|

 

Research.chicagobooth.edu

Informational Shocks, O-Label Prescribing and the Eects of Physician Detailing Bradley T. Shapiro∗ This Version September 1, 2015 (Preliminary Version. Comments Welcome.) Promotional strategies employed by pharmaceutical rms to convince physicians to prescribe their products are the subject of considerable regulatory scrutiny. In partic- ular, regulators worry that rms may use sales reps to try to convince physicians to prescribe drugs for uses that have not been approved by the Food and Drug Administra- tion. Since 2004, 31 federal cases alleging o-label promotional practices have settled, totaling over $12 billion. In this paper, I study the eects of detailing on physician pre- scribing in the anti-psychotic category, which was the category most heavily targeted by the federal government for o-label promotion. I identify the eects of detailing using within-physician variation along with two studies that disseminated new information that drastically changed the nature of competition. Detailing eects are modest, but statistically signicant. While detailing lifts both on-label and o-label prescriptions, I nd evidence that it disproportionately increases on-label prescriptions, tilting the distribution of prescriptions towards on-label. I nd that the source of this dispropor- tionality is that detailing to physicians who primarily prescribe o-label is ineective.
The combination of small eects of detailing along with the disproportional eect on on-label prescribing suggests that regulators may not nd it worthwhile to pursue such cases. Additionally, the risk of extremely large settlements and lack of eectiveness of o-label promotion suggests that rms would be better o avoiding such practices.
∗University of Chicago Booth School of Business. I thank Kenneth Wilbur, Ginger Jin, Joshua Gottlieb, Jonathan Kolstad and Neale Mahoney as well as seminar participants at UCSD Rady, the Bates-White Life Sciences Symposium, the iHEA Congress and the 8th Workshop on the Econimcs of Advertising and Marketing for their helpful comments. I thank LuAnn Patrick at AlphaImpactRx for assistance with data resources. I acknowledge generous nancial support from the Beatrice Foods Co. Faculty Research Fund at the University of Chicago Booth School of Business.
In the healthcare industry, physicians control over $2 trillion of largely price-insensitivedollars (Arrow 1963). As such, it should come as no surprise that pharmaceutical rms useadvertising directly to physicians, or detailing, as their primary marketing tool to increaseprots. In 2012, pharmaceutical rms spent roughly $15 billion on detailing activities.1 Thistype of promotion has been the subject of considerable regulatory scrutiny, as regulatorswould like to ensure that rms give scientically justiable information. Indeed, promotinga drug for any use that has not been approved by the FDA (o-label uses) is illegal and suchpromotional activities have been vigorously pursued by regulators. Since 2004, 31 federalcases alleging o-label promotional activities of pharmaceutical rms have settled for morethan $12 billion.
While regulators are concerned that promoting o-label might not be in the best interest ofpatients, a rm must take into consideration not only the cost associated with hiring a salesforce, but also the possibility of large regulatory nes. Estimating the demand response todetailing is very important, but rms must also consider how they are detailing and whatimpact that has on the liklihood of litigation and if the incremental impact of particulareorts to promote o-label are worth the risk.
This paper addresses these questions in the context of antipsychotic drugs. These drugsare used to treat severe psychosis, particularly schizophrenia and bipolar disorder. Notonly is the antipsychotic class important for treating a serious illness, it is a multi-billiondollar category that has drawn substantial regulatory attention. Branded antipsychoticshave brought in at least $3.5 billion in revenue every year since 2001, and in 2013, thehighest grossing drug in the United States was the antipsychotic, Abilify, which grossed over$7 billion by itself. In the meantime, the category has faced nearly $5 billion in regulatorynes from charges of marketing products o-label in addition to more than $2 billion in nesfor failure to disclose adverse eects. There is little existing empirical evidence of the eectsof promotional activities on o-label prescribing despite the enormous regulatory nes.
To estimate the eect of detailing on prescribing behavior in the anti-psychotic category, Ileverage within-physician variation as well as new scientic discoveries to generate random-ness in the timing of detail visits for each physician. In particular, the early 2000s saw theintroduction of two main pieces of scientic information from clinical studies. In the rstinformation revelation, the market leader Zyprexa was found to be signicantly worse thanthe rest of the category in terms of adverse eects, generating a positive shock for the otherproducts, particularly the brand Seroquel. Next, a comparative eectiveness study foundthat two products had comparable main eects and superior side eect proles to the rest ofthe products. The two products receiving good news were Seroquel and an older generationgeneric drug, perphenazine. Both information shocks led to a huge increase in detailingactivities of Seroquel to primary care physicians. By taking the timing of the informationalshocks as exogenous, I use the associated increases in detailing at the physician level to iden-tify the eects of detailing. I quantify the direct eect of the informational shock by usingthe generic drug, perphenazine, which received the same positive shock to information with-out any associated increase in detailing. I nd that one detail visits generates approximately0.15 prescriptions in the month of the visit and a total of 0.3 prescriptions over time.
I leverage the high level of detail in the data, which is at the patient visit level and includes adiagnosis code, to address questions about o-label prescribing. Whether a patient recievinga prescription for an antipsychotic was, for example, diagnosed with schizophrenia (on-label)or insomnia (o-label) is observable in the data. Using this information, I compare theshare of the detailing eect attributable to o-label prescriptions with the o-label share ofprescriptions. I nd that while detailing increases o-label prescriptions, it increases on-labelprescriptions far more and in fact shifts the distribution more towards on-label prescribing.
I nd that this distributional shift is attributable to detailing being ineective to primarilyo-label physicians rather than shifting the share of o-label prescriptions within a givenphysician. This nding calls into question both manager and regulator decisions. First,given that regulators are willing to levy such large nes and that the eects of promotionon o-label are so small, promoting o-label might not be a protable decision. Second,given that the main eects of detailing are modest and that they do not inuence o-labelprescribing considerably, regulator eorts might be better spent elsewhere.
The main contributions of this paper are twofold. First and foremost, I am able to di-rectly speak to the specic, very important regulatory problem of o-label promotion in theantipsychotic category. This adds to the literature at the intersection of advertising andregulation, relevant both to rms and policy makers alike. O-label prescribing is a verypopular topic in regulatory circles, but the academic research speaking to o-label concerns isespecially sparse. Second, I estimate these eects using a new identication strategy combin-ing within-physician variation to control for persistent dierences in physician willingness toprescribe with exogenous changes in scientic knowledge to generate exogenous variation inthe timing of detail visits. While the exact methodolgy using these specic studies is specialto this application, other drug classes have had discrete changes in information that mightalso aect detailing levels, making this approach potentially useful for other applications.
The main contribution of this study is to examine the interaction of advertising and regu-lation in a particularly important category. Most industries are allowed to speak as theywish in their advertisements conditional on avoiding false claims. Rao (2015) examines theeects of false claims on sales, nding that they are extremely useful even after paying settle-ments or stopping their advertisements. Liu et al. (2014) show that combination therapiesin HIV/AIDS can drive positive spillovers of detailing onto other drugs and argue that suchspillovers can cause strange incentives in the presence of regulation. Larkin et al. (2013),Stremersch (2009) and Anderson et al. (2015) all look directly at changes in detailing regu-lations and their eects on demand, all nding reasonably dierent eects depending on theexact context of the policy changes. The size of the nes levied on the antipsychotic categoryfor o-label promtion indicate just how important the question is to regulators. In partic-ular, regulators are worried that a large portion of o-label prescribing of antipsychotics isto seniors in nursing homes with insomnia and dementia (Ray et al. (1980), Gurwitz et al.
(2000)). These populations are seen as particularly vulnerable to adverse eects and mighbe less able to decline treatment. While o-label prescribing need not be welfare reducing(Bradford et al. 2015), rms are not legally allowed to promote such uses, and empiricalevidence on the question is scarce.
