Pharmaceuticals, Incremental Innovation andMarket Exclusivity

Nina Yin∗

Department of Economics and the Murphy InstituteTulane University

April 23, 2015

Abstract

This paper assesses the welfare gains from incremental innovation in phar-maceuticals. Such innovation can yield consumer gains through improved qual-ity, but the additional market exclusivity granted to innovators may also delaygeneric entry, a practice referred to as “evergreening”, and reduce consumer sur-plus. Quantifying this tradeoff is vital in determining the optimal patent policyand regulatory treatment of incremental innovation. To shed light on this prob-lem, I focus on incremental innovations in selective serotonin reuptake inhibitor(SSRI) anti-depressant drugs. Based on individual-level prescription data, I es-timate the patients’ demands for antidepressants with a conditional logit model,and backed up firms’ marginal costs with Bertrand-Nash model. By evaluating thewelfare changes from status quo to the counterfactual scenarios in which marketexclusivity and incremental innovation become no longer available, I found thatthe consumer surplus loss due to market exclusivity extensions far exceeds theconsumer surplus benefits from incremental innovation. Without taking into ac-count the innovation costs, the innovators benefit from incremental innovation by$1.55 billion even without market exclusivity. Overall, the evidence indicates thatthe market exclusivity granted for incremental innovations in SSRIs has resultedin a $37 billion social welfare loss from 1996 to 2011. The result suggests thatpolicy makers may need to revisit the provisions for granting market exclusivityto incremental innovations.

Keywords: pharmaceuticals, incremental innovation, market exclusivity, patents.JEL Codes: I18, K11, L65, L10.

∗I deeply appreciate my advisor Margaret Kyle for her invaluable supports and suggestions in thiswork. And I would like to thank Peter Arcidiacono, Andrew Ching, Pierre Dubois, Matt Higgins,Yinghua He, Laura Lasio, Thierry Magnac, Fanny Camara, Ying Fan, Martin O’Connell, Mary Olson,Alexander Sandukovskiy, Yuya Takahashi, Sergio Urzua, Heidi Williams, Takuro Yamashita, and Xun-dong Yin for their illuminating discussions and thank all the participants in EARIE-2012 Conferencein Rome, applied microeconomics workshop, IP/IT workshop, PhD workshop and Brown Bag Seminarat TSE, 10th CEPR/JIE School on Applied Industrial Organization, Conference “The Economics of In-tellectual Property, Software and the Internet”, 2013 China Meeting of the Econometric Society, HealthPolicy Working Group at Tulane, 2013 Health Econometrics Workshop at UNC-Chapel Hill and 2014ASHEcon meeting. All remaining errors are my own. Email: nyin@tulane.edu.

1

1 Introduction

“Incremental Innovations are not small achievements. Heat stable version of anti-retroviral drugs may not be critically important to HIV patients in large cities wherethere is easy access to electricity and refrigeration, but they are surely important topeople in rural areas.”

Greg Kalbaugh, Director of US-India Business Council1

“Over the next few years, a number of blockbuster drugs face patent expiration. ...it is estimated that by 2012, brands with more than $30 billion in sales will face newcompetition from generics. As more brands face patent expiration, many manufacturerswill face the dilemma of how to grow revenue and minimize operational cutbacks asreliance on the new drug pipeline is unrealistic. One tactic is to develop an extendedrelease formulation of an existing brand. Whether you call it extended release (ER, XR),long-acting (LA), or extra-long (XL), the modified formulation is intended to simplifydosing, improve compliance and extend the life of the patent. (emphasis added)”

Kelly Renfro, Marketing Manager for McKesson Patient Relationship Solutions2

Innovation drives growth, but competition may undermine the incentives forinnovation due to the nonrivalrous nature of ideas. Market exclusivity, providedby intellectual property rights and data exclusivity provisions, aims to incentivizeinnovation by allowing firms to recoup their R&D expenditures and extract thereturns to investments in the marketplace. These policies are controversial: Howcan we minimize dead-weight loss due to monopoly pricing without underminingincentives to innovate? This paper examine the welfare effects of incremental in-novation and delayed generic entry that results from granting additional marketexclusivity to incremental pharmaceutical innovations.

“Incremental innovation” in pharmaceuticals, in contrast to the radical inno-vation of a new molecule to treat diseases, involves improvements over existingdrugs such as the discovery of a new therapeutic use, new formulation, additionalpediatric use, or improved efficacy and safety. The Hatch-Waxman Act (1984)and the FDA Modernization Act (FDAMA, 1997) grant innovators additional mar-ket exclusivity for incremental innovations (details in Section 2). Although incre-mental innovation may generate social benefits, additional market exclusivity mayimpose social costs. Market exclusivity implies an extension of monopoly, whichhurts consumer surplus: budget-constrained patients may be unable to purchasea treatment at all, or too little; this may in turn affect health status measures. Imeasure the value of incremental innovation from an economic perspective. I es-timate demand based on individual prescription-level data for antidepressant use,and then calculate welfare under various counterfactual policy scenarios. Specifi-cally, I am interested in determining whether the value of incremental innovationto patients exceeds the costs of extending market exclusivity.

1http://www.thaindian.com/newsportal/health1/allow-patents-for-incremental-innovation-of-medicines-study 100238527.html

2http://www.pharmaphorum.com/2011/03/18/battling-patent-expiration-by-building-brand-loyalty/

2

This study contributes to policy debates in the regulation of pharmaceuticals,healthcare, and intellectual property. Many have accused the pharmaceutical in-dustry of “evergreening, ” or extending patent protection through the introduc-tion of new products that represent minor advances over older drugs. An oft-citedexample is the case of Prilosec, a best-selling prescription heartburn drug. Justbefore the expiration of Prilosec’s patent in April 2002, the producer of Prilosec,AstraZeneca got approval for another drug, Nexium, in February 2001 and mar-keted it from March 2002. Nexium is one-half of the Prilosec molecule. Mostclinical studies found no increase in efficacy. However, AstraZeneca obtained an-other patent that will expire in November 2018. But others point to examples ofincremental innovation with important benefits. The anti-retroviral drug Norvir,a first-line treatment of HIV/AIDS, was approved by FDA in 1996 but requires re-frigeration. In 2010, a heat-stable version of Norvir was introduced. While this in-cremental advance may not be important for patients in developed country, it maymake treatment far more accessible to patients in countries without widespreadrefrigeration.

Even if incremental innovation yields real clinical benefits, the costs associ-ated with extending a monopoly – either through new patents or through Hatch-Waxman exclusivity extensions – may be significant. Many developed countriesare facing rising health care expenditure, a large portion of which is due to phar-maceuticals. Regulators use a variety of mechanisms to control pharmaceuticalexpenditure, including price controls and the promotion of generics. As ? hasemphasized, the introduction of new drugs may reduce total health expenditureand create other benefits, such as fewer inpatient visits and lost workdays due toillness. A full examination of these trade-offs is essential in assessing efforts tobalance cost, innovation and access. Market exclusivity extensions are anotherpolicy tool that allows regulators to balance incentives for incremental innovationthat increases welfare against the static gains from earlier generic entry ( ? ?; ?).

In the intellectual property area, there is a debate about the standards of patent-ability, particularly whether incremental innovation satisfies the novelty and nonob-viousness requirement of most patent systems. Some countries, including India,refuse to grant patents for some types of incremental innovation. Giving innova-tors the same period of patent protection on an incremental advance as on a newmolecule may distort incentives. Although this study doesn’t explicitly considerthe shift between incremental innovation and radical innovation, it might shedlight on the future research related to the substitution effects of these two types ofinnovation.

In this study, I focus on incremental innovations in selective serotonin reup-take inhibitor (SSRI) anti-depressant drugs, including the pediatric use of exist-ing drugs, which has been granted a six-month exclusivity extension over exist-ing patents, and two new formulations of existing branded drugs (Lexapro andPaxil CR), which have been granted new patents. I estimate the patients’ demandsfor antidepressants with a conditional logit model, based on individual-level pre-scription drug data. I then recover the marginal costs for the branded and genericfirms based on the Bertrand Nash model. Before calculating the welfare changefrom the status quo to counterfactual scenarios, I estimate counterfactual equilib-rium prices from the firms profit maximizing model, given the simulated demands

3

over counterfactual alternative choice sets (for example, incremental innovationproducts might be removed, or the generic entry might occur one year earlier dueto the withdrawal of market exclusivity). The change in consumer surplus is mea-sured using the compensating variation between two scenarios.

By comparing scenarios of either withdrawing or allowing market exclusivityfor incremental innovations with scenarios of withdrawing or retaining incremen-tal innovations, I found that the overall social benefits from all the incremental in-novations are overwhelmed by the welfare loss due to market exclusivity: the netwelfare loss amounts to $37 billion. Both consumers and generic firms are worseoff from granting market exclusivity to incremental innovation, each of which suf-fer a loss around 39.2 billions and 6.92 billions respectively; while the innovatorsgain $8.45 billion. When considering the removal of a single incremental innova-tion rather than all incremental innovation in SSRIs, the major results still hold:the consumer surplus loss from extending market exclusivity to incremental in-novation is much larger than the consumer benefits from incremental innovation.It suggests that current patent policies and data exclusivity regulations requirefurther examination.

Besides the important policy implications, this paper also contributes the IOliterature by providing a novel way to address the endogeneity of price. This isthe first study to use this method. Unlike the traditional way in which IVs are uti-lized in the second stage estimation to estimate the price coefficient (such as BLPmodel), I replace the market prices with predicted individual level copay prices.On the one hand, due to the individual level variation in the copay prices, theprice coefficient could be estimated consistently in the conditional logit model bycontrolling the product-year fixed effects3. On the other hand, replacing marketprices with individual copay prices results in a more accurate estimate for priceelasticity, as we know in the short run patients tend to respond to the copay pricesrather than the full market prices paid to the retailers. This method could bewidely applied to the demand estimation with individual-level data for healthcare markets where insurance coverage is widespread and the copay rate varies byindividuals.

This paper is structured as following: Section 2 explain market exclusivitypolicies for pharmaceuticals and the antidepressant market. Section 3 describesthe data and summary statistics. Section 4 details the empirical strategy. Section5 presents the estimation results, and Section 6 concludes.

2 Legislation Background and the Anti-depressantMarket

2.1 Background of Market Exclusivity

The pharmaceutical industry is an innovation-intensive sector, with firms invest-ing a very high proportion of annual sales into R&D; Additionally, their profitabil-

3Once controlling the product-year fixed effects, most of the unobserved product and market factorshave been captured and therefore there is little endogeneity

4

ity relies heavily on patent protection. This salient feature of the industry resultsfrom the characteristics of innovation and the production process of the pharma-ceuticals. The development of a successful branded drug is highly risky. To getapproval from government regulatory authorities such as FDA in the US, the me-dian time duration from the start of clinical trials to New Drug Application(NDA)issuance was 98.9 months. In each phase, the attrition rate is surprisingly high(around 25%, 52%, 36% for phase I, II, III, respectively, see ?), resulting in largecosts for clinical trials.

In comparison, the cost to copy a branded drug is remarkably low, not onlyfor production but also for approval from the government authority. Accpordingto the Hatch-Waxman Act passed in 1984, a generic producer could file an Ab-breviated New Drug Application(ANDA) based only on roof of bio-equivalence tothe branded drug as long as its data exclusivity has expired. Once generic drugsenter into the market, the sales of branded drugs tend to drop about 80% over thenext year. In this way, the length of patent protection governs the profitability andeffective life of pharmaceuticals.

Policy makers are continually trying to balance the need to develop new drugsand improving patients’ access to existing drugs while maintaining affordablehealth expenditures. By allowing generic producers to easily get approvals andmarket share, patients have access to blockbuster drugs that otherwise might notbe affordable. However, it greatly reduces the profits of innovators and thereforeundermines the incentives for them to carry out R&D for a new generation ofdrugs. The objective of government regulator is to develop an optimal policy thatmaximizes total social welfare while taking into account all these considerations.

Market exclusivity is influenced by a complex interaction of several factors (?).In short, it is determined by the complementary action of patents and data ex-clusivity. Patents granted by USPTO (The United States Patent and Trademark Of-fice) protect drug innovations from pure price competition, enabling drug innova-tors to recover research and development expenditures and thereby encouragingfurther research investment. Data exclusivity approved by the FDA restores thepatent period loss during the regulatory review and clinical trial period, as well asprotects clinical data from being utilized by generic competitor to file ANDAs4.

To illuminate the importance of data exclusivity, one should examine the so-phisticated nature of pharmaceutical innovation and the market regulatory sys-tem. To ensure the safety and efficacy of marketing drugs, the FDA sets a series ofregulatory requirements throughout the whole development process from the mo-ment a new chemical entity is synthesized until final FDA approval of a NDA. Theprocess is divided into three main stages: preclinical research, clinical investiga-tion, and NDA approval. It is well-known that the requirements are rigorous anddemanding. However, the patent term for pharmaceuticals remains unchanged,allowing 20 years after the filing date for patents applied for after 1995; and 17years after the issue date for patents applied for before 1995.

