Effects of Direct-to-Consumer Advertising on
`Medication Choice: The Case of Antidepressants
`
`Julie M. Donohue and Ernst R. Berndt
`
`Although direct-to-consumer advertising (DTCA) has generated substantial controversy, little is known
`about its effects on consumer and physician behavior. In this article, the authors examine the impact
`of DTCA and physician detailing on the choice of antidepressant medication. The authors find that
`detailing has a much greater effect on medication choice in the antidepressant market than does
`DTCA.
`
`Pharmaceutical promotion has traditionally been aimed
`
`at physicians, the “learned intermediaries” who are
`responsible for prescribing medications. From the
`mid-twentieth century, when federal regulations began
`requiring a doctor’s prescription for many pharmaceuticals,
`to the 1990s, pharmaceutical firms relied primarily on
`“detailing” by pharmaceutical sales representatives and
`advertising in medical journals to promote prescription
`drugs. Pharmaceutical marketing strategies have become
`more diversified in recent years. In addition to detailing and
`medical journal advertising, firms now promote their prod-
`ucts to medical professionals through educational events
`and directly to the public through mass media advertising.
`Spending on direct-to-consumer advertising (DTCA)
`increased from $266 million in 1994 to $2.6 billion in 2002,
`making this form of pharmaceutical marketing the object of
`substantial controversy (IMS Health 2003a). In 1997, a
`Food and Drug Administration (FDA) policy change made
`broadcast advertising of prescription drugs more feasible
`and may have contributed to the increase in the use of
`consumer-directed advertising by the pharmaceutical indus-
`try. One study of DTCA suggests that it increases demand
`for prescription drugs, accounting for roughly 12% of the
`increase in prescription drug sales between 1999 and 2000
`(Rosenthal et al. 2003).
`That DTCA increases prescription drug sales indicates lit-
`tle about the effect of advertising on consumer welfare or on
`
`Julie M. Donohue is an assistant professor, Department of Health
`Policy and Management, University of Pittsburgh Graduate School of
`Public Health (e-mail: jdonohue@pitt.edu). Ernst R. Berndt is Louis
`B. Seley Professor of Applied Economics, National Bureau of Eco-
`nomic Research, Sloan School of Management, Massachusetts Insti-
`tute of Technology (e-mail: erberndt@mit.edu). Dr. Donohue was a
`research fellow in pharmaceutical policy at Harvard Medical School.
`Dr. Donohue gratefully acknowledges financial support from the
`Henry J. Kaiser Family Foundation, the National Institute of Mental
`Health (training grant T32 MH19733-08), and the Harvard Pilgrim
`Health Care Foundation. Rusty Tchernis and Alan Zaslavsky provided
`valuable statistical advice for the article. The authors also wish to
`thank Rena Conti, Haiden Huskamp, Laura Eselius, Shanna Shul-
`man, Tom McGuire, and two anonymous JPP&M reviewers for help-
`ful comments on previous drafts of this article.
`
`competition in pharmaceutical markets. An important ques-
`tion is whether DTCA increases individual product market
`share, expands total class sales, or both. The weight of evi-
`dence to date suggests that DTCA has a significant impact
`on total class sales but little influence on individual product
`market share (Ling, Berndt, and Kyle 2003; Narayanan,
`Desiraju, and Chintagunta 2003; Rosenthal et al. 2003;
`Wosinska 2002). This result is not surprising, given the
`agency relationship between physicians and patients.
`Whereas consumer surveys show that DTCA motivates
`people to talk to their physicians about prescription drugs,
`the choice of whether and what to prescribe is ultimately up
`to the treating physician. Thus, the effect of DTCA is likely
`mediated by physician preferences, which may in turn be
`influenced by physician detailing or other forms of pharma-
`ceutical promotion. However, much of the pharmaceutical
`advertising to consumers is brand specific, and consumer
`requests for specific drug brands may influence physicians’
`prescribing decisions. Studies suggest that patient requests
`have a substantial impact on physician behavior (Soumerai,
`McLaughlin, and Avorn 1989).