This paper adds to the literature thinking about the eects of detailing on demand. In addition to documenting a detailing elasticity for an additional class of drugs, this study usesa novel identication strategy leveraging shifts in the scientic knowledge surrounding thesedrugs. Identifying the causal eects of detailing is challenging largely due to the fact that itis individually targeted by a presumably prot maximizing rm. Individual targeting createsboth signicant opportunity and signicant challenges for rms: opportunity in the abilityto inuence the inuentiable without wasting money on those who are not, but challenges inestimating the eects of rm actions and optimally allocating sales force. Previous literaturehas attempted to address this challenge using three main approaches: structural approaches[Manchanda et. al (2009), Dong et al. (2011), Montoya et al. (2010), Iyengar et al. (2011),and many others], instrumental variable approaches [Berndt et al. (1995, 1997)] and panelmethods with xed eects [Datta et al. 2011, Mizik et al. (2004)]. Another stream ofliterature examines the eects of policy changes relating to detailing on prescribing behavior[Stremersch et al. (2009), Larkin et al. (2013), Anderson et al. (2015)] without explicitlyestimating a detailing eect. In this paper, I leverage the introduction of new scienticinformation in the form of a comparative eectiveness study. In addition, the data is aphysician level panel, allowing the study to control for unobserved physician-specic factorsthat lead to prescriptions.
2 The Setting - Antipsychotics 2.1 Psychosis and Background of the Class The specic setting of antipsychotic drugs will be important both as a vehicle for identifyingthe eects of detailing, but also because it is a very large and important market per se aswell as the category most scrutiniezd by regulators for o-label promotional practices. An-tipsychotics are approved to treat psychosis, in particular schizophrenia and bipolar disorder.
Figure 1 shows that the antipsychotic market is huge in terms of revenue per year. The topve branded antipsychotic drugs, Zyprexa, Rispderdal, Seroquel, Abilify and Geodon repre-sented approximately $3.5 billion in 2001 and grew to over $7 billion by 2006. In addition,this is a market that has undergone shifts in preferences particularly driven by scientic discovery in the mid-2000s. Figure 2 shows the revenues of the top ve brands between 2001and 2006. The market leader at the beginning of the sample, Zyprexa, lost favor to surgingshares from Seroquel and Abilify. Each of these drugs, with the exception of Geodon, madeit into the top 20 highest grossing drugs at various points, and Seroquel, Zyprexa and Abilifyall made it into the top 10. By 2013, Abilify was the overall top grossing drug, bringing inmore than $7 billion on its own even though many of the other popular drugs had becomeavailable in cheaper generic form. With these levels of revenues, the antipsychotic industryis roughly on par with box oce revenues in the motion picture industry.2 In addition, asthese drugs are small molecules' in capsule and tablet form, the marginal cost of productionis miniscule.
Prior to the invention of these new, highly successful brands in the late 1990s, psychosis waslargely treated with what are called typical' antipsychotics. Typical antipsychotics wererst introduced in the 1950s and all have long since lost patent protection. As such, theyare inexpensive and widely available from numerous generic manufacturers. These drugslargely were considered to be very risky in terms of adverse eects. In particular, use ofthe drugs is associated with both metabolic syndrome (weight gain) and antipyramidal sideeects (dicult to control body movements). Some studies even linked the drugs to thedevelopment of diabetes and heart disease (Arana 2000).
The rst atypical' antipsychotic to be discovered was Clozapine in 1971, but it quicklywas removed from the market due to adverse eects. It later returned to the market, butnever to the kind of commercial success that the newer generation of drugs attained. Atypicalantipsychotic drugs were introduced with the hopes of reducing the adverse eects associatedwith the treatment of psychosis. In 1993, Janssen Pharmaceuticals won FDA approval forRisperdal as a treatment for psychosis. That was followed in 1996 by Eli Lilly's Zyprexa, in1997 by AstraZeneca's Seroquel, in 2001 by Pzer's Geodon and in 2002 by Otsuka's Abilify,which was jointly marketed in the United States with Bristol-Myers Squibb. These drugswere all thought to have better side eect proles than typical antipsychotics while stillbeing eective for treating psychosis. While the revenues they generated were enormous, thequantities dispensed were not quite as impressive, as the category is characterized by very high prices. From 2001-2006, manufacturers charged between $225 and $380 per prescriptionfor atypical antipsychotics on average. In 2013, the average price received by Otsuka forAbilify was $650 per prescription. While it was the highest grossing drug in 2013, Abilifywas only the 23rd most prescribed drug3.
2.2 Regulatory Controversy As these drugs became more popular, they were sometimes used to treat illness for whichthe drugs were not indicated. A drug receives an indication by clinically testing its ecacyagainst placebo in a randomized controlled trial registered with the Food and Drug Admin-istration. The results of that study are then evaluated by the FDA for ultimate marketingapproval. Any prescription for a drug being used to treat a condition for which it did notreceive FDA approval is considered an o-label prescription. While it is perfectly legalfor a physician to prescribe a drug o-label, it is illegal for a pharmaceutical manufacturerto market the drug as eective in treating something for which it is not approved by theFDA. In addition, the welfare eects of o-label prescriptions are not clear and could bepositive (Bradford et al. 2015). Antipsychotics became popularly used o-label, primarilyfor treating dementia and insomnia. While no cases ever came to trial, there have beennumerous nes paid to settle charges that rms illegally promoted these o-label uses. InSeptember 2007, Bristol-Myers Squibb (BMS) agreed to pay over $515 million to resolvea wide variety of illegal marketing charges. In particular, .the Government alleged that,from 2002 through the end of 2005, BMS knowingly promoted the sale and use of Abilify, anatypical antipsychotic drug, for pediatric use and to treat dementia-related psychosis, botho-label uses.4 In fact, the FDA had even mandated that Abilify carry a black box warn-ing against its use in treating dementia. The Food and Drug Administration has approvedAbilify to treat adult schizophrenia and bi-polar disorder, but has not approved the use ofAbilify for children and adolescents or for geriatric patients suering from dementia-relatedpsychosis. Further, the DoJ charged, BMS also created a specialized long term care salesforce that called almost exclusively on nursing homes, where dementia-related psychosis is far more prevalent than schizophrenia or bipolar disorder. In April 2010, AstraZeneca wasned $520 million by the United States Department of Justice for o-label promotion. Ac-cording to a DoJ statement, the company recruited doctors to serve as authors of articlesthat were ghostwritten by medical literature companies and about studies the doctors inquestion did not conduct. AstraZeneca then used those studies and articles as the basis forpromotional messages about unapproved uses of Seroquel.5 In 2009, Eli Lilly pled guiltyto a criminal misdemeanor charge of illegally promoting Zyprexa o-label and paid a neof $1.4 billion.6 In 2013, Johnson & Johnson, Janssen's parent company, paid more than$2.2 billion to settle several cases charging o-label promotion of Risperdal.7 In particular,authorities emphasized that Janssen focused their o-label detailing practices on the mostvulnerable populations: elderly nursing home residents, children and individuals with men-tal disabilities. This was one of the largest settlements to date in a drug marketing case.
These nes, accounting for nearly $5 billion, were all imposed for the messages contained inphysician detail visits, allegedly encouraging physicians to prescribe o-label. With the totalsettlements for all drug classes being around $12 billion, the antipsychotic class was clearlythe most heavily scrutinized and ned, largely due to the fear that the adverse eects madeinappropriate prescribing especially costly.
In addition to the o-label promotion controversy, rms were also ned for failing to disclosesevere side eects discovered during clinical trials, notably metabolic side eects includingweight gain and the onset of diabetes. Between 2006 and 2007, Eli Lilly spent $1.2 billionto settle over 26,000 lawsuits from patients who claimed to have developed diabetes or otherdiseases while taking Zyprexa. In April of 2012, a jury found Johnson & Johnson guilty ofdownplaying several risks of the drug Risperdal and a judge ned the company $1.2 billion,though this ne would later be reversed by the Arkansas Supreme Court.8 While the data in this study cannot speak to whether or not sales reps hid risks, it can speak to the question of o-label marketing. In particular, as diagnosis code is observable in thedata, I can speak to the eect detailing had on prescriptions for o-label use.
