharmacogenetic testing in depression. Therapeutic drug monitoring (TDM) using probe drugs, or phenotyping as it is also called, has the potential of delivering the same treatment improvements as pharmacogenetic testing [7]. The results from phenotyping are more precise, but not stable as weight, drinking, and smoking habits, etc., can alter an individuals’ metabolizing capacity, thus necessitating the conduction of phenotyping tests at each new disease episode. TDM can also be used with antidepressants, but then only after pharmacotherapy has been initiated and then plasma levels are measured at repeated intervals. A debate about whether pharmacogenetic testing is superfluous if routine TDM is used exists among practitioners. As our results indicate, respondent’s value the outcomes associated with the pharmacogenetic test but not the test per se. Thus, the demand for pharmacogenetic testing observed in the present study may not reflect the demand in real life if TDM (in either shape) is an option. Future studies investigating preferences for TDM versus pharmacogenetic testing are therefore recommended.
Conclusion Respondents have preferences for reductions in the number of changes in antidepressant products as well as reductions in the time with dosage adjustments due to adverse side effects and/or lack of effect. Assuming that a pharmacogenetic test is able to offer a 10% chance of one change less in antidepressant products and a reduction of 1 month in the total time with dosage adjustments, respondents are willing to pay an amount which is greater than the actual price of the pharmacogenetic test and the introduction of such a test will be improving welfare. The value of reductions in the number of changes in antidepressants is higher than reductions in the time with dosage adjustments, and 2 months of dosage adjustments are preferred if it can prevent one change in antidepressant medication. This is valuable information when doctors choose their prescription strategy.
566 Subgroup analyses indicate that preferences for this kind of pharmacogenetic testing for CYP2D6 are consistent across groups with different levels of knowledge of depression, suggesting that the hypothetical DCE exercise was well-understood by the respondents. The fact that respondents are insensitive to the scope of the probability attribute does raise concern. It is however reassuring that the valuation of the pharmacogenetic test is robust as WTP-estimates were relatively unaffected by choice of model specification. Source of financial support: Center for Pharmacogenomics at the University of Copenhagen as well as the Danish Institute for Health Services Research
References 1 Mathers CD, Bernard C, Moesgaard KI, et al. Global Burden of Disease 2002: Data Sources, Methods and Results (54th ed.). Geneva: WHO, 2003. 2 Mors NPO, Børglum AD. Status og perspektiver inden for psykiatrisk genomisk medicin [status of and perspectives on psychiatric genomic medicine]. Ugeskr Læger 2005;167:2191–3. 3 Kirchheiner J, Nickchen K, Bauer M, et al. Pharmacogenetics of antidepressants and antipsychotics: the contribution of allelic variations to the phenotype of drug response. Mol Psychiatry 2004;9:442–73. 4 Bijl MJ, Visser LE, Hofman A, et al. Influence of the CYP2D6*4 polymorphism on dose, switching and discontinuation of antidepressants. Br J Clin Pharmacol 2008;65:558–64. 5 Linder MW, Keck PE, Jr Standards of laboratory practice: antidepressant drug monitoring. National Academy of Clinical Biochemistry. Clin Chem 1998;44:1073–84. 6 Balant-Gorgia AE, Balant LP, Garrone G. High blood concentrations of imipramine or clomipramine and therapeutic failure: a case report study using drug monitoring data. Ther Drug Monit 1989;11:415–20. 7 Hiemke C. Clinical utility of drug measurement and pharmacokinetics—therapeutic drug monitoring in psychiatry. Eur J Clin Pharmacol 2008;64:159–66. 8 Nierenberg AA. The medical consequences of the selection of an antidepressant. J Clin Psychiatry 1992;53(Suppl.):S19–24. 9 Anderson IM. Meta-analytical studies on new antidepressants. Br Med Bull 2001;57:161–78. 10 Brook O, van Hout H, Nieuwenhuyse H, Heerdink E. Impact of coaching by community pharmacists on drug attitude of depressive primary care patients and acceptability to patients: a randomized controlled trial. Eur Neuropsychopharmacol 2003;13: 1–9. 11 Zackrisson AL, Holmgren P, Gladh AB, et al. Fatal intoxication cases: cytochrome P450 2D6 and 2C19 genotype distributions. Eur J Clin Pharmacol 2004;60:547–52. 12 Corey-Lisle PK, Birnbaum HG, Greenberg PE, et al. Identification of a claims data “signature” and economic consequences for treatment-resistant depression. J Clin Psychiatry 2002;63:717– 26. 13 Ruhé HG, Huyser J, Scholten RJPM, et al. Antidepressant doseincrease for non-responders to a standard dose for depression. (Protocol) Cochrane Database of Systematic Reviews 2007, Issue 1. Art. No.: CD004045. DOI: 10.1002/14651858. CD004045.pub2. 14 Berto P, D’Ilario D, Ruffo P, et al. Depression: cost-of-illness studies in the international literature, a review. J Ment Health Policy Econ 2000;3:3–10. 15 Cuijpers P, Smit F, Oostenbrink J, et al. Economic costs of minor depression: a population-based study. Acta Psychiat Scand 2007;115:229–36. 16 Luppa M, Heinrich S, Angermeyer MC, et al. Cost-of-illness studies of depression: a systematic review. J Affect Disord 2007; 98:29–43.
Herbild et al. 17 Sobocki P, Lekander I, Borgstrom F, et al. The economic burden of depression in Sweden from 1997 to 2005. Eur Psychiatry 2007;22:146–52. 18 Russell JM, Hawkins K, Ozminkowski RJ, et al. The cost consequences of treatment-resistant depression. J Clin Psychiatry 2004;65:341–7. 19 McCombs JS, Nichol MB, Stimmel GL, et al. The cost of antidepressant drug therapy failure: a study of antidepressant use patterns in a Medicaid population. J Clin Psychiatry 1990;51 (Suppl.):S60–9. 20 Sobocki P, Ekman M, Agren H, et al. The mission is remission: health economic consequences of achieving full remission with antidepressant treatment for depression. Int J Clin Pract 2006;60: 791–8. 21 Kirchheiner J, Brøsen K, Dahl ML, et al. CYP2D6 and CYP2C19 genotype-based dose recommendations for antidepressants: a first step towards subpopulation-specific dosages. Acta Psychiat Scand 2001;104:173–92. 22 Mulder H, Herder A, Wilmink FW, et al. The impact of cytochrome P450-2D6 genotype on the use and interpretation of therapeutic drug monitoring in long-stay patients treated with antidepressant and antipsychotic drugs in daily psychiatric practice. Pharmacoepidemiol Drug Saf 2006;15:107–14. 23 Rau T, Wohlleben G, Wuttke H, et al. CYP2D6 genotype: impact on adverse effects and nonresponse during treatment with antidepressants-a pilot study. Clin Pharmacol Ther 2004;75:386– 93. 24 Andersson T, Flockhart DA, Goldstein DB, et al. Drugmetabolizing enzymes: evidence for clinical utility of pharmacogenomic tests. Clin Pharmacol Ther 2005;78:559–81. 25 Medicin.dk. Medicin.dk [online Danish medicine catalogue]. Medicin dk; Available from: URL: http://www.medicin.dk/ (41ezgpnlrwn5gbrmdj55au55)/show.aspx [Accessed April 9, 2007]. 26 Biosam. Farmakogenetik i Danmark [Pharmacogenetics in Denmark]. Biosam 2003;15. 27 Brøsen K. Some aspects of genetic polymorphism in the biotransformation of antidepressants. Therapie 2004;59:5–12. 28 Brøsen K. Klassisk farmakogenetik [Classical pharmacogenetics]. Ugeskr Laeger 2005;167:2143–6. 29 Caraco Y. Genes and the response to drugs. N Engl J Med 2004;351:2867–9. 30 Heller T, Kirchheiner J, Armstrong VW, et al. AmpliChip CYP450 GeneChip: a new gene chip that allows rapid and accurate CYP2D6 genotyping. Ther Drug Monit 2006;28:673–7. 31 Ingelman-Sundberg M. Genetic polymorphisms of cytochrome P450 2D6 (CYP2D6): clinical consequences, evolutionary aspects and functional diversity. Pharmacogenomics J 2005;5:6–13. 32 Ingelman-Sundberg M. The human genome project and novel aspects of cytochrome P450 research. Toxicol Appl Pharmacol 2005;207(Suppl. 2):52–6. 33 Kawanishi C, Lundgren S, Agren H, Bertilsson L. Increased incidence of CYP2D6 gene duplication in patients with persistent mood disorders: ultrarapid metabolism of antidepressants as a cause of nonresponse. A pilot study. Eur J Clin Pharmacol 2004; 59:803–7. 34 Raimundo S, Fischer J, Eichelbaum M, et al. Elucidation of the genetic basis of the common “intermediate metabolizer” phenotype for drug oxidation by CYP2D6. Pharmacogenetics 2000; 10:577–81. 35 Rasmussen JO, Christensen M, Svendsen JM, et al. CYP2D6 gene test in psychiatric patients and healthy volunteers. Scand J Clin Lab Invest 2006;66:129–36. 36 Matchar DB, Thakur ME, Grossman I, et al. Testing for Cytochrome P450 Polymorphisms in Adults with Non-Psychotic Depression Treated with Selective Serotonin Reuptake Inhibitors (SSRIs) (146 ed.). Duke Evidence-based Practice Center, Durham, NC: Agency for Healthcare Research and Quality, U.S. Department of Health and Human Services, 2006. 37 Wedlund PJ, de Leon J. Cytochrome P450 2D6 and antidepressant toxicity and response: what is the evidence? Clin Pharmacol Ther 2004;75:373–5.