`We add to the literature on the demand effects of DTCA
`by focusing on the antidepressant class. Prior research on
`antidepressants suggests that DTCA increases the number of
`people receiving drug treatment for depression, lending fur-
`ther support to the notion that DTCA increases class sales
`(Donohue et al. 2004). We examine the effect of DTCA on
`the choice of antidepressant observed at the individual
`patient level. There are three advantages to using individual-
`level data. First, we can account for differences in diagnosis
`that affect the choice of medication, which is important for
`antidepressant medications because they are used to treat a
`variety of conditions. Second, individual-level claims data
`contain more precise information on the out-of-pocket price
`paid by the consumer for prescription drugs. Third, the use
`of individual-level data enables us to treat aggregate adver-
`tising expenditures as exogenous to individual drug choice,
`a more tenuous assumption in studies that use aggregate-
`level data on prescription drug sales and marketing.
`We organize this article as follows: In the next section,
`we review the empirical work on the effects of pharmaceu-
`tical promotion. We then provide background on the anti-
`depressant class and depression. Subsequently, we lay out
`the conceptual framework for drug choice, explain the
`
`Vol. 23 (2) Fall 2004, 115–127
`
`Journal of Public Policy & Marketing 115
`
`Exhibit 2178
`Page 01 of 13
`
`

`

`116 Direct-to-Consumer Advertising and Antidepressants
`
`econometric methods used in the analyses, and describe the
`data sources for this study. Finally, we provide the results
`and discuss their implications.
`Empirical Literature on Pharmaceutical
`Promotion
`Physician Promotion
`The bulk of pharmaceutical promotion has been aimed at
`physicians, and thus much of the empirical work on pre-
`scription drug promotion has focused on physician-directed
`marketing efforts, such as detailing. Many previous studies
`have found that promotion to physicians raised entry costs
`into a particular therapeutic class and decreased price com-
`petition by increasing perceived product differentiation
`(Bond and Lean 1997; Hurwitz and Caves 1988; Leffler
`1981; Vernon 1981). Two studies of antihypertensive and
`antiulcer medications find that physician promotion reduces
`the absolute values of price elasticities of demand (King
`2000; Rizzo 1999). Another study of antiulcer medications
`finds that product marketing to physicians increases sales
`for the advertised product (Berndt et al. 1997). Total thera-
`peutic class marketing to physicians also has been found to
`increases class sales, though this effect generally declines
`with the number of products introduced.
`Effects of DTCA
`Consumer surveys suggest that prescription drug advertising
`motivates people to visit their physicians for a range of
`chronic conditions, some of which are newly diagnosed
`(FDA 1999a; Jim Lehrer 2000; Slaughter and Schumacher
`2001; Weissman et al. 2003). A recent study of the impact
`of DTCA on aggregate sales of prescription drugs in five
`therapeutic classes with high DTCA expenditures finds that
`though DTCA is effective in generating increased sales of
`the therapeutic class as a whole, it has no impact on market
`share (Rosenthal et al. 2003). Own DTCA for H2-antagonist
`drugs (before their switch to over-the-counter status) has
`been found to have a smaller impact on market share than do
`physician-directed marketing efforts (Ling, Berndt, and
`Kyle 2003). Similarly, in a study of nonsedating antihista-
`mines, Narayanan, Desiraju, and Chintagunta (2003) find a
`smaller positive effect of DTCA on market share in that
`therapeutic class than that of detailing. Most studies on
`detailing and DTCA use aggregate data on sales and mar-
`keting and thus do not take into account the effects of indi-
`vidual characteristics on the demand for prescription drugs.
`These studies also rely on aggregate measures of price and
`therefore do not account for the enormous variation in prices
`of prescription drugs across different types of consumers or
`for the presence of insurance (Frank 2001).
`Using data from the National Ambulatory Medical Care
`Survey and Competitive Media Reporting data on DTCA
`spending, Iizuka and Jin (2004) find that DTCA has no
`effect on physicians’ choice of medication. Similarly, using
`individual-level data on medication choice, Wosinska
`(2002) finds that advertising for cholesterol-lowering drugs
`has a small positive impact on drug choice but only for
`drugs with a preferred status on the health plan’s formulary.
`In addition, Wosinka finds that detailing has a much more
`significant effect on drug choice than does DTCA. That
`
`study does not evaluate whether the effects of DTCA on
`drug choice are mediated by individual-level factors such as
`diagnosis, age, or gender. The effects of DTCA are likely to
`vary across therapeutic classes because of the differences in
`the diagnosis and treatment of the condition, the level of dis-
`ability associated with the condition, and the differences in
`the features of the medications in that class. We examine the
`effects of pharmaceutical promotion in the antidepressant
`class.