The data for this study come from AlphaImpactRx, a pharmaceutical market research com-pany. The data follow a monthly panel of 1762 primary care physicians (PCPs) and 239psychiatrists from 2001 through 2006 and include physician identiers. AlphaImpactRx re-cruits this panel largely from those physicians in the 40th percentile or higher of antipsychoticprescribing. The reason for this non-respresentativeness is that physicians below the 40thpercentile are highly unlikely to be detailed. These physicians record the number and typesinteractions with sales reps on a daily basis. In addition, the number of minutes spent onthe product and the other products promoted in that visit are included in the data.
The physician panel and detailing data are connected with patient treatment information.
All patient visit information for each physician is recorded for two days each week. Thespecic days are rotated across time so as to avoid making the treatment data skewed byday of the week eects. Included in each visit observation is information about the patientdiagnosis in the form of an ICD-9 code. In addition, patient race, age, insurance status anddiagnosis severity are included in each observation (though there are no patient identiers).
Further, whether or not there was a prescription, the drug the prescription was written forand whether or not the diagnosis was new or recurring is included in the data. As physiciansare not restricted to prescribe for only on-label diagnoses, the data include every diagnosiscode for which an anti-psychotic was ever written. As such, whether a prescription was on-label (used to treat schizophrenia or bipolar disorder) or o-label (used to treat anythingelse) is observable in the data.
Two main challenges of the data are measurement error and representativeness. As treatmentdata is only observed for two days per week for each physician, the measure of total prescrip-tions is 2.5 times the observed number of prescriptions. This may over or under state thetrue number of prescriptions. As prescriptions will be a left hand side variable throughout the study, measurement error will reduce the precision in estimation but not induce bias.
In terms of representativeness, physicians who are above the 40th percentile in prescribingare over-represented in the data. Further, as the panel was recruited by AlphaImpactRx,there might be selection on willingness to participate in market research, which might becorrelated with treatment eects. To the extent that these items are a concern, they willaect how we should think about the counterfactuals. That is, we should not think that in-creasing detailing to those below the 40th percentile should necessarily have the same eectsas estimated here. Similarly, those willing to participate in the sample might have a higheror lower sensitivity to marketing activities than the general population of physicians.
Table 1 shows summary statistics before and after the informational shocks for both PCPsand psychiatrists. It is notable that PCPs have a large distinct jump in detailing and anincrease in prescribing as the shocks occur. That is much less pronounced for psychiatrists.
Table 2 shows summary stats for PCPs on the types of prescriptions they are writing,including on-label vs. o-label and distributions by age, severity of illness and insurancestatus.
Figure 3 presents psychiatrist prescribing of branded Seroquel, generic perphenazine, andthe dierence between the two. As can be seen, perphenazine is prescribed at a very lowrate, and overall prescriptions have a modest upward trend. In contrast, Figure 4 presentsPCP prescribing of the same products. As with the psychiatrists, PCPs hardly ever prescribeperphenazine, however, there is a strong upward trend in PCP prescriptions of Seroquel.
4 Identication Strategy and Estimation 4.1 The endogeneity problem and previous literature The individually targeted nature of detailing makes for signicant managerial opportunitiesbut also provides challenges in terms of estimating the eects of detailing. In particular, notonly is detailing set at the individual physician level, it is also not random. A rm optimallytargeting their details would necessarily direct more detailing activities to those who pro- vide the most potential prot: those who are most responsive and those who prescribe themost. As documented previous literature (Manchanda et al. 2004) and learned in informalconversations with managers, decile rules play a signicant role in detailing allocation. Thatis, sales reps tend to work independently and can visit physicians as many times as theyplease. However, the rm provides them with recommendations of how often to visit eachphysician, often generated by a third party analytics rm, while putting pressure on reps tomake a minimum total number of physician contacts. These recommendations are stronglyinuenced by the physician's volume of prescribing in the class of drugs, in particular thephysician's decile of category prescribing. Given this information, sales reps make choicesof whom to visit, the rm attributes sales to visits and rewards the sales reps with bonusesbased on performance. Given this structure, it is important that physician specic char-acteristics be controlled for as best as is possible, preferably with a physician xed eect.
Otherwise, the researcher runs the risk of attributing sales to detailing when that prescriberwould have prescribed with or without the sales rep visit.
This type of endogeneity problem is also very dicult to solve in aggregate data. Someauthors (Berndt et al. 1995, 1997) employ instrumental variables methods, using time untilpatent expiration as an excluded instrument. Firms advertise less as the patent expirationdate approaches. The diculty in this approach is that a number of supply and demandchanges happen at the same time as patent expiration approaches, so it is dicult to dis-entangle these eects. The most common approach to controlling for the endogeneity ofrm detailing decisions is a structural approach (e.g. Dong et al. 2009, Kalra et al. 2011,Montoya et al. 2010, Manchanda et al. 2004, Stremersch et al. 2013) whereby the re-searcher writes down the rm's objective function and requires that the rm optimize it.
If the theory is correct, the structural approach will control for the factors that determinedetailing decisions that might contaminate the estimated demand eects. This approachrequires reasonably strong assumptions on both the objective function of the rm and therm's sophistication in maximizing that objective function. This is especially true in aggre-gate data, but is also a complication in physician level data without using physician xedeects. Using xed eects can be computationally unattractive in complicated non-linearestimation problems, but failing to account for unobserved physician-specic characteristics might induce signicant positive bias.
A smaller literature has used physician level data and xed eects to assess the eects ofdetailing. One strand of such literature looks at the eects of policy changes involvingdetailing on physician behavior (Anderson et al. 2015, Larkin et al. 2013, Stremersch etal. 2009). Using these policy changes, it is possible to directionally sign the strategiesthat are being outlawed, but without detailing data, it is dicult to extract manageriallymeaningful implications. In particular, Anderson et al. (2015) show that conict of interestdisclosure policies have no signicant eect on the prescribing of antispychotics while Larkinet al. (2013) show that direct restrictions on detailing activities decreased antipsychoticprescriptions to all populations. Two other papers use physician xed eects together withphysician level detailing data (Datta et al. 2014, Mizik et al. 2004). While these approachescan control for a strict application of decile rules, they require the timing of detail visitsto be random. If sales reps can anticipate when demand will be high, they might detailmore during those months, and that would induce an upward bias, even with physicianxed eects. It is notable, however, that those studies with the smallest point estimates ondetailing elasticities are exactly those which employ physician level xed eects, suggestingthat physician characteristics play a large role in the allocation of detailing and in particular,those who would prescribe the most in the absence of detailing are the ones who are detailedthe most.
This study will employ a reduced form approach, using changes in scientic knowledge asa quasi-exogenous shock to the timing of detailing for one drug as well as physician xedeects to control for physician-specic unobservables that drive prescribing behavior.
4.2 Informational Shocks and Antipsychotics This study will leverage exogenous informational shocks that caused one branded product,Seroquel, to advertise much more to physicians. The informational shocks also providedequally good news for a generic competitor, perphenazine, which had no marketing response.
As both Seroquel and perphenazine received positive information, the bump in sales ofperphenazine caused by the shock will be attributed to the direct eect of the informational shock on demand and the dierence between the perphenazine and the Seroquel eect,conditional on a time trend, will be attributed to sales rep activities. The panel natureof the data will be exploited using physician-specic xed eects to control for unobservedphysician-specic factors that lead to prescribing. As such, the level of variation will be thetiming of detail visits to individual physicians. The identifying assumption will be that thetiming of those visits, driven by the introduction of scientic information, is exogenous.
When atypical antipsychotics were rst discovered and widely used in the late 1990s, theywere thought to be signicantly better than the older style typical antipsychotics in termsof their side eect proles. In particular, the atypicals were thought to carry a signicantlylower risk of metabolic side eects: signicant weight gain and development of diabetes.
