This PDF is a selection from a published volume
`from the National Bureau of Economic Research
`
`Volume Title: Frontiers in Health Policy Research,
`Volume 6
`
`Volume Author/Editor: David M. Cutler and Alan
`M. Garber, editors
`
`Volume Publisher: MIT Press
`
`Volume ISBN: 0-262-03309-7
`
`Volume URL: http://www.nber.org/books/cutl03-1
`
`Conference Date: May 29, 2002
`
`Publication Date: January 2003
`
`Title: Demand Effects of Recent Changes in Prescription
`Drug Promotion
`
`Author: Meredith B. Rosenthal, Ernst R. Berndt,
`Julie M. Donohue, Arnold M. Epstein, Richard G.
`Frank
`
`URL: http://www.nber.org/chapters/c9862
`
`

`
`1 D
`
`emand Effects of Recent Changes in
`Prescription Drug Promotion
`
`Meredith B. Rosenthal, Harvard School of Public Health
`Ernst R. Berndt, Massachusetts Institute of Technology and NBER
`Julie M. Donohue, Harvard Medical School
`Arnold M. Epstein, Harvard School of Public Health
`Richard G. Frank, Harvard Medical School and NBER
`
`Executive Summary
`
`The release of clarified Food and Drug Administration (FDA) guide-
`lines and independent changes in consumer behavior provide an
`opportunity to study the effects of direct-to-consumer advertising
`(DTCA) in the prescription drug market alongside the effects of various
`physician-oriented promotions. We examine the effects of DTCA and
`detailing for brands in five therapeutic classes of drugs, using monthly
`aggregate U.S. data from August 1996 through December 1999. In
`terms of impact of DTCA on demand, we provide evidence on two
`issues: (1) do increases in DTCA increase the market size of an entire
`therapeutic class? and (2) does DTCA increase within-class market
`share of advertised drugs? Our findings suggest that, for these classes
`of drugs, DTCA has been effective primarily through increasing the
`size of the entire class. Overall, we estimate that 13 to 22 percent of
`the recent growth in prescription drug spending is attributable to the
`effects of DTCA.
`
`Advertisements contain the only truth to be relied on in a newspaper.
`
`Thomas Jefferson1
`
`Advertising is a racket . . . its constructive contribution to humanity is exactly minus
`zero.
`
`F. Scott Fitzgerald1
`
`I think I was wrong. . . . On the whole, I think there is a lot of educational benefit [to
`direct to consumer prescription drug advertising].
`David A. Kessler, M.D., former FDA Commissioner2
`
`

`
`2
`
`Rosenthal, Berndt, Donohue, Epstein, and Frank
`
`Since 1994, total spending on consumer-directed promotion for pre-
`scription drugs has grown nearly tenfold (Henry J. Kaiser Family Foun-
`dation 2000). Consumers and their physicians report that prescription
`drug advertisements are increasingly influential. In surveys of consum-
`ers, the share of people reporting that they have seen an advertisement
`on television or heard an advertisement on the radio for a prescription
`drug more than doubled between 1993 and 2000, reaching 81 percent
`by 2002 (Alperstein and Peyrot 1993, Newshour with Jim Lehrer/
`Henry J. Kaiser Family Foundation / Harvard School of Public Health
`2000, Aikin 2002). Some estimates indicate that as many as 25 percent
`of Americans have asked their doctors about a medication as a result of
`seeing an advertisement (Prevention Magazine 2001). Similarly, many
`physicians report that their patients have asked them about drugs as
`a direct consequence of consumer advertising (Borzo 1997, Kaufman-
`Sherr and Hoffman 1997).
`The rise of direct-to-consumer advertising (DTCA) of prescription
`drugs represents a departure from the industry's historical concentra-
`tion on promotion to physicians, hospitals, and other health care orga-
`nizations.3 This phenomenon is highly targeted toward a minority of
`products. In one recent study of 391 major branded drugs in 1999, only
`18 percent had positive DTCA expenditures, whereas 95 percent of
`brands sent "detailers" to visit physicians' offices (Neslin 2001). Tradi-
`tional physician-oriented forms of promotion remain important. Even
`among those products that employ DTCA, professional promotion
`continues to command a larger share of marketing budgets (table 1.1).
`The change in marketing mix by pharmaceutical manufacturers is
`likely a response to recent changes in market and regulatory condi-
`tions.4 Consumers are increasingly seeking active participation in their
`own health care, aided in part by the wealth of information available
`on the Internet. At the same time, physicians may have less discretion
`over choice of brand-name drugs than they once did as a result of direct
`and indirect constraints placed on their prescribing behavior by man-
`aged care. In addition, in 1997, the FDA released explicit tentative
`guidelines (finalized in 1999) regarding advertising to consumers via
`electronic media that may have facilitated more widespread use of this
`promotional strategy. In particular, the new FDA guidelines clarified
`the requirements for adequate disclosure of information concerning in-
`dications, and risks and effects of a drug, thus removing a major barrier
`to television and radio advertising (Rosenthal et al. 2002).
`
`