This means that the effective patent life of pharmaceuticals is substantiallyshorter due to the government requirements. When a firm discovers a promis-ing new chemical compound, the first thing to do is to file for a patent. From

4? discusses the importance of patent term restoration to pharmaceutical innovation.

5

that moment, the patents 20-year term is on the clock, well before knowing if thecompound can be developed into a marketable medicine. The government thenrequires substantial chemical, animal and human testing, followed with FDA re-view process of a NDA. The testing and approval process usually takes about 7 to13 years ( ?). Therefore, for pharmaceutical products, the 20-year patent term hasbecome a legislative figment. In reality, the effective patent life for pharmaceuti-cals has been eroded, and on average has only 6.8 years. The incentives to investin pharmaceutical R&D have been reduced substantially.

The Hatch-Waxman Act (1984) was enacted by Congress to alleviate this prob-lem by restoring market exclusivity to pharmaceutical innovations. In this act, twoseemingly contradictory goals have guided the federal government’s legislation:encouraging pioneer companies to continue developing innovative technologieswhile also making inexpensive generic pharmaceuticals available to consumers.

According to Hatch-Waxman Act, the generic manufacturer is allowed to usethe clinical data of the patented drug to prepare its own FDA application priorto expiration of the patent rights, thus providing an abbreviated process for FDAapproval of generic drug applications. The generic manufacturers can file AN-DAs without their own clinical trials data as long as they provide proof of bio-equivalence to the branded drug. This substantially facilitates the entry of genericdrugs (from 3 or more years to a few months, see ?). On the other hand, the Actprovides a partial restoration of the patent life of the research-based drug com-pany (the “pioneer”) by adding 5 years of data exclusivity without surpass themaximum effective patent life of 14 years. The 5-year exclusivity is the periodin which generic firms are forbidden from filing any ANDAs based on pioneer’sclinical trial data. Besides the NCE exclusivity, Hatch-Waxman Act also providesother terms of data exclusivity:

1. Three years of exclusivity granted for a change in an approved drug product.The changes include new indications, dosage strength, dosage form, routeof administration, patient population, and conditions of use. The changesrequire new clinical investigations and the exclusivity could prevent effectiveapproval (but not submission) of ANDAs.

2. Seven years of orphan-drug exclusivity. A company that develop such a drug(to treat a rare medical condition which implies the condition affects fewerthan 200, 000 people in the US) could be protected from competition forseven years.

3. Six months of pediatric exclusivity. Unlike other exclusivity, the pediatricexclusivity could be attached to existing periods of exclusivity and patentprotection. It can be extended to all approved formulations, dosage formsand indications for products that contain the same active ingredient so longas they are protected by an exclusivity or patent. More than one period ofpediatric exclusivity is possible, e.g., new indication.

Within these three exclusivity periods, generic firms could file ANDAs, butthey cannot get approval from the FDA. Only after the expiration of the dataexclusivity, their generic products could be approved. However, only when thepatent expires, the approved generic ANDAs could be marketed.

6

In this sense, data exclusivity adds a new hurdle for the generic firms to enterthe market. Generic firms could also file and get approval of new drug applicationbased on their own safety and efficacy data, however, that implies they shouldcarry out their own clinical trial, and it introduces an immense amount of costsand uncertainty.

For pharmaceutical manufacturers, whether innovator or generic, their objec-tive is to exploit every market potential and extract maximum returns. For the in-novator, they are protecting their product from competition, restoring their patentterms and obtaining as long a data exclusivity as possible. However, for the genericfirm, they are trying every shot to challenge current patents, and apply for AN-DAs as early as possible to capitalize on market share. According to the Act, thefirst-to-file ANDA generic can obtain a 180-day exclusivity in which FDA will notapprove a subsequently filed ANDA for the same product. Therefore, the compe-tition of the branded and the generic passed on to the arena of patent and dataexclusivity. As quoted at the beginning of our paper, the incremental innovationis one of the important strategies for the innovator to win in this battle.

2.2 The Anti-depressant industry and SSRIs

In this paper, we focus on the antidepressant industry, more specifically, the Se-lective Serotonin Reuptake Inhibitor antidepressants (SSRIs). The reason to con-centrate on this market lies in the following: molecules in this class are relativelyhomogeneous with individually idiosyncratic therapeutic responses and side ef-fects. Therefore, it is a differentiated but mutually exclusive market. The prod-ucts with patent protection, incremental innovations, and generic entry coexist. Inthis sense, the market structure forms the indispensable ingredients (incrementalinnovation, market exclusivity, consumer loss without generic entry) for answer-ing our question. Market is competitive with balanced share across brands, whichfacilitates the demand estimation.

The antidepressant manufacturing industry is the largest prescription industryin the United States, with more than 200 million prescriptions dispensed annuallyfor 2007-2011 (IMS Health).5 Trends will likely continue upward with the risingdiagnosis rates and increasing public awareness of the disease. In fact, the WorldHealth Organization forecasts that unipolar major depression will displace heartdisease as the heaviest disease burden by 2020(See ?). Furthermore, anxiety disor-ders affect about 40 million American adults in any given year. They include panicdisorder, obsessive-compulsive disorder, post-traumatic stress disorder, general-ized anxiety disorder, and phobias. For these anxiety disorders, antidepressantsare standard treatment.

Antidepressants are almost equal in their successes in relieving depression.The differences determining patient choices, come down to safety, effectiveness,costs, side effects and the presence of other medical conditions that could affectthe drug’s safety and effectiveness. Among them, side effects are especially cru-cial considerations. Common antidepressant side effects are nausea, weight gain,

5See the Top-line Market Data “Top Therapeutic Classes by U.S. Dispensed Prescriptions” inwww.imshealth.com.

7

sexual dysfunction, and anxiety. Moreover, antidepressants are addictive, and dis-continuing use may cause withdrawal symptoms in patients. Reducing these sideeffects is a vital area of research for pharmaceutical companies.

The diagnosis of depression involves more subjective criteria than other com-mon diseases such as arthritis, cancer, or diabetes. As a result, Direct-To-Consumer(DTC) advertising has historically played a large role in creating popular recog-nition of depressive symptoms, “growing market” and fostering a demand forspecific medications. Studies show that only half of people with depression aretreated (See ?). Thus, the success of new antidepressant treatments hinges on theability of manufacturers to effectively market their products to the public.

According to action mechanisms, the antidepressants can be categorized intoseveral therapeutic subdivisions, 6 in which SSRIs are the most commonly used asthe first-line treatment of depression because of their favorable side effect profileand low toxicity. SSRIs also amount to the top prescribed antidepressants in theUS retail market in 2010.7 They work by preventing the reuptake of serotoninby the presynaptic neuron, thus maintaining a higher level of serotonin in thesynapse. This allows the brain to better transmit signals, thus improving mood.

Prozac was the first SSRI marketed by Eli Lilly with FDA approval in 1987,later followed with other SSRIs including Zoloft (1991), Paxil (1992), Celexa (1998),Lexapro (2002) and updated versions, such as Prozac Weekly, Paxil CR etc.. Afterthe added six-month of data exclusivity for Prozac due to its pediatric studies ex-pires in Aug. 2001, the generic competitors enter. Generic counterparts for Paxil,Celexa and Zoloft then entered into the market in 2003, 2004 and 2006. The mostrecent SSRI Lexapro also began to face its generic competitor after Mar. 2012.

Unlike most markets where consumers hold the full discretion in making choices,demand for pharmaceuticals relies not only on ultimate patients’ tastes (the effi-cacy and side effects response, brand loyalty), but also on the behaviors of physi-cians who prescribe these drugs and pharmacists who dispense the prescriptions.Fortunately, unlike other therapeutic classes of drugs8, anti-depressants leave lit-tle scope to physicians when they decide what to prescribe. The efficacy and side-effect responses of anti-depressants are quite idiosyncratic across individuals. Thetraditional way to prescribe anti-depressants is to try one or two molecules to seewhich works with the fewest adverse reactions. According to patients’ responses,doctors then write the prescription afterwards. In addition, people usually be-lieve that physicians have incomplete or no information about relative prices (?),however, for the treatment of chronic disease, exceptions happen (?). Therefore,in our study, agency problems play little role in choice decisions, which justifiesthe rationale of applying discrete choice model in our problem.

6The antidepressants mainly include the monoamine oxidase inhibitors (MAOIs), tricyclic antide-pressants (TCAs), tetracyclic antidepressants (TeCAs), selective serotonin reuptake inhibitors (SSRIs),and Serotonin-norepinephrine reuptake inhibitors (SNRIs).

7From the statistics in “2010 top 200 generic drugs by total prescriptions” and “2010 top 200branded drugs by total prescriptions” provided by www.drugtopics.com.

8Such as several chemically distinct but similarly working H2 antagonists used to treat duodenalulcers; Several ACE inhibitors used in the treatment of hypertension; And also numerous chemicallydistinct antibiotics.

8

3 Data

3.1 Data Sources

Three data sources are employed for this study: the consolidated individual data,medical condition data, and prescribed medicine event data from the Medical Ex-penditure Panel Survey (MEPS) from 1996 to 2011, the information about newdrug application and patent/data exclusivity for drugs from the U.S. Food andDrug Administration (FDA), and self-collected data about the drug characteristicsfrom package insert labels.

MEPS provides nationally representative estimates of health care uses, expen-ditures, sources of payment, and health insurance coverage for the U.S. civiliannon-institutionalized population. It follows all the individuals in randomly se-lected US families for 2 years with 5 rounds of questions, information recordedincludes respondents’ health status, demographic and socio-economic character-istics, health insurance, medical expenditure, etc.. More importantly, it supple-ments the survey data by contacting medical providers and pharmacies to acquiredetailed and accurate consumption and billing information. For example, if thepatient reports purchasing Norvir from a pharmacy, the pharmacy is contactedand required to provide the purchasing history of Norvir for the patient. Overall,there are 276,702 individuals including in the survey from 1996 to 2011.

Two datasets from the FDA (Drug@FDA and Orange Book) provide exhaustiveinformation on the drug approvals, supplemental approvals, patent, and data ex-clusivity. For each approval, whether it is a radical innovation (new molecule en-tity) or incremental innovation (new formulation, new indication, and new com-bination) is documented. The subsequent supplemental approvals (including thesafety, efficacy, new indication, new strength, new formulation, label change, pa-tient population change) following each approved drug could be tracked in theFDA database. The information is critically indispensable for our study to iden-tify incremental innovation from radical innovation.

Moreover, the drug characteristics (the strength, dosage form, active ingre-dient, producer, approval date, patent expiration date, data exclusivity expira-tion date, type of data exclusivity, etc.) are all available from the FDA. Based onthis information, we could generate several important drug characteristics such asdummy of generics and age of molecule for each drug. The only flaws is that itonly provides the unexpired data exclusivity information, i.e., once the data ex-clusivity expires, the information about expiration date, exclusivity type becomesabsent from the available dataset. Fortunately, we could infer the expiration dateof the expired drugs from the entry of generic competitors (ANDAs approval ormarketing date). As a complementary source, USPTO provides more detailed in-formation on the patent and their expiration. We use National Drug Code Direc-tory (NDC) data to link the MEPS prescibed medicine data with FDA drug data.

A unique self-collected data is utilized in our study. By reviewing the packageinsert label for each drug, we construct several variables of drug attributes. Thesevariables mainly lie in two aspects: the indications and side effects9. Indications

9There is also information about FDA safety alarms to health professionals and patients about drugs.The alarms usually happen several years later after the approvals, when the patients utilization of the

9

are noted in each label and remain the same until the supplement approval hasbeen granted for new indication. We generate several dummies of indications foreach drug. We also collect the side effects information for each drug from theclinical trial data. Due to the heterogeneity of the clinical trial across drugs inthe sense of the composition of the patients, the drug strength, the length of theclinical studies, etc., the original statistics is not comparable across drugs. To makethe statistics meaningful across drugs, we generate variables for each symptom:the ratio of the occurrence rate of the symptom for the patients taking drugs overthe occurrence rate for the patients taking placebo10.

3.2 Products and Incremental Innovations

Our focus is on four active ingredients in selective serotonin reuptake inhibitor(SSRI) anti-depressant drugs, denoted in ATC code as N06AB-. 11 It involves fournew molecule entities following with 10 incremental innovations and subsequentgenerics (See Table 1). 12 Prozac Weekly, Sarafem, as incremental innovations ofProzac are not included in this study due to their limited number of observations.

