Do Pharmaceutical Sales Respond to
`Scienti c Evidence?
`
`Pierre Azoulay
`Columbia University, New York, NY 10027-6902
`pierre.azoulay@columbia.edu
`
`I investigate how different sources of information inuence the diffusion of
`pharmaceutical innovations. In prescription-drug markets, both advertising
`and scientic information stemming from clinical trials can affect physicians’
`prescription choices. Using novel indices of clinical-research output, I nd
`that both marketing and scientic evidence directly inuence the diffusion
`process in the antiulcer-drug market, with marketing having a more pro-
`nounced inuence. I also nd evidence that clinical outputs are important
`drivers of rms’ marketing efforts, affecting sales indirectly. Taken together,
`the direct and indirect effects of science on demand imply strong private
`incentives for clinical research. I conclude that product-market competition
`in the pharmaceutical industry is shaped by both advertising rivalries and
`scientic rivalries. Moreover, drug advertising may perform an important
`informative function.
`
`Introduction
`1.
`How do different types of information inuence the diffusion of phar-
`maceutical innovation? The spread of technological advances is lim-
`ited by the extent to which relevant information is available among
`potential adopters. Furthermore, the information necessary for the dif-
`fusion of pioneer products may be different from that required for the
`market penetration of subsequent innovations.
`In most industries, one would expect underinvestment in the
`production of knowledge to limit the availability of objective sources
`of
`information about product characteristics, safety, and efcacy
`(Arrow, 1962). However, in prescription-drug markets, two features
`of the institutional environment—extensive, government-mandated
`
`For useful suggestions and support, I would like to thank audience participants at
`the MIT IO Lunch and the NBER Productivity Lunch, as well as Dan Ackerberg,
`Richard Caves, Peter Davis, John DeFigueiredo, Sara Ellison, Stan Finkelstein, Jeff Fur-
`man, David Genesove, Jerry Hausman, David Hsu, Rebecca Henderson, Kip King, Bob
`Pindyck, Robert Rubin, Otto Toivanen, and especially Scott Stern and Ernie Berndt. The
`usual disclaimer applies.
`
`© 2002 Massachusetts Institute of Technology.
`Journal of Economics & Management Strategy, Volume 11, Number 4, Winter 2002, 551–594
`
`IMMUNOGEN 2280, pg. 1
`Phigenix v. Immunogen
`IPR2014-00676
`
`

`

`552
`
`Journal of Economics & Management Strategy
`
`testing requirements, and the structure of incentives in academic
`medicine—provide a context in which privately valuable information
`is made publicly available through the publication of clinical studies
`in medical journals. In addition, pharmaceutical companies promote
`their products extensively,
`though disagreements remain among
`economists and policy makers concerning the role of drug adver-
`tising. For some, marketing activities foster the rapid dissemination
`of product information about potentially life-saving products, while
`others emphasize its strategic use by sellers of incumbent brands to
`jam information channels that could be used by new entrants (Lefer,
`1981; Hurwitz and Caves, 1988).
`A nding that pharmaceutical sales do not respond to scien-
`tic information (holding advertising intensity constant) would be
`consistent with the jamming hypothesis. In contrast, a positive sci-
`ence elasticity of demand would imply that a more nuanced view
`of the relationships between advertising, scientic information, and
`demand is needed. Moreover, boundaries between science and adver-
`tising in pharmaceutical markets are blurry, since much advertising
`refers explicitly to clinical results. Thus, the pharmaceutical indus-
`try provides a unusual setting in which to compare the informative
`as well as persuasive functions of advertising: Are rms’ promotion
`efforts sensitive to changes in the supply of objective, scientic infor-
`mation contained in published clinical studies?
`I explore these questions using data pertaining to a partic-
`ular subset of the antiulcer-drug market: the therapeutic class of
`histamine2-receptor antagonists, commonly referred to as H2 antago-
`nists or simply H2 blockers. It enjoyed explosive growth from 1977,
`the year of the pioneer drug’s introduction, until the early 1990s,
`when there were four related molecules in this class vying for mar-
`ket dominance.1
`Importantly, product-market competition in this
`therapeutic market was marked by the overthrow of an established
`monopolist (Tagamet) by a subsequent entrant (Zantac). As noted by
`Suslow (1997), this change in market dominance could be the result
`of intense price competition, advertising rivalry (both persuasive
`and informative), or a battle to offer the most attractive package of
`nonprice attributes. In this paper, I argue that among these nonprice
`attributes, published clinical results contributed signicantly to this
`turnover in market leadership.
`
`1. During the time spanned by the dataset, none of these drugs went off patent or
`moved to the over-the-counter (OTC) market. Therefore, I can safely ignore important
`issues such as substitution with generics and market segmentation between distribution
`channels.
`
`IMMUNOGEN 2280, pg. 2
`Phigenix v. Immunogen
`IPR2014-00676
`
`