Pharmacogenetics and Population Preferences 38 Chou WH, Yan FX, de Leon J, et al. Extension of a pilot study: impact from the cytochrome P450 2D6 polymorphism on outcome and costs associated with severe mental illness. J Clin Psychopharmacol 2000;20:246–51. 39 Ingelman-Sundberg M, Rodriguez-Antona C. Pharmacogenetics of drug-metabolizing enzymes: implications for a safer and more effective drug therapy. Philos Trans R Soc Lond B Biol Sci 2005;360:1563–70. 40 Ingelman-Sundberg M. Email correspondence: Preferences for genetic screening for CYP-polymorphisms, 2007. 41 Kolonien Filadelfia. Laboratoriet i Dianalund [The laboratory at Dianalund]. Kolonien Filadelfia Laboratoriet; Available from: URL: http://www.laboratoriet.epilepsihospitalet.dk/ [Accessed August 9, 2007]. 42 Bateman IJ, Carson RT, Day B, et al. Economic Valuation with Stated Preference Techniques. A manual. London: Edward Elgar, 2002. 43 Hensher D, Rose JM, Greene WH. Applied Choice Analysis—A Primer. Cambridge: Cambridge University Press, 2005. 44 Herbild L. Working Paper: Designing a DCE to Outlay Patients’ and The Publics’ Preferences for Genetic Screening in The Treatment of Depression. Odense: University of Southern Denmark, Institute of Public Health—Health Economics, 2007. 45 Ryan M, Skatun D. Modelling non-demanders in choice experiments. Health Econ 2004;13:397–402. 46 Herbild L, Gyrd-Hansen D, Bech M. Patient preferences for pharmacogenetic screening in depression. Int J Technol Assess Health Care 2008;24:96–103. 47 Ratcliffe J. The use of conjoint analysis to elicit willingness-to-pay values. Proceed with caution? Int J Technol Assess Health Care 2000;16:270–5. 48 Carson RT, Mitchell RC. The issue of scope in contingent valuation studies. Am J Agric Econ 1993;75:1263–7. 49 Slovic P, Lichtenstein S. Relative importance of probabilities and payoffs in risk taking. J Exp Psychol 1968;78:1–18. 50 Kuhfeld WF. Marketing Research Methods in SAS Experimental Design, Choice, Conjoint and Graphical Techniques (TS-722 ed.). Cary, NC: SAS Institute, 2005. 51 Efron B, Tibshirani RJ. Introduction to the Bootstrap (1st ed.). Boca Raton, FL: Chapman&Hall/CRC, 1993.
567 52 Briggs AH, Wonderling DE, Mooney CZ. Pulling costeffectiveness analysis up by its bootstraps: a non-parametric approach to confidence interval estimation. Health Econ 1997;6: 327–40. 53 Hole AR. A comparison of approaches to estimating confidence intervals for willingness to pay measures. Health Econ 2007;16: 827–40. 54 Swait J, Louviere J. The role of the scale parameter in the estimation and comparison of multinomial logit models. J Marketing Res 1993;30:305–14. 55 Daugaard J. Analyse Vedr. Depressionsmidler [Analysis with Regards to the Treatment of Depression]. Copenhagen: Gallup, 2007. 56 Cairns J, van der Pol M, Lloyd A. Decision making heuristics and the elicitation of preferences: being fast and frugal about the future. Health Econ 2002;11:655–8. 57 Burrows C, Brown K. Are any numbers better than no numbers? The sorry state of willingness-to-pay and some major methodological shortcomings. Aust Health Rev 1992;15:135–44. 58 Lloyd AJ. Threats to the estimation of benefit: are preference elicitation methods accurate? Health Econ 2003;12:393–402. 59 Tversky A, Kahneman D. The framing of decisions and the psychology of choice. Science 1981;211:453–8. 60 Slovic P, Lichtenstein S. Preference reversals: a broader perspective. Am Econ Rev 1983;73:596–605. 61 Beattie J, Covey J, Dolan P, et al. On the contingent valuation of safety and the safety of contingent valuation: part 1-caveat investigator. J Risk Uncertainty 1998;17:5–25. 62 Dubourg WR, Jones-Lee M, Loomes G, et al. Imprecise preferences and survey design in contingent valuation. Economica 1997;64:681–702. 63 Lichtenstein S, Slovic P. Reversals of preference between bids and choices in gambling decisions. J Exp Psychol 1971;89:46– 55. 64 Lichtenstein S, Slovic P, Zink D. Effect of instruction in expected value on optimality of gambling decisions. J Exp Psychol 1969;79:236–40. 65 Danmarks Statistik. Befolkningens brug af Internet 2007 [Population use of Internet in 2007] (42 ed.). Copenhagen: Danmarks Statistik, 2007.