`
`Effects of Changes in Prescription Drug Promotion
`
`The release of clarified FDA guidelines and independent changes in
`consumer behavior provide an opportunity to study the effects of
`DTCA in the prescription drug market as well as the effects of various
`physician-oriented promotions. In this paper, we examine the effects
`of two types of promotional spending for brands in five therapeutic
`classes of drugs, and we use monthly aggregate U.S. data from August
`1996 through December 1999. Specifically, we provide evidence on two
`issues: (1) do increases in DTCA increase the market size of an entire
`therapeutic class? and (2) does relative DTCA within a given therapeu-
`tic class affect market shares within that therapeutic class? In both
`cases, we examine the effects relative to traditional physician-oriented
`promotional efforts.
`
`I. Review of Related Literature
`
`Our research builds on and relates to several other studies. Early eco-
`nomic studies of physician-oriented marketing of prescription drugs
`by Bond and Lean (1977), Hurwitz and Caves (1988), Leffler (1981),
`and Vernon (1981) considered evidence regarding whether this market-
`ing was more "persuasive" than "informative," although the distinc-
`tion between the two was ambiguous. This distinction reflects an
`earlier, more general literature that viewed advertising alternatively as
`changing consumers' preferences (Kaldor 1950), creating or exaggerat-
`ing product differentiation and thereby increasing barriers to entry
`(Bain 1956), or providing information about a product's characteristics
`and its price (Stigler 1961). A common finding from the empirical litera-
`ture was that professional promotion of prescription drugs increased
`entry costs and decreased price competition by increasing perceived
`product differentiation. A related medical study on physician detailing
`by Avorn, Chen and Hartley (1982) advocated that counterdetailing be
`utilized to offset the unbalanced information provided by prescription
`drug detailers.5
`More recent research by King (2000) on anti-ulcer medications finds
`that a brand's own marketing reduces (in absolute value) a brand's
`own price-elasticities of demand, but that total industry marketing
`(i.e. the sum of marketing expenditures for each brand in the class) re-
`duces the extent of product differentiation. Berndt et al. (1997) dis-
`tinguish between "industry expanding" and "rivalrous" marketing
`efforts, and find that, for anti-ulcer medications, both medical journal
`
`