However, these beliefs about comparative eectiveness and side eects were not clinicallyproven. As time progressed, the scientic community learned more about how these drugscompared with each other and with the older typical antipsychotics. In February of 2004,there was a consensus statement released by the American Diabetes Association, the Amer-ican Psychiatric Association, the American Association of Clinical Endocrinologists and theNorth American Association for the Study of Obesity in the journal, Diabetes Care. Thestatement was meant to summarize the results of a consensus development conference thattook place in November of 2003. In particular, the statement said, Clozapine and olanza-pine [generic name for Zyprexa] are associated with the greatest weight gain and highestoccurrence of diabetes and dyslipidemia. Risperdione [Risperdal] and quetiapine [Seroquel]appear to have intermediate eects. Aripiprazole and ziprasidone are associated with lit-tle or no signicant weight gain, diabetes, or dyslipidemia, although they have not beenused as extensively as other agents. The guidelines recommended both metabolic baselinescreening and follow-up monitoring of patients prescribed a second generation antipsychotic.
This information was a positive shock for all products other than Clozapine and Zyprexa,as Zyprexa was previously the market leader in antipsychotics and received clear negativenews. Indeed, Figure 2 shows that Zyprexa revenues took a signicant hit in 2004 andbeyond, losing its status as a market leader, while other products gained ground.
While the consensus statement provided some information about the comparative eective- ness and side eects of these products, it was well short of denitive in setting the standardof care. It was not the culmination of a randomized control trial, but rather the summa-tion of a number of other studies, some more suggestive than others. The eld went a bitfurther in commissioning The Clinical Antipsychotic Trials of Intervention Eectiveness(CATIE) in the early 2000s. The purpose of CATIE was to compare four new atypical an-tipsychotics (Zyprexa, Seroquel, Risperdal and Geodon) and one old typical antipsychotic(perphenazine) on the dimensions of tolerability, ecacy and side eect proles. It shouldbe noted that none of these products are molecularly equivalent: they are therapeutic sub-stitutes. The study was conducted from January 2001 through December of 2004 and theresults were disseminated in early 2005, to be published in the New England Journal ofMedicine in September of that year. The study found the big winners to be branded Sero-quel and generic perphenazine (which costs around $10 per prescription). Both products hadthe lowest incidence of metabolic and antipyramidal side eects while having similar ecacyand tolerability to other treatments. Given that there was more or less equal informationfor a very inexpensive generic and a very expensive brand, it would seem that perphenazinewould take over as the rst line medication for psychosis. Indeed, the National Institutesof Mental Health published a press release following the publication of CATIE explainingthat while the conventional wisdom that the older generation drug would have a worse sideeect prole, the study showed this not to be the case and that physicians should take thatinformation into account when making treatment decisions.9 However, these new results were met with a marketing response by Seroquel. Figure 5 showsthe average number of Seroquel detail visits received by each primary care physician in thesample after partialing out physician xed eects. There are distinct jumps in the amountof detailing in February of 2004 and January of 2005. As perphenazine is a very low margingeneric drug, it did not nd an expensive detailing campaign worth doing, even in light of thepositive information. This is unsurprising as it is extroadinarily rare for generic products tobe detailed. Figure 6 shows the average number of Seroquel detail visits for psychiatrists permonth, partialing out physician xed eects. Interestingly, there is no such distinct jumpin marketing activities to psychiatrists. The lack of a detailing response to psychiatrists suggests that the large number of visits received by psychiatrists are possibly for reasonsorthogonal to this particular new scientic information. With no distinct jumps at the datesof information revelation to psychiatrists, the remainder of this study will focus only onprimary care physicians, as the research design will not have power to identify eects onpsychiatrists.10 Are PCPs even an important target for antipsychotics? Figure 9 shows the total numberof detail visits in the sample inated to represent the number of PCPs and psychiatrists inthe United States. It appears that after the release of the CATIE study, more than half ofall detail visits are going to PCPs. This suggests that at least to the rm, PCPs are a veryimportant target. PCPs represent about a third of the antipsychotic prescriptions in thissample, though this sample is slightly skewed towards PCPs.
These jumps in marketing were not met with an overwhelmingly obvious jumps in prescribingbehavior. Figure 7 plots the Seroquel prescriptions per physician per month after partialingout physician-specic xed eects. While there could be a jump, it is not nearly as clear asthe distinctive jump in detailing. Figure 8 plots the perphenazine prescriptions per physicianper month after partialing out physician-specic xed eects. The vertical axis is notable,as PCPs rarely prescribe perphenazine at all, before or after the shocks.
A potential alternative explanation to dierential take-up of Seroquel over perphenazine byPCPs would be that they had already previously tried perphenazine before and decided thatthe side eects were too severe. Even though the CATIE study showed in the populationthat the side eect prole of perphenazine was just as good as that of Seroquel, perhaps thephysicians trust their past experience over the clinical study. However, Figure 4 and Figure8 show that very few PCPs ever prescribe perphenazine, suggesting that learning from pastexperience about perphenazine is unlikely. Additionally, each PCP tends to have low levels ofantipsychotic prescribing generally, with less than half of a prescription per month, makinginference based on experience very dicult. Finally, the CATIE study and the press releasefrom the National Institutes of Mental Health directly address the fact that Seroquel does 10To the extent that rms use decile rules to target detailing, a xed eects design will still be valid for psychiatrists under the assumption that the timing of visits is essentially random. A xed eects analysis of pyshciatrists is provided in the appendix.
not have a superior side eect prole to perphenazine. This should reduce the possibilitythat PCPs prescribe Seroquel over perphenazine due to an expectation of a lower incidenceof side eects. If such expectations exist among PCPs, it seems highly likely that thoseexpectations are driven by detailing rather than directly by the new scientic information.
Are the direct eects of the information shocks large, as measured by the trend-break intake-up of perphenazine at the information shock times? I rst address this question byestimating: P erphenazineRxit = αi + γ ∗ time + β1P ostShock1 + β2P ostShock2 + it Table 3 presents the results of estimating this equation including only the rst shock orusing both shocks. Neither shock induces a trend break in the prescribing of perphenazineby PCPs, even though the information shocks contain information that is highly benecial toperphenazine. The result is consistent with the pictures in Figures 4 and 8 which show thatPCPs rarely prescribe perphenazine. Given this result, I will continue under the assumptionthat a direct information eect without detailing is not inuencing PCP demand for Seroquel.
To the extent that there is an informational eect that is not detected by a product whichgets identical news, it will lead to an over-statement of the detailing eect.
In this study, the timing of these shocks to scientic knowledge will be treated as quasi-random, causing new detailing visits for Seroquel. Further, the panel nature of the dataallows for the control of unobserved physician-specic factors that aect prescribing in-dependent of marketing actions. As such the identifying variation will be the timing ofwithin-physician detailing visits generated by informational shocks.
4.3 Empirical Specication Given the structure outlined above, for physician i, in month t, the basic xed-eects speci-cation is: SeroquelRxit = αi + β ∗ SeroquelDetailit + γ ∗ time + it where there is a time trend to control for the trend in Seroquel prescribing as seen in Fig-ure 4. As prior to the informational shocks, PCPs receive almost no detailing visits, thisspecication will largely pick up the intuition of using the informational shocks as the ex-ogenous variation driving Seroquel detail visits. The further identifying assumption is thatconditional on a time trend, the direct eect of the informational shock on demand is zero,as detailed in Table 3 and the previous section. Specications including rival detailing willalso be included.
As the previous specication does not explicitly only use variation from the shocks, thefollowing specication does exactly that: SeroquelRxit = αi + ψ∗1(t ≥ tshock) + β ∗ SeroquelDetailit1(t ≥ tshock) + γ ∗ time + it.
I use this specication in two ways: one using the rst information shock of the consesnusstatement as the date of interest and one using the second shock induced by the CATIEstudy. The results of this specication are not expected to dier greatly from those in thexed-eects specication, as prior to the rst shock, there is very little PCP detailing as canbe seen in Table 5.
As detailing might represent a stock variable that depreciates over time, it is important toaccount for carry-over, both to to avoid contaminating the estimates of the eect of detailingow as well as to account for all of the prescriptions caused by a detail visit. I account forthe depreciation of detailing stock in two ways. First, I assume a geometric persistenceparameter and perform sensitivity analysis around that assumed parameter. In followingthe literature, I use specications with δ = 0.6 and δ = 0.8. Second, I non-parametricallyestimate the eect of detailing in previous periods on prescriptions in the current period.