`

`Do Pharmaceutical Sales Respond to Scientic Evidence?
`
`553
`
`Using a brand-level, discrete-choice model of product differen-
`tiation, I examine the impact of scientic information embodied in
`randomized controlled trials (RCTs) on the sales of these four drugs.
`I attempt to use the fact that RCTs can use either a placebo or a com-
`peting drug as a control group to isolate the effects of these two
`types of scientic information on drug sales, contingent on market
`structure. The results show that both marketing and science directly
`inuence the diffusion process, with marketing having a more pro-
`nounced inuence. I also examine the possibility of an indirect inu-
`ence of scientic information on demand by estimating advertising
`response functions, and I nd some evidence that clinical-research
`outputs indeed drive rms’ marketing expenditures. Plugging back
`the advertising equation into the demand system, the sum of the
`direct and indirect effects yields total demand elasticities of science
`of between 0.3 and 0.5 for the pioneer drug and its challenger.
`Overall,
`these results are consistent with a view that sees
`product-market competition outcomes in the pharmaceutical industry
`as the result of rms’ rivalrous efforts in marketing and applied sci-
`ence. They cast doubt on the validity of the belief, widespread in the
`medical community, that drug advertising totally jams other conduits
`of professionally sanctioned information, such as the results of RCTs
`(Wade et al., 1989). Finally, these ndings help explain the grow-
`ing involvement of industry in the conduct and funding of clinical
`research. Not only do clinical expenditures contribute to meet safety
`and efcacy requirements (thereby securing regulatory approval for
`entry), they also constitute investments marked by long-lived and
`direct economic payoffs on the product market.
`The remainder of the paper proceeds as follows. Section 2
`reviews the literature on drug advertising and the diffusion of phar-
`maceutical innovations. Section 3 provides a short background on
`the antiulcer-drug market, in addition to describing the dataset and
`constructing clinical-output variables. Section 4 presents the econo-
`metric results for the discrete-choice model, while Section 5 provides
`estimates of advertising response functions. I offer some concluding
`remarks and suggestions for future research in Section 6.
`
`2. Literature Review
`The diffusion of pharmaceutical
`innovations is a complex social
`process and is subject to multiple inuences. Because drugs are expe-
`rience goods, the impact of entry is limited by physicians’ switching
`costs and herding around the most popular products in a given
`therapeutic class (Coscelli, 2000; Berndt et al., 2000). As a result,
`
`IMMUNOGEN 2280, pg. 3
`Phigenix v. Immunogen
`IPR2014-00676
`
`