`
`Background on Depression and
`Antidepressant Treatment
`The antidepressant class has been characterized by a high
`level of innovation and rivalry in recent years. Technologi-
`cal innovation and increased product variety, along with
`increased marketing expenditures for these medications,
`have resulted in dramatic growth in the sales of anti-
`depressants (Berndt et al. 2002). Antidepressant medica-
`tions ranked third in total sales and second in total number
`of prescriptions in the United States in 2002 (IMS Health
`2003b).
`Several features of depression and antidepressant medica-
`tions make these agents good candidates for DTCA from the
`pharmaceutical firm’s perspective. First, depression is a
`highly prevalent condition that results in substantial func-
`tional impairment (Ormel et al. 1994; Spitzer et al. 1995).
`Despite the availability of a wide range of effective phar-
`macological and psychosocial treatments, roughly half of
`the people with depression receive no treatment (Kessler et
`al. 2003). Thus, a large potential market exists for anti-
`depressant medications. For various reasons, including
`greater awareness and acceptance of drug treatment, the pro-
`portion of people treated for depression who received med-
`ication increased from 37.3% to 74.5% between 1987 and
`1998 (Olfson et al. 2002).
`Second, newer antidepressants are good candidates for
`DTCA because they are relatively safe. Newer medications
`have been found to be as effective as older antidepressants,
`such as tricyclic antidepressants, and are considered more
`“user friendly” because they have milder side effect profiles
`and require less titration by clinicians (Anderson and
`Tomenson 1994, 1995). As a result of FDA regulations
`regarding risk disclosure in advertising, drugs with fewer or
`less serious side effects and contraindications may be more
`likely to be advertised.1 Moreover, because there is substan-
`tial variety within the antidepressant class with respect to
`side effects, contraindications, and approved indications,
`pharmaceutical firms have an incentive to promote these
`newer products heavily (Berndt et al. 2002).
`Third, advertising may have a substantial role in anti-
`depressant use because of the complex nature of the condi-
`tions the medications are used to treat. Not only are these
`medications effective for many different conditions, but
`each condition also is highly heterogeneous. For example,
`depression encompasses several Diagnostic and Statistical
`Manual IV diagnoses and subtypes. Studies of major depres-
`
`1The FDA (2000) requires advertisements that include both the drug
`name and the therapeutic indication to disclose all major side effects and
`contraindications in the advertisement.
`
`Exhibit 2178
`Page 02 of 13
`
`

`

`sive disorders reveal heterogeneity with respect to biology,
`family history, pharmacologic response, genetics, and
`course of illness (Depression Guideline Panel 1993a).
`Although the effectiveness of antidepressants is similar at
`the population level, the effects of the medications vary
`widely at the individual patient level (Huskamp 2003;
`Kroenke et al. 2001). Because the effectiveness of any given
`medication for a particular patient is uncertain, advertising
`has great potential to influence medication choice.
`
`Data and Methods
`Conceptual Framework
`In analyzing choices of antidepressants, we borrow from tra-
`ditional models of demand for health care and prescription
`drugs (Newhouse 1993). We assume that the choice of anti-
`depressant is influenced by three sets of factors: (1) charac-
`teristics of the person choosing the medications, (2) features
`of the medications, and (3) physician preferences.
`
`Individual-Level Factors
`We assume that medication choice will vary by demo-
`graphic characteristics such as age and gender. In general,
`older age is correlated with greater use of medications and
`thus a greater risk of drug interactions. There is variation in
`the antidepressant class with respect to contraindications
`and the risk of drug interactions. For example, Prozac (flu-
`oxetine) and Paxil (paroxetine) have a higher risk of some
`drug interactions than does Zoloft (sertraline) (Spina and
`Scordo 2002). Thus, we expect the probability of choosing
`Prozac and Paxil to be lower among older people. We also
`expect antidepressant choice to vary by clinical factors,
`including mental illness diagnosis. For example, we expect
`people with anxiety disorders to be more likely to fill pre-
`scriptions for Zoloft, Paxil, and Effexor (venlafaxine),
`because these products have been approved by the FDA for
`treating anxiety disorders.