Estimating the Danish Populations’ Preferences for Pharmacogenetic Testing Using a Discrete Choice Experiment. The Case of Treating Depression Louise Herbild, MSc,1 Mickael Bech, PhD,2 Dorte Gyrd-Hansen, PhD2 1
Danish Institute of Health Services Research, Copenhagen, Denmark and Institute of Public Health, University of Southern Denmark, Odense, Denmark; 2Institute of Public Health, University of Southern Denmark, Odense, Denmark and Odense University Hospital, Odense, Denmark [Correction added after online publication 20-November-2008: Author Affiliations have been updated]
vhe_465
560..567
A B S T R AC T Objective: The objective is to estimate willingness-to-pay (WTP) for pharmacogenetic testing in the treatment of depression. Methods: In a web-based discrete choice questionnaire, four attributes were included: 1) number of changes in antidepressants before symptom relief; 2) time with dosage adjustments due to adverse side effects and/or lack of effects; 3) cost of pharmacogenetic testing; 4) probability of benefits from pharmacogenetic testing. Respondents were asked to choose between two scenarios; 1) pharmacogenetic testing; and 2) an opt-out option reflecting a scenario without pharmacogenetic testing. The indirect utility model was assumed to be multiplicative in probability of benefits and reduced time with dosage adjustments as well as reduced number of antidepressant changes. Results: Most coefficients had the expected signs and were statistically significant. WTP for avoidance of one change in antidepressant medication
is 1571 Danish Krone (DKK), whereas WTP for reducing the period with dosage-adjustments by 1 month is DKK604. Both were statistically significantly different from zero. Conclusion: If diagnosed with depression, peoples’ WTP for pharmacogenetic testing appears to exceed its price as long as there is a reasonable probability for improvements in treatment (in the present case 10%). Utility is associated with outcomes only. Hence, other modes of provision of similar improvements in treatment may be valued equally highly. WTP estimates and the associated policy implications appear to be robust because they were unaffected by estimation model. Keywords: depression, discrete choice experiment, pharmacogenetics, preferences.
Introduction
The lack of robust evidence with regards to the optimal choice of drug, as well as starting and maintenance dosages has led to clinicians using an iterative approach to prescribing, whereby therapy is started at low dosages which are then gradually increased to decrease the risk of severe life-threatening toxic effects [4,11]. Consequently, this often leads to waiting time for alleviation of symptoms in addition to the initial minimum of 2 to 3 weeks with pharmacotherapy before symptom relief. Frequently subsequent dose-increases ends with a change in antidepressant medication and a new waiting process because treatment was either ineffective or associated with intolerable ADRs [4,7,12,13]. During this phase of treatment, patients may continue to suffer and experience severe and uncontrolled symptoms and/or ADRs [2]. The costs of depression are immense [14–17] and in the Swedish population, which has some resemblance with the Danish, annual total costs were found to equal €11,000, of which direct costs constituted €3800 [17]. Substantial costs are associated with treatment failure [18–20] as well as personal cost for patients in terms of suffering and lost quality of life.
Depression was rated as the leading cause of worldwide nonfatal disability by the World Health Organization in 2003 [1]. In Denmark, the prevalence of people needing treatment for depression is estimated to be 3.5%, although the risk of developing the disease in a lifetime perspective is 15% [2]. Depression is difficult to treat, and only 50% to 70% of patients respond to initial pharmacotherapy [2–4]. Recommended antidepressant dosages is based on comprehensive plasma concentration levels, which is sought by taking account of factors such as age, sex, other concomitant pharmaceuticals, as well as alcohol consumption, coffee, cigarette, or drug abuse, etc. For a large part of drugs, these recommended plasmaconcentration levels encompass rather wide intervals and concentration–response curves are not well-established [5,6]. Also, the consequences of choosing one drug as first treatment as opposed to another is hard to assess [7–9]. Different antidepressants are therefore tested successively to find a potent drug with a minimum of associated adverse drug reactions (ADRs)—a well-known phenomena resulting in severe problems with nonadherence to pharmacotherapy for as many as 30% to 60% of patients [2,10]. Examples of common ADRs are dry mouth, constipation, blurred vision, nausea, tremor, insomnia, and cardiac disturbance [5]. Address correspondence to: Louise Herbild, Danish Institute of Health Services Research, PO Box 2595, Dampfaergevej 27-29, DK-2100, Koebenhavn OE, Denmark. E-mail: [email protected] 10.1111/j.1524-4733.2008.00465.x
560
Taking Genetics into Account Some studies have identified that genetic variation in the drug metabolizing enzyme, CYP2D6, may be associated with individuals’ response to certain antidepressants (as well as some antipsychotics), in terms of both risk of treatment failure as well as prevalence of adverse drug reactions and observed plasma concentration levels [21–23]. Knowledge about patients’ genetic status with respect to this enzyme might reduce the length as well
© 2008, International Society for Pharmacoeconomics and Outcomes Research (ISPOR)
1098-3015/09/560
560–567
561
Pharmacogenetics and Population Preferences Table 1
Examples of antidepressants primarily metabolized by CYP2D6
Fluoxetine, Fluvoxamine, Paroxetine, Citalopram, Doxepin, Amitriptyline, Imipramine, Trimipramin, Nortryptyline, Maprotilin, Clomipramine, Venlafaxin, Mirtazapine, Mianserin [25].
as number of periods with dosage adjustments as patients’ metabolic capacity can be taken into consideration along with age, sex, lifestyle, and concomitant pharmaceuticals from the very beginning when the first dosage of an antidepressant agent is prescribed. This kind of personalized medicine—testing for genetic differences between individuals when prescribing medicines—is termed pharmacogenetic testing. The benefits of pharmacogenetic testing for the CYP2D6 gene when prescribing antidepressants is compromised by the fact that far from all antidepressants are metabolized this way [21,24]. Table 1 provides a list of substrates primarily metabolized by the CYP2D6 [25]. Aside from that, the partly unknown and complex nature of psychiatric diseases complicates the prediction of response to pharmaceuticals, as this is also affected by individual differences in pathways, receptors, disease pathology, as well as lifestyle, age, sex, and concomitant pharmaceuticals [21,26–29]. Hence, in spite of a relatively high prevalence among Caucasians where 20% to 30% have either no functioning CYP2D6 enzymes, reduced number of enzymes, or varying levels of duplications of the enzyme relative to the dominant part of Caucasians [21,30–35], no evidence on the benefits of pharmacogenetic testing is available [4,36,37]. Chou et al. [38] found trends toward greater number of ADRs and prolonged hospital stays as well as increased costs among psychiatric patients with genetically increased or decreased enzymatic activity. A large population-based cohort study [4], which examined the influence of genetically decreased enzymatic activity on intolerability of antidepressants by studying dosages, frequency of switching to another antidepressant, or discontinuation of pharmacotherapy, found an increased risk of switching between antidepressants among patients with decreased metabolic capacity as well as lower mean antidepressant maintenance dosages in these patients [4]. Studies indicate that pharmacogenetic testing for metabolizing enzymes in general might benefit between 15% and 40% of drug treatments, but numbers vary between studies [30,31,39]. A conservative assumption based on expert opinion is that only 2% to 5% of depressed patients tested would benefit from pharmacogenetic testing of the CYP2D6 [40]. Ultimately, the benefits of testing will depend on the most frequently used antidepressant (and the way this is metabolized) as well as the population under investigation as the frequency of the “non-normal” deviant metabolizers differs across ethnic groups [31]. Possible benefits of pharmacogenetic testing are reductions in the time-to-target-dose as well as reductions in the number of changes in antidepressant medication as patients can be characterized by whether they are expected to be able to tolerate ordinary dosages or whether dosages should either be decreased or increased, or if a pharmaceutical not primarily dependent on the CYP2D6 enzyme should be chosen as first choice. Pharmacogenetic testing for changes in the CYP2D6 gene has been introduced at one psychiatric hospital in Denmark and is available at the price of DKK1630 (€219) [41]. A single blood sample (or mouth swab) is required; and because our genes do not change, the test will never need to be conducted again. Other psychiatric hospitals are likely also to consider implementing the pharmacogenetic test in the future. We therefore wanted to see if pharmacogenetic testing for CYP2D6 and the
potential consequences for the treatment of depression would be of any value to potential patients and if this could be modeled by means of a discrete choice experiment. The purpose of this study was to elicit the Danish populations’ preferences for the introduction of routine pharmacogenetic testing among newly diagnosed depressed patients about to be treated with antidepressants. The aim was to estimate the monetary use value of the pharmacogenetic test.