As shown in Table 1, following the marketing of four new molecule entities:Celexa (1998), Paxil (1992), Prozac (1987), Zoloft (1991) 13, subsequent improvedgenerations are innovated. In our paper, to consider the incremental innovation,we mainly focus on the pediatric use of SSRIs and newly formulated drugs withnew drug application to the FDA, i.e., Lexapro and Paxil CR. New indication useswill not be considered in this paper because the data exclusivity granted to thenew indications in our study fell within the period of patent protection and didnot extend the market exclusivity. Our results, as demonstrated below, illustratethat the utility from new indications is not significant and also ignorable. 14

If we define “product” as the drug with constant characteristics, then an up-dated version implies a new product; therefore, overall there are 12 products plusan outside option in our analysis. The indications for each drug are listed in Table

drug reveals more safety problems. For this category (SSRI), all the drugs have been received the sameFDA alarm and therefore, we cannot identify this variable without variation and therefore we didn’tinclude it into our study.

10The details to construct side effect variables are illustrated in Appendix A.2.11According to the WHO ATC index, there are nine substances which have ever been approved as

medicines. (See http://www.whocc.no/atc ddd index/ and search N06AB.) However, three of themhave been withdrawn due to safety problem, leaving only 6 molecules in the US market: citalopramhydrobromide, escitalopram oxalate, fluvoxamine maleate, fluoxetine hydrochloride, paroxetine hy-drochloride, and sertraline hydrochloride. Among them, fluvoxamine maleate is omitted in our analy-sis because of its limited market share. Finally we only focus on the remaining five active ingredients.

12Escitalopram oxalate is basically a new formulation of citalopram hydrobromide, and therefore, wewill treat it as an incremental innovation instead of a new molecule entity.

13The numbers in the parenthesis indicate the years of first approval.14The market exclusivity granted to new indication or new formulation only protects the updated

drug from generic competition, however, the exclusivity for original new molecule entities will expireas usual. We can see from Table 1 that, while the marketing of new version of branded drugs, the gener-ics entered as well. Nevertheless, even with market exclusivity extension on the new indication, thedrug producer would generally need a new dosing strength or formulation to make this commerciallyreasonable since doctors could, and would, write prescriptions for the old version for the new indica-tion. In this sense, it can be expected that the generics could obtain the new indication characteristicsas the branded one as long as the drug strength or formulation are the same.

10

10 and the side effects for each molecule are also shown in Table 11. 15 Unlike mostof the literature which treat drugs in different formulation and different strengthas different products, in my study, I treat them as one product, since most de-pression drugs are administered once daily, I use weighted price across strengthand formulation as the product prices.16 Treating these simialar drugs as differentproducts will make the estimation computationally burdensome without gainingbenefits.

Products from branded firms and generic producers are distinguished. But wefind it from the Table 11 and 10 that the characteristics of the branded drug andthe generic one are similar in terms of indications and side effects because of thecurrent legislation (Hatch-Waxman Act: Generic drugs can provide only the proofof bio-equivalence to branded drugs to get abbreviated new drug approval fromFDA without carrying on clinical trials but building their safety and efficacy onthe clinical data of the branded counterpart). As shown in Table 11, side effectsdata remain the same for all drugs with the same molecule across years.

The market exclusivity granted to the incremental innovations in SSRIs in-cludes:17

• Prozac was approved for additional six month pediatric exclusivity for itspediatric use which extend their market exclusivity from Feb. 2001 to Aug2001;

• Paxil obtained 6-month exclusivity for its pediatric studies extend its marketexclusivity from Sep. 2015 to Mar. 2016;

• Zoloft obtained 6-month exclusivity for its pediatric studies extend its mar-ket exclusivity from Dec. 2005 to June 2006;

• Lexapro obtained 6-month exclusivity for its pediatric studies extend itsmarket exclusivity from Sep. 2011 to Mar. 2012.18

3.3 Sample Selection

As shown in Table 2 and 4, among the 276,702 respondents-year observations inMEPS, only those with depression (48,317) 19 20 and those who has no depression

15Side effects data keep the same for all drugs with the same molecule which is obtained from theclinical data of branded drug. After Hatch-Waxman Act in 1984, generics could enter into the marketbased on the clinical data of branded counter-parts with only providing the bio-equivalence to thebranded one. Therefore, we could expect that the side-effects of the generics remain the same as thebranded one.

16Alternatively, the average prices of the recommended initial dosing per day are also considered.The results are not sensitive to these two measures of average market prices.

17The data exclusivity granted to incremental innovations in our market includes only those for pedi-atric studies. Three years of data exclusivity granted to the firms for new indications were overlappedwith patent protection and therefore they have played no role in extending market exclusivity.

18We will not investigate the withdrawal of this exclusivity since the time span is out of our datascope.

19According to the ICD-9-CM diagnosis codes, if the respondent has condition with code 296(Episodic mood disorders), 300 (Anxiety, dissociative and somatoform disorders) or 311 (Depressivedisorder not elsewhere classified), then we regarded them as having depression.

20The conditions are recorded in MEPS normally for the following reasons: 1. reported by the house-hold respondent for a particular medical event (hospital stay, outpatient visit, emergency room visit,

11

but purchased SSRI anti-depressants (2,749) are included in our analysis: 22,203(43.48%) of the whole sample have ever purchased SSRI drugs. The rest of thepatient-year records (28,863, 56.52%), who were diagnoised as having depressionin the year but did not purchase SSRI drugs, either had never purchased the drugsor purchased other anti-depressants. The individuals who purchased SSRIs con-stitute the demand of these products, and the patients who had condition butdidn’t purchased SSRI drugs make up the potential market size, we regard themas choosing the outside option.21

As we know, depression is a chronic disease which requires long-term medi-cation treatments. Therefore, refills and repeated purchasing are very common inthe survey. From MEPS, 117,100 purchasing events for the 14,555 individuals aredocumented. 79,485 observations are refills and 15,412 are repeated purchasingwithin the year, which are dropped from our sample and left with 22,203 obser-vations. Among them, 1,410 patients were switching drugs within the year andresulted in 3,045 purchasing records. To better exploit the information from thedata, we keep these switching choices. For most of the observation, we keep themat the individual-year level. Remind that MEPS normally follows respondents fortwo years, which implies that in our analysis sample, we might have recordedthe choices of the same individual in the two consecutive year. As the individualhealth conditions, access to health care (such as insurance coverage, better infor-mation, etc.) and the market conditions may change across time, we believe thatwe could capture more information in this way. Therefore, finally there are 22,203purchases plus 28,863 records of outside option at the patient-year level kept inour study. As evidenced in Table 3, the deletion of refills and repeated purchasinghas little effect on our demand estimation since the market share remains more orless the same between the MEPS survey sample and our analytic data.

3.4 Variables

The dependent variable in this paper is defined as the treatment choice dummyfor each patient in a period. The alternative choices include all the available drugsproducts listed in Table 1 in the year and the outside option. 22

The details of data construction for the individual characteristics and drugattributes are described in Appendix 1. The individual demographics that we em-ployed include Ageit, Maleit, Adultit 23. The socio-economic variables includeY ears of educationit and Family incomeit. Family income is deflated in 1996 dol-lars. For children and adolescents, I use their parents’ education as the educationlevel since we want to see whether education helps people obtain better health

home health episode, prescribed medication purchase, or medical provider visit); 2. reported as thereason for one or more episodes of disability days. 3. Reported by the household level respondent as acondition ”bothering” the person during the reference period.

21There is another potential group of patients unobserved in our sample due to their mild conditionfor which they never sought treatment, and are not reported in the survey. Therefore, we cannot includethem. We believe this approach is reasonable as depression is the type of condition that only becomesa disease when it disturbs the mood of the patients and severely affects their daily life.

22If an individual purchase the drugs, he is assumed to be taking the medicine, i.e., choiceit = 1.Patients’ compliance to the medication isn’t considered in our context.

23We define Adultit = 1 if Ageit ≥ 18

12

service.The health-related variables are composed of two parts: health insurance vari-

ables and health status variables. As we all know, besides drug price, the insurancecoverage plays an important role in the patients choice making as well. Accord-ing to the patients’ insurance coverage, we defined several insurance dummies:Having Medicareit,Having Medicaidit, P rivate insuranceit andOther public insuranceit.If the patient were covered by any type of health insurance in the year, we treatthem as Having medical insuranceit. Even after we control the insurance type,it is still unsatisfactory to fully capture the insurance effects as we know that foreach insurance type, the generosity for each type of medical services (e.g., hospi-talization, prescription drugs) varies across plans. Fortunately, even if the detailedinsurance coverage for prescription drugs is not observable in the survey, howeverwe observe both the out-of-pocket prices and the full prices for each purchases inMEPS. Therefore, we are able to construct Out of pocket rateit for each individ-uals based on their purchasing history. 24 The health status variables include asubjective perceived health status variable P erceived healthit and some diagnosedhealth conditions indicators. We consider six major diseases: Respiratory diseases,asthma, hypertension, cardiovascular health diseases, diabetes, and depression.

The drug attribute variables mainly include P riceit,Age of Moleculeit,Genericit,P ediatric Useit, indication dummies and side effect variables25. P riceit is deflatedto 1996 price level using CPI in Managed Care Commodities category26. Age ofMoleculeit is the same for both the branded and the generics in the same year. Itmeasures the drug age since the market approval of the new molecule entity tillthe year when purchasing happens. 27

SSRIs are primarily used to treat major depressive disorder (MDD), besidesthat, they can also be used to treat obsessive compulsive disorder (OCD), panicdisorder (PD), post-traumatic stress disorder (PSD), premenstrual dysphoric dis-order (PDD), social anxiety disorder (SAD), bulimia nervosa (BN). As we can seein Table 10, when a drug is initially developed, its clinical trial data only supportsthe safety and efficacy of one or two indications. As time goes by, the manufac-turers carried on subsequent investigation and then more and more indications issupported after the supplemental approval by FDA. This process is one type ofincremental innovation that we are interested in.

4 Estimation Strategy

The empirical estimation in this paper consists of three steps: First, with con-ditional logit model, I estimate the individual preferences over drug attributes

24For individuals who have multiple purchases in the year, we use the average out of pocket rate.And for individuals who never purchase in the year, we impute it with the average out of pocket rateof the sample.

25The detailed discussion about indication dummies and side effect variables could be found in theAppendix 1 and they are listed in Table 10 and Table 11.

26CPI source: US Bureau of Labor Statistics.27A dummy variable outside is generated to denote whether the individual chooses SSRIs (outside =

0) or outside option (outside = 1). The variable is introduced to facilitate the estimation for outsideoption.

13

(branded/generics, drug age, indications and side effects) allowing it varying withobserved individual characteristics. The estimated parameters enable us to con-struct the demand curves confronting each drug products. Second, with the es-timated demand curves, a supply side estimation is implemented by assuming aBertrand-Nash equilibrium to back up the marginal costs for each firms acrossyears. Third, in the counterfactual scenarios that we remove either market exclu-sivity or both ME and II, virtual demand function are modeled with the estimateddemand parameters, and then the optimal price decisions are made for each coun-terfactual scenario assuming that firms are profit-optimizing. In the third step, weassume that the marginal costs remain unchanged across different market scenar-ios, therefore, for each drug producer confronting the demand curves as a functionof undetermined price, their optimal choices are to set the prices that maximizethe final profits. In the following, we are going to elaborate the detailed estimationstrategies in each step.

4.1 Demand Side Estimation

The novelty of the method utilized in this paper appears to be that the price en-dogeneity problem is addressed by replacing market prices with individual levelcopay prices for which the coefficient is identifiable even after we control for theunobserved product level factors with drug-year fixed effects in the demand sidediscrete choice models. This framework applies to both the conditional logit andrandom coefficient logit model with individual level data.

The intuition is below: in the regular BLP model, once we include the product-market level fixed effects in the individual’s indirect utility function, the estimatesfrom the choice model are deemed to be consistent as the unobserved factors thatmight be correlated with the product level attributes have been captured by theproduct-market fixed effects. The reason that we need to address the endogeneityproblem in the BLP model lies in that the product prices are invariant within mar-ket and therefore unidentifiable. Therefore, in the second stage estimation, treat-ing the estimated product means utility as dependent variables, we estimate theproduct price coefficient with IV. With the help of fruitful individual level infor-mation from our dataset and the nature of pharmaceutical purchase, I argue thatthe prices that matter for the patients are the copay prices rather than the averagemarket prices. Since different patients have access to different insurance cover-age and therefore different out of pocket price, by replacing the average marketprices with patient copay prices varying across individuals, the price coefficient isidentifiable in our model in the first stage.

In the following, we are going to discuss the estimation framework on the de-mand side. Similar to the BLP model, we consider that a utility-maximizing pa-tient i, where i = 1, . . . , It , in a given time period t, where t = 1, . . . ,T , faces Jt + 1alternatives: Jt different antidepressant drugs and the option of not purchasingany of the drugs, the outside option, j = 0. As in ?? and ?, the indirect utility ofany patient i, in market t choosing drug j can be expressed as

uijt = βiptpcijt +

∑k

xjktβikt + ξjt + εijt .