`

`554
`
`Journal of Economics & Management Strategy
`
`diffusion is rooted in learning, word-of-mouth, and other dynamic
`phenomena occurring within the population of potential adopters. In
`their landmark study of tetracycline’s diffusion, Coleman et al. (1966)
`emphasized these demand-pull forces by documenting the heterogene-
`ity of the physician population with regard to patterns of information
`consumption, and highlighted the role of “medical opinion leaders”
`who were both among the early adopters of this novel antibiotic
`and closely tied with the academic medical community. On the other
`hand, diffusion paths are also inuenced by technology-push forces,
`in particular the approval by the Food and Drug Administration of
`additional indications for existing drugs (or of additional therapies
`within a given therapeutic market). These decisions result in the
`fall of quality-adjusted prices over time, triggering the adoption of
`inframarginal consumers.
`While there exists numerous sources of information that might
`inuence the adoption of pharmaceutical innovations at the individ-
`ual physician level, at a more aggregate level information regarding
`product quality is made available to potential adopters through two
`primary information channels: advertising by pharmaceutical rms
`and published clinical results regarding the safety and efcacy of drug
`therapies.2
`Beginning with Bond and Lean’s (1977) FTC study, economists
`have extensively studied the role of drug advertising. In experience-
`goods markets, the mere fact that a product is advertised can signal
`to customers that it is of high quality (Nelson, 1974; Milgrom and
`Roberts, 1986). In this perspective, advertising can be interpreted
`as performing mostly a persuasive role, since it conveys informa-
`tion only implicitly. The medical literature has further argued that
`advertising swamps the effect of professionally sanctioned sources
`of information (Avorn et al., 1982; Manning and Denson, 1980) and
`has deleterious effects on medical practice (Wade et al., 1989). Phar-
`maceutical rms promote their products heavily, with advertising
`expenditures typically amounting to between 12% and 15% of sales.
`The most heavily used form of promotion—known as detailing—
`consists of visits to physicians by the sales representatives of the
`producers of branded pharmaceuticals. Another instrument for bring-
`ing product information to the attention of prescribing physicians is
`medical-journal advertising. Relative to detailing, journal advertising
`expenditures are modest, although the mix of promotion methods
`varies substantially across products and rms (Berndt et al., 1997).
`
`2. At least, this was the situation during the period examined in this paper. The lift-
`ing of the ban on direct-to-consumer advertising and the advent of the World Wide Web
`in the mid-1990s have further expanded the number of relevant information sources.
`
`IMMUNOGEN 2280, pg. 4
`Phigenix v. Immunogen
`IPR2014-00676
`
`

`

`Do Pharmaceutical Sales Respond to Scientic Evidence?
`
`555
`
`Despite the intensity of promotion, the overall concern and dis-
`trust for commercial messages is surprising, since the advertising of
`ethical drugs is quite stringently regulated by the FDA.3 Comanor
`(1986) observes that the hypothesis of wasteful or jamming adver-
`tising is insufciently formalized, and that evidence on its behalf
`is largely impressionistic, relying on comments, letters and editori-
`als of a self-appointed group of physicians and health professionals.
`Indeed, other scholars have claimed that drug advertising performs an
`eminently informative function. Peltzman (1975) proposes that adver-
`tising helps to achieve an efcient rate of diffusion—where the ben-
`et from increasing the rate just pays the costs required to do so.
`Lefer (1981) shows that product promotion has a signicant positive
`effect on the entry success of new drugs yielding important thera-
`peutic gains. However, this evidence must be pitted against results
`demonstrating the role of advertising outlays in building up brand-
`name recall effects that favor established products facing new competi-
`tion by generic entrants (Hurwitz and Caves, 1988). In a similar vein,
`Stern and Trajtenberg (1998) nd that physicians who prescribe a nar-
`row set of therapies for a given condition are more likely to prescribe
`highly advertised drugs.
`In one of the most detailed studies of pharmaceutical advertis-
`ing, Berndt et al. (1997) examine the effect of marketing investments
`on the growth and changing composition of the antiulcer-drug mar-
`ket.4 The authors nd that the effect of these investments was sub-
`stantial and long-lived, although it partly spilled over to competing
`drugs. They also show that the second entrant’s intense promotion
`efforts were instrumental in overthrowing the market-share leader-
`ship of the incumbent. Finally, they hint—but do not explicitly test
`empirically—that advertising was more effective when it interacted
`with a superior bundle of product-quality attributes, such as lower
`dosage or fewer side effects.
`Market power in prescription-drug markets seems to rest as
`much upon habit persistence as upon fears that serious adverse con-
`sequences (such as a malpractice lawsuit) will follow an inappropriate
`
`3. Any material distributed by pharmaceutical companies must carry the “full pack-
`age insert,” i.e., the complete product information reviewed by the agency as part of the
`drug approval process. Also, the advertising of drugs for nonapproved indications is
`prohibited, and comparative advertising must be supported by well-controlled clinical
`studies. Finally, comparison of side-effect proles is not allowed, because most drug
`studies are not designed to assess the incidence of adverse interactions (Kessler and
`Pines, 1990).
`4. I am indebted to Ernie Berndt for providing their data, which is used in this
`paper.
`
`IMMUNOGEN 2280, pg. 5
`Phigenix v. Immunogen
`IPR2014-00676
`
`