Methods
Discrete Choice Experiment A discrete choice experiment (DCE) was used to elicit preferences for pharmacogenetic testing among a panel of Danes, 18 to 65 years old. In a DCE, respondents are faced with hypothetical choices between treatment scenarios that differ in terms of specified attributes and attribute levels. The value of the good (in this case the pharmacogenetic test) is determined by the magnitude of these attributes, and by varying the levels of these in hypothetical scenarios, respondents are forced to make trade-offs. Those incremental changes in attributes which are relevant for respondents’ choices can then be identified. On the basis of the relative weighting of the attributes, a compensation variation measure that conforms with demand theory is derived [42,43].
Survey Design Seven hundred one respondents were e-mailed a link to the study questionnaire sent out by an independent research agency (Gallup) in August 2007. The sample was stratified by age, sex, geographic location, and education. Participants in the research panel from which respondents were sampled were recruited randomly from telephone interviews and invited to participate in the Internet-based research panel. Participation in the research panel is rewarded by participation in a lottery. The questionnaire was developed on the basis of a literature review, expert opinion and three focus group interviews with depressive patients. The focus group interviews were conducted in an iterative manner from which problems related to the pharmaceutical treatment of depression were specified in more detail and related to those elements of the pharmaceutical treatment that the pharmacogenetic test could have an impact upon. A more detailed description of the development of the questionnaire and results from the focus group interviews are available in English elsewhere [44]. From the focus group interviews, it was clear that patients clearly prefer if the correct antidepressant could be administered at the right dosage from the very beginning of their pharmaceutical treatment. Two attributes found to be of importance to patients’ experience of the pharmaceutical treatment which could be affected by pharmacogenetic testing were identified: 1) the number of changes in antidepressant medication before symptom relief; and 2) the time with dosageadjustments due to lack of effect and/or unacceptable ADRs. These two attributes are hereafter called the effect-attributes and both relates to the experience of the health improvement associated with the treatment. The number of changes in antidepressant medication was important to patients because each change represents a realization that the current antidepressant will not be effective or is associated with intolerable ADRs, that time has been wasted, and that a new course of treatment with associated risks of ADRs must be initiated. Although changing antidepressants was associated with a negative utility due to frustration and the facing of new risks, the time with dosage adjustments was found to be frustrating because of the waiting time and the lack
Herbild et al.
562 of symptom relief and/or intolerable ADRs, as well as the uncertainty regarding the response of an increased or decreased dosage. In the questionnaire, it was clearly explained to respondents that the number of changes and the total time with dosage adjustments (across all medication attempts) could vary independently. This is reality in practice, where a prescribing doctor may choose to switch medications early in the process; or alternatively opt for longer periods with dosage adjustment in an attempt to avoid switching. Hence, a reduced number of changes may involve longer or shorter total time with dosage adjustments, and a longer dosage-adjustment period may involve fewer or more medicine changes. Clinically, the relevance of the two attributes is supported by evidence on increased number of antidepressant changes as well as difference in maintenance dosages among patients with genetically altered enzymatic capacity relative to patients with normal capacity [4]. The pharmacogenetic tests’ ability to affect the two effectattributes depends upon whether the “first choice” prescribed antidepressant is metabolized by the CYP2D6 and whether the patient has a non-normal metabolic capacity. Other factors such as specific pathways, lifestyle, concomitant pharmaceuticals, etc., may also affect the effects of the pharmacogenetic test. Hence, a significant proportion of patients will not gain from pharmacogenetic testing, but at the time the test is performed, one cannot determine who these individuals are. Consequently, we included a third attribute representing the probability of getting any benefit (i.e., any reduction in the time with dosage adjustment and any reductions in the number of antidepressant changes) from the test. Finally, a price of the test was included, allowing for the calculation of marginal willingness-to-pay (WTP) estimates. Each respondent was presented with eight choice-sets, each containing two scenarios; one scenario represented treatment outcomes with the pharmacogenetic test and the other represented a status quo scenario with treatment outcomes as standard practice is today without the pharmacogenetic test. This status quo scenario represents an opt-out option [45]. In this case, the opt-out option is specified in accordance with the expected outcomes of current practice, and the respondent is thereby provided with the necessary information required to make an informed choice. Respondents were given a thorough description of depression and the symptoms characterizing the disease as well as the pharmacological treatment and the related problems. The attributes were carefully explained and respondents were told to think of the test as an insurance-policy “you do not know whether you will benefit, but you will have to pay with certainty. If you do not benefit your treatment outcome will be equivalent to the scenario without the test.” Respondents were faced with the following question: “Imagine that you have just been diagnosed with depression for the first time. Which of the two treatment possibilities would you choose?” An example of a choice set is shown in Table 2. The treatment scenario without the pharmacogenetic test was constructed as a fixed status quo scenario and represented a treatment profile for a group of moderately to severely depressed patients’ first episode of depression. Levels for the two effectattributes characterizing this scenario were based on median first episode values obtained in a previous study among moderately to severely depressed patients treated without the test in accordance with current practice [46]. The levels were four changes in antidepressants and 5 months with dosage adjustments. A positive price-attribute was associated with the pharmacogenetic test scenario. Respondents were informed that on an average, having the test would not affect out-of-pocket pharmaceutical expenses.
Table 2 Example of a choice set presented for the respondents in the discrete choice experiment Imagine you have just been diagnosed with depression for the first time. Which of the two treatment possibilities would you choose? Likelihood of getting improvements from the test: Price of the test (expenses for medicine remains constant): Number of changes in antidepressant medicine: Time with dosage adjustments due to lack of effect and/or unacceptable adverse side effects:
Genetic test
No genetic test
50% DKK18.000
DKK0
2 changes
4 changes
3 months
5 months
DKK, Danish Krone.
In the pharmacogenetic treatment scenario, attribute levels were conservatively reduced by one or two changes in antidepressants, and the period with dosage adjustments was reduced by 2 or 4 months. These values were entered into the regression analyses by way of difference coding as reductions of one and two, and two and four, respectively. To secure that our priceattribute covered both a lower as well as an upper bound of WTP [47], eight levels were chosen for the price attribute in the pharmacogenetic treatment scenario, ranging from DKK200 (€27) to DKK18,000 (€2420). The price of the test was stated as an out-of-pocket, once in a lifetime payment. The prevalence of genetically non-normal metabolizers in Denmark is almost 20% [35]. In the present study, we assumed that half of these could benefit from pharmacogenetic testing. Hence, the base-case probability of benefiting from the test was set at 10%. In other stated preference studies, it has been found that respondents are frequently insensitive to scope, and this problem is enhanced when respondents face probabilities [48,49]. We therefore chose to present respondents with another probability level to test if individuals are indeed sensitive to the level of probability. This level was set at 50% to distinguish it from the base-level. The probability attribute was thus coded as a dummy variable (which equaled 1 if the probability was 50%; otherwise zero) and the associated coefficient was used to test for sensitivity to scope. Attributes and attribute levels as well as coding are shown in Table 3.