`
`Features of the Medications
`Surveys show that roughly two-thirds of people who ask
`their physician for a prescription for a brand they have seen
`advertised have their request honored (FDA 1999a; Jim
`Lehrer 2000; Slaughter and Schumacher 2001). Thus, we
`hypothesize that antidepressants with higher DTCA spend-
`ing are more likely to be chosen. We assume that various
`other features of the medications influence drug choice
`directly and/or through their interaction with marketing or
`individual-level characteristics, including price, length of
`time on the market, therapeutic indications, and side effects.
`Because our study population had insurance coverage, we
`use prescription drug copayment as the price faced by
`patients for the antidepressant medications. We expect a
`drug’s choice probability to decrease with the copayment
`amount, ceteris paribus. In addition, we hypothesize that
`drugs that have been on the market longer are more likely to
`be chosen because they are more familiar to consumers and
`physicians. We assume that drugs with a greater number of
`FDA-approved therapeutic indications are more likely to be
`chosen. In addition, we expect drugs with a high incidence
`of side effects to have a lower probability of being chosen.
`
`Journal of Public Policy & Marketing 117
`
`Physician Preferences
`Because of their agency relationship with patients, physi-
`cians exercise a significant amount of influence over
`demand for medical care (McGuire 2001). We assume that
`detailing expenditures significantly affect physicians’ pre-
`scribing behavior and hypothesize that a drug’s choice prob-
`ability will increase with spending on detailing to physi-
`cians. Because we had data on product-specific spending on
`detailing, we include this form of promotion as a character-
`istic of each drug.
`Overview of Analytical Strategy
`There has been substantial variation in the marketing strate-
`gies for antidepressants with respect to the use of DTCA.
`Our study attempts to connect the cross-sectional and tem-
`poral variation in marketing strategy to medication choice.
`The time period for this study, January 1997 through
`December 2000, encompasses the change in FDA policy
`that made broadcast advertising of prescription drugs more
`feasible. In August 1997, the FDA (1999b) clarified its pol-
`icy on broadcast advertising of prescription drugs by issuing
`a draft guidance to the industry. Before 1997, it was difficult
`to air product-claim advertisements that mentioned the
`name of the product and the condition it was meant to treat
`because of rules on the provision of the approved product
`labeling information that contained information on risks and
`benefits. As a result, most television advertisements for pre-
`scription drugs were “reminder advertisements,” which pro-
`vided the name of the drug but not the condition it was
`meant to treat, or “help-seeking advertisements,” which dis-
`cussed a condition but did not mention any specific treat-
`ments. The policy change led to a shift in the composition of
`television advertisements from primarily reminder and help-
`seeking
`advertisements
`to mainly product-claim
`advertisements.
`We focused on six antidepressants in three categories of
`medications: selective serotonin reuptake
`inhibitors
`(SSRIs), which include Prozac (fluoxetine), Zoloft (sertra-
`line), Paxil (paroxetine), and Celexa (citalopram); serotonin
`norepinephrine reuptake inhibitors (SNRIs), which include
`Effexor (venlafaxine); and serotonin antagonist and reup-
`take inhibitors (SARIs), which include Serzone (nefa-
`zodone). The FDA has approved all of the study drugs for
`the treatment of depression, and some of the drugs have
`received FDA approval to treat other mental disorders. None
`of the drugs’ patents had expired before the end of the study
`period. We did not have access to promotional spending
`data on (and thus did not include) SSRIs that did not have an
`indication for depression (i.e., Luvox [fluvoxamine]); anti-
`depressants that had generic equivalents at the time of the
`study (i.e., Desyrel [trazodone]); older-generation medica-
`tions, such as tricyclic antidepressants; or products that rep-
`resented a small share of the antidepressant market or prod-
`ucts used primarily to treat conditions other than depression
`(i.e., Remeron [mirtazapine] and Wellbutrin [buproprion],
`respectively). None of these medications was advertised to
`consumers, and thus we do not include them in the study.
`Econometric Method
`Discrete choice analyses often use a conditional logit model
`(sometimes called a multinomial logit model). The condi-
`
`Exhibit 2178
`Page 03 of 13
`
`

`

`the nested logit, in which ρ is restricted to equal 1. If 0 < ρ <
`1, the nested logit model is the preferred specification. We
`used a likelihood ratio test to determine the proper model
`specification (Hausman and McFadden 1984). The likeli-
`hood ratio is –2(Lr – Lu), where Lr is the log-likelihood
`value of the conditional logit model, and Lu is the log-
`likelihood value of the nested logit.