`

`556
`
`Journal of Economics & Management Strategy
`
`choice. However, this power may be diminished when objective infor-
`mation about the quality of competing products is available (Scherer,
`1990). In the pharmaceutical industry, government-mandated testing
`requirements coalesce with the incentives of academic clinicians
`to assist in the production of impartial
`information—through the
`publication of clinical-trial results in medical journals. While many
`studies have sought to inform the debate between the persuasive and
`informative nature of pharmaceutical advertising, none has explicitly
`considered published clinical
`trials as an alternative information
`conduit. This paper is the rst to use both advertising data and
`clinical-research output measures to examine the effect of advertising
`and science on diffusion, as well as the complex relationship between
`promotion efforts and scientic sources of information.
`
`3. The Antiulcer-Drug Market
`
`3.1 History
`The antiulcer-drug market has been extensively studied by applied
`economists over the past ve years, so I will provide only a brief
`review here. The interested reader should refer to Berndt et al. (1997)
`and Suslow (1996) for a more complete exposition. H2 antagonists
`heal ulcers by reducing the secretion of acids in the stomach, and are
`effective in several contexts. Originally approved to cure gastric ulcers
`(located in the stomach), and subsequently for the treatment of peptic
`and duodenal ulcers, their introduction (starting in 1977 with cime-
`tidine) suppressed the need for costly surgeries, allowing treatment
`on an outpatient basis. Later, the FDA approved H2 antagonists as
`preventive treatments, and most importantly, for the treatment of gas-
`troesophageal reux disease (GERD)—a nonulcerous condition whose
`mild manifestation is more commonly known as heartburn. Finally,
`the liquid formulation of these drugs is also used by hospitals for the
`treatment of patients severely burned or bleeding.
`The antiulcer-drug market can be segmented into three distinct
`submarkets: antacids, H2 antagonists, and proton-pump inhibitors.
`Antacids (Mylanta, Maalox) were the rst drugs introduced on the
`market, and are still considered good pain relievers (they relieve
`symptoms within minutes). They do not, however, inhibit acid secre-
`tion, and are therefore poor substitutes for the therapeutic class
`considered here.
`Beginning in 1989, a new generation of drugs, known as proton-
`pump inhibitors, appeared on the market. Proton-pump inhibitors
`operate by completely shutting down acid secretion, and seem to pro-
`vide an improvement on the performance of H2 antagonists. Today,
`
`IMMUNOGEN 2280, pg. 6
`Phigenix v. Immunogen
`IPR2014-00676
`
`

`

`Do Pharmaceutical Sales Respond to Scientic Evidence?
`
`557
`
`TABLE I.
`H2-Antagonist Drugs
`FDA Indications
`
`Drug
`
`Molecule
`
`Firm
`
`Entry Duodenal Duod. Ulcer Gastric
`Date
`Ulcer Maintenance Ulcer GERD
`
`Tagamet Cimetidine SmithKline Aug. 77 Aug. 77
`Zantac
`Ranitidine Glaxo
`June 83
`June 83
`Pepcid
`Famotidine Merck
`Nov. 86 Aug. 86
`Axid
`Nizatidine Eli Lilly
`May 88 May 88
`
`Apr. 80
`May 86
`Oct. 86
`Apr. 88
`
`Dec. 82 Mar. 91
`June 85 May 86
`Oct. 88 Dec. 91
`July 91
`
`H2 antagonists are available over the counter, and Prilosec has become
`the main prescription drug for the treatment of infections of the gas-
`trointestinal tract.5
`
`3.2 Overview of the Data
`Given the competitive history described above, the empirical exercise
`will be limited to the period beginning in August 1977 (date of entry
`on the US market for the pioneer drug), and ending in May 1993
`(before Prilosec’s rise to market dominance and Tagamet’s imminent
`patent expiration). The rst H2 antagonist, Tagamet, was launched
`by SmithKline in 1977, and soon became one of the most popular
`prescription drug ever sold. Since then, three alternative H2-blocker
`medications have entered the market: Zantac (Glaxo) in 1983, Pepcid
`(Merck) in 1986, and Axid (Eli Lilly) in 1988. A brief synthesis of the
`main product characteristics for these four drugs appears in Table I.
`The date for a specic indication corresponds to the time of FDA
`approval.
`The dataset draws upon two distinct sources of information.
`First, product-level data on monthly sales, prices, advertising lev-
`els, and other product characteristics originates with the market
`research rm IMS (Philadelphia, Pennsylvania) and is discussed at
`greater length in Section 3.4.6 Second, product-level indices of clinical
`
`5. Three other forms of treatment deserve brief mention. Carafate was introduced in
`1981, after Tagamet but before all the other H2 antagonists. Because the required dosage
`is four times a day, it has remained a marginal player. Reglan entered the market in
`1984, but is only approved for GERD. Finally, Cytotec entered in 1988, but is only
`indicated for the prevention of ulcers induced by nonsteroidal antiinammatory drugs
`(NSAIDs).
`6. This data is also described extensively in Azoulay (2001) and in Berndt et al.
`(1997), both in Section 7.3 (pp. 282–295) and in the appendix (pp. 314–321). Although
`both drugstore and hospital markets are covered by IMS audits, the analysis below will
`
`IMMUNOGEN 2280, pg. 7
`Phigenix v. Immunogen
`IPR2014-00676
`
`