Statistical Design A fractional factorial experimental design optimizing Defficiency was generated in SAS [50] using the MktEx macro, resulting in a design with 32 hypothetical alternatives. These 32 Table 3 Attributes and attribute levels used in the discrete choice experiment
Attribute Likelihood of getting improvements from the test: Price of the test
Number of changes in antidepressant medicine: Time with dosage adjustments due to lack of effect and/or unacceptable adverse side effects:
Attribute levels (coding) in the scenario representing pharmacogenetic testing 10% (0), 50% (1) DKK200; DKK600; DKK1000; DKK1500; DKK3000; DKK6000; DKK9000; DKK18,000 2 changes (2), 3 changes (1) 1 months (4), 3 months (2)
563
Pharmacogenetics and Population Preferences alternatives were distributed across four blocks of eight using the MktBlock macro in SAS that creates a blocking factor, which is uncorrelated with every attribute of every alternative while ensuring the maximum degree of design efficiency. The 32 alternatives were each paired with the fixed status quo alternative such that a total of 32 choice sets were constructed. The design is available from the authors on request. In the design, we allowed for two-way interaction effects to be estimated between the probability of getting an effect and the attributes on the effect i.e., reductions in the number of antidepressant changes and time with dosage adjustments.
Theory of Expected Utility Using the two attributes on “number of changes in medicine” and “time with dosage adjustments before symptom relief” to describe the pharmacogenetic test (the effect-attributes), we implicitly assume that reductions in these represent the total observable value of the pharmacogenetic test, i.e., its utility-generating properties. As the underlying assumption of the DCE is that individuals are utility-maximizing, we expect respondents to maximize utility with respect to the two effect-attributes subject to the price of the pharmacogenetic test. The utility gain depends upon the probability by which the effects will be provided. According to the theory of expected utility (EUT), we assume that the expected utility is a multiplicative function of probability and effect. Respondents are thus expected to trade among treatment scenarios in a way that maximizes their utility from the effects of the test subject to the probability of getting these and the price. Hence, we assume that the utility function can be represented by the following specification:
ΔU test = αˆ test ASC + βˆ price Δx1 + βˆ change Δx2 + βˆ time Δx3 + βˆ change ∗ prob Δx2 Δd50 + βˆ time ∗ prob Δx3Δd50
(1)
The coefficient βˆ price represents the marginal disutility of the price of the test (because this has to be paid in addition to existing costs of treatment without pharmacogenetic testing), and βˆ change and βˆ time represents the utility of marginal changes in any of the two effect-attributes when the probability of effect is 10% (reference level). Dx2Dd50 and Dx3Dd50 constitute added utility associated with changes in the effect-attributes when the probability level is 50% (d50 equals one if the probability is 50%; else, zero). The coefficient αˆ test for the alternative specific constant (ASC) captures any utility associated with the test per se.
version 9.2 (Stata Corp., College Station, TX). WTP-estimates were calculated as marginal rates of substitution (MRS) between the coefficients of the effect-attributes and the coefficient on price, e.g., MRS = −Δβˆ change Δβˆ price. Confidence intervals (CI) for WTP-estimates were calculated using the parametric bootstrapping procedure specified by Krinsky-Robb with 10,000 replications [51–53]. To test for preference heterogeneity across relevant groups without introducing problems of scale [54], we compared WTPestimates in a number of stratified analyses or included interaction terms between the attributes and dummy variables representing sub-groups. We tested 1) if WTP differed across income groups with a household income above or below DKK350,000 (€47,000); 2) if current or former pharmaceutical treatment for depression among respondents affected WTP; 3) if prior diagnosis of depression and/or having friends or relatives with present or past diagnosis of depression and experience with antidepressants affected WTP; 4) if there was a difference in WTP across sex, age, and education; and 5) if perceived difficulty in answering the DCE-questions was associated with differences in WTP-estimates.
Results A total of 701 questionnaires were sent out. Four respondents were noneligible to participate, 74 did not wish to participate, 39 did not complete any of the DCE questions, and 261 did not reply at all. A total of 323 completed questionnaires were returned, providing a response rate of 46%.
Respondent Descriptives Males represented 47% of respondents. Compared with the general Danish population, respondents aged 18 to 24 years were slightly under-represented and respondents with a high level of education were over-represented [55]. Nineteen respondents did not like the idea of pharmacogenetic testing or/and would never want to have the test. Some of these were among the 43 nontraders always choosing the no-test scenario. As we wanted to estimate the use value of the test, these respondents were excluded because we interpret these as being nonusers within the given spectrum of attribute levels. Three respondents had already had the pharmacogenetic test and were also excluded. This left us with a total sample of 277 individuals each answering eight choice sets.
Preferences for the Pharmacogenetic Test Econometric Analysis The data were analyzed using the conditional logistic regression procedure, clogit, in the statistical software package of STATA Table 4
Results of the conditional logistic regression analysis are presented in Table 4. As expected, the coefficient on the price attribute is negative, indicative of decreasing utility as the price of
Results from the DCE assuming a multiplicative utility function in accordance with expected utility theory
ASC Price Changes (reductions in medicine changes) Changes*d50 (added utility if probability of effect increases to 50%) Time (reduction in the time with dosage adjustments) Time*d50 (added utility if probability of effect increases to 50%) N (observations/respondents) AIC* LR chi-square Pseudo R2
Coefficient
t-value
0.3999 -0.0003 0.3999 -0.022 0.1528 0.1655
0.47 -19.29 3.42 -0.15 2.62 2.22
Prob > |t| 0.637 0.000 0.001 0.884 0.009 0.027 4432/277 2354 730.12 0.2377
WTP (DKK) 430 — 1571 -70 604 645
*AIC = 2 k - 2 ln (L), where k is the number of parameters in the model and L is the maximized value of the likelihood function for the estimated model. AIC, Akaike information criterion; CI, confidence interval; DKK, Danish Krone; WTP, willingness-to-pay.
90% CI (DKK) -1007 — 809 -1055 230 162
1860 — 2331 903 986 1135
Herbild et al.
564 the test increases. The coefficients of the effect-attributes are positive, signifying a utility increase associated with decreases in the number of antidepressant changes or reductions in time with dosage adjustments. This is all in accordance with a priori expectations and the coefficients are statistically significant (P > |t| less than 0.01). The coefficient of the interaction term between changes in antidepressant medicine and probability is unexpectedly negative, but not statistically significantly different from zero, although the interaction between the time-attribute and probability is positive and significantly different from zero (P = 0.027), though smaller than expected (according to EUT). Preferences for reductions in the number of changes in antidepressant products do not seem to be affected by the probability level at which improvements are given. The constant represents the value of the pharmacogenetic test per se, i.e., irrespective of its effects on treatment outcome. It is statistically insignificant, which implies that the value of the test is associated with outcomes only.
Marginal WTP Estimates of WTP for a marginal change in attributes are also shown in Table 4, including 90% CIs. The WTP estimate for a 10% probability of a reduction of one in the number of antidepressant changes is DKK1571 (90% CI DKK809–2331) and the WTP for a reduction of 1 month in the time with dosage adjustments with 10% probability is DKK604 (90% CI DKK230– 986). WTP for reductions in a number of changes is not statistically higher when the probability of obtaining this gain is 50%, although WTP for an increase in the chance of obtaining a month of reduction in the time with dosage adjustment to 50% is DKK645 (90% CI DKK162–1135). The total value of 1 month reduction in the time with dosage-adjustments with 50% probability is thus DKK1249. The CI for the WTP for the ASC is wide and not statistically significantly different from zero, signifying that on average people only value the pharmacogenetic test by the reductions in the attributes, although some might value the more intangible benefits of the test as well because it is slightly positive. Overall, the results imply that if the pharmacogenetic test with a chance between 10% and 50% can reduce the number of antidepressant changes by one and shorten the dosage adjustment period by 1 month, the WTP for the pharmacogenetic test exceeds the price of the test (current price is DKK1630).