`
`41
`
`.ft
`
`0.00
`0.03
`0.01
`0.05
`0.00
`
`0.06
`0.15
`0.11
`0.00
`0.00
`
`0.02
`0.00
`0.00
`
`0.01
`0.00
`0.01
`0.00
`0.01
`0.00
`
`0.02
`0.11
`0.12
`0.13
`0.18
`
`0.08
`0.30
`0.26
`0.44
`0.68
`
`0.08
`0.15
`2.05
`
`0.09
`0.14
`0.16
`0.21
`0.35
`0.52
`
`0.02
`0.07
`0.07
`0.07
`0.07
`
`0.04
`0.17
`0.09
`0.18
`0.19
`
`0.06
`0.10
`0.84
`
`0.06
`0.10
`0.10
`0.09
`0.15
`0.21
`
`0.00
`0.04
`0.05
`0.06
`0.11
`
`0.04
`0.14
`0.17
`0.26
`0.49
`
`0.02
`0.05
`1.21
`
`0.03
`0.05
`0.05
`0.12
`0.21
`0.31
`
`1991
`1991
`1996
`1991
`1993
`
`1993
`1995
`1996
`1994
`1996
`
`1989
`1995
`1999
`
`1987
`1991
`1992
`1993
`1994
`1998
`
`ratio
`to sales
`3-year DTC
`
`ratio'
`to sales
`promotion
`physician
`Total
`
`sales ratio
`sampling to
`3-year
`
`sales ratio
`detailing to
`3-year
`
`date
`approval
`FDA
`
`Merck
`Merck
`Pfizer
`Bristol Myers Squibb
`Novartis
`
`Schering Plough
`Pfizer
`Aventis
`Celltech Pharms
`Wallace
`
`Astrazeneca
`Tap Pharm
`Eisai
`
`Eli Lilly
`Pfizer
`GlaxoSmithKline
`Wyeth Ayerst
`Bristol Myers Squibb
`Forest Labs
`
`Manufacturer
`
`Mevacor
`Zocor
`Lipitor
`Pravachol
`Lescol
`Cholesterol
`Claritin
`Zyrtec
`Allegra
`Semprex-D
`Astelin
`Antihistamines
`Priosec
`Prevacid
`Aciphex
`PPIs
`Prozac
`Zoloft
`Paxil
`Effexor XR
`Serzone
`Celexa
`Antidepressants
`
`Manufacturer, approval date, and promotion to sales ratios for five therapeutic classes, 1997-1999
`Table 1.1
`
`

`
`01
`
`0.00
`0.00
`0.12
`0.09
`0.00
`0.23
`
`0.00
`0.40
`0.20
`0.47
`0.32
`0.48
`
`'Total physician promotion includes detailing and retail value of free samples. It does not includejournal advertising.
`from the Food and Drug Administration's Orange Book.
`data from IMS Health, 2001; DTCA data from Competitive Media Reporting, Strategy Report, 2001; manufacturerand FDA approval date obtained
`Sources: Detailing data from Scott-Levin Personal Selling Audit, 2001; sales data from Scott-Levin, SourcePrescription Audit, 2001; sampling
`Beconase
`Vancenase
`Flonase
`Nasacort
`IThinocort
`Nasonex
`Nasal Sprays
`
`GlaxoSmithKline
`Schering Plough
`GlaxoSmithKline
`Aventis
`Astrazeneca
`Schering Plough
`
`0.00
`0.37
`0.14
`0.37
`0.20
`0.26
`
`0.00
`0.03
`0.06
`0.11
`0.12
`0.22
`
`Prior to 1982
`Prior to 1982
`1994
`1991
`1994
`1997
`
`

`
`6
`
`Rosenthal, Berndt, Donohue, Epstein, and Frank
`
`and physician detailing marketing stocks positively affect own-brand
`sales. They also report that the impact of total class marketing efforts
`on total class sales is positive, and generally (but not always) declines
`with the number of products on the market. Rizzo (1999) reports that
`for antthypertensive drugs, both stocks and flows of detailing expendi-
`tures decrease the price elasticity through the development of greater
`brand loyalty.
`Using a proprietary data set from a pharmaceutical manufacturer
`that incorporates physician-specific marketing measures for one of its
`brands, Manchanda, Chintagunta and Gertzis (2000) find that detailing
`has a significant positive impact on the number of prescriptions written
`for that drug by the physician; that this marginal impact increases
`when free product samples are also provided to the physician; and that,
`for the majority of physicians in their sample, diminishing (though still
`positive) returns to detailing had already set in. Gonul et al. (2001)
`report similar diminishing returns to physician detailing, but find that
`detailing and free samples increase price sensitivity, where price is
`measured as the average retail price for the drug.
`The empirical studies cited above focused on physician detailing and
`in some cases on medical journal advertising, but did not examine
`DTCA.6 Unlike physician-oriented promotions, to the extent DTCA
`might raise awareness among previously untreated consumers of the
`existence of potentially effective treatments, DTCA could bring more
`patients into physician offices. Whether the effect of such increased
`"physician office foot traffic" is greater on overall therapeutic class
`sales, or on the share of sales for the specific brand named in the adver-
`tisement, is unclear a priori and is an empirical question addressed in
`this study.
`Prior to 1997, DTCA was permitted if a medical condition was men-
`tioned but the brand was not; or if the brand was mentioned, no men-
`tion was made of the medical condition for which it was intended, and
`instead the ad encouraged the individual to see her or his physician
`regarding the brand. Berndt et al. (1995) obtained advertising agency
`DTCA data for branded anti-ulcer (H2-antagonist) prescription drugs
`through May 1994, along with detailing and medical journal advertis-
`ing data from other sources. For the entire IJ2 therapeutic class, adver-
`tising demand elasticities were 0.55 for detailing, 0.20 for medical
`journal advertising, and 0.01 for this type of DTCA; the sum of these
`elasticities is 0.76, suggesting decreasing returns to scale for overall
`advertising. Within this therapeutic class, although a brand's own de-
`
`