`
`Data
`The data set we used in the analysis consists of health insur-
`ance claims for the use of medical services and prescription
`drugs, marketing data on pharmaceutical promotion, and
`information on various characteristics of the study medica-
`tions. The medical claims data were obtained from The
`Medstat Group’s MarketScan database. MarketScan con-
`tains medical and pharmacy claims for beneficiaries of a
`group of large, self-insured companies. The data set for
`1997 to 2000 contains enrollment information and claims
`records for 5,718,683 people from 30 large employers
`located throughout the United States. The data set also
`includes information on the benefit designs of the more than
`100 indemnity and managed care plans used by these large
`employers.
`We used product-specific monthly data on DTCA
`(including print, radio, and television advertising) and
`detailing to physicians. We obtained monthly data on DTCA
`spending from Competitive Media Reporting, which tracks
`local and national advertising campaigns. We obtained
`information on monthly spending on detailing to physicians
`from Scott-Levin Inc., an independent medical information
`company that conducts market research on the pharmaceuti-
`cal industry. Scott-Levin imputes spending on detailing
`from a panel of more than 11,000 office and hospital physi-
`cians who track their encounters with pharmaceutical repre-
`sentatives. The panel is geographically representative,
`includes members of 31 clinical specialties, and accounts for
`approximately 2% of the U.S. physician population.
`
`Study Sample
`We identified all claims for the six study drugs from the pre-
`scription drug claims data file in the MarketScan database.
`Because drug choice is likely to be affected by previous
`experience with a particular medication, we limited the sam-
`ple to the first prescription for each person observed in our
`data collection period. To prevent censoring of observa-
`tions, we required patients to be enrolled in a MarketScan
`health plan for at least six months before the first prescrip-
`tion drug claim for an antidepressant. To identify new pre-
`scriptions, we imposed a six-month pretreatment period,
`during which there could be no prescriptions for the study
`drugs. Therefore, all prescription drug claims included in
`the analysis were filled between July 1, 1997, and Decem-
`ber 31, 2000. People for whom health plan information was
`unavailable or who lacked coverage for prescription drugs
`were excluded from the analysis.
`
`Explanatory Variables
`Our main explanatory variables were monthly spending on
`DTCA and monthly detailing spending for each of the study
`medications. Previous studies of drug marketing have found
`
`118 Direct-to-Consumer Advertising and Antidepressants
`
`tional logit model will yield only correct estimates of the
`effect of promotion on antidepressant choice if the six med-
`ications are viewed as equally substitutable (or not substi-
`tutable) for one another. This requirement is related to the
`independence of irrelevant alternatives (IIA) property of the
`conditional logit, which assumes that the ratio of probabili-
`ties of choosing any two alternatives is independent of the
`attributes of any other alternatives in the choice set (McFad-
`den 1981). If, however, some of the drugs are viewed as
`closer substitutes for one another than other drugs in the
`antidepressant class (e.g., SSRIs), a modeling procedure that
`relaxes the IIA assumption is more appropriate. To evaluate
`the extent to which these six antidepressants had similar
`cross-elasticities of substitution, we modeled antidepressant
`choice using both a conditional logit analysis and a nested
`logit model.
`For the nested logit analysis, we imposed a hierarchical
`structure on the drug choice process by grouping the med-
`ications (Figure 1). Providers would theoretically choose to
`prescribe a SSRI, a SNRI, or a SARI and then choose among
`drugs within each subcategory.2 The nested logit model
`allows the variance to differ across groups while the IIA
`assumption is maintained within the groups. In the nested
`logit model, the probability that person i chooses drug t is
`equal to
`
`=
`
`)
`
`P
`
`|t c
`
`×
`
`P
`c
`
`, and
`
`,
`
`ρ
`I
`c c
`
`ρ
`I
`c c
`
`e
`
`=
`
`1
`
`e
`
`cC
`
`∑
`
`×
`
`
`
`
`
`|1X
`
`=
`
`α
`
`
`
`( )1
`
`=
`
`P
`
`n
`it
`
`Pr(
`
`D
`
`it
`
`X
`
`e
`
`it
`
`α
`
`X
`
`e
`
`it
`
`1
`
`tJ
`
`1
`=
`
`=
`
`P
`
`nn
`it
`
`∑
`
`where Pc is the probability of choosing drug class c, Jc is the
`number of drugs in class c, and Ic = 1n{ΣJι
`t = 1eXitα}. The
`parameter ρ, called the inclusive value, is a measure of the
`cross-elasticity of substitution within the nests and is esti-
`mated in the nested logit model. McFadden (1981) shows
`that theoretically the value of ρ falls between 0 and 1. If ρ =
`1, all six drugs have the same degree of substitutability for
`one another, and the conditional logit model is the appropri-
`ate specification. The conditional logit is a special case of
`
`2This grouping is relevant only for the SSRI nest, which has more than
`one choice.