Subgroup Analyses Including an interaction term between a dummy (equal to one for high-income households; else, zero) and the price attribute indicated that WTP for the test differed between respondents with different levels of income. This was expected and indicates that respondents have taken budget constraints into consideration when they completed the DCE. Almost 17% of the respondents had been diagnosed with depression by their physician. Most of these (81%; n = 35) had also been treated with antidepressants. As the choice sets were designed with two hypothetical scenarios and these respondents still constituted potential users of pharmacogenetic testing, the respondents were not excluded. We hypothesized that preferences for the pharmacogenetic test might be affected by previous experience with antidepressants. Inclusion of an interaction term between the attribute on number of antidepressant changes and a dummy variable (equal to one if respondents had been treated with antidepressants; else, zero) demonstrated no statistically significant differences between those with and without experience. We also tested for differences in preferences across groups with a presumed different level of
knowledge of depression. Sixty-five percent of the respondents knew somebody among their nearest friends or family who had been treated pharmacologically for depression, and some of these were among the 17% who had been diagnosed with depression themselves. Including an interaction term between the timeattribute and a dummy variable (equal to one if knowledge of depression; else zero), suggested no statistically significant differences in preferences because of knowledge. We also tested for sex-differences, age-related differences, and educational differences but found no impact on preferences. Finally, we tested if reported self-perceived difficulty in answering the choicequestions could be taken as a marker of preference heterogeneity among respondents. Difficulty in answering the choice questions was associated with insignificant WTP-estimates for both effectattributes and a clear tendency to choose on the basis of price alone. In sum, we found no evidence of heterogeneity in preference structures because of personal experience of antidepressant medication, knowledge of the disease, or sex, age, and education.
Discussion
Internal Validity The results of the discrete choice exercise, conducted among a random sample of the Danish population who is to imagine having been recently diagnosed with depression, suggest that the stated expected use value of pharmacogenetic testing exceeds the current price of the test, even if the test offers relatively small expected gains. If, for example, the pharmacogenetic test provides a 1-month reduction in time with dosage-adjustments as well as one change less in antidepressant products, with a probability between 10% to 15%, the mean WTP among potential users exceeds the cost of the pharmacogenetic test (current price is DKK1630). The price of the test does however not cover the entire costs of the test because blood samples have to be taken and the results interpreted. Such analyses are associated with marginal resource consumption in terms of labor costs. Consequently, pharmacogenetic testing will only be associated with a welfare gain if the marginal labor costs associated with performing and analyzing the test are small, or if the benefits of the pharmacogenetic test exceed the 10% probability of reducing the number of antidepressant changes by one and the dosage adjustment period by 1 month. The small or statistically insignificant coefficients associated with the interaction terms between the effect attributes and the probability dummy variable (equal to one if 50%; else, zero) suggests that individuals are either insensitive to the level of probability, or that respondents’ preferences in some way do not accord with EUT. There has been much criticism of EUT and its underlying assumptions [56,57], and studies have shown that when respondents face gambles, they tend to use decision heuristics that are not in accordance with EUT [49,58–64]. Some of these studies have shown that when facing gambles, respondents tend to focus either on outcomes or probabilities rather than expected value. Our study appears to suffer from similar problems. We therefore tested if a model in which outcome and probability entered the utility function as additive components, rather than in a multiplicative manner provided a better fit to the data. This was not the case (Akaike Information Criterion of 2377 compared with 2354). With the additive model specification, all coefficients except the ASC were highly statistically significant (P > |t| less than 0.001). The WTP for an increase in the probability of an effect from 10% to 50% was estimated to be DKK1984 (SD 403). Our WTP-estimates were relatively unaffected by choice of model and our results seem to be fairly robust.
565
Pharmacogenetics and Population Preferences We conducted a number of stratified analyses to compare WTP for the pharmacogenetic test across subgroups as well as test for preference homogeneity. These showed a large degree of homogeneity across strata except for that of income, which confirmed decreasing marginal utility of income and increasing price-insensitivity with rising income, as would be expected theoretically. For a subgroup of respondents reporting difficulties with the DCE questions, choices seemed to be based primarily on the level of the price attribute. This consequently led to a higher WTP for the test itself (the ASC) and lower WTP-estimates for marginal changes in the effect-attributes.
External Validity Respondents aged 18 to 24 were slightly underrepresented, and people with higher education and females were over-represented relative to the Danish population. This result may be due to the fact that the questionnaire was distributed electronically via email. Nevertheless, in Denmark, the majority of the population has access to the Internet in their homes, and email is a common source of communication [65]. Hence, we do not think that the distribution method in itself has introduced a severe bias in representation. The observed bias is more likely to be due to interest, ability, and availability of time. Only if the younger and/or highly educated people have markedly different preference structures in relation to pharmacogenetic testing does the small bias in representation introduce bias in the presented results. Subgroup analyses showed that sex, youth, and a high level of education had no impact on preference structures. The levels of the attributes applied in the status quo scenario were based on estimates obtained from a similar DCE conducted among patients, where respondents were asked about their experienced number of changes in antidepressants and experienced time with dosage adjustment for their first episode of depression [46]. A majority of the interviewed respondents were likely to be moderately to severely formerly or currently depressed patients. Assuming the marginal utility of reductions in the effectattributes is affected by the absolute levels of the effectsattributes presented in the status quo scenario, the results presented here are only representative of preferences for pharmacogenetic testing among moderately to severely depressed patients.
Design of the Scenarios The price of the test was presented as an out-of-pocket once-ina-lifetime payment. Treatment in the Danish health-care system is primarily free of charge and collectively financed by taxes, but there are relatively high copayments on dentist services and pharmaceuticals. Because the pharmacogenetic test is very closely linked to pharmacotherapy, the choice of an out-of-pocket payment vehicle is unlikely to severely impair the theoretical realism of the presented scenarios. We observed 43 nontraders always choosing the opt-out alternative, and it is likely that some of these have been protest-voters, unwilling to choose the test because of an aversion to out-of-pocket payment. An alternative payment vehicle such as increases in income taxes could have been applied to avoid protest bidding, but this would have led to other problems. Payment via income tax would involve paying annually over a lifetime, for what is actually a once-in-a-lifetime investment. Hence, even minute levels of annual tax-payments would have constituted large and most likely overestimated WTP values for the pharmacogenetic test. Thus, we chose the more conservative approach, acknowledging that the WTP values based on out-of-pocket payment may provide WTP-estimates that underestimate the value of the pharmacogenetic test because of protest bidding.