14

withβict = βuc +

∑r

zirtβocr , where c = p,k. (1)

The payoff of patient i purchasing drug j at date t depends on the drug pricepcijt(copay price) and drug characteristics, including xjkt (observable) and ξjt (un-observable). We allows patients to have heterogeneous preferences over observeddrug attributes xjkt and drug price pcijt through the coefficients βict. The coeffi-cients vary with observable patient characteristics zirt, indexed r = 1, · · · ,R. Fornotational simplicity, in the following we omit the market notation t.

By grouping the drug-specific common term together, we get:

uij = δj + βuppcij +

∑r

pcijzirβopr +

∑kr

xjkzirβokr + εij , (2)

with δj =∑k xjkβ

uk + ξj , we call it as the mean utility component of drug j, which

is common to all patients in market t.Similarly, the indirect utility the individual has by choosing outside option

(i.e., not to purchase any of these drugs) is expressed as

ui0 = δ0 +∑r

zirβ1r + εi01.

where zir for r = 1, · · · ,R, is a vector of observable individual characteristics. Weinteract the payoff provided by selecting the outside option with observable in-dividual characteristics to allow for heterogeneity in choices to buy or not. Theparameter δ0 capture the baseline payoff from the outside option and for each r,β1r captures the variation in payoffs across individuals due to zir .

Individual i chooses drug j which gives him/her the highest payoff, i.e., uij >uil ,∀l , j, l ∈ {0, · · · , J}. Then, by assuming the random utility component εij tobe independent and identically distributed across both drugs and patients andfollows an extreme value distribution, we can write the probability of individual ichoosing drug j as

P ri(j | z,x,δ,β) =exp(δj + βupp

cij +

∑r p

cijzirβ

opr +

∑kr xjkzirβ

okr )∑J

l=0 exp(δl + βuppcil +∑r p

cilzirβ

opr +

∑kr xlkzirβ

okr ). (3)

With conditional logit model28 , the estimates for parameters δj , βup , βopr , β

okr could

be obtained. Additionally, we need to identify what drug attributes determine themean utility for drug j in year t, δj . Because only in that way, we could constructthe mean utility of drug j in year t′ for which year drug j is originally not availablein status quo. The OLS model could be described as:

δjt = λ+ Γ · xjt + yt +mj + ζjt (4)

28The reason that I did not adopt the random coefficient model lies in the fact that the randomcoefficient model with individual-level data is computationally burdensome. Although the randomcoefficient model allows more flexible fit of the model with the observed data, however, from the resultsof random coefficient model and conditional logit model, we can see the gains is little. Therefore, forthe main estimation and counterfactual analysis, I will adhere to conditional logit model.

15

The drug mean utility (δjt) depends on the drug characteristics (xjt), year fixed ef-fects (yt) and molecule fixed effects(mj ). By plugging equation (4) and (6) into(3), the patient demands are pinned down by drug attributes (such as marketprices, drug age, generic, pediatric use etc.), individual characteristics (e.g., in-come, health status, health insurance), molecule FE and year FE. These estimatedcoefficients define the individual preferences and are critical in constructing thecounterfactual demands contingent on different drug attributes and individualcharacteristics.

4.1.1 Drug Prices and Copay

Different from other industry where consumers directly respond to the marketprices, in the health care industry, because of wide-spread insurance coverage,patients tend to be less responsive to the market prices of medical products andservices. In that sense, higher market prices may not lead to lower demands, andtherefore, treating market listed prices as the major determinant of patents’ choicemaking may resulted in biased estimates of price coefficients. ? and ? have shownthat the implementation of Medicare Part D improved overall drug utilization andgenerics use; ? has found that Medicare Part D coverage gap was associated withmodest reductions in the use of antidepressants and the coverage for generic drugsreduced the brand-name antidepressant prescriptions. These studies suggest thatconsumers did respond to the prescription drug insurance and make differentchoices due to the varied copay rate.

In recent years, researchers have realized this issue when uncovering phar-maceutical price elasticity. And they tend to replace market prices with copayprices in the demand estimation (??). They estimated market-level copay pricesby assuming a non-linear relationship with market listed prices. Due to the datalimitation, under their framework, the copay prices are invariant within marketacross patients. Different from their work, I am going to replace market priceswith copay prices paid by each individual. As we could see from Figure 6, withthe individual-year level sample, the out-of pocket rates heterogeneously variesfrom 0 to 1 with an average at .48. Among which, 22% of patients paid the fullprices, 10% of patients paid nothing, and the payment from the rest of patientsalmost uniformly distributed between 0 to 1 with a slightly emphasis on lowerrate. Therefore, it makes sense to estimate the copay prices at the individual levelas we described below.

From MEPS, we could observe the copay prices for the alternatives the patientschose, but not the copay prices for the alternatives that patients did not choose.However, for the purpose of demand side estimation, it is prerequisite to knoweach individual’s copay price for each alternative before evaluating their indirectutilities from different options. On the other hand, due to the heterogeneity ofthe insurance coverage across individuals and drugs, simply inferring the out ofpocket prices for one drugs from another brands or molecule might be mislead-ing.29 Therefore, estimating copay prices based on the comprehensive information

29Different individuals may paid different prices out of pocket for the same drug in the same market;The same individual may pay different drugs with totally different copay rate. To promote genericsubstitution, insurers tend to set lower copay rate for the generics, and higher copay rate for the most

16

from both the individual characteristics and drug attributes become more appro-priate. That’s to say, the copay price of drug j for individual i at time t, pcijt maydepends on the market listed prices of drug j, the out of pocket rate observed fromindividual i’s other purchases, the individual’s income, health status and healthinsurance dummies, the drug molecule fixed effects, i.e.,

pcijt = βpLog(Market P ricesjt) +α1Out of P ocket Rateit +α2Log(Family Incomeit)

+ Θ1 ·Diagnosed Disease Dummiesit +Θ2 · Insurance Category Dummiesit+ Θ3 ·Molecule Dummiesjt +Θ4 ·Molecule Dummiesjt ∗Genericjt +Y earFE + ηijt ,

(5)where Θ1, · · · ,Θ4 are parameter vectors.

Market listed prices are the basis of copay prices, and the relationship is as-sumed to be non-linear (therefore we take log on the market prices). Out of pocketrate of purchases for other drug products could in some extent shed light on thegenerosity of the insurance plans. The individuals with higher income tend to pur-chase more generous insurance plans; while on the other hand, they may not careabout the financial risks of the medical expenditure and therefore purchase littleinsurance coverage. Patients with severe health diseases tend to purchase gener-ous insurance provisions and the retailers or insurance payers may have differentdiscounts for different molecules or different producers (branded v.s. generics) forthe same molecule. Year fixed effects is controlled to capture the time trend.

With the estimated coefficients vector α̂1, α̂2, Θ̂1, · · · , Θ̂4 and β̂p, the copayprice for each individual i and drug j at time t, pcijt, could be predicted as

p̂cijt = β̂plog(Market P ricesjt) + α̂1Out of P ocket Rateit + α̂2Log(Family Incomeit)

+ Θ̂1 ·Diagnosed Disease Dummiesit + Θ̂2 · Insurance Category Dummiesit+ Θ̂3 ·Molecule Dummiesjt + Θ̂4 ·Molecule Dummiesjt ∗Genericjt +Y earFE

(6)One advantage of this specification to predict copay prices lies in that with the

estimated parameters it could consistently estimate the copay prices for the imag-inary incremental innovated drugs or actually non-existent generics in the coun-terfactual scenarios since they inherit the same molecule and branded /genericnature as their precedent generations. 30

4.2 Supply Side Estimation

Market is a complicated system in that any single change in one factor would trig-ger the responses from other parts of the system and results in a totally differentequilibrium. When considering the effects of incremental innovation, newly im-proved features on the existing drugs may reduce the patients’ price elasticity, theproducer may find it to be profitable to charge a higher price due to the reduced

innovative branded drugs.30For example, if we would like to estimate the copay prices for generic Paxil CR in counterfactual

scenarios, as generic Paxil CR has the same molecule with generic Paxil, and similar to generic Paxil,it’s also generic. Therefore, generic Paxil CR and generic Paxil would share a similar component in thecopay prices.

17

elasticity; the other competitors may response accordingly due to the altered sub-stitution between their product and the innovated one. Therefore, to measure thecounterfactual outcomes, it is inevitable to estimate the firms’ pricing under newmarket conditions (e.g., new available alternatives, new patients’ demands). Butbefore that, we first need to back up the supply side primitives, i.e., the marginalcosts for each product.

That’s to say, to evaluate the welfare effects of removing the market exclusivityand incremental innovations, we have to accomplish two estimations in steps onthe supply side:

• First, to uncover the cost structure of the producers;

• Second, to estimate the optimal price decisions in counterfactual scenariosbased on the estimated marginal costs.

The frameworks used to conduct these two estimations are essentially the same.Following the traditional approach in the literature and assuming that firms con-front oligopolistic competition with differentiated products, myopic firms choosesthe optimal prices to maximize their instant total profits, which are representedby the total revenues net the variable costs. The framework is formalized as below.At each period t, assume firm f produces subset Jt of the J total products31. It’sprofit function could be expressed as:

Πf =∑j∈Jf

(pj −mcj )Dj (7)

whereDj =∑i P ri(j | z,x,δ,β), is the sum of all the individual’s estimated demands

for product j . Traditionally, we assume that the marginal cost mcjt is constant forthe product j at time t. Firms’ optimization assumption requires that the firstorder condition must hold. That’s to say:

F.O.C. : Dj +∑l∈Jf

(pl −mcl)∂Dl∂pj

= 0. (8)

Rewrite it in matrix form, we have:

p −mc = ∆(p,X,β)−1D(p,X,β), (9)

where the (j, l) element of ∆ is

∆j,l = {−∂Dl∂pj

if j and l are produced by the same firm;

0 otherwise.

As the first step, to uncover the cost structure of the firms, based on equation (9),with observed average market prices, p and estimated consumer demands, D(·),

31In the SSRIs market, most of the firms have only one product, that’s to say, Jt = 1. The exceptionsinclude Forest Labs (producer of Celexa and Lexapro) and Glaxosmithkline (producer of Paxil and PaxilCR). Before the patent expiration of Celexa, Forest Labs developed a new transforms of CitalopramHydrobromide (active ingredient of Celexa): Escitaloproam Oxalate. When pricing Celexa (Lexapro),Forest Labs should consider the substitive effects of Lexapro to Celexa.

18

the marginal costs could be backed up.Subsequently, to estimate the counterfactual prices, we simulate the consumer

demands for the counterfactual scenarios as a function of market prices32, andplug them into equation (9) to simulate the equilibrium prices .

One thing should be noted here is that we are unable to identify drug produc-ers for each generics purchase.33 For example, in 2007, there are two incumbentsand 17 entries for generic Zoloft, and 229 purchasing records for generic Zoloft(sertraline hydrochloride). We couldn’t tell which producer supplies the purchaseof individual A in data. To solve this issue, taking advantage of the fact that gener-ics for one brand are homogenous products and producers are usually indifferentfor the patients, I allow all generic competitors for one molecule equally dividethe market share (For example, in 2007, each generic competitor receives 229/19demands.) and randomly assign patients who purchased this generic drug to themas patients don’t care much about the name of generic producers when they decideto choose generics over brands.

4.3 The Counter-factual Scenarios and Welfare Analysis

To evaluate whether and how much the patients would benefit from the removalof market exclusivity, we need to compare the status quo (the real world) withthe scenarios in which the regulators no longer reward incremental innovationwith market exclusivity. Without the financial incentives from the exclusivity, onenatural response from the innovator side is to reduce supply of the incrementalinnovations as the reward from devoting into R& D for incremental innovationdeclines without the patent or data exclusivity protection. Firms could save theresources for other type of R& D (e.g., radical innovation) or some marketing ac-tivities (e.g., advertising and promotion). The extent to which the firms wouldreduce their incremental innovation is unknown and current dataset does not al-lowed us to exploit this issue. Therefore, considering the extreme case, we assumethat there would be no supply of incremental innovations given no market exclu-sivity. The assumption allows us to circumvent the issue about the responsivenessof innovation activities to the exclusivity policy provision and also defines the up-per bound of the benefits from granting exclusivity to incremental innovations.Therefore, the welfare effects we are interested in could be expressed as below:

∆ ≡W (II,ME)−W (no II,no ME), (10)

where II denotes Incremental Innovations,ME denotes Market Exclusivity,W (II,ME)implies the social welfare when we allow granting exclusivity to incremental in-novations, and (no II , no ME) denotes the scenario when regulators do not grantmarket exclusivity to incremental innovations. The social welfare includes both

32In the next section, we will discuss in more detail about how to conduct counterfactual analysis.33The National Drug Code information in MEPS used to link the FDA drug producer data and MEPS

purchase data has lots of missings. Based on the drug names recorded in MEPS, we could identifywhether the drug is branded or generic. And each branded drug is produced only by its innovator inSSRI market. While for the generic competitors, once market exclusivity expired, numerous genericcompetitors poured into the market (See Table 13). In this sense, we could identify the purchase for thebranded innovators but not for generic producers.