`

`558
`
`Journal of Economics & Management Strategy
`
`TABLE II.
`Description of Variables in Sample
`
`Variable
`
`Qit
`shareit
`interactionsit
`indicationsit
`dosageit
`pit
` owdetailingit
` owdetailing it
`
` owdetailingcoit
`
`stkdetailingit
`
` owjournalit
`
` owjournal it
`
` owjournalcoit
`
`stkjournalit
`science1it
`science2it
`
`science it
`
`Denition
`
`Number of monthly patient-days for drug i, in millions
`Market share of drug i (total market includes outside good)
`Number of adverse drug interactions for drug i
`Number of approved FDA indications for drug i
`Daily frequency of administration for drug i
`Real price per daily dose of duodenal ulcer therapy for drug i
`Flow of monthly detailing minutes for drug i, in millions
`Flow of monthly detailing minutes for competitors of drug i,
`in millions
`Flow of monthly detailing minutes for rm selling drug i,
`in millions
`Stock of monthly detailing minutes for drug i,
`in millions (d
`5%)
`Flow of monthly journal advertising expenditures for drug i,
`in millions of real dollars
`Flow of monthly journal advertising expenditures for
`competitors of drug i, in millions
`Flow of monthly journal advertising expenditures for rm
`selling drug i, in millions
`Stock of monthly journal advertising expenditures for drug i,
`in millions (d
`5%)
`Stock of market-expanding citations for drug i,
`in hundreds (d
`0%)
`Stock of comparative citations for drug i,
`in hundreds (d
`0%)
`Total stock of citations for competitors of drug i,
`in hundreds (d
`0%)
`
`research output are constructed in Section 3.3 below, using abstracts of
`clinical studies published in the medical literature. The complete list
`of variables and their denitions can be found in Table II. Descriptive
`statistics are displayed in Table III.
`
`3.3 Measuring Clinical-Research Output
`In order to study the response of sales to scientic information,
`as well as the sensitivity of promotion efforts to published clinical
`results, it is necessary to construct measures of relevant scientic
`activity. In the Data Appendix, I describe in detail the construction
`of indices measuring the ows and stocks of scientic information in
`
`rely exclusively on the drugstore-market data, since it accounts for 90% of total dollar
`sales, and hospital use is very different from outpatient use, both in purpose and in
`presentation.
`
`IMMUNOGEN 2280, pg. 8
`Phigenix v. Immunogen
`IPR2014-00676
`
`