`
`Figure 1.
`
`Antidepressant Choice Decision Tree for Nested
`Logit Model
`
`Choice
`Level 1
`
`Choice
`Level 2
`
`SSRI
`
`SNRI
`
`SARI
`
`Prozac Zoloft Paxil Celexa
`
`Effexor
`
`Serzone
`
`Exhibit 2178
`Page 04 of 13
`
`

`

`that though the effects of promotion last beyond the period
`during which marketing expenditures are incurred, the
`effects diminish over time (Gonul et al. 2001; Ling, Berndt,
`and Kyle 2003; Narayanan, Desiraju, and Chintagunta
`2003). Therefore, we constructed cumulative measures of
`spending on advertising to consumers and physician promo-
`tion and treated both forms of promotion as depreciating
`assets. We used promotional spending for the month in
`which the prescription was filled plus the discounted sum of
`spending from the previous six months. We applied a
`monthly depreciation rate of 20% based on estimates from
`previous analyses of pharmaceutical promotion (Narayanan,
`Desiraju, and Chintagunta 2003). We used a natural loga-
`rithm transformation for the promotional variables to adjust
`for the skewed distribution of the data. To assess whether
`the effect of DTCA was moderated or mediated by the level
`of spending on physician detailing, we included an interac-
`tion term in the model: DTCA × detailing.
`To examine whether the effect of pharmaceutical promo-
`tion on medication choice varied in response to the FDA’s
`policy change regarding broadcast advertising, we created a
`binary variable coded as 1 if the prescription was filled after
`December 1997 (several months after the draft guidance
`was released) and interacted it with DTCA and detailing.
`We took two alternative approaches to modeling the
`effects of drug characteristics on antidepressant choice. In
`the first approach, we explicitly analyzed the effects of drug
`characteristics such as the amount of time on the market and
`number of indications. This approach assumes that all of the
`variation in a drug’s choice probability is attributable to the
`characteristics we identified in the analysis. As an indicator
`of the time a drug had been on the market, we used the num-
`ber of months between the FDA approval date and the
`month in which the antidepressant prescription was filled.
`We obtained data on initial FDA approval dates from the
`FDA’s (2003b) Orange Book. We obtained data on the num-
`ber of indications from the Physicians’ Desk Reference
`(Medical Economics Co. 2002) and the FDA’s (2003a) Web
`site, which posts product labeling changes. No previous
`study has compared the incidence of side effects across all
`six of the study medications. Instead, we obtained informa-
`tion on the side effect profiles of the study drugs from the
`Depression in Primary Care Guidelines developed by the
`Agency for Health Care Policy and Research, the Physi-
`cians’ Desk Reference, and other sources (Delgado and
`Gelenberg 2001; Depression Guideline Panel 1993b).
`Although newer antidepressants have similar incidences of
`many side effects, they appear to differ with respect to the
`risk of sedation or activation side effects and sexual dys-
`function. The side effects variable was coded as 1 for drugs
`with a higher incidence of these side effects.
`We constructed a measure of relative price for anti-
`depressants based on the claims data by estimating out-of-
`pocket prices for the medications not chosen. We used the
`median copayment for each antidepressant for patients in
`the same health plan during the year in which the prescrip-
`tion was filled to approximate the price the patient would
`have paid for the medications not chosen.
`Our second approach to modeling the effects of drug
`characteristics on drug choice uses fixed effects for the
`drugs. We included DTCA, detailing, and out-of-pocket
`
`Journal of Public Policy & Marketing 119
`
`price in the analysis, along with indicator variables for each
`drug except Prozac (fluoxetine), which we used as the refer-
`ence drug. This approach requires fewer assumptions about
`the key attributes of the medications in the choice set and the
`relationship between the main explanatory variables and
`drug choice.