19

consumer surplus (CS) and firms’ profits (Ps), therefore ∆ ≡ ∆CS +∆P s.

4.3.1 Calculating the Changes of Firms’ Profits and Consumer Surplus

Among the two parts of the social welfare, the firms’ profits are relatively straight-forward to calculate. As the fixed costs don’t varies with production level, there-fore, the firms’ profit changes could be easily calculated by the changes of totalrevenues net the variable costs for all the producers j and year t, i.e.,

∆P s =∑jt

(P bjt −mcjt) ·Dbjt −

∑jt

(P ajt −mcjt) ·Dajt (11)

where the letter a, b denote different scenarios.On the consumer side, we measure changes in consumer welfare by the com-

pensating variation (CV), defined as the additional expenditure that consumersneed in order to achieve the same utility level as the case before the product qual-ity and price change. By defining the individual i’ utility from consume productj at scenario b as ubij = V bij + εbij , the compensating variation for individual i fromscenario b to scenario a could be estimated as:

CV bai =ubi −u

ai

βip= ∆CSbai

where ubi = maxj Vbij , βip is defined in equation (1), and scenario b and a corre-

spond to different product characteristics (including prices, indications, and otherdrug attributes).

Therefore, the total change of consumer surplus in the market could be calcu-lated as:

∆CS =∑i

CV bai ==∑i

ln[∑Jbj=0 exp(V bij )]− ln[

∑Jaj=0 exp(V aij )]

βip. (12)

Here, I adopt the way which ? used to estimate consumer surplus and assume thatpatients will take the SSRI antidepressant pills once daily. Their compliance rateis assumed to be .75 and they will continue to purchase the drugs for 365 days inthe year.

4.3.2 Decomposing the Welfare Effects

To better interpret the welfare effects in equation (10), we introduce the socialwelfare of the intermediate counterfactual scenario, W(II, no ME), and therefore,equation (10) can be rewritten as

∆ = {W (II,ME)−W (II,no ME)}︸ ︷︷ ︸Deadweight Losses from Market Exclusivity

+ {W (II,no ME)−W (no II,no ME)}︸ ︷︷ ︸Benefits of Incremental Innovations

(13)

In this way, the total social welfare is decomposed into two terms: dead-weightloss resulting from granting market exclusivity to incremental innovation and thebenefits of incremental innovations. As we described in Section 4.3.1, social wel-

20

fare includes both firms’ profits and consumer surplus, therefore, it implies thatthe two terms in equation (13) should also be calculated separately for the con-sumers and the firms. 34

If ∆ is positive, it implies that granting additional market exclusivity to incre-mental innovation is valuable since its losses are offset by its benefits. The firstterm represents the welfare effects of allowing incremental innovation obtainingextended market exclusivity (from the consumer side, it mainly involves the dead-weight loss from monopoly pricing; for the innovators, they could gain from theirprolonged market power). The second term denotes the welfare changes fromintroducing the improved products with higher quality without altering their ex-clusivity status. For the consumers, improved products (such as, additional in-dications, less adverse effects) without being charged for higher prices is alwaysfavorable comparing to the case without incremental innovations. While for theinnovators, the effects are ambiguous due to two opposite driven forces: Improvedquality may attract more demands while without additional exclusivity rewards,generic entry may be more aggressive and results in price competition.

4.3.3 Definition of Counterfactual Scenarios

To measure the values in equation (13), we need to establish the counterfactualcases: (II, no ME) and (no II, no ME), while case (II, ME) indicates the status quo,the real world we observed. As shown in Table 1, scenario (no II, no ME) impliesthat column 2 and 3 will be withdrawn, which include both the improved productswith new formulation/pediatric use and the patents and pediatric exclusivitiesgranted to these products. Similarly, for scenario (II, no ME), we will withdrawthe exclusivity noted in red in column 2 and 3. To remove the improved products,I assume that improved attributes introduced in the status quo become no longeravailable.35 The way to remove market exclusivity should be discussed case bycase according to how patents and data exclusivity function.36

To tease out the welfare effects from different exclusivity, we will proceed fromwithdrawal of the entire incremental innovations to eliminating each single incre-mental innovation separately,37 which gives us four cases of counterfactual anal-ysis:38

34Our results in Section (5.4) will be illustrated in this way.35For example, removing the pediatric use of Prozac in 2003 requires that the pediatric use indication

would not apply in counterfactual case since 2003.36Particularly, pediatric exclusivity applies to all the products which have the same molecule, includ-

ing all formulations, indications and dosing. Therefore, removing the pediatric exclusivity granted toProzac implies that the generic Prozac will enter the market six-month earlier. While for the patentgranted to Lexapro, to construct case (no II, no ME), since Lexapro has been removed, therefore re-moving the patent has no effects in the market; while for case (II, no ME), it implies that Lexapro isnot protected since its approval, therefore Generic Lexapro was allowed to enter the market as soon asLexapro was approved in 2002.

37This approach was motivated by the observation that market exclusivity, provided by Hatch-Waxman Act and intellectual property law, are specific to each type of incremental innovation. Byconsidering separately the introduction of each innovation and its market exclusivity, the welfare ef-fects from each type of provision could be evaluated.

38For each case, ∆ will be estimated based on equation (13).For example, for case 3, ∆ is the sum ofconsumer gains from pediatric use of the SSRIs without the market exclusivity extension and the lossof granting pediatric exclusivity to all the products that share the same molecule with the one that has

21

1. To withdraw all incremental innovations and the corresponding additionalmarket exclusivity;

2. To withdraw new formulated version of Celexa, Lexapro, and its patent;

3. To withdraw new formulated version of Paxil with safety improvement, PaxilCR, and its patent;

4. To withdraw pediatric use indication for all products and their 6-month mar-ket exclusivity;

Table 5 illustrates the way we establish the counterfactual scenarios. One thingto be clarified here is that I didn’t model generic entry in the counterfactual sce-narios39 but assuming that the generic products which exist in status quo followsthe similar pattern of entry with earlier/later timing in counterfactuals. For exam-ple, 4 generic Prozac producers entered into the market in 2001 and 15 entered in2002. When remove the pediatric exclusivity, we assume that 4 producers enteredin 2000 and 15 entered in 2001. Since our data is in year level while the pediatricexclusivity is only six months, I will treat the exclusivity as one year and once wecalculate the welfare losses about the pediatric exclusivity, we will divide it bytwo to get the estimates for six months. And for the generic products which arenon-existent in status quo, I assume they follows the same patterns as the gener-ics of the original products. For example, 12 generic Celexa producers enteredinto the market in 2004 and 3 another producers entered in 2005 in status quo.To consider the case that generic Lexapro after removing the patent for Lexapro,we assume that 12 generic Lexapro entered in 2004 and 3 entered in 2005. Thesimilar method applies to the rest of the year and other products.40

4.3.4 Conducting the Counterfactual Analysis

In this section, we are going to discuss the detailed routine of conducting the coun-terfactual analysis. The primary changes that took place from the status quo to thecounterfactual scenarios lie in that the product characteristics were altered, andthe market structure (i.e., number of generic competitors) became different. Giventhe changed product attributes, with the estimated individual preference param-eters in equation (3), (4) and (6) at Section 4.1, we could simulate the individualdemands for each product as a function of its own price and the other productprices, denoted as

Dcjt(p1t , · · · ,pJt) =∑i

P rci (j;p1t , · · · ,pJt |z,xc, α̂, Γ̂ , β̂, λ̂,Θ̂), (14)

been conducted pediatric studies.39To internalize generic entry, we need to construct a dynamic programming model to reflect the

generic producers’ entry decision by maximizing their expected present discounted profits of remainingperiod, t = 1, · · · ,T , which requires T to be large enough. One concern that prevent me doing this isthat our data has only 16 periods from 1996 to 2011 in year level. If we dis-aggregate the sample intoquarters, the sample size and the variations across quarters may not support this dynamic estimation.

40The assumption is appropriate in the sense that when Mylan got approval to produce generic Celexain 2004, as Lexapro and Celexa are very close in chemical structure in that Escitalopram (Lexapro) is anactive form contained in Citalopram (Celexa), Mylan may find it efficient to produce generic Lexaproas well, as long as Lexapro was not under patent protection.

22

where c indicates “counterfactual”. Plug (14) into (9), with the m̂cjt estimated inSection 4.2, the equilibrium prices could be simulated. In the next step, plug backthe simulated equilibrium prices into the equation (11) and (12) in Section 4.3.1,the welfare effects could be derived.

5 Results

5.1 Descriptive Statistics

Figure 1 shows the market share of SSRIs by the branded and the generics. The de-mand for SSRI antidepressants shifted gradually from the branded to the genericstarting from 2001, the first year when generic Prozac became available. The mar-ket share of generics surpassed that of the branded in 2009. From the evolutionof Celexa, Lexapro in market demand (see Figure 2), we can find an interestingphenomenon. That is, two or three years before the patent expiration of Celexa(2004), Lexapro is marketed in 2002. Over that time, the demand for Celexa grad-ually shifted to Lexapro. When the generic Celexa entered into market in 2004,the overall use of Celexa had already gone down significantly, leaving only lit-tle market share for the generic counterparts. This observation is consistent withthe evidence of ?, who find large decreases in overall use after patent expirationthat begin in the two years before generic entry and continue in the years follow-ing. Furthermore, they suggest that it might be due to advertising which shiftsdemand from the now cheaper original molecule to another patented molecule.

The price trend of SSRI antidepressants across the years are provided in Figure3. The bar graph indicates the number of generic firms producing the moleculein each year. With the number of generic entries increasing, both prices for thebranded and generics go down, although the price reduction doesn’t happen im-mediately after the first generic entry.41 The turning point is in 2007, when mostof the SSRIs prices dropped dramatically. It can be viewed as the joint effects fromMedicare Part D and the accelerated generic entry. The prices for the brandeddrugs restored again in 2010, which suggests that the patients who confront the in-novators become price inelastic and it allow the innovators charge patients higherprices.

The individual demographic statistics of the sample are provided in Table 4comparing to the national representative sample. Column 1 and 2 shows the meanand the standard deviation for the whole sample in MEPS and the analytic samplein our study; column 3-5 show the statistics of the subjects in the sample by threemutually exclusive groups: for those who reported that he/she has depressionbut he/she didn’t purchase SSRI drugs; for those who have depression and everpurchased SSRIs; and for those who have no depression condition but purchasedSSRIs, respectively.

Table 4 reveals that those in the sample are quite distinct from the nationalpopulation. Compared to the national representative sample, our study sampleconsists of individuals who are older ( 46.21 vs. 33.73), more likely to be female,

41The Hatch-Waxman Act grants 180-days of market exclusivity to the first generic entry which ex-clude the generic competitors in the short run.

23

have a lower perceived health status (2.96 vs. 2.21), have a higher prevalence ofrespiratory diseases (0.10 vs. 0.04), asthma (0.10 vs. 0.05), hypertension (0.29 vs.0.14), cardiovascular heart disease (0.19 vs. 0.08), diabetes (0.13 vs. 0.06) andare, of course, more depressive (0.95 vs. 0.09). Our sample has higher insurancecoverage (0.89 vs. 0.83), lower out of pocket rate (0.48 vs. 0.54). Our samplehas more Medicare (0.25 vs. 0.13), Medicaid (0.25 vs. 0.23) and other publicinsurance coverage (0.03 vs. 0.02) than the national sample. Our sample havelower household income ($59,620 vs. $65,570) but higher household income percapita ($25,240 vs. $22,550) which implies that they have fewer dependants inthe family. Overall, it shows considerable variation in most of the demographicvariables.

Within our sample, there are also significant differences across the three groups.For those who reported of having depression, those who chose SSRIs are signifi-cantly different from those who chose an outside option for most of the demo-graphic variables. For those who purchased SSRIs and are diagnosed as beingdepressive are also significantly different from the individuals who are not de-pressive.