`

`Do Pharmaceutical Sales Respond to Scientic Evidence?
`
`559
`
`TABLE III.
`Descriptive Statistics
`
`188)
`(A) For Tagamet (T
`Mean
`Std. Dev.
`
`27.685
`0.142
`7.096
`2.745
`2.516
`1.057
`0.094
`0.155
`0.254
`1.727
`0.253
`0.337
`0.623
`4.749
`24.929
`2.071
`7.301
`
`8.904
`0.047
`3.617
`1.089
`1.416
`0.328
`0.036
`0.149
`0.095
`0.656
`0.161
`0.335
`0.248
`1.149
`4.291
`1.973
`6.477
`
`Correlation Matrix
`science1it
`1.000
`0.903
`0.947
`0.533
`
`(B) For Zantac (T
`
`117)
`
`Mean
`30.477
`0.150
`0.145
`3.248
`1.342
`1.771
`0.133
`0.221
`0.166
`2.233
`0.361
`0.405
`0.661
`6.618
`6.381
`2.962
`3.150
`
`Std. Dev.
`14.241
`0.068
`0.354
`1.231
`0.476
`0.238
`0.036
`0.098
`0.038
`0.739
`0.195
`0.258
`0.376
`1.256
`2.479
`1.209
`3.891
`
`Correlation Matrix
`science1it
`1.000
`0.842
`0.911
`0.231
`
`Min.
`
`3.923
`0.021
`0.000
`1.000
`1.000
`0.672
`0.019
`0.000
`0.128
`0.263
`0.010
`0.000
`0.146
`0.701
`12.220
`4.720
`0.000
`
`Min.
`4.190
`0.022
`0.000
`1.000
`1.000
`1.309
`0.048
`0.060
`0.065
`0.378
`0.056
`0.010
`0.057
`1.558
`2.030
`0.040
`8.650
`
`Max.
`
`46.424
`0.244
`10.000
`5.000
`4.000
`1.700
`0.199
`0.464
`0.732
`2.576
`1.019
`1.386
`1.317
`7.230
`31.490
`0.000
`21.600
`
`science2it
`
`1.000
`0.856
`0.227
`
`Max.
`54.271
`0.276
`1.000
`4.000
`2.000
`2.129
`0.212
`0.456
`0.299
`3.046
`0.940
`1.179
`1.696
`8.621
`13.070
`3.970
`3.300
`
`science2it
`
`1.000
`0.778
`0.406
`
`Continued
`
`Variable
`
`Qit
`shareit
`interactionsit
`indicationsit
`dosageit
`pit
` owdetailingit
` owdetailing it
` owdetailingcoit
`stkdetailingit
` owjournalit
` owjournal it
` owjournalcoit
`stkjournalit
`science1it
`science2it
`science it
`
`science1it
`science2it
`stkdetailingit
`stkjournalit
`
`Variable
`Qit
`shareit
`interactionsit
`indicationsit
`dosageit
`pit
` owdetailingit
` owdetailing it
` owdetailingcoit
`stkdetailingit
` owjournalit
` owjournal it
` owjournalcoit
`stkjournalit
`science1it
`science2it
`science it
`
`science1it
`science2it
`stkdetailingit
`stkjournalit
`
`IMMUNOGEN 2280, pg. 9
`Phigenix v. Immunogen
`IPR2014-00676
`
`

`

`560
`
`Journal of Economics & Management Strategy
`
`TABLE III.
`continued
`(C) For Pepcid (T
`77)
`
`Variable
`
`Mean
`
`Std. Dev.
`
`Qit
`shareit
`interactionsit
`indicationsit
`dosageit
`pit
` owdetailingit
` owdetailing it
` owdetailingcoit
`stkdetailingit
` owjournalit
` owjournal it
` owjournalcoit
`stkjournalit
`science1it
`science2it
`science it
`
`science1it
`science2it
`stkdetailingit
`stkjournalit
`
`Variable
`Qit
`shareit
`interactionsit
`indicationsit
`dosageit
`pit
` owdetailingit
` owdetailing it
` owdetailingcoit
`stkdetailingit
` owjournalit
` owjournal it
` owjournalcoit
`stkjournalit
`science1it
`science2it
`science it
`
`science1it
`science2it
`stkdetailingit
`stkjournalit
`
`9.157
`0.045
`0.000
`2.961
`1.000
`1.620
`0.075
`0.348
`0.506
`1.193
`0.173
`0.600
`1.404
`3.360
`1.978
`0.876
`0.283
`
`3.739
`0.018
`0.000
`0.715
`0.000
`0.114
`0.023
`0.069
`0.126
`0.406
`0.198
`0.240
`0.735
`1.275
`0.470
`0.459
`5.121
`
`Correlation Matrix
`science1it
`1.000
`0.894
`0.828
`0.304
`
`(D) For Axid (T
`
`59)
`
`Mean
`4.935
`0.024
`1.000
`2.390
`1.000
`1.683
`0.114
`0.332
`0.310
`1.611
`0.091
`0.579
`0.176
`1.531
`0.815
`0.428
`1.811
`
`Std. Dev.
`2.337
`0.011
`0.000
`0.492
`0.000
`0.165
`0.024
`0.065
`0.056
`0.546
`0.067
`0.244
`0.067
`0.284
`0.181
`0.240
`4.024
`
`Correlation Matrix
`science1it
`1.000
`0.461
`0.581
`0.283
`
`Min.
`
`1.729
`0.009
`0.000
`2.000
`1.000
`1.402
`0.031
`0.205
`0.053
`0.226
`0.000
`0.124
`0.253
`1.021
`0.350
`0.000
`5.800
`
`Min.
`0.715
`0.004
`1.000
`2.000
`1.000
`1.484
`0.069
`0.199
`0.200
`0.316
`0.000
`0.117
`0.071
`0.583
`0.720
`0.640
`6.410
`
`Max.
`
`16.134
`0.076
`0.000
`4.000
`1.000
`1.851
`0.131
`0.505
`0.797
`1.641
`0.906
`1.310
`3.023
`5.532
`2.590
`1.390
`9.710
`
`science2it
`
`1.000
`0.784
`0.487
`
`Max.
`9.207
`0.047
`1.000
`3.000
`1.000
`1.943
`0.217
`0.441
`0.476
`2.268
`0.308
`1.099
`0.372
`2.036
`1.220
`0.000
`6.760
`
`science2it
`
`1.000
`0.941
`0.302
`
`IMMUNOGEN 2280, pg. 10
`Phigenix v. Immunogen
`IPR2014-00676
`
`