The pharmacogenetic test has not yet been tested in larger randomized controlled trials, and the efficacy of the test is unknown. The chosen level for the probability attribute was consequently based on assumptions with regards to the percentage of patients taking antidepressants metabolized by CYP2D6 as well as the number of patients with genetic and nongenetic characteristics affecting the benefits of pharmacogenetic testing. In clinical settings where the predominantly used antidepressants are selective serotonin reuptake inhibitors (SSRI), the likelihood of benefiting from the test is likely to be lower than the 10% assumed in this study (because the SSRIs are metabolized by CYP2D6 to a lesser extent). In contrast, settings where tricyclic antidepressants products are predominantly used would most likely attain a larger probability of benefit than the 10% (because of larger dependence on CYP2D6-metabolism among these products). Also, in populations with a larger frequency of “non-normal” metabolizers, e.g., in Africans, benefits will be larger, although the opposite is the case in populations with a lower frequency, e.g., in Asian populations. Nevertheless, clinical and genetic research within the area is expanding so fast that even if the present “true” benefit level were available it would most likely be outdated within a few years. In addition, pharmacogenetic testing for CYP2D6-polymorphisms is likely to be beneficial in relation to pharmaceutical treatments for several conditions because other pharmaceuticals than antidepressants are metabolized by the CYP2D6-enzyme. Finally the effects of pharmacogenetic testing for other genetic changes than those of the CYP2D6, i.e., in other metabolizing enzymes and/or clinical relevant pathways, is likely to also influence the potential benefits of pharmacogenetic testing in depression. Therapeutic drug monitoring (TDM) using probe drugs, or phenotyping as it is also called, has the potential of delivering the same treatment improvements as pharmacogenetic testing [7]. The results from phenotyping are more precise, but not stable as weight, drinking, and smoking habits, etc., can alter an individuals’ metabolizing capacity, thus necessitating the conduction of phenotyping tests at each new disease episode. TDM can also be used with antidepressants, but then only after pharmacotherapy has been initiated and then plasma levels are measured at repeated intervals. A debate about whether pharmacogenetic testing is superfluous if routine TDM is used exists among practitioners. As our results indicate, respondent’s value the outcomes associated with the pharmacogenetic test but not the test per se. Thus, the demand for pharmacogenetic testing observed in the present study may not reflect the demand in real life if TDM (in either shape) is an option. Future studies investigating preferences for TDM versus pharmacogenetic testing are therefore recommended.
Conclusion Respondents have preferences for reductions in the number of changes in antidepressant products as well as reductions in the time with dosage adjustments due to adverse side effects and/or lack of effect. Assuming that a pharmacogenetic test is able to offer a 10% chance of one change less in antidepressant products and a reduction of 1 month in the total time with dosage adjustments, respondents are willing to pay an amount which is greater than the actual price of the pharmacogenetic test and the introduction of such a test will be improving welfare. The value of reductions in the number of changes in antidepressants is higher than reductions in the time with dosage adjustments, and 2 months of dosage adjustments are preferred if it can prevent one change in antidepressant medication. This is valuable information when doctors choose their prescription strategy.
566 Subgroup analyses indicate that preferences for this kind of pharmacogenetic testing for CYP2D6 are consistent across groups with different levels of knowledge of depression, suggesting that the hypothetical DCE exercise was well-understood by the respondents. The fact that respondents are insensitive to the scope of the probability attribute does raise concern. It is however reassuring that the valuation of the pharmacogenetic test is robust as WTP-estimates were relatively unaffected by choice of model specification. Source of financial support: Center for Pharmacogenomics at the University of Copenhagen as well as the Danish Institute for Health Services Research
References 1 Mathers CD, Bernard C, Moesgaard KI, et al. Global Burden of Disease 2002: Data Sources, Methods and Results (54th ed.). Geneva: WHO, 2003. 2 Mors NPO, Børglum AD. Status og perspektiver inden for psykiatrisk genomisk medicin [status of and perspectives on psychiatric genomic medicine]. Ugeskr Læger 2005;167:2191–3. 3 Kirchheiner J, Nickchen K, Bauer M, et al. Pharmacogenetics of antidepressants and antipsychotics: the contribution of allelic variations to the phenotype of drug response. Mol Psychiatry 2004;9:442–73. 4 Bijl MJ, Visser LE, Hofman A, et al. Influence of the CYP2D6*4 polymorphism on dose, switching and discontinuation of antidepressants. Br J Clin Pharmacol 2008;65:558–64. 5 Linder MW, Keck PE, Jr Standards of laboratory practice: antidepressant drug monitoring. National Academy of Clinical Biochemistry. Clin Chem 1998;44:1073–84. 6 Balant-Gorgia AE, Balant LP, Garrone G. High blood concentrations of imipramine or clomipramine and therapeutic failure: a case report study using drug monitoring data. Ther Drug Monit 1989;11:415–20. 7 Hiemke C. Clinical utility of drug measurement and pharmacokinetics—therapeutic drug monitoring in psychiatry. Eur J Clin Pharmacol 2008;64:159–66. 8 Nierenberg AA. The medical consequences of the selection of an antidepressant. J Clin Psychiatry 1992;53(Suppl.):S19–24. 9 Anderson IM. Meta-analytical studies on new antidepressants. Br Med Bull 2001;57:161–78. 10 Brook O, van Hout H, Nieuwenhuyse H, Heerdink E. Impact of coaching by community pharmacists on drug attitude of depressive primary care patients and acceptability to patients: a randomized controlled trial. Eur Neuropsychopharmacol 2003;13: 1–9. 11 Zackrisson AL, Holmgren P, Gladh AB, et al. Fatal intoxication cases: cytochrome P450 2D6 and 2C19 genotype distributions. Eur J Clin Pharmacol 2004;60:547–52. 12 Corey-Lisle PK, Birnbaum HG, Greenberg PE, et al. Identification of a claims data “signature” and economic consequences for treatment-resistant depression. J Clin Psychiatry 2002;63:717– 26. 13 Ruhé HG, Huyser J, Scholten RJPM, et al. Antidepressant doseincrease for non-responders to a standard dose for depression. (Protocol) Cochrane Database of Systematic Reviews 2007, Issue 1. Art. No.: CD004045. DOI: 10.1002/14651858. CD004045.pub2. 14 Berto P, D’Ilario D, Ruffo P, et al. Depression: cost-of-illness studies in the international literature, a review. J Ment Health Policy Econ 2000;3:3–10. 15 Cuijpers P, Smit F, Oostenbrink J, et al. Economic costs of minor depression: a population-based study. Acta Psychiat Scand 2007;115:229–36. 16 Luppa M, Heinrich S, Angermeyer MC, et al. Cost-of-illness studies of depression: a systematic review. J Affect Disord 2007; 98:29–43.