5.2 Demand Side

5.2.1 Copay Prices

Based on the OLS regression model (5), the copay prices are predicted in Table 6.By gradually introducing the controls (such as market prices, out of pocket rate,family income, and the dummies), we have achieved a good predict for the indi-vidual copay with adjusted R-squared, .767 (See column 4). Higher market pricesleads to higher copay: for every 10% increase in market price , on average out ofpocket price will increase $.1133; every 0.1 increase in individual’s out of pocketrate leads to the increase of copay price on average by $.1917; individuals withhigher family income tends to have lower copay prices, however, the magnitudeis negligible. We could see that out of pocket rate plays an important role in ex-plaining the copay prices: when we include it in Model 2, the prediction becomesmuch better, with adjusted R2 jumping from .23 to .762. 42 The insurance vari-ables are introduced in Model 4. Given the market prices and the out of pocketrate from other purchases for two individuals remain constant, individuals withprivate insurance coverage on average tends to pay $.11 lower comparing to unin-sured, those with Medicare tends to pay $.04 higher. The difference of copay pricesbetween the uninsured (the default category) and Medicaid/other public insureesis not significantly different. It suggests that the private insurance payers tend tobargain with sellers for a better discounts comparing to other insurance payers.Individuals with higher incomes tend to pay less: Every 10% increase in familyincome leads to $.03 lower on average. This evidence indicates that the wealthypeople may have better information in searching for lower costs after controllingtheir insurance types and generosity of the plans.

42Someone may wonder whether the out of pocket rate is highly correlated with the insurance dummyvariables, such as private insurance, Medicare, Medicaid, etc. However, the correlations between out ofpocket rate and the insurance variables have been checked and they are not high.

24

After we use the estimated coefficients to construct virtual copay prices for thealternatives the individuals didn’t choose, I plot the distribution of the marketprices and copay prices in Figure 6. The left panel shows the percentage distri-bution of the market prices for all the choice scenarios: It appears that the pricesare mainly centered in two areas which represents the separate equilibrium of thedrug prices for the branded and the generics respectively. The drug prices for thegenerics mainly lie in the price range lower than $.8, while those for the brandedare in the range between $1.7 to $4.6. The right panel illustrates the predictedcopay prices based on the results from Column 4 of Table 6. Comparing to the leftpanel, we could see that the actual payments from the patients are significantlyreduced. Many of them lies in the range between $0 to $1.3, with some excep-tions around $2. the highest payment has been greatly reduced from close to $5to less than $3 due to the insurance coverage. The red line denotes the mean ofthe prices: the mean for market prices is around $1.94, while the mean for thepredicted copay prices is around $.95.

5.2.2 Demand Estimates

With the predicted copay prices, we estimate the equation (3) with conditionallogit model. The full set of parameters are shown in the Appendix Table (12) andTable (14). This model contains 115 drug-year fixed effects and coefficients for 49explanatory variables (including copay price effect and 48 individual-drug char-acteristics interaction terms). I report the results for selected variables in Table(7). Model 1 is based on the copay prices predicted by Model 3 in Table 6 andModel 2 is based on the prediction of Model 4 in Table 6. From the log likelihood,we can see that Model 2 achieves better fit for the data. From both models, theprice coefficients are significantly negative. Our price coefficient, -5.278, based oncopay prices tend to be more price elastic comparing to the study from?(its esti-mated price coefficient is -1.108) and are closer to the results from ? (-4.5683),both of which focus on anti-depressants market. It indicates that the estimates forprice elasticity based on copay prices which reflect the patient’s actual responseto their out of pocket payment are more elastic than the estimates based on themarket prices. Therefore, the welfare analysis from price changes based on themarket price elasticity is not accurate and is suffering from bias.

Using estimates of mean utility (drug-year fixed effects) derived from the con-ditional logit model (Model 2 in Table 7), the OLS estimation of mean utility ondrug characteristics is presented in Table 8. After controlling the molecule fixedeffects and year fixed effects, the generic drugs bring lower utility to the patients;The drugs with pediatric use indication bring higher utility; The older generationdrugs are more favored. The estimates from Model 4 will be used in the follow-ing counterfactual analysis to calculate the simulated demands in counterfactualscenarios.

5.3 Supply Side

For the Bertrand-Nash equilibrium, the recovered marginal costs satisfy equation(9): the estimates of marginal costs are built upon the observed market prices and

25

the estimates of demands and their derivatives. The estimated marginal costs andmargins ( Margin = (P rice −MC)/P rice.) for Celexa (radical innovation), GenericCelexa(generics) and Lexapro (incremental innovation) are shown in Figure 6 asan illustration for the whole set of results. The first three panels illustrate the mar-ket prices and estimated MC of Celexa, Lexapro and Generic Celexa respectively.Generic Celexa entered into the market in 2004. According to Hatch-Waxman Act,the first generic company enjoyed 180-day exclusivity. Therefore, its initial priceis as high as the branded Celexa. Not until 2007, the prices for generic Celexa ex-perienced a dramatic drop from $2 per unit to less than $.6 per unit. After 2008,the prices for the generics leveled off and stayed at the level around $.2. The pricesfor branded Celexa go through a similar downward trend in 2006 - 2009, however,from 2010 the prices climb up to as high as $2.8 in 2011. The price drop in 2006-2007, which is also observed in other SSRIs, is viewed as the joint results of theimplementation of Medicare Part D and intensified competition from the gener-ics. To better understand the estimated MC, the margins for these three drugs areplotted in the fourth panel. Although the prices fluctuated and changed too much,the predicted margins are relatively stable. The margins for the branded are muchhigher, at around .2-.3, than the generics which is only about .1.

Once we back up marginal costs for each firm, we could simulate the counter-factual prices optimized by the firms.43 The method to simulate the counterfac-tual prices has been described in Section 4.2 and 4.3. In Figure 7-10, I showed thesimulated prices for case 2 and case 4 described in Section 4.3.3.44

In Figure 7, comparing the scenario 3 in which no Lexapro is available, in-troducing Lexapro without patent protection in scenario 4, doesn’t affect pricingof the original products Celexa and generic Celexa, which suggests that the priceelasticity of these two products isn’t influenced by the presence of Lexapro. How-ever, the prices of Paxil and Prozac tend to be lower in the earlier years of Lexapro’sentry and then go up in the later years. If granting patent to Lexapro (see Figure8), the prices for Lexapro increase a little bit in the years right after the generic en-try and then drop dramatically afterward from more than $2 per unit to less than$.5. It suggests that granting Lexapro patent entail the producers a great deal ofmonopoly power. Similar to the case in Figure 7, the prices for Celexa and genericCelexa still remain unchanged.

In Figure 9, we illustrate the price changes from the scenario in which both pe-diatric use indication and pediatric exclusivity are not allowed to the scenario inwhich we introduce pediatric use indication while no pediatric exclusivity awarded.It’s easy to see that pediatric use indication almost brings little to the market pricechanges. However, if we impose additional 1-year of pediatric exclusivity45 on the

43For the product in counterfactual scenarios that are not available in the status quo, i.e. genericLexapro and generic Paxil CR, I assume that their marginal costs are equal to generic Celexa and genericPaxil as the former is the incremental innovation of the latter.

44For case 1 in which we remove all the incremental innovations and market exclusivity, it is noteasy for us to identify the effects from different type of exclusivity. And since the Paxil CR has verylimited market share and thus less important, therefore, we will focus on case 2 and case 4 to presentthe results.

45As we discussed in the welfare analysis part, we first assume 1-year of pediatric exclusivity becauseour data is in year level, and then we will divide the welfare changes by 2 to obtain the estimates for6-months.

26

drugs for their pediatric studies, the price competition is delayed for one year. Aswe could see in the figure, the major drop of prices are delayed from 2006 to 2007for Celexa, generic Celexa, Paxil, generic Paxil, Prozac, generic Prozac, Zoloft andgeneric Zoloft. The results highlight the striking effects from pediatric exclusivityin postponing generic entry and price competition.

5.4 Welfare Analysis

Based on the method stated in Section 4.3, we are able to separate the welfare ef-fects of granting incremental innovation market exclusivity into two parts: thewelfare benefits from improved quality and the deadweight loss from delayedgeneric competition. The welfare effects are presented in Table 9. Imaging a worldwhere all the incremental innovation and market exclusivity are no longer avail-able in the SSRI market (scenario 1), suppose now all the incremental innovationbecome available for use without patent and data exclusivity extension (scenario2), consumers’ accumulated monetary benefits from the improved products arearound $72.8 billion from 1996-2011. The branded firms also gain from introduc-ing the improved products by $1.55 billion. Only generic firms are worse off andsuffer a loss at around $4.75 billion. This result is interesting and meaningful tothe policy debates which suggests that even if the regulators don’t allow marketexclusivity for incremental innovation, the innovators may also have the incen-tive to invest in R& D as long as the expected investment costs is lower than theexpected profits, in our context, say, $1.55 billion.

Furthermore, imagine that in the next step we will impose additional marketexclusivity to the incremental innovation, that’s to say, the branded firms will begranted monopoly power in pricing with no changes on product characteristics,the results show that consumer surplus suffer severely from the delayed genericentry and price competition at a cost around $112 billion; generic producers alsodislike it with accumulated profit change, $-1.54. Only the innovators acclaim forthat with $6.9 billion. Notice that the innovators benefit more from granting ex-clusivity than to introduce improved products. Overall, the accumulated welfareloss46 from granting market exclusivity to incremental innovation in SSRI anti-depressant market is around $37 billion.

By re-considering the above counterfacutal analysis with alternative cases: towithdraw only one single incremental innovation and its associated market exclu-sivity (see row 2-4 in Table 9), the above conclusion persists. The only exceptionis for the welfare effects of introducing Paxil CR into the market. The consumersbenefit only $.67 billion; and the effects on the innovator and generic producersare reversed. The innovator suffers a loss at around $.25 billion, while genericproducers win around $1.26 billion. The results may due to the fact that overall,Paxil CR is not a successful practice for the innovator. From the data, we couldsee that the market share for Paxil CR is very low with only a negligible part ofpatients purchased it.

46The estimated result is the lower bound for this value since we assume that the innovators will notinvest in R& D for incremental innovation if market exclusivity is not allowed.

27

6 Conclusion

In this paper, I address the following research question: Should additional marketexclusivity be granted to incremental innovations to allow innovators to recoupR&D investment cost from monopoly pricing, despite the fact that excluding com-petition could harm the consumer surplus?

By comparing scenarios of either withdrawing or allowing market exclusivityfor incremental innovations with scenarios of withdrawing or retaining incremen-tal innovations, the consumer surplus loss due to market exclusivity extensionsfor incremental innovation exceed the consumer surplus benefits from incremen-tal innovation. Overall, the evidence indicates that the market exclusivity grantedfor incremental innovations in SSRIs has resulted in a $37 billion social welfareloss from 1996 to 2011. The result suggests that policy makers should reconsiderthe provisions for granting market exclusivity to incremental innovations.

This paper contributes to the literature in the following ways. First, to myknowledge, this is the first study to measure the value of innovated products whilequantifying loss from market exclusivity, which offers a number of interestingand critical policy implications. Second, as we discussed in the introduction, Iaddressed the price endogeneity problem with a novel method which has not beenexplored in the literature to my best knowledge. In the demand estimation, Ireplace the market price with individual-level copay price. Because of its variationacross individuals which would not be confounded by the drug-year fixed effects,the price coefficient is identified and estimated consistently without adopting anyIVs.

There are two points to be clarified. First, someone may concern about the sub-stitution effect of market exclusivity—motivating firms to invest their resources inincremental innovation results in decreased investment resources in radical inno-vation. Although we didn’t address this issue in this paper, however, even withoutconsider this negative effects (about “granting market exclusivity to incrementalinnovation”) we have already found out that granting market exclusivity is wel-fare - deteriorating. Therefore, this ignorance doesn’t affect our conclusion. Sec-ond, the responsiveness of incremental innovation to market exclusivity has notbeen investigated in this paper. We assume that without market exclusivity, noincremental innovation would be innovated. If we relax this assumption, it im-plies that some of incremental innovation would be produced in either way (nomatter granting exclusivity or not). That implies that the benefits from the ex-clusvity induced incremental innovation are even smaller, which would enhanceour conclusion that the overall loss is larger than the benefits.

However, our results should be interpreted with caution. Our results are drawnfrom the incremental innovations of SSRIs antidepressants. For other therapeu-tic classes of drugs, the conclusion may not hold. Development of innovativemedicines is full of uncertainty, as is its value to the patients, which varies acrossdiseases and treatments.

28

Table 1: Incremental Innovations in SSRI Anti-depressant Drugs

New Molecule New Formula-tion

Pediatric Use Generics

Celexa Lexapro 2002 Generic Celexa1998

(patent till2012)

2004

Paxil Paxil CR 1999(safety) six months

Generic Paxil

1992(patent till2017)

2003

Prozac Prozac Generic Prozac1987 2003 2001

six monthsGeneric Prozac

2003 (Ped.)Zoloft

six monthsGeneric Zoloft

1991 2006Note: 1. 4 new molecule entity, 3 incrementally innovated drugs and 5generics; 2. Safety– less side effects here; 3. Ped.– pediatric use; 4. Thepatents and data exclusivities granted to the incremental innovations aremarked in red. “Six months” indicates pediatric exclusivity.