`
`Effects of Changes in Prescription Drug Promotion
`
`7
`
`tailing and medical journal advertising stocks positively affected mar-
`ket shares, this was not the case for DTCA.
`Two recent studies of DTCA by Wosinska (2001) and Ling, Berndt,
`and Kyle (2002) incorporate data after the FDA's 1997 clarification
`of DTCA guidelines. Using 1996-1999 prescription drug claims data
`for 4,728 patients who filled a total of 11,529 new prescriptions for
`cholesterol-reducing drugs in the Blue Shield of California medical
`plans, along with national data on physician detailing, samples, and
`DTCA, Wosinska finds that DTCA positively affects total therapeutic
`class sales, but affects an individual brand positively only if that brand
`has a preferred status on the third-party payer's formulary
`Unlike the cholesterol-reducing drugs, the H2-antagonist drugs are
`sold in prescription (Rx) form and, since 1995-1996, also as over-the-
`counter (OTC) drugs. Both OTC and Rx versions of these brands have
`utilized DTCA, and thus various spillovers between Rx and OTC
`DTCA can be assessed, both between and within brands. Ling, Berndt,
`and Kyle find that DTCA marketing of OTC brands has no spillover
`to the same brand in the Rx market. Within the Rx market, own-brand
`physician-oriented detailing and medical journal advertising efforts
`have positive and long-lived effects on own Rx market share, while
`DTCA of the Rx brand has no significant impact on own Rx market
`share. Within the OTC market, not only are own-brand effects of DTCA
`on the OTC brand significantly positive and long-lived, but physician-
`oriented Rx marketing efforts have positive own-brand spillovers to
`the OTC share. DTCA efforts for Rx brands have no significant impact
`on same-brand OTC shares.
`
`II. Theoretical Considerations
`
`The codfish lays ten thousand eggs,
`The homely hen lays one.
`The codfish never cackles
`To tell you what she's done.
`And so we scorn the codfish,
`While the humble hen we prize,
`Which only goes to show you
`That it pays to advertise.1
`The theoretical foundations underlying the economics of advertis-
`ing rely in large part on Dorfman and Steiner (1954), who showed
`that for a profit-maximizing monopolist facing a downward-sloping
`
`