The geometric decay specication is: it = αi + β ∗ δt−τ SeroquelDetailiτ + γ ∗ time + it While the nonparametric specication is: βτ SeroquelDetailiτ + γ ∗ time + it In the non-parametric specication, lags up to four months prior to the current month areincluded.
4.3.2 Detailing and O-Label Prescribing The main problem of this study is to assess the regulatory question of o-label promotioneorts. To assess this question, two analyses will be run: one for on-label vs. o-labeland one testing the eect on dierent age groups, as the primary concern about o-labelmarketing in this category is the potential for exploitation of vulnerable children or theelderly. In each analysis, the sample will be split into a set of mutually exclusive groups, g.
For the o-label analysis g ∈ {on−label, off −label}. To get a further sense of this eect asit relates to seniors in nursing homes or pediatric patients, we can also assess whether or notthere is a disproportionate eect based on age, using g to index age, g ∈ {under 18, Adult ≤ 60, Adult > 60}. For each analysis, the specication for group g is: SeroquelRxgit = αig + βg ∗ SeroquelDetailit + γg ∗ time + git The total of the treatment eects add up to the estimated treatment eect from the baselinespecication. To see if detailing has a disproportionate aect on group g, ˆβg will be compared with SeroquelRxg . That is, we will compare the share of the detailing eect attributable to group g to the share of the prescriptions that group g makes up. In this way, we willbe comparing the distribution of detailing eects with the distribution of prescriptions, or comparing the detailing marginal' patient with the average' patient. If detailing has adisproportionate eect on o-label prescriptions, for example, then ˆβofflabel > SeroqueRxofflabel , or the proportion of the detailing eect on o-label prescriptions will be greater than theproportion of prescriptions which is o-label. If the reverse is true, detailing would have hadthe eect of shifting the distribution more to on-label. If these two ratios are the same, weconclude that promotion lifts the number of prescriptions, but does not aect the distributionof where those prescriptions are coming from with respect to the focal patient characteristic.
If there is a disproportionate eect of detailing on o-label prescriptions, it could comefrom two main sources: within-physician changes in the share of o-label prescriptions orbetween-physician heterogeneity in the treatment eect. The intuition of the rst source isreasonably straightforward and will be estimated by regressing the within-physcian share ofo-label prescriptoins on detailing, conditional on a physician xed eect and time trendusing: )it = αi + β ∗ SeroquelDetailit + γ ∗ time + it For intuition of the second potential source of distribution shifting, suppose that there aretwo types of physicians: mainly on-label and mainly o-label. If the treatment eect onthe mainly o-label type is larger than that on the mainly on-label type, the populationdistribution of prescriptions would tilt towards o-label even if detailing had no eect onthe within-physician share of o-label prescriptions. To test this intuition, I will groupthe physicians by share of o-label prescriptions. In particular, there are 244 physicians(14% of the physicians in the sample) in the data who prescribe exclusively o-label. Thosephysicians will be labeled as o-label physicians in the following estimating equation: SeroquelRxit = αi+β1∗SeroquelDetailit+β2∗SeroquelDetailit∗Of f LabelP hysician+γ∗time+it If β2 > 0 , the eect of detailing on o-label physicians is greater than the eect on on-labelphysicians, tilting the distribution towards o-label. If β2 < 0, the opposite is true. Thetotal eect on o-label physicians is β1 + β2.
Results to the baseline analysis are presented in Table 4. Column 1 presents a naiveregression in which Seroquel prescriptions are the dependent variable and Seroquel detailvisits the independent variable. It has neither a time trend nor physician xed eects. Thisspecication suggests that each detail visit is associated with 0.29 additional prescriptionsof Seroquel, resulting in an implied elasticity of demand with respect to detailing ow ofabout 0.2. Column two adds in physician xed eects, but continues to omit a time trend.
As can be seen in Figure 4, Seroquel prescriptions for PCPs are trending up through timeeven before the rst information shock, so this estimate would still be expected to be biasedupward. Indeed, it suggests that each Seroquel detail visit is associated with around 0.19prescriptions concurrently resulting in an elasticity of 0.13. In column 3, a time trend isadded, and the resulting eect of detailing is about 0.15, with an elasticitiy of about 0.1.
Columns 4 and 5 show specications where the period post information shock is treated asan experimental period. In column 4, the post period is everything after the joint statementin 2004. In column 5, the post period is just following the CATIE release. Neither of thesespecications is signicantly dierent from the xed eects and time trend specication incolumn 3. As such, the remainder of the study will make use of the xed eects specicationwith the time trend. The nearly 100% upward bias in column 1 suggests that detailing is infact focused on those physicians most likely to prescribe, even in the absence of a detailingvisit, further solidifying the need to control for unobserved physician factors. Columns 6-8include rival product detailing. The coecient on rival detailing is small and insignicantand does not signicantly change the coecient on Seroquel detailing.
5.1.1 Stock Measures and Depreciation Results to the stock/ow analysis are presented in Table 5. The rst column transposescolumn 3 from Table 4 for comparison. Column 2 presents the non-parametric specicationof the long-lasting eects of detailing. Detailing from one and two months prior has a signicant eect on prescribing smaller than the instantaneous eect. In months three andfour, the eect depreciates to nearly zero. An average geometric persistence given thesenon-parametric parameters would be about 0.66 assuming geometric decay. Adding up thetotal eect from this specication gives that a single detailing visit results in about 0.3 totalprescriptions through time. Column 3 shows the eect with an assumed geometric persistenceparameter of 0.6. The point estimate on concurrent detailing decreases a bit from baseline,but implies about 0.31 prescriptions resulting from a detail visit through time. Column 4shows the eect with a persistence parameter of 0.8. Again, the point estimate on concurrentdetailing goes down from the baseline and the 0.6 case, but implies about 0.45 prescriptionsresulting from a detail visit through time. As the non-parametric specication and the 0.6specication are reasonably consistent with each other in terms of long run eects, a long-runeect of about 0.3 will be used going forward. Additionally, as the eects of past detailingare signicant, all further analysis will use a depreciated stock of detailing with persistenceparameter of 0.6 as the variable of interest.
Given these eects of detailing, it is instructive to ask if they make sense given the amountof detailing observed in the data. For the sake of a back-of-the-envelope calculation, assumeeach sales rep earns roughly $150,000 per year. Per Figure 7, assume that roughly half ofeach rep's time is allocated to PCPs. This translates to roughly $6,250 per month per salesrep on PCPs. Over the course of this sample, the average price per prescription of Seroquelis $225. Given this price, each sales rep needs to generate around 28 new PCP prescriptionsof Seroquel per month in order to break even. Given the estimate of 0.3 prescriptions perdetail visit, 28 new prescriptions would require about 100 PCP visits over the course of amonth, or about ve visits per work day. Discussions with a former sales rep indicate to methat each sales rep is expected to make between eight and ten total visits per day. If half ofthese visits went to PCPs, it would put the rm very close to the break-even point. Whilethe rm likely would like to do better than simply breaking even on sales rep activities,these numbers indicate that the rm is not way out of the ballpark in terms of sales force allocation and that these numbers are plausible.
The small size of the detailing eect relative to the existing literature is also notable. Here Iemphasize that in this particular context, a small eect of detailing is perhaps unsurprising.
Indeed, prior to the rst information shock, there was almost no detailing to PCPs. If rmsthought that detailing would be hugely productive in this sample, surely they would havebeen detailing the PCPs before. The pattern of detailing is consistent with an upward shift inthe expected quality of the product in the eyes of PCPs, making detailing to these physicianschange from being not worthwhile to being worthwhile to the rm.
The small eect is also interesting when considering potential contamination in the estimatedeect. In particular, if the reader is concerned that the informational eect is not fullycontrolled for, the detailing eect will be over-estimated, making it even smaller.
5.2 Regulatory Concerns 5.2.1 O-label prescribing Table 6 shows the results of the o-label analysis, using the depreciated stock of detailingas the main variable of interest. In the rst column is the eect of detailing on o-labelprescriptions and in the second column is the eect of detailing on on-label prescriptions.