`We also included individual-level variables such as age
`and gender in the analysis. We included age as a binary vari-
`able (less than or equal to 44 years [the mean age] or more
`than 44 years). We identified whether people had been diag-
`nosed with major depression within six months (before or
`after) of the index prescription (based on the presence of an
`outpatient claim with a diagnosis of major depression cur-
`rent episode [International Classification of Diseases (ICD)-9
`code 296.2] or major depression recurrent episode [ICD-9
`code 296.3]). We also identified people who had been diag-
`nosed with an anxiety disorder within six months (before or
`after) of the antidepressant prescription (based on the pres-
`ence of an outpatient claim with a diagnosis of anxiety
`[ICD-9 code 300.0], phobic disorders [ICD-9 code 300.2],
`obsessive-compulsive disorder [ICD-9 code 300.3], or pro-
`longed post-traumatic stress disorder [ICD-9 code 309.81]).
`The effects of these two variable types (attributes of the
`medications and attributes of the people in the sample) were
`specified differently in both the conditional logit and the
`nested logit model. We included the individual-specific
`variables (e.g., gender), which did not vary across the med-
`ication choices, as interaction terms. Prozac × (individual-
`level parameter) served as the reference category for each
`individual-level characteristic. Therefore, the parameter
`estimates for the individual-specific variables correspond to
`the probabilities of a person choosing each medication rela-
`tive to the probability of choosing Prozac. In contrast, the
`parameter estimates for the medication-specific variables
`(e.g., DTCA, detailing) reflect how these characteristics
`affect the overall choice probabilities.
`We examined whether the effects of DTCA and detailing
`on medication choice varied across patients and products.
`Because of the heterogeneity among consumers who fill
`prescriptions for antidepressants and the differences in the
`marketing strategies of the drugs in our study, we tested
`whether the effects of DTCA varied across mental illness
`diagnoses. We interacted DTCA spending with the major
`depression and anxiety disorder indicator variables and
`added these parameters to the nested logit model. We also
`interacted the promotional spending variables with months
`after the approval date to assess whether the effects of
`DTCA and detailing varied on the basis of how long a par-
`ticular drug had been on the market.
`Results
`
`Descriptive Results
`We identified 25,716 subjects who filled at least one pre-
`scription for one of the six study medications between July
`1997 and December 2000. Of those, 27.3% filled prescrip-
`tions for Zoloft (sertraline), 25.9% for Prozac (fluoxetine),
`25.0% for Paxil (paroxetine), 10.2% for Celexa (citalo-
`pram), 7.1% for Effexor (venlafaxine), and 4.5% for Ser-
`zone (nefazodone). The newer antidepressants gained mar-
`ket share over the time period (Figure 2).
`
`Exhibit 2178
`Page 05 of 13
`
`

`

`120_Direct-to-Consumer Advertising and Antidepressants
`
`
`Figure 2. Antidepressant Choice in MarketScan
`
`
`
`45% PercentageofPrescriptions
`Filled
`
` —@- Prozac — Zoloft —4— Paxil —<— Effexor —kK— Serzone —®— Celexa
`
`
`supported, and we conclude that the nested logit model is
`the appropriate specification for the analysis of drug choice.
`The implication of the likelihood ratio test result is that sub-
`jects view the drugs in each nest as more substitutable for
`one another than for drugs in anothernest. Thus, throughout
`the “Results” section, we refer to the results from the nested
`logit model unless otherwise stated.
`
`Multivariate Results
`
`Table 1 presents descriptive statistics for the anti-
`depressant medications and the subjects in the sample. All
`the drugs in the study were approved to treat depression
`between 1987 and 1998. The medications vary with respect
`to the number of FDA-approved indications other than
`major depression. Prozac, Zoloft, Paxil, and Effexor have a
`higher incidence of both sedation/activation side effects and
`sexual dysfunction than do Serzone and Celexa. The aver-
`age copayments for the six study drugs were between $10
`and $14. Figures 3 and 4 show product-level spending on
`Characteristics of the Drugs
`DTCAanddetailing during the study period.
`Wediscussthe results from the analysis in which drug char-
`More than two-thirds of the subjects in the sample were
`acteristics are modeled explicitly (Table 2, column 2).
`women,and more than 38% hadan outpatientvisit in which
`Direct-to-consumer advertising had nostatistically signifi-
`a depression diagnosis was recorded within six months of
`cant effect on the choice of antidepressant. Pharmaceutical
`filling a prescription for an antidepressant. A little less than
`company spending on detailing to physicians, in contrast,
`half