`
`8
`
`Rosenthal, Berndt, Donohue, Epstein, and Frank
`
`linear demand curve, the optimal advertising expenditure to dollar
`sales ratio equaled the ratio of two elasticities: EQAthe elasticity of
`quantity demanded with respect to advertising efforts, and £QAthe
`elasticity of quantity demanded with respect to price (in absolute
`value). The equation for this relationship is shown below:
`
`Advertising dollars
`Sales dollars
`
`QA
`
`EQP
`
`(1 1
`
`The Dorfman-Steirier theorem is static because it assumes that adver-
`tising efforts last only one time period, but it can be generalized readily
`to a dynamic case in which the effects of advertising efforts persist
`several periods into the futurethe so-called carryover effects (Schma-
`lensee 1972). When there are several marketing instruments (and con-
`stant unit marketing media costs), under reasonable conditions the
`optimal ratio of expenditures for any two media equals the ratio of
`their marketing elasticities (Palda 1969).
`In terms of demand marketing elasticities and advertising-sales ra-
`tios, it is useful to consider the taxonomy of Nelson (1970, 1974), who
`has distinguished search and experience goods as polar opposites. If
`a consumer can determine a product's quality and impact prior to pur-
`chase merely by visual, tactile, or analytical inspection, the product
`is said to have search qualities. Examples of search goods are many
`electronic goods, tools, and credit cards. If a customer must consume
`the product to predict its quality and impact, the good is said to have
`experience qualities. Examples of experience goods include cosmetics,
`restaurants, and cereals. In practice precise demarcation of goods into
`search versus experience is not possible, particularly when multi-
`attribute goods have both search and experience qualities. In general,
`however, goods with dominant experience attributes have greater
`advertising-sales ratios than do goods with dominant search qualities.
`To the extent prescription pharmaceuticals have idiosyncratic and un-
`predictable effects (differential efficacy, side effects, and adverse inter-
`actions with other drugs), pharmaceuticals would appear to have more
`experience than search qualities. For those pharmaceuticals having
`highly predictable outcomes, however, search qualities may dominate.
`Carlton and Perloff (1994) suggest that producers of search goods are
`more likely to use informational advertising, while experience goods
`producers are more likely to use persuasive advertising. They note,
`however, that the division is not perfect, nor is there unanimity in what
`constitutes information versus persuasion. They interpret the greater
`
`

`
`Effects of Changes in Prescription Drug Promotion
`
`9
`
`advertising-sales ratios of experience goods as possibly reflecting the
`fact that "images (used in persuasive advertising) are forgotten more
`quickly than facts (used in informative advertising). Thus, consumers
`may learn and remember that a particular good has fewer calories (is
`"less filling") in one or a few exposures to an advertisement, but need
`to be bombarded with repeated exposures to be convinced that a prod-
`uct "tastes great."7
`The effects of advertising on the welfare of individuals have long
`been analyzed and debated. Carlton and Perloff summarize this litera-
`hire by stating that "the welfare effects of advertising are complex and
`depend on the type of product and type of advertising," and therefore
`"are generally ambiguous."8 Brand loyalty, for example, may reduce
`price responsiveness of demand, but can also reduce consumers' search
`costs.9 Bagwell (2001) provides a useful collection of economics articles
`dealing with the theoretical foundations and empirical analyses of ad-
`vertising as information versus persuasion; search versus experience;
`and the relationships among advertising, product quality and market
`structure.'°
`Finally, in addition to affecting demand, advertising can be used as
`a strategic tool to affect possible entry into a product market. Ellison
`and Ellison (2000) hypothesize and empirically assess branded phar-
`maceutical firms' advertising strategies in the context of affecting po-
`tential generic entry as the brand's patent protection expires. Their
`model predicts and they find empirical evidence supporting the notion
`that branded drugs with medium-size markets reduce their advertising
`intensities to a greater extent prior to patent expiration than do drugs
`with either very small or very large markets.
`
`III. The Marketing of Prescription Pharmaceuticals: Descriptive
`Data
`
`Pharmaceutical companies currently employ several promotional strat-
`egies for prescription drugs designed to target physicians and consum-
`ers, respectively (table 1.2). Because physicians have long been the key
`decision makers when it comes to choosing a prescription drug, phar-
`maceutical companies traditionally have concentrated most of their
`marketing efforts on physicians, and still do so today. These physician-
`oriented marketing efforts include visits or phone calls by pharmaceu-
`tical sales representatives to physicians (detailing), free samples, print
`advertising, and sponsorship of medical education events. In 2000, the
`
`