Here, on-label is dened to be a diagnosis of schizophrenia or bipolar disorder. O-labelprescriptions are primarily dementia and insomnia diagnoses, but range from joint pain topost-traumatic stress disorder (PTSD) to anger management. Seroquel is not approved bythe FDA to treat any of these illnesses, so they are coded as o-label. The estimated eectof detailing on o-label prescriptions is positive, suggesting that detailing does cause anincrease in o-label prescribing. However, on-label prescribing is much more aected. Whileo-label prescriptions make up more than 44 percent of the total prescriptions, they onlymake up about 29 percent of the eect of detailing, a dierence that is both economicallyand statistically signicant. As such, while detailing increases o-label prescribing, it shiftsthe distribution of prescribing more towards on-label than o-label. These results are furtherillustrated in Figure 12.
Table 7 highlights the source of the dispropotionate eect on on-label prescribing. Column1 shows that detailing is not changing the within-physician share of prescribing that isattributable to o-label prescriptions. Meanwhile, column 2 shows that the eect of detailingto physicians that primarily prescribe o-label is much lower than the average eect ofdetailing. In fact, the eect of detailing on the o-label type physicians is not distinguishablefrom zero. Since the eect on those physicians who never prescribe on-label is zero whilethe eect on those that do is positive while holding the within-physician share of on-labelprescriptions xed, the population share of on-label prescriptions increases.
A primary concern about o-label promotion in antipsychotics relates to seniors in nursinghomes being treated for dementia and insomnia as well as pediatric patients. Table 8 andFigure 13 show the results of the eects of detailing on dierent age groups. While thoseover age sixty make up roughly 33% of the total Seroquel prescriptions, they are only respon-sible for about 24% of the detailing eect. The dierence is signicant and indicates thatdetailing in fact shifts the distribution of Seroquel prescriptions away from seniors, ratherthan towards. Pediatric patients, on the other hand, make up a very small share of both theprescriptions and the share of the detailing eect and the dierence between the two is notsignicant.
What do these results imply for managers? Firms are targeting physicians who are primarilyprescribing antipsychotics for o-label use. That detailing is not only ineective, it carrieswith it the risk of huge nes from the government. Firms might think twice about trying topromote o-label use given that the penalty is high if caught and the benet is reasonablylow if uncaught.
What should a regulator take from this? It depends upon the perceived seriousness of o-label prescribing. If regulators believe that any o-label prescribing is unacceptable, then itmight be worth sacricing the new on-label prescriptions to those who need them in order toeliminate the new o-label prescriptions by banning detailing. However, if the main concernis that promotion is really about promoting o-label uses, the data seems not to play thatout, as detailing shifts the distribution toward on-label prescribing. This is consistent withthe story that detailing is informing physicians about the relatively low adverse eects of the drug, and this leads to the physician being more comfortable in prescribing, and this isespecially pronounced in the on-label diagnoses.
The results above paint a reasonable sanguine regulatory picture of antipsychotic detailingof Seroquel in terms of its eects on primary care providers. While it seems clear that thesales reps did not bother to mention that there was a very cheap alternative (as evidencedby the lack of increase in perphenazine demand), there is also no evidence to support thatdetailing was disproportionately aecting o-label prescriptions, seniors or children. Indeed,there is evidence to support detailing as shifting the prescription distribution more to on-label and away from seniors while leaving prescriptions for pediatric patients unchanged.
Given these results for primary care physicians, the billions of dollars in settlements maybe seen in a dierent light. Regulators might in fact be less concerned about psychiatristsbeing unduly inuenced, as they are likely to know the important information and be awareof the risks involved in o-label prescribing. The results shed less light on the rms' intent.
Indeed, while their eorts have no eect on primarily o-label physicians, those physiciansstill receive visits from sales reps in the data. Since the eects of detailing on o-labelprescribing are so small, perhaps managers should be extra careful not to promote that way,as the risk of the regulatory nes is enormous. Conversely, perhaps the regulator time andeort could be allocated more eciently than on litigating o-label promotion.
Some caveats to these conclusions are necessary. First, the estimated eects in this study areall the eects of Seroquel detailing on Seroquel demand. It is possible that other productshave signicantly dierent detailing eorts with dierent eects on the regulatory concernsof interest. Indeed, Zyprexa and Risperdal did have much higher nes imposed upon themfor their marketing eorts. Second, these eects apply only to primary care physicians ratherthan psychiatrists. It is curious that pyschiatrists get so many detail visits when it is likelythat they already know the relevant scientic information. In fact, the new discoveries didnot lead to any signicant change in detailing activities to psychiatrists in the data. Finally,even given the reasonably sanguine regulatory results presented here, very few physicians ever adopt the very inexpensive and eective perphenazine. While further regulation of detailingis not likely to lead to mass adoption of perphenazine, it is notable that the observed marketoutcomes might signicantly dier from the social planner's rst best outcome. However,the results of this study suggest that detailing eorts to primary care physicians perhapshave fewer adverse market eects than regulators might previously have thought.
In this study, I study the eects of promotion on o-label prescribing decisions of physiciansin the antipsychotic category, which was the category most implicated in o-label promotionlitigation by the federal government. To estimate this, I employ two scientic informationalshocks to the antipsychotic category together with a physician xed-eect design are used toidentify the eects of detailing on the prescribing of antipsychotics by primary care providers.
In particular, I address the important question of o-label marketing in a category that hasseen incredibly large regualtory nes for its promotional activities. I nd that the shortterm eect of detailing to PCPs is about 0.15 with a long-run eect of about 0.3. Thesemagnitudes are modest and from a regulatory perspective, these additional prescriptionsare disproportionately likely to be on-label rather than o-label prescriptions. The dispro-portionality stems from detailing having no eect on physicians that primarily prescribeo-label rather than from within-physician changes in the share of o-label prescriptions. Inaddition, detailing shifts the age distribution of the prescribed away from the elderly ratherthan towards, alleviating the regulatory concern about o-label prescribing to seniors.
Meanwhile the lack of any eect of the informational shocks on detailing to psychiatristssuggests that those visits are orthogonal to this particular scientic information. Whilethat may not be surprising, as psychiatrists might be very likely to know the informatoinpre-release, they receive signicant numbers of sales rep visits both before and after theinformational shocks. The intention and eects of those visits are less clear and are certainlyworthy of further research.
[1] Anderson, T. S., Huskamp, H. A., Epstein, A. J., Barry, C. L., Men, A., Berndt, E. R., . & Donohue, J. M. (2015). Antipsychotic Prescribing: Do Conict of Interest PoliciesMake a Dierence?. Medical care, 53(4), 338-345.
[2] Arana, G. W. (2000). An overview of side eects caused by typical antipsychotics.
Journal of clinical psychiatry, 61(8), 5-11.
[3] Arcidiacono, P., Ellickson, P. B., Landry, P., & Ridley, D. B. (2013). Pharmaceutical followers. International Journal of Industrial Organization, 31(5), 538-553.
[4] Arrow, K. J. (1963). Uncertainty and the welfare economics of medical care. The Amer- ican economic review, 941-973.
[5] Berndt, E. R., Bui, L., Reiley, D. R., & Urban, G. L. (1995). Information, marketing, and pricing in the US antiulcer drug market. The American Economic Review, 85(2),100-105.
[6] Berndt, E. R., Bui, L., Reiley, D. R., & Urban, G. L. (1997). The roles of price, quality, and marketing in the growth and composition of the antiulcer drug market. The eco-nomics of new goods (NBER Studies in Income and Wealth, vol. 58), Bresnahan, T. F.,& Gordon, R. J. (Eds.). (1997), (Vol. 58). University of Chicago Press., pp. 277-322.
[7] Berndt, E. R., Pindyck, R. S., & Azoulay, P. (2003). Consumption externalities and dif- fusion in pharmaceutical markets: antiulcer drugs. The Journal of Industrial Economics,51(2), 243-270.
[8] Bradford, W. D., Turner, J. L., & Williams, J. W. (2015). O-label use of pharmaceuticals: a detection controlled estimation approach. Available at SSRN:http://ssrn.com/abstract=2230976 or http://dx.doi.org/10.2139/ssrn.2230976 [9] Ching, A., & Ishihara, M. (2010). The eects of detailing on prescribing decisions under quality uncertainty. Quantitative Marketing and Economics, 8(2), 123-165.