Table 2: Sample Selection

Conditions No Conditions;No SSRIs SSRIs SSRIs

No. of Obs. 28,863 19,454 2,749Percentage 56.52% 38.10% 5.38%

Total 51,066

Note: (No)Conditions: The respondents reportedthat they have (no) depressive condition. (No)SSRIs: The respondents purchase (no) SSRI an-tidepressants. No Drugs: The respondents didn’tpurchase any antidepressants. Other Drugs: Therespondents purchased other class of antidepres-sants.

29

Table 3: Manufacturers and Market Share

Products Analytic Sample Survey SamplePurchasing Obs. Market Share Purchasing Obs. Market Share

CELEXA 1,476 6.65 7,464 6.37LEXAPRO 3,244 14.61 17,565 15.00PAXIL 3,124 14.07 15,631 13.35PAXIL CR 395 1.78 1,877 1.60PROZAC 2,659 11.98 13,948 11.91ZOLOFT 4,830 21.75 25,482 21.76Generic CELEXA 1,620 7.30 8,246 7.04Generic PAXIL 1,176 5.30 7,110 6.07Generic PROZAC 2,111 9.51 11,498 9.82Generic ZOLOFT 1,568 7.06 8,279 7.07Total 22,203 100 117,100 100

30

Figure 1: Market Share of SSRIs across year

050

01,

000

1,50

02,

000

Num

ber

of In

divi

dual

s

1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

Branded Generics

Figure 2: Market Share of Celexa, generic Celexa and Lexapro across Year

020

040

060

080

0N

umbe

r of

Indi

vidu

als

1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

CELEXA Generic_CELEXA

LEXAPRO

31

Table 4: Individual Demographics

Variables All MEPS Sample Having Having Nosample Employed Conditions Cond.; Cond.;(96-11) in Study No SSRIs. SSRIs SSRIs

Age 33.73 46.21*** 44.6 48.23*** 48.74[22.57] [18.69] [19.18] [17.64] [19.01]

Male 0.48 0.32*** 0.35 0.28*** 0.29[0.50] [0.47] [0.48] [0.45] [0.45]

Adult(age>18) 0.7 0.93*** 0.91 0.95*** 0.93***[0.46] [0.26] [0.28] [0.21] [0.25]

Years of Education 9.4 11.86*** 11.57 12.27*** 11.98***[5.61] [3.58] [3.77] [3.23] [3.56]

Family Income Per Capita(1,000$) 22.55 25.24*** 24.03 26.97*** 25.73**[23.80] [25.84] [25.57] [26.39] [23.93]

Family Income(1,000$) 65.57 59.62*** 56.77 62.94*** 66.03**[60.06] [58.35] [57.30] [59.52] [59.24]

Perceived Health Status 2.21 2.96*** 2.91 3.04*** 2.87***[0.95] [1.04] [1.04] [1.04] [1.04]

Having Medical Insurance 0.83 0.89*** 0.86 0.92*** 0.94***[0.37] [0.32] [0.35] [0.27] [0.24]

Private Insurance 0.57 0.57 0.53 0.61*** 0.67***[0.50] [0.50] [0.50] [0.49] [0.47]

Having Medicare 0.13 0.25*** 0.23 0.27*** 0.28[0.34] [0.43] [0.42] [0.44] [0.45]

Having Medicaid 0.23 0.25*** 0.27 0.23*** 0.19***[0.42] [0.43] [0.44] [0.42] [0.39]

Other Public Insurance 0.02 0.03*** 0.03 0.03 0.03*[0.14] [0.16] [0.16] [0.16] [0.18]

Out of Pocket Rate 0.54 0.48*** 0.51 0.44*** 0.45[0.27] [0.36] [0.31] [0.41] [0.41]

Having Respiratory Diseases 0.04 0.10*** 0.09 0.11*** 0.08***[0.20] [0.29] [0.28] [0.31] [0.27]

Having Asthma 0.05 0.10*** 0.09 0.11*** 0.08***[0.22] [0.30] [0.29] [0.31] [0.27]

Having Hypertension 0.14 0.29*** 0.27 0.33*** 0.28***[0.35] [0.46] [0.45] [0.47] [0.45]

Having Cardiovascular Heart Diseases 0.08 0.19*** 0.17 0.21*** 0.2[0.27] [0.39] [0.38] [0.41] [0.40]

Having Diabetes 0.06 0.13*** 0.12 0.15*** 0.12***[0.23] [0.33] [0.32] [0.35] [0.32]

Having Depression 0.09 0.95*** 1 1 0***[0.17] [0.26] [0.00] [0.00] [0.00]

Observations 276702 51066 28863 19454 2749

Note: Standard Deviation is enclosed in the bracket. Income is deflated at 1996 dollar level and inthousands. The stars indicate the significance of ttest comparing to the previous column, i.e., starsin Column 5 compare Column 4 and 5. ∗ ∗ ∗p < .01,∗ ∗ p < .05,∗p < .1.

32

Figure 3: Price Trend with Generic Entry

01

23

45

05

1015

2025

# G

ener

ic E

ntry

1995 2000 2005 2010Year

# Generic Entry CELEXA Price

Generic_CELEXA Price LEXAPRO Price

Celexa

01

23

45

Pric

e

05

1015

2025

1995 2000 2005 2010Year

# Generic Entry PAXIL Price

Generic_PAXIL Price PAXIL_CR Price

Paxil

01

23

45

05

1015

2025

# G

ener

ic E

ntry

1995 2000 2005 2010Year

# Generic Entry PROZAC Price

Generic_PROZAC Price

Prozac

01

23

45

Pric

e

05

1015

2025

1995 2000 2005 2010Year

# Generic Entry ZOLOFT Price

Generic_ZOLOFT Price

Zoloft

Figure 4: Distribution of Out of Pocket Rate for Patients Who Purchased SSRIs

05

1015

2025

Per

cent

0 .2 .4 .6 .8 1Out of Pocket Rate

Note: It is based on the 51,066 individual-year observations. The red line denotes the mean of out ofpocket rate, which is around .48.

33

Table 5: Market Exclusivity Changes across Counterfactuals

W (II, no ME) W (no II, no ME)1. All incremental innova-tions

S2: Generic Paxil, Prozacand Zoloft entered one yearearlier, Generic Lexapro andGeneric Paxil CR enteredwith Generic Celexa andGeneric Paxil

S1: Generic entries for Paxil,Prozac and Zoloft are oneyear earlier,

2. Lexapro as a new formula-tion of Celexa

S4: Generic Lexapro enteredwith Generic Celexa

S3: No Lexapro developedand no generic Lexapro en-tering

3. Paxil CR with improvedsafety

S6: Generic Paxil CR enteredwith Generic Paxil

S5: No Paxil CR developedand no Generic Paxil CR en-tering

4. Pediatric use S8: Generic Paxil, Prozac andZoloft entered one year ear-lier

S7: Generic Paxil, Prozac andZoloft entered one year ear-lier

Note: “S” denotes Scenario. Prices are calculated in simulation, but are subjected to changeswith exclusivity removal. For example, if the exclusivity of Prozac is reduced for one year,competitive market present one year earlier from 2001 to 2000. Then the price curve shiftone year earlier.

34

Table 6: OLS Regression of out of pocket prices (copay) on average full prices

Model 1 Model 2 Model 3 Model 4Coef./S.E. Coef./S.E. Coef./S.E. Coef./S.E.

Log(Market Price) 0.829*** 1.132*** 1.133*** 1.133***(0.012) (0.006) (0.008) (0.008)

Having Private Insurance -0.423*** -0.111***(0.007) (0.004)

Having Medicare 0.040*** 0.048***(0.006) (0.004)

Having Other Public Insurance 0.204*** -0.013(0.041) (0.023)

Having Medicaid -0.810*** 0.002(0.008) (0.004)

Log(Family Income) -0.014*** -0.016*** -0.015*** -0.003***(0.002) (0.001) (0.001) (0.001)

Out Of Pocket Rate 1.914*** 1.913*** 1.917***(0.003) (0.004) (0.004)

Intercept 0.408*** -1.046*** -1.037*** -1.076***(0.027) (0.015) (0.017) (0.017)

N 117100 117100 117100 117100Adjusted R2 .231 .762 .764 .767Year FE Yes Yes Yes YesMolecule FE Yes Yes Yes YesMolecule*Generic FE No No Yes YesInsur. Cat. FE Yes No No YesDiag. Dis. Var. No No Yes Yes

Note: 1. * p<0.10, ** p<0.05, *** p< .01. 2. Insurance Category include: Medicare, Medicaid,Private Insurance, Other Public Insurance, No insurance. 3. Diagnoised disease variables includeCardiovascular diseases, Hypertension, Diabetes, Depression, Respiratory diseases.

Table 7: Conditional Logit Regression of Drug Choices

Model 1 Model 2Coef./S.E. Coef./S.E.

Price -4.269*** -5.278***(0.437) (0.436)

N 447483 447483Log likelihood -68702 -68421DrugYear FE Yes Yes* p<0.10, ** p<0.05, *** p< .01.

35

Figure 5: Distribution of Average Market Prices and Predicted Copay Prices for Pa-tients Who Make the Choices

05

1015

Per

cent

0 1 2 3 4 5Average Market Prices at 1996$

05

1015

Per

cent

0 1 2 3 4 5Predicted Copay Prices

Note: It is based on the 447,483 individual-year-choice level data.

Table 8: OLS Regression of Mean Utility on Drug Characteristics

Model 1 Model 2 Model 3 Model 4Coef./S.E. Coef./S.E. Coef./S.E. Coef./S.E.

Generic -2.509*** -2.056*** -2.104*** -1.816***(0.361) (0.298) (0.283) (0.228)

Pediatric 0.633* -0.599 0.500* 0.880**(0.352) (0.487) (0.274) (0.421)

Age of Molecule -0.171 -1.178*** 1.153*** 2.635***(0.312) (0.332) (0.291) (0.580)

Intercept -1.606** 1.000 -2.895*** 3.826***(0.736) (1.323) (0.933) (1.288)

N 116 116 116 116Adjusted R2 .314 .573 .598 .76Year FE No No Yes YesMolecule FE No Yes No Yes* p<0.10, ** p<0.05, *** p< .01.

36

Figure 6: Estimated MC and Margin for Celexa, Lexapro and generic counterparts

0.0

1.0

2.0

3.0

1995 2000 2005 2010Year

Price MC

CELEXA

0.0

1.0

2.0

3.0

1995 2000 2005 2010Year

Price MC

LEXAPRO0.

01.

02.

03.

0

1995 2000 2005 2010Year

Price MC

Generic_CELEXA

0.0

0.2

0.4

0.6

1995 2000 2005 2010Year

CELEXA LEXAPRO

Generic_CELEXA

Predicted Margin

Figure 7: Price Changes from Scenario 3 to Scenario 4

.51

1.52

2.5

1995 2000 2005 2010Year

S 3 S 4

CELEXA

0.51

1.52

2.5

1995 2000 2005 2010Year

S 3 S 4

Generic_CELEXA

.51

1.52

2.5

1995 2000 2005 2010Year

S 3 S 4

PAXIL

012345

1995 2000 2005 2010Year

S 3 S 4

Generic_PAXIL

01234

1995 2000 2005 2010Year

S 3 S 4

PROZAC

012345

1995 2000 2005 2010Year

S 3 S 4

Generic_PROZAC

01234

1995 2000 2005 2010Year

S 3 S 4

ZOLOFT

0.51

1.52

2.5

1995 2000 2005 2010Year

S 3 S 4

Generic_ZOLOFT

Note: Scenario 3 refers to the case that Lexapro is no longer available; Scenario 4 refers to the case thatLexapro become available without patent protection.

37

Figure 8: Price Changes from Scenario 4 to Status Quo

0123

1995 2000 2005 2010Year

S 4 S. Q.

CELEXA

0.51

1.52

2.5

1995 2000 2005 2010Year

S 4 S. Q.

Generic_CELEXA

.51

1.52

2.5

1995 2000 2005 2010Year

S 4 S. Q.

LEXAPRO

.51

1.52

2.53

1995 2000 2005 2010Year

S 4 S. Q.

PAXIL

0.51

1.52

2.5

1995 2000 2005 2010Year

S 4 S. Q.

Generic_PAXIL

012345

1995 2000 2005 2010Year

S 4 S. Q.

PROZAC

012345

1995 2000 2005 2010Year

S 4 S. Q.

Generic_PROZAC

01234

1995 2000 2005 2010Year

S 4 S. Q.

ZOLOFT

0.51

1.52

2.5

1995 2000 2005 2010Year

S 4 S. Q.

Generic_ZOLOFT

Note: Scenario 4 refers to the case that Lexapro become available without patent protection; Status quorefers to the real world where Lexapro is available with patent protection.