`

`Do Pharmaceutical Sales Respond to Scientic Evidence?
`
`561
`
`pharmaceutical markets, an original methodological contribution of
`this paper. I briey summarize the process below.
`I select 483 randomized controlled trials (RCTs) pertaining to the
`four H2-antagonist drugs published in 16 prominent general medicine
`and gastroenterology academic journals (these journals are listed in
`the Data Appendix). I examine the control group used in the trial. If
`a placebo or any active substance other than the four H2 blockers is
`used, I assign to the article the label market-expanding science. In the
`case of a comparative drug study between two or more of the H2
`antagonists, the label comparative science is assigned. Conditional on
`the label, I score each RCT using a three-step Likert scale ( 1, 0,
`1)
`to assess the negative, neutral, or positive impact of the article:
`1
`(respectively
`1) is assigned if the treatment effect is signicant and
`favors (respectively does not favor) the drug studied. A score of 0 is
`assigned if the treatment effect fails to reach statistical signicance. In
`order to capture variation in quality across clinical studies, I weight
`the treatment effect score by the cumulative number of forward cita-
`tions to the original article as of May 2001.
`3.3.1 Flows and Stocks of Scienti c Information. The
`nal step is to dene the variables characterizing the monthly ows
`and stocks of scientic information in the H2-blockers therapeutic
`class. For each market-expanding (respectively comparative) study
`s, pertaining to drug i, published during month t, I dene  ow1it
`(respectively  ow2it) as  ow1it
`s ws tesit, where tesit is the score
`received by drug i in trial s published during month t, and ws is the
`weight assigned to study s. The variable  owit
` ow1it
` ow2it
`lumps together market-expanding and comparative science for each
`drug.
`However, one would not expect the monthly ow of scientic
`information to have an effect on sales. Figure 1 graphs  owit for
`Tagamet and Zantac. Except for the spikes, which correspond to large-
`scale trials published in prestigious journals, it is difcult to discern
`a trend by studying month-to-month variations. Since RCTs provide
`information about the existence and/or usefulness of a molecule, one
`would expect their effect to be long-lived, decaying slowly until better
`evidence appears in the literature. Hence, the effect of scientic infor-
`mation on sales is likely to be better proxied by a stock rather than
`a ow variable. Since clinical results start to accumulate before entry
`on the product market and do not diffuse instantaneously, the origin
`on the time axis was set at m0
`36 (where m0 denotes the month of
`entry).
`Finally, I allow for the possibility that the stock of clinical output
`(to be estimated
`decays over time with monthly depreciation rate d
`
`IMMUNOGEN 2280, pg. 11
`Phigenix v. Immunogen
`IPR2014-00676
`
`S
`