[10] Chintagunta, P. K., & Desiraju, R. (2005). Strategic pricing and detailing behavior in international markets. Marketing Science, 24(1), 67-80.
[11] Chintagunta, P. K., Goettler, R. L., & Kim, M. (2012). New drug diusion when forward-looking physicians learn from patient feedback and detailing. Journal of Mar-keting Research, 49(6), 807-821.
[12] Conti, R. M., Huskamp, H. A., & Berndt, E. R. (2011). The Eect of FDA Advisories on Branded Pharmaceutical Firms' Valuations and Promotion Eorts (No. w17528).
National Bureau of Economic Research.
[13] Datta, A., & Dave, D. M. (2013). Eects of Physician-Directed Pharmaceutical Promo- tion on Prescription Behaviors: Longitudinal Evidence (No. w19592). National Bureauof Economic Research.
[14] David, G., Markowitz, S., & Richards, S. (2009). The eects of pharmaceutical mar- keting and promotion on adverse drug events and regulation (No. w14634). NationalBureau of Economic Research.
[15] Dong, X., Manchanda, P., & Chintagunta, P. K. (2009). Quantifying the benets of individual-level targeting in the presence of rm strategic behavior. Journal of MarketingResearch, 46(2), 207-221.
[16] Dong, X., Chintagunta, P. K., & Manchanda, P. (2011). A new multivariate count data model to study multi-category physician prescription behavior. Quantitative Marketingand Economics, 9(3), 301-337.
[17] Ellison, G., & Ellison, S. F. (2011). Strategic entry deterrence and the behavior of pharmaceutical incumbents prior to patent expiration. American Economic Journal:Microeconomics, 1-36.
[18] Fischer, M., & Albers, S. (2010). Patient-or physician-oriented marketing: what drives primary demand for prescription drugs?. Journal of Marketing Research, 47(1), 103-121.
[19] de Frutos, M. A., Ornaghi, C., & Siotis, G. (2013). Competition in the pharmaceutical industry: How do quality dierences shape advertising strategies?. Journal of healtheconomics, 32(1), 268-285.
[20] Gönül, F. F., Carter, F., Petrova, E., & Srinivasan, K. (2001). Promotion of prescription drugs and its impact on physicians' choice behavior. Journal of Marketing, 65(3), 79-90.
[21] Gurwitz, J. H., Field, T. S., Avorn, J., McCormick, D., Jain, S., Eckler, M., . & Bates, D. W. (2000). Incidence and preventability of adverse drug events in nursing homes.
The American journal of medicine, 109(2), 87-94.
[22] Huckfeldt, P. J., & Knittel, C. R. (2011). Pharmaceutical use following generic entry: Paying less and buying less (No. w17046). National Bureau of Economic Research.
[23] Iyengar, R., Van den Bulte, C., & Valente, T. W. (2011). Opinion leadership and social contagion in new product diusion. Marketing Science, 30(2), 195-212.
[24] Janakiraman, R., Sismeiro, C., & Dutta, S. (2009). Perception spillovers across com- peting brands: a disaggregate model of how and when. Journal of Marketing Research,46(4), 467-481.
[25] Janakiraman, R., Dutta, S., Sismeiro, C., & Stern, P. (2008). Physicians' persistence and its implications for their response to promotion of prescription drugs. ManagementScience, 54(6), 1080-1093.
[26] Kalra, A., Li, S., & Zhang, W. (2011). Understanding responses to contradictory infor- mation about products. Marketing Science, 30(6), 1098-1114.
[27] Kolsarici, C., & Vakratsas, D. (2010). Category-versus brand-level advertising messages in a highly regulated environment. Journal of Marketing Research, 47(6), 1078-1089.
[28] Larkin, I., Ang, D., Avorn, J., & Kesselheim, A. S. (2014). Restrictions On Pharmaceu- tical Detailing Reduced O-Label Prescribing Of Antidepressants And AntipsychoticsIn Children. Health Aairs, 33(6), 1014-1023.
[29] Lieberman, J. A., Stroup, T. S., McEvoy, J. P., Swartz, M. S., Rosenheck, R. A., Perkins, D. O., . & Hsiao, J. K. (2005). Clinical Antipsychotic Trials of InterventionEectiveness (CATIE) Investigators Eectiveness of antipsychotic drugs in patients withchronic schizophrenia. N Engl J Med, 353(12), 1209-1223.
[30] Liu, H., Liu, Q. & Chintagunta, P.K., Advertising spillovers: drug detailing in com- bination therapy (2014). Available at SSRN: http://ssrn.com/abstract=2406363 orhttp://dx.doi.org/10.2139/ssrn.2406363 [31] Majumdar, S. R., Almasi, E. A., & Staord, R. S. (2004). Promotion and prescribing of hormone therapy after report of harm by the Women's Health Initiative. Jama, 292(16),1983-1988.
[32] Manchanda, P., Rossi, P. E., & Chintagunta, P. K. (2004). Response modeling with nonrandom marketing-mix variables. Journal of Marketing Research, 41(4), 467-478.
[33] Mizik, N., & Jacobson, R. (2004). Are physicians easy marks? Quantifying the eects of detailing and sampling on new prescriptions. Management Science, 50(12), 1704-1715.
[34] Montoya, R., Netzer, O., & Jedidi, K. (2010). Dynamic allocation of pharmaceutical detailing and sampling for long-term protability. Marketing Science, 29(5), 909-924.
[35] Moon, S., Kamakura, W. A., & Ledolter, J. (2007). Estimating promotion response when competitive promotions are unobservable. Journal of Marketing Research, 44(3),503-515.
[36] Nair, H. S., Manchanda, P., & Bhatia, T. (2010). Asymmetric social interactions in physician prescription behavior: The role of opinion leaders. Journal of Marketing Re-search, 47(5), 883-895.
[37] Narayanan, S., & Manchanda, P. (2009). Heterogeneous learning and the targeting of marketing communication for new products. Marketing Science, 28(3), 424-441.
[38] Narayanan, S., Manchanda, P., & Chintagunta, P. K. (2005). Temporal dierences in the role of marketing communication in new product categories. Journal of MarketingResearch, 42(3), 278-290.
[39] Narayanan, S., Desiraju, R., & Chintagunta, P. K. (2004). Return on investment impli- cations for pharmaceutical promotional expenditures: The role of marketing-mix inter-actions. Journal of marketing, 68(4), 90-105.
[40] Ray, W. A., Federspiel, C. F., & Schaner, W. (1980). A study of antipsychotic drug use in nursing homes: epidemiologic evidence suggesting misuse. American Journal ofPublic Health, 70(5), 485-491.
[41] Rao, A., & Wang, E. (2015). Demand for healthy products: false claims in advertising.
University of Chicago working paper.
[42] Ross, J. S., Nazem, A. G., Lurie, P., Lackner, J. E., & Krumholz, H. M. (2008). Up- dated estimates of pharmaceutical company payments to physicians in Vermont. JAMA,300(17), 1998-2000.
[43] Stremersch, S., & Lemmens, A. (2009). Sales growth of new pharmaceuticals across the globe: The role of regulatory regimes. Marketing Science, 28(4), 690-708.
[44] Stremersch, S., Landsman, V., & Venkataraman, S. (2013). The relationship between DTCA, drug requests, and prescriptions: Uncovering variation in specialty and space.
Marketing Science, 32(1), 89-110.
[45] Venkataraman, S., & Stremersch, S. (2007). The debate on inuencing doctors' deci- sions: Are drug characteristics the missing link?. Management Science, 53(11), 1688-1701.
[46] Zoltners, A. A., & Sinha, P. (2005). Sales territory design: Thirty years of modeling and implementation. Marketing Science, 24(3), 313-331.