Herbild et al. 17 Sobocki P, Lekander I, Borgstrom F, et al. The economic burden of depression in Sweden from 1997 to 2005. Eur Psychiatry 2007;22:146–52. 18 Russell JM, Hawkins K, Ozminkowski RJ, et al. The cost consequences of treatment-resistant depression. J Clin Psychiatry 2004;65:341–7. 19 McCombs JS, Nichol MB, Stimmel GL, et al. The cost of antidepressant drug therapy failure: a study of antidepressant use patterns in a Medicaid population. J Clin Psychiatry 1990;51 (Suppl.):S60–9. 20 Sobocki P, Ekman M, Agren H, et al. The mission is remission: health economic consequences of achieving full remission with antidepressant treatment for depression. Int J Clin Pract 2006;60: 791–8. 21 Kirchheiner J, Brøsen K, Dahl ML, et al. CYP2D6 and CYP2C19 genotype-based dose recommendations for antidepressants: a first step towards subpopulation-specific dosages. Acta Psychiat Scand 2001;104:173–92. 22 Mulder H, Herder A, Wilmink FW, et al. The impact of cytochrome P450-2D6 genotype on the use and interpretation of therapeutic drug monitoring in long-stay patients treated with antidepressant and antipsychotic drugs in daily psychiatric practice. Pharmacoepidemiol Drug Saf 2006;15:107–14. 23 Rau T, Wohlleben G, Wuttke H, et al. CYP2D6 genotype: impact on adverse effects and nonresponse during treatment with antidepressants-a pilot study. Clin Pharmacol Ther 2004;75:386– 93. 24 Andersson T, Flockhart DA, Goldstein DB, et al. Drugmetabolizing enzymes: evidence for clinical utility of pharmacogenomic tests. Clin Pharmacol Ther 2005;78:559–81. 25 Medicin.dk. Medicin.dk [online Danish medicine catalogue]. Medicin dk; Available from: URL: http://www.medicin.dk/ (41ezgpnlrwn5gbrmdj55au55)/show.aspx [Accessed April 9, 2007]. 26 Biosam. Farmakogenetik i Danmark [Pharmacogenetics in Denmark]. Biosam 2003;15. 27 Brøsen K. Some aspects of genetic polymorphism in the biotransformation of antidepressants. Therapie 2004;59:5–12. 28 Brøsen K. Klassisk farmakogenetik [Classical pharmacogenetics]. Ugeskr Laeger 2005;167:2143–6. 29 Caraco Y. Genes and the response to drugs. N Engl J Med 2004;351:2867–9. 30 Heller T, Kirchheiner J, Armstrong VW, et al. AmpliChip CYP450 GeneChip: a new gene chip that allows rapid and accurate CYP2D6 genotyping. Ther Drug Monit 2006;28:673–7. 31 Ingelman-Sundberg M. Genetic polymorphisms of cytochrome P450 2D6 (CYP2D6): clinical consequences, evolutionary aspects and functional diversity. Pharmacogenomics J 2005;5:6–13. 32 Ingelman-Sundberg M. The human genome project and novel aspects of cytochrome P450 research. Toxicol Appl Pharmacol 2005;207(Suppl. 2):52–6. 33 Kawanishi C, Lundgren S, Agren H, Bertilsson L. Increased incidence of CYP2D6 gene duplication in patients with persistent mood disorders: ultrarapid metabolism of antidepressants as a cause of nonresponse. A pilot study. Eur J Clin Pharmacol 2004; 59:803–7. 34 Raimundo S, Fischer J, Eichelbaum M, et al. Elucidation of the genetic basis of the common “intermediate metabolizer” phenotype for drug oxidation by CYP2D6. Pharmacogenetics 2000; 10:577–81. 35 Rasmussen JO, Christensen M, Svendsen JM, et al. CYP2D6 gene test in psychiatric patients and healthy volunteers. Scand J Clin Lab Invest 2006;66:129–36. 36 Matchar DB, Thakur ME, Grossman I, et al. Testing for Cytochrome P450 Polymorphisms in Adults with Non-Psychotic Depression Treated with Selective Serotonin Reuptake Inhibitors (SSRIs) (146 ed.). Duke Evidence-based Practice Center, Durham, NC: Agency for Healthcare Research and Quality, U.S. Department of Health and Human Services, 2006. 37 Wedlund PJ, de Leon J. Cytochrome P450 2D6 and antidepressant toxicity and response: what is the evidence? Clin Pharmacol Ther 2004;75:373–5.
Pharmacogenetics and Population Preferences 38 Chou WH, Yan FX, de Leon J, et al. Extension of a pilot study: impact from the cytochrome P450 2D6 polymorphism on outcome and costs associated with severe mental illness. J Clin Psychopharmacol 2000;20:246–51. 39 Ingelman-Sundberg M, Rodriguez-Antona C. Pharmacogenetics of drug-metabolizing enzymes: implications for a safer and more effective drug therapy. Philos Trans R Soc Lond B Biol Sci 2005;360:1563–70. 40 Ingelman-Sundberg M. Email correspondence: Preferences for genetic screening for CYP-polymorphisms, 2007. 41 Kolonien Filadelfia. Laboratoriet i Dianalund [The laboratory at Dianalund]. Kolonien Filadelfia Laboratoriet; Available from: URL: http://www.laboratoriet.epilepsihospitalet.dk/ [Accessed August 9, 2007]. 42 Bateman IJ, Carson RT, Day B, et al. Economic Valuation with Stated Preference Techniques. A manual. London: Edward Elgar, 2002. 43 Hensher D, Rose JM, Greene WH. Applied Choice Analysis—A Primer. Cambridge: Cambridge University Press, 2005. 44 Herbild L. Working Paper: Designing a DCE to Outlay Patients’ and The Publics’ Preferences for Genetic Screening in The Treatment of Depression. Odense: University of Southern Denmark, Institute of Public Health—Health Economics, 2007. 45 Ryan M, Skatun D. Modelling non-demanders in choice experiments. Health Econ 2004;13:397–402. 46 Herbild L, Gyrd-Hansen D, Bech M. Patient preferences for pharmacogenetic screening in depression. Int J Technol Assess Health Care 2008;24:96–103. 47 Ratcliffe J. The use of conjoint analysis to elicit willingness-to-pay values. Proceed with caution? Int J Technol Assess Health Care 2000;16:270–5. 48 Carson RT, Mitchell RC. The issue of scope in contingent valuation studies. Am J Agric Econ 1993;75:1263–7. 49 Slovic P, Lichtenstein S. Relative importance of probabilities and payoffs in risk taking. J Exp Psychol 1968;78:1–18. 50 Kuhfeld WF. Marketing Research Methods in SAS Experimental Design, Choice, Conjoint and Graphical Techniques (TS-722 ed.). Cary, NC: SAS Institute, 2005. 51 Efron B, Tibshirani RJ. Introduction to the Bootstrap (1st ed.). Boca Raton, FL: Chapman&Hall/CRC, 1993.
567 52 Briggs AH, Wonderling DE, Mooney CZ. Pulling costeffectiveness analysis up by its bootstraps: a non-parametric approach to confidence interval estimation. Health Econ 1997;6: 327–40. 53 Hole AR. A comparison of approaches to estimating confidence intervals for willingness to pay measures. Health Econ 2007;16: 827–40. 54 Swait J, Louviere J. The role of the scale parameter in the estimation and comparison of multinomial logit models. J Marketing Res 1993;30:305–14. 55 Daugaard J. Analyse Vedr. Depressionsmidler [Analysis with Regards to the Treatment of Depression]. Copenhagen: Gallup, 2007. 56 Cairns J, van der Pol M, Lloyd A. Decision making heuristics and the elicitation of preferences: being fast and frugal about the future. Health Econ 2002;11:655–8. 57 Burrows C, Brown K. Are any numbers better than no numbers? The sorry state of willingness-to-pay and some major methodological shortcomings. Aust Health Rev 1992;15:135–44. 58 Lloyd AJ. Threats to the estimation of benefit: are preference elicitation methods accurate? Health Econ 2003;12:393–402. 59 Tversky A, Kahneman D. The framing of decisions and the psychology of choice. Science 1981;211:453–8. 60 Slovic P, Lichtenstein S. Preference reversals: a broader perspective. Am Econ Rev 1983;73:596–605. 61 Beattie J, Covey J, Dolan P, et al. On the contingent valuation of safety and the safety of contingent valuation: part 1-caveat investigator. J Risk Uncertainty 1998;17:5–25. 62 Dubourg WR, Jones-Lee M, Loomes G, et al. Imprecise preferences and survey design in contingent valuation. Economica 1997;64:681–702. 63 Lichtenstein S, Slovic P. Reversals of preference between bids and choices in gambling decisions. J Exp Psychol 1971;89:46– 55. 64 Lichtenstein S, Slovic P, Zink D. Effect of instruction in expected value on optimality of gambling decisions. J Exp Psychol 1969;79:236–40. 65 Danmarks Statistik. Befolkningens brug af Internet 2007 [Population use of Internet in 2007] (42 ed.). Copenhagen: Danmarks Statistik, 2007.