`
`10
`
`Rosenthal, Berndt, Donohue, Epstein, and Frank
`
`Table 1.2
`Spending on physician-directed promotion and promotion to sales ratios, 1996-2000
`
`Dollars (thousands)
`
`Detailing
`Journal advertising
`Retail value of samples
`Total physician promotion
`Direct-to-consumer promotion
`Total Promotion
`
`Promotion to sales ratios
`
`1996
`
`3,010
`459
`4,904
`8,373
`791
`9,164
`
`1996
`
`1997
`
`3,365
`510
`6,047
`9,922
`1,069
`10,991
`
`4,057
`498
`6,602
`11,157
`1,317
`12,474
`
`1997
`
`1998
`
`1998
`
`1999
`
`2000
`
`4,803
`484
`7,954
`13,241
`2,467
`15,708
`
`2000
`
`4,320
`470
`7,230
`12,020
`1,848
`13,868
`
`1999
`
`Detailing
`Journal advertising
`Retail value of samples
`Total physician promotion
`Direct-to-consumer promotion
`Total promotion
`
`0.047
`0.046
`0.050
`0.043
`0.043
`0.007
`0.007
`0.006
`0.005
`0.004
`0.084
`0.076
`0.081
`0.071
`0.071
`0.129
`0.138
`0.137
`0.118
`0.118
`0.012
`0.016
`0.015
`0.018
`0.022
`0.141
`0.153
`0.153
`0.136
`0.140
`Sources: Physician promotion spending data are from IMS Health, Integrated Promotional
`ServicesTM, June 2001; sales data are from Pharmaceutical Research and Manufacturers
`of America, Annual Survey, 2001; direct-to-consumer promotion spending data are from
`IMS Health and Competitive Media Reporting, June 2001.
`
`vast majority of spending on physician-oriented promotion (about 81
`percent) was concentrated on detailing (30.6 percent) and samples
`(50.6 percent).
`Consumer-oriented promotion, which includes advertising in both
`print and electronic media, was nearly nonexistent as an approach to
`promotion of pharmaceuticals in the United States as of 1980. Begin-
`ning in the 1980s and early 1990s, a limited amount of DTCA began
`appearing. By 1994 a rapidly increasing trend in DTCA spending
`became apparent (figure 1.1). The release of the clarified FDA guide-
`lines in 1997 occurred in the midst of this trend and may have accel-
`erated it. By 2000, DTCA comprised 15.7 percent of total promotion
`expenditures.'1
`As shown in table 1.1, there is very substantial heterogeneity in
`DTCA and physician-oriented promotion to sales ratios. For new prod-
`ucts, this ratio can be as high as 2.05 (Aciphex, a proton pump iiihibi-
`tor), and as low as 0.0 (Mevacor, a cholesterol-reducing drug). Within
`classes, there is also substantial heterogeneity in DTCA, detailing, and
`sampling to dollar sales ratios (e.g., the antidepressants and nasal spray
`classes in table 1.1).
`
`

`
`Effects of Changes in Prescription Drug Promotion
`
`11
`
`FDA Guidelines Released
`
`1994
`
`1995
`
`1996
`
`1997
`
`Year
`
`1998
`
`1999
`
`2000
`
`3,000
`
`2,500
`
`g 2,000
`
`1,500
`
`1,000
`
`0 Ca
`
`C)
`
`500
`
`0
`
`Figure 1.1
`Trend in direct to consumer advertising spending, 1994-2000
`
`Source: IMS Health and Competitive Media Reporting
`
`IV. Empirical Implementation
`
`Our aim is to estimate the impact of DTCA on consumer demand for
`prescription drugs. The literature on the demand for prescription drugs
`uses several approaches to model specification, including implementa-
`tion of an almost ideal demand system (AIDS), Cobb-Douglas specifi-
`cations, and logit models (Ellison et al. 1997; Rizzo 1999; King 2000;
`Ling, Berndt, and Kyle 2002; Frank and Hartman 2002). Each specifica-
`tion has its advocates in the literature. None has yet been shown to
`be superior in estimating demand models in markets for prescription
`drugs.
`Following others, in the demand analysis pursued here we estimate
`the demand for prescription drug products in the context of multistage
`budgeting. That is, we estimate models of the impact of promotional
`spending (DTCA and detailing) at the level of the entire therapeutic
`class (e.g., selective serotonin reuptake inhibitors [SSRI] antidepres-
`sants) and at the level of the individual product within the class (e.g.,
`Prozac among the SSR[s).'2 This multistage structure is illustrated in
`figure 1.2. At the top level of the tree, which represents the therapeutic
`
`