Tables and Figures Figure 1: Revenues Over Time Figure 2: Revenues over Time by Firm Figure 3: Psychiatrist Prescribing Figure 4: PCP Prescribing Table 1: Summary Statistics Pre and Post Information Table 2: Summary Statistics: PCP by Type of Rx Figure 5: Probability of Seroquel Detail Visit for PCP PCP seroquel visits Physician Fixed Eects partialed out Figure 6: Probability of Seroquel Detail Visit for PSY PSY seroquel visits Physician Fixed Eects partialed out Figure 7: Seroquel Prescriptions for PCP Physician Fixed Eects and Time trend partialed out Figure 8: Perphenazine Prescriptions for PCP Physician Fixed Eects and Time trend partialed out Figure 9: Total Detailing Visits By Specialty Detailing from this sample extrapolated to population of physicians Table 3: Eects of Informational Shocks on Perphenazine Demand *** p<0.001, ** p<0.01, * p<0.05 Physician clustered standard errors in paranetheses. All specications in- clude physician-specic xed eects, and time trends.
Figure 10: Eects of Detailing on O-Label Prescribing 0.169*** (0.0271) 0.00884 (0.00909) 0.155*** (0.0237) 0.00642 (0.00917) 0.146*** (0.0226) 0.00556 (0.00923) 0.168*** (0.0271) 0.145*** (0.0226) 62,849 ** dummies 0.188*** (0.0235) 62,849 *** theses. ws.o 0.289*** (0.0323) Table 5: Stock Measures 0.145*** 0.115*** 0.126*** 0.0902*** Implied Elasticity *** p<0.01, ** p<0.05, * p<0.1 Physician clustered standard errors in parentheses. Physician-specic xed eects and time trends are included in all specications. Elasticities are computed at the sample means and are with respect to advertising ows.
Table 6: Eects of Detailing on O-Label Prescribing Seroquel O-Label Rx Seroquel On-Label Rx *** p<0.001, ** p<0.01, * p<0.05 Physician clustered standard errors in paranetheses. Seroquel Visits are depreciated stocks with persistence parameter 0.6. All specications include physician-specic xed eects and time trends.
Table 7: How Does Detailing Shift O-Label Distribution? Seroquel O-Label Share Seroquel Rx Visits x (Only O-Label) *** p<0.001, ** p<0.01, * p<0.05 Physician clustered standard errors in paranetheses. Seroquel Visits are depreciated stocks with persistence parameter 0.6. All specications include physician-specic xed eects and time trends.
Table 8: Heterogeneity by Age 0.0934*** 0.0307*** Pct of Total Eect *** p<0.001, ** p<0.01, * p<0.05 Physician clustered standard errors in paranetheses. All specications include physician-specic xed eects and time trends.
Figure 11: Heterogeneity by Age Appendix A - Eects of Detailing on Psychiatrist O-Label Prescribing The informational shocks do not aect detailing to psychiatrists. As such, the proposedidentication strategy using the informational shocks to generate quasi-exogenous timing indetail visits will not have power to identify the eects of detailing on psychiatrist prescribing.
However, if we are willing to accept that the timing of visits to psychiatrists are essentiallyrandom, as would be the case if rms were employing a decile rule, we could implementa simple xed eects estimator for psychiatrists to obtain an estimate of the eectivenessof detailing for driving prescriptions, both on and o label. In this section, I explore theeects of detailing on psychiatrist prescribing using only a xed eects approach. Most of thedirectional conclusions are similar as with the PCPs, noting that the identication requiresthe additional assumption of random timing of detail visits.
Table 9 shows the eects of the two informational shocks on psychiatrist prescribing ofthe generic drug, perphenazine. Similar to the PCPs, the informational shocks do notdrive psychiatrists to prescribe the eective and inexpensive drug. If anything, it appears Table 9: Eects of Informational Shocks on Perphenazine Demand for Psychiatrists *** p<0.001, ** p<0.01, * p<0.05 Physician clustered standard errors in paranetheses. All specications in- clude physician-specic xed eects, and time trends.
that following the CATIE study, psychiatrists in fact prescribe less perphenazine. Table10 provides the basic xed eects results. Unlike in the PCP case, including rival detailingmakes a signicant dierence for psychiatrists. This may be due to the fact that psychiatristsare highly likely to be detailed by many companies while PCPs are more likely only tobe detailed by one. It appears as though rival detail visits are positively correlated withSeroquel prescriptions. That is, detailing to psychiatrists appears to be category expansive,and provide a positive spillover to rivals. This could also be driven by non-random timingof visits to each physician. If sales reps were able to predict the months with high levels ofprescribing and visit in those months, prescribing would be positively correlated with bothown and rival detailing, even though the correlation was not causal. In order to interpretthese eects as causal, we need to assume that such non-random timing does not occur.
Table 11 presents results using a depreciated stock of detailing. Similar to PCPs, we canreject that past detailing does not matter and a persistence parameter of about 0.6 seemsreasonable and will be used going forward. It is notable that while the marginal eect of adetail visit is considerably larger for psychiatrists, at about 2.4 prescriptions (as comparedwith 0.3 for PCPs), the elasticity is roughly half as big: only around 0.05. This comes fromthe fact that psychiatrists start from a much higher base of prescribing. This elasticity is extremely small, and in fact smaller than any other study in the detailing literature. Afailure in the assumption of random timing would lead to an over-estimate of the elasticity.
As such, we should view this 0.05 as an upper bound, and a very small one at that.
Tables 12 and 13 provide the o-label analysis for psychiatrists. Before any regression anal-ysis, it is notable that psychiatrists prescribe a much smaller share o-label than do PCPs,at about 22% of prescriptions as opposed to 43% by PCPs. Even so, detailing still providesa disproportionately large eect on on-label prescriptions, tilting the distribution towardson-label. While this disproportionality is not as pronounced as it is in the PCP sample, itremains signicant.
The rst qualitative dierence we see in the psychiatrists is that detailing to them seemsnot to shift the age distribution of prescribing at all. Visits do not disproportionately aectseniors, children or non-elderly adults in any statistically signicant way.
A.1 Discussion and Policy Implications While the marginal eect of detailing on psychiatrists is higher than it is for PCPs (as shouldbe expected with their higher base rate of prescribing), the detailing elasticity remains verylow. While the identication of these eects is not as clean due to the potential for non-random timing, the potential bias here would bias the eect in the upward direction. Thesemain eects are much smaller than those found elsewhere in the literature. Just given thesmall overall eects, regulators might nd the social returns on their litigation disappoint-ing. They will have spent time and energy that could have been employed elsewhere whilepotentially causing distortions in the product market. Conversely, managers might nd thereturns on their detailing eorts to be disappointing. Perhaps the growing inuence of payersand the inherent knowledge of the physicians has lowered the inuence of sales reps.
Of the small detailing eect, most of it is attributable to on-label prescribing. In fact, o-labelprescribing is much lower for psychiatrists than PCPs and detailing shifts the distributioneven more towards on-label prescriptions. This makes even smaller the regulatory concernabout o-label promotion leading to o-label prescriptions to those most vulnerable. Thisreinforces the managerial implication that while it might seem attractive to try and promote o-label use through sales reps, the return is almost surely not worth the expected cost oflitigation.
Figure 12: Eects of Detailing on O-Label Prescribing 0.961*** (0.203) 0.736** (0.113) Table 11: Stock Measures 0.961*** 0.796*** 0.932*** Implied Elasticity *** p<0.01, ** p<0.05, * p<0.1 Physician clustered standard errors in parentheses. Physician-specic xed eects, rival visits and time trends are included in all specications. Elasticities are computed at the sample means and are with respect to advertising ows.
Table 12: Eects of Detailing on O-Label Prescribing Seroquel O-Label Rx Seroquel On-Label Rx *** p<0.001, ** p<0.01, * p<0.05 Physician clustered standard errors in paranetheses. Seroquel visits are depreciated stocks with persistence parameter 0.6. All specications include physician-specic xed eects, rival detail visits and time trends.
Table 13: Heterogeneity by Age Pct of Total Eect *** p<0.001, ** p<0.01, * p<0.05 Physician clustered standard errors in paranetheses. All specications include physician-specic xed eects and time trends.
Figure 13: Heterogeneity by Age

Source: https://research.chicagobooth.edu/~/media/7CB817FF82E34EA3AD98273E60B1CF62.pdf