Figure 9: Price Changes from Scenario 7 to Scenario 8

.51

1.52

2.5

1995 2000 2005 2010year

S 7 S 8

CELEXA

1.82

2.22.4

1995 2000 2005 2010year

S 7 S 8

LEXAPRO

0.51

1.52

1995 2000 2005 2010year

S 7 S 8

Generic_CELEXA

.51

1.52

2.5

1995 2000 2005 2010year

S 7 S 8

PAXIL

012345

1995 2000 2005 2010Year

S 7 S 8

Generic_PAXIL

01234

1995 2000 2005 2010year

S 7 S 8

PROZAC

012345

1995 2000 2005 2010year

S 7 S 8

Generic_PROZAC

.51

1.52

2.5

1995 2000 2005 2010Year

S 7 S 8

ZOLOFT

.51

1.52

2.5

1995 2000 2005 2010Year

S 7 S 8

Generic_ZOLOFT

Note: Scenario 7 refers to the case that all the pediatric exclusivity has been removed and the pediatricuse indication become no longer available; Scenario 8 refers to the case that the pediatric use indicationis available while no pediatric exclusivity granted.

38

Figure 10: Price Changes from Scenario 8 to Status Quo

0123

1995 2000 2005 2010Year

S 8 S. Q.

CELEXA

0.51

1.52

2.5

1995 2000 2005 2010Year

S 8 S. Q.

Generic_CELEXA

1.82

2.22.42.6

1995 2000 2005 2010Year

S 8 S. Q.

LEXAPRO

.51

1.52

2.53

1995 2000 2005 2010Year

S 8 S. Q.

PAXIL

012345

1995 2000 2005 2010Year

S 8 S. Q.

Generic_PAXIL

012345

1995 2000 2005 2010Year

S 8 S. Q.

PROZAC

0123

1995 2000 2005 2010Year

S 8 S. Q.

Generic_PROZAC

.51

1.52

2.5

1995 2000 2005 2010Year

S 8 S. Q.

ZOLOFT

0.51

1.52

2.5

1995 2000 2005 2010Year

S 8 S. Q.

Generic_ZOLOFT

Note: Scenario 8 refers to the case that the pediatric use indication is available while no pediatricexclusivity granted; Status quo refers to the real world where pediatric use indication is available andthe pediatric exclusivity has been granted to Paxil, Prozac, Zoloft.

39

Table 9: Welfare Changes (Billion Dollars)

II W (II, ME)-W (II, no ME) W (II, no ME)-W (no II, no ME)Consumer Branded Firm Generic Firm Consumer Branded Firm Generic Firm

1. All incremental innovations -112 6.9 -1.54 72.8 1.55 -4.752. Lexapro as a new formulation ofCelexa

-105 6.25 -1.08 37.5 0.24 -3.21

3. Paxil CR with improved safety -92.5 5.1 -2.56 0.67 -0.25 1.264. Pediatric use -95 5.4 -1.87 26.6 0.99 -1.74

40

7 Appendix

7.1 Data Construction

7.1.1 Individual-level Data

The individual characteristics are constructed from MEPS Full Year Consolidatedfiles. The Full Year Consolidated file includes demographic and labor market in-formation, sample weights, health status and a rich set of health insurance cov-erage information. MEPS provides the date of birth for each respondents whichcan be used to calculate the Age. Y ears of Education is collected for each re-spondents at the first round of interviews. We generate a Adult dummy for eachindividual, Adult = 1 if his/her age is greater than or equal to 18 years old. Fam-ily Yearly Income is summarized over all household members from three compo-nents: person’s total income; person’s refund income; person’s sale income. InMEPS, total person-level income is the sum of all income components with theexception of person’s refund income and person’s sale income to match as closelyas possible the CPS definition of income; For our purpose, we will sum them upto obtain the measure of total individual income. By averaging family incomewithin household members, we obtain Family Income Per Capita. After deflatedto 1996 dollars level, we define the income variable as Log(Family Income P erCapita/1000 + 1) to avoid the scale problem in MLE. In the survey, respondentsare asked the following question during each round: “In general, compared to otherpeople of the same age, would you say that your health is excellent, very good, good,fair, or poor?” Based on this question, the perceived health status of each respon-dent are evaluated from 1 (excellent) to 5 (poor). The variable P erceived HealthStatus in our study is the mean over rounds. MEPS constructs a medical insurancevariable which summarizes health insurance coverage for the person in each year.Our medical insurance variable Having Medical Insurance equals to one if thesummarized insurance variable indicates the person has either public or privatehealth insurance. Whether the individual participates inMedicare orMedicaid ornot is provided by MEPS. All the purchasing observations for SSRI drugs are pre-cisely recorded in MEPS Prescribed Medicines Component Files. The PrescribedMedicines Component collects information from the actual pharmacies where sur-vey participants obtain their prescriptions. The information obtained from phar-macies includes the national drug code and name of the drug, the strength andquantity obtained, for what condition the drug is prescribed, the total price, aswell as the amounts paid by different insurance sources and the patient. Based onthe price information provided by MEPS Prescribed Medicines Component Files,we can easily construct Out of P ocket Rate by dividing the price paid by patientswith the total price. Most of the individuals have purchasing records (not re-stricted to SSRI drugs) in the data which can be used to generate this variable.Less than 10 % of the individuals have no information of Out of P ocket Rate. Themean of Out of P ocket Rate in that year is utilized for these observations. A largepart of the observations in our study are those who have depression condition.Before we describe the sample construction, the first thing to know is the depres-sion condition dummy. Condition information is provided in Medical Condition

41

Files. The condition can be included in MEPS condition roster only for the fol-lowing reasons: reported by the household respondent for a particular medicalevent (hospital stay, outpatient visit, emergency room visit, home health episode,prescribed medication purchase, or medical provider visit); reported as the reasonfor one or more episodes of disability days; or reported by the household level re-spondent as a condition ”bothering” the person during the reference period. Wedefine a respondent as having depression if the ICD-9-CM code equal to one ofthe following: 296 for episodic mood disorders, 300 for anxiety, dissociative andsomatoform disorders, code 311 for depressive disorder not elsewhere classified.By merging the Medical Condition File with Full Year Consolidated File, we couldobtain the sample that has depression. Combining all those who purchased SSRIanti-depressants with those who have depression conditions gives us the analyticsample.

7.1.2 Drug-level Data

One of the key drug characteristics is price, which we constructed by averagingprices across strengths and dosage forms by sales weight. All prices have been ad-justed to unit level, i.e., price per pill; and are deflated to 1996 dollar level by CPIfor Managed Care Commodities category (Source: US Bureau of Labor Statistics).The information about drug characteristics is obtained from package insert labels.At the section of Indications and Usage in labels, indications of drugs are enumer-ated. We generate seven indication dummies for each product, including MajorDepressive Disorder (MDD), Obsessive Compulsive Disorder (OCD), Panic Disor-der (PD), Post-traumatic Stress Disorder (PSD), Premenstrual Dysphoric Disorder(PDD), Social Anxiety Disorder (SAD), and Bulimia Nervosa (BN). Indications ofeach product are listed in Table 13. As we see from the table, when a generic drugenters into market, it automatically obtains the new indication of the branded aslong as they are therapeutically equivalent, for example Generic Prozac 2001 andProzac 2000. The reason lies in that the branded firm cannot forbid it without pro-viding new indications within a new strength or a new formulation since physi-cians could and would prescribe generics for the new indication of the brandedone (as long as they are the same). If the safety and efficacy of a drug for pediatricuse are provided, then we can regard the drug as having pediatric use. However,the difficulty is that in the label, the safety and efficacy data for pediatric usageare specific to each indication, for example, the efficacy of ZOLOFT for the treat-ment of OCD was demonstrated in clinical trials. However, safety and efficacy inthe pediatric population other than pediatric patients with OCD have not beenestablished. If we consider the pediatric usage for each specific indication willincrease the number of dummies several times which brings about an estimationproblem, therefore, here as long as the safety and efficacy data for one indicationhave been established, we will regard this drug with P ediatric Use = 1. The sideeffects information is obtained from the Adverse Reactions Section in the labels.We choose the 10 most common symptoms which bother patients most into ourdata: Headache, Asthenia, Nausea, Diarrhea, Anorexia, Insomnia, Anxiety, Som-nolence, Rash, and Abnormal Ejaculation (these do not need to be capitalized).For each symptom, the side effect ratio is defined as the occurrence rate of the pa-

42

Figure 11: Part of the Table in Section Adverse Reactions in the label of Prozac

tients taking medicine with respect to the occurrence rate of the patients takingplacebo. To illuminating the generating process, the example of Prozac’s label isillustrated below: As shown in Figure 15, for symptom nausea, the incidence ofnausea is: 21 % of 1728 Major Depressive Disorder patients taking Prozac vs. 9%of 975 subjects feeling nausea who have taken placebo; 26 % of 266 patients takingProzac to treat OCD felt nausea, while 13 % of 89 OCD patients taking placebofelt nausea; 29% of 450 Bulimia Nervosa patients taking Prozac vs. 11% of 267 BNpatients taking placebo; 12 % of 425 Panic Disorder patients taking Prozac vs. 7 %of 342 PD patients taking placebo. Then the hazard rate ratio variable for nauseacan be computed as:

1728∗.21+266∗.26+450∗.29+425∗.121728+266+450+425

975∗.09+89∗.13+267∗.11+342∗.07975+89+267+342

= 2.34

7.2 Appendix Figures

7.3 Appendix Tables

43

Figure 12: Market Share of SSRIs over Brands across Year

050

01,

000

1,50

02,

000

Num

ber

of In

divi

dual

s

1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

CELEXA Generic_CELEXA

Generic_PAXIL Generic_PROZAC

Generic_ZOLOFT LEXAPRO

PAXIL PAXIL_CR

PROZAC ZOLOFT

Figure 13: Annual Sales of SSRIs over Brands across Year

020

040

060

080

0A

nnua

l Sal

es($

1,00

0)

1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

CELEXA Generic_CELEXA

Generic_PAXIL Generic_PROZAC

Generic_ZOLOFT LEXAPRO

PAXIL PAXIL_CR

PROZAC ZOLOFT

44

Table 10: Drug Indications

Ingredient Brand Name Entry MDD OCD PD PSD PDD SAD BN Ped.CITALOPRAM CELEXA 1998 1 0 0 0 0 0 0 0HYDROBROMIDE Generic CELEXA 2004 1 0 0 0 0 0 0 0ESCITALOPRAM LEXAPRO 2002 1 0 0 0 0 0 0 0OXALATE LEXAPRO 2003 1 0 0 0 0 1 0 0FLUOXETINE PROZAC 1987 1 1 0 0 0 0 1 0HYDROCHLORIDE PROZAC 2000 1 1 0 0 1 0 1 0

Generic PROZAC 2001 1 1 0 0 1 0 1 0PROZAC 2003 1 1 0 0 1 0 1 1

Generic PROZAC 2003 1 1 0 0 1 0 1 1PROZAC 2006 1 1 1 0 1 0 1 1

Generic PROZAC 2006 1 1 1 0 1 0 1 1PAROXETINE PAXIL 1992 1 1 1 0 0 0 0 0HYDROCHLORIDE PAXIL 1999 1 1 1 0 0 1 0 0

PAXIL 2001 1 1 1 1 0 1 0 0PAXIL CR 1999 1 0 0 0 0 0 0 0PAXIL CR 2002 1 0 1 0 0 0 0 0PAXIL CR 2003 1 0 1 0 1 1 0 0

Generic PAXIL 2003 1 1 1 1 0 1 0 0SERTRALINE ZOLOFT 1991 1 1 1 0 0 0 0 1HYDROCHLORIDE ZOLOFT 2002 1 1 1 1 1 0 0 1

ZOLOFT 2003 1 1 1 1 1 1 0 1Generic ZOLOFT 2006 1 1 1 1 1 1 0 1

Note: Entry: Approved or Supplemental approved by FDA; MDD=Major Depressive Disor-der; OCD=Obsessive Compulsive Disorder; PD=Panic Disorder; PSD=Posttraumatic Stress Disor-der; PDD=Premenstrual Dysphoric Disorder; SAD=Social Anxiety Disorder; BN=Bulimia Nervosa;Ped.=The safety and efficacy of the drug for pediatric usage have been established.

45

Table 11: Side effects of drugs

Ingredient Brand Name Headache Asthenia Nausea Diarrhea Anorexia Insomnia Anxiety Somnolence Rash AbnEjaCITALOPRAM Celexa 1 1 1.5 1.6 2 1.1 1.33 1.8 1 6HYDROBROMIDEPAROXETINE Paxil 1.38 2.82 3.51 1.89 1 2.33 1.68 3.84 2.6 29.24HYDROCHLORIDEFLUOXETINE Prozac 1.1 1.83 2.34 1.57 3.33 1.9 2 2.4 1.33 3.42HYDROCHLORIDESERTRALINE Zoloft 1.08 1.71 2.27 2 3 1.9 1.33 1.86 1.5 14HYDROCHLORIDEESCITALOPRAM Lexapro 1 1 2.14 1.6 1 2.25 1 3 1 10OXALATEPAROXETINE