`
`12
`
`Rosenthal, Berndt, Donohue, Epstein, and Frank
`
`Figure 1.2
`Multi-tier demand structure
`
`class of drugs, we estimate the impact of DTCA spending and detailing
`in the context of a Cobb-Douglas demand specification (double loga-
`rithmic). In the analysis of competition at the individual product level
`within each class, we specify and estimate three alternative models:
`(1) an AIDS type specification; (2) a logit model with log of quantity
`share divided by (1 - quantity share) on the left-hand side, and prices
`and promotional spending on the right-hand side; and (3) a Cobb-
`Douglas model in log levels.
`
`Data Sources
`
`We examine monthly data from August 1996 to December 1999 for
`five therapeutic classes of drugs: recent vintage anti-depressants (SSRIs
`plus serotonin/norepinephrine reuptake inhibitors [SNR[s}), antthy-
`perlipidemics, proton pump inhibitors, nasal sprays, and antihista-
`mines (table 1.2). These classes were selected on the basis of the
`following criteria: (1) presence of at least one product with high DTCA
`expenditures during the time period, (2) within-class variation in
`DTCA, and (3) within-class variation in the life cycles of the drugs. The
`classes treat a wide variety of ailments, are indicated for different pa-
`tient populations, and are prescribed by several different clinical spe-
`cialties. Data were collected on all of the drugs in each of these five
`classes.
`Sales and promotion data for the selected products were obtained
`from marketing research firms. Data on DTCA spending were obtained
`
`

`
`Effects of Changes in Prescription Drug Promotion
`
`13
`
`from Competitive Media Reporting (formerly known as Leading
`National Advertisers), which tracks local and national advertising
`campaigns in major media, such as television and radio, for brands
`with at least $25,000 in annual advertising expenditures. Three ma-
`jor components of professional promotional spending are reported
`here: detailing to office-based physicians, detailing to hospital-based
`physicians, and the value of free samples left with physicians. Drug-
`specific data on professional journal advertising, which represents a
`small percentage of overall promotional expenditures for products in
`the six drug classes, was not included.
`Spending data on detailing were obtained from Scott-Levin Inc.,
`a pharmaceutical market research firm. Scott-Levin imputes spending
`on detailing from a panel of roughly 12,000 office- and hospital-based
`physicians, which comprises roughly 2-3 percent of the U.S. physician
`population. The physicians track their contacts with pharmaceu-
`tical sales representatives. Data on the retail value of free samples
`provided by pharmaceutical companies were obtained from IMS
`Health, which uses a panel of 1,265 front office staff in medical prac-
`tices to monitor the volume of samples dropped off by sales repre-
`sentatives. Sales data were obtained from Scott-Levin, which audits
`over 35,000 retail pharmacies and projects total sales based on
`an independent estimate of total U.S. retail sales of prescription
`drugs.
`Market shares were constructed from product-level data on sales for
`the drugs in each of the five classes. Price data were constructed by
`Scott-Levin using their sales data. We then created a quantity sold vari-
`able based on dividing sales by price for a particular month. Note that
`this price variable is not the average consumer copayment for that
`drug, but instead is the average price received by wholesalers from
`their customers.
`Patent information was collected from the Food and Drug Adminis-
`tration's Orange Book. A variable was created to indicate the number
`of months left on a product's patent. Order of entry within a class was
`determined based on the FDA approval date available through the
`Orange Book.
`Table 1.2 identifies the therapeutic class and individual product com-
`binations that are analyzed below. Together these five classes ac-
`counted for 30 percent of all DTCA in 1999. Table 1.2 also reports the
`year each drug was approved, the three-year average promotion to
`sales ratio for detailing, sampling, and DTCA.
`
`