`

`562
`
`800
`
`600
`
`400
`
`200
`
`0
`
`-200
`
`-400
`
`-600
`
`-800
`Jan
`
`-75
`
`Journal of Economics & Management Strategy
`
`Tagamet/Cimetidine
`
`Zantac/Ranitidine
`
`Jul-76
`
`Jan
`
`-78
`
`Jul-79
`
`Jan-81
`
`Jul-82
`
`Jan-84
`
`Jul-85
`
`Jan-87
`
`Jul-88
`
`Jan-90
`
`Jul-91
`
`Jan-93
`
`FIGURE 1. CIMETIDINE AND RANITIDINE FLOW OF SCIENTIFIC
`INFORMATION, 1975–1993
`
`below). A perpetual inventory model is used to dene stock variables
`corresponding to each ow variable dened above:
`
`scienceit
`
`(1
`
`d )sciencei, t 1
`
` owit
`
`t
`
`d ) t ¿  owi ¿
`
`.
`
`(1
`¿ m 0 36
`3.3.2 Quality of the science Variables. Several issues
`can be raised regarding the method used to compute the science
`indices. First, one could have designed many alternative scoring and
`weighting schemes. One such alternative would have been to choose
`a measure of efcacy (such as the treatment effect itself) as the grade.
`For example, Cockburn and Anis (2001) use clinical studies to build
`regressors in the construction of hedonic price indices for arthritis
`drugs. They collect measures of efcacy, toxicity, and side effects
`for each study and use them to build variables measuring quality
`changes. However, it is difcult to replicate this approach in the con-
`text of the antiulcer-drug market because no homogeneous efcacy
`measure is available for the three conditions studied. For instance,
`ulcer treatment studies refer to healing rates, ulcer maintenance
`studies refer to relapse rates, and GERD studies most often record
`the percentage of patients for whom the symptoms disappeared.
`Measurement devices (endoscopes, pH meters) vary across articles.
`Though the Likert-scale approach is a simplication, it circumvents
`these issues.
`
`IMMUNOGEN 2280, pg. 12
`Phigenix v. Immunogen
`IPR2014-00676
`
`S
`

`

`Do Pharmaceutical Sales Respond to Scientic Evidence?
`
`563
`
`Second, are the articles selected authentically scientic? This is
`a concern in light of the evidence that industry is funding clinical tri-
`als, and that frontiers between advertising and research have become
`blurred. Bero and Rennie (1996) document the growing prominence
`of industry sponsorship of clinical studies concurrent with the relative
`shrinking of government support for research.7 This trend is alleged
`to result in a growing distrust of academic medicine by practitioners.
`To examine the relevance of this claim in the data, I reviewed fund-
`ing sources for a subsample of 21 studies included in this analysis.
`Eight articles did not report their funding. The remainder of the stud-
`ies either were totally industry-funded (14%), were partially industry-
`funded (14%), or beneted from some form of industry support (38%),
`for example, through the supply of drugs, placebo tablets, advice on
`experimental design, or help with the statistical analysis. It is difcult
`to ascertain the precise effect of industry funding on the quality of
`clinical trials. The imposition of rigid selection criteria for the journals
`and articles alleviates the concern that science only represents dif-
`ferent measures of advertising: all included studies report the results
`of randomized, controlled trials, and are published by peer-reviewed
`journals with good standing in their eld.
`Third, is the weighting scheme adopted justiable? Adams and
`Griliches note that the ow of ideas is in general difcult to quan-
`tify. Nevertheless, they assert that “the best that can be done at the
`moment is to count papers and patents and adjust them for the wide
`dispersion in their quality by using measures of citation frequency”
`(1996, p. 12664). This is accomplished here through the use of forward-
`citation weights. This method implicitly assumes that the academic
`clinical community acts as a catalyst of the diffusion process. If one
`believed instead that physicians make prescription choices on the
`basis of their individual reading of the clinical literature, then a better
`alternative would have been to weight studies by journal circulation.
`Unfortunately, time-series circulation data is regarded as proprietary
`by publishing houses and is not available.
`
`3.4 Descriptive Statistics
`In Figure 2, I plot the quantity of US drugstore sales (in patient days)
`over time for the four H2 antagonists. Starting from 0 in 1977, total
`
`7. Between 1980 and 1989, 61% of the clinical trials conducted in the US were fully
`funded by the pharmaceutical industry, whereas this practice was unheard of between
`1945 and 1969. Furthermore, in 1992, the $10.9 billion in research expenditures reported
`by the drug industry exceeded the entire NIH budget of $10.1 billion (the industry
`gure includes drug discovery research).
`
`IMMUNOG