`
`14
`
`Rosenthal, Berndt, Donohue, Epstein, and Frank
`
`Basic Models
`
`We now set out the basic estimation models used in the analysis. As
`noted above, the Cobb-Douglas formulation is used for both the class
`level demand model as well as the individual product demand model.
`Equation (1.2) is the Cobb-Douglas specification for the product spe-
`cific analyses:
`in q = a +
`
`in DTCA + 2 In Det +
`
`(1.2)
`
`where Det is detailing and the X are other explanatory variables.
`Equation (1.3) represents the general specification of the modified
`AIDS model.
`
`S = a +
`
`In () + in DTCA +
`
`In Det +
`
`(1.3)
`
`where 5, is the dollar revenue share of the ith drug within a therapeutic
`class, and P1 and P1 are prices of drugs i and] (in our model, P1 is actually
`the share-weighted price index for other competitors in the class, since
`there are more than two drugs in every class). Finally, we use the same
`regressors in estimating model specifications where the dependent
`variable is specified as the logit of quantity shares for the individual
`drug products. Several variants of these specifications are also esti-
`mated, as discussed below.
`
`Specification and Measurement of Key Variables
`
`There are two important sets of measurement and specification issues
`with our demand models of prescription drug promotion: (1) the mea-
`surement and specification of the price variables, and (2) the specifica-
`tion of promotional spending. The appropriate price to measure in a
`traditional consumer demand model is the out-of-pocket cost of the
`drug to the consumer. For various reasons, this desired measure is not
`available. The Scott-Levin data we analyze measure price as the pay-
`ment made by drugstores to wholesalers for each drug. Because this
`measure takes into account discounts and charge-backs to pharmaceu-
`tical manufacturers, these prices are not simple list prices. Neverthe-
`less, these observed prices do not account for the rebates given to health
`plans and other third-party payers for prescription drugs, nor the
`structure of beneficiary cost sharing for prescription drugs by health
`
`

`
`Effects of Changes in Prescription Drug Promotion
`
`15
`
`plans. Both the level of rebates obtained by different payers and the
`structure of copayments are quite heterogeneous, even within the cus-
`tomers of a single drugstore (Frank 2001). Thus, the observed prices
`are measured with error and are probably not closely correlated with
`the desired consumer out-of-pocket price. For these reasons, we take
`two approaches in empirical implementation. In one set of models, we
`include the mismeasured price variables. In another set of models, we
`omit the price variables and instead include a more extensive set of
`indicator variables to account for time trends and unobserved cross-
`sectional effects.
`The specification of measures of promotional spending has generally
`taken one of three forms in the literature. Promotion has been treated
`as a simple flow variable that is measured by current levels of promo-
`tional spending (Wosinska 2001). That approach assumes that current
`buying behavior depends largely on current exposure to promotional
`activities. A second approach specifies promotional activity as affecting
`consumer. choice in terms of a lag structure (Rizzo 1999, Wosinska
`2001). The assumption here is that the appropriate measure is still
`viewed as a flow, only there is a lagged response to the promotional
`activities. The third approach is to treat promotional activity as a stock
`that depreciates at a constant rate over time (Berndt et al. 1995, 1997).
`Our point of departure is to treat promotional spending as a simple
`flow. We also explored creating a stock. Our historical data on DTCA
`and detailing is quite limited, however. Therefore, we experimented
`with a variable based on three-month cumulative spending. When we
`used this variable in the specifications, our time series was shortened,
`and the resulting parameter estimates did not differ markedly from
`those assuming a simple flow. Therefore, we chose to focus on the re-
`sults that are based on treating promotional spending as a simple flow.
`We account for the possibility that spending on DTCA and physician
`promotion and product sales are jointly determined by estimating in-
`strumental variables (IV) models where all three variables are assumed
`to be endogenous. Three sets of variables serve as the basis for our
`exclusion restrictions in the IV specifications for the product specific
`demand equations. The first is the time left on the patent for each drug
`(and its square). This is based on the notion that for products reaching
`the end of their patent protection period, there is little incentive to in-
`vest in promotion (Frank and Salkever 1992; Berndt, Kyle, and Ling
`2002). The length of the patent life remaining is determined far in ad-
`vance of current sales decisions, and there is no direct effect on sales
`
`

`
`16
`
`Rosenthal, Berndt, Donohue, Epstein, and Frank
`
`because there is no generic competition yet in place for these drugs.
`The second instrument reflects the timing of the FDA's clarification of
`the conditions governing direct to consumer advertising on television
`(and the interaction of that indicator with time). In 1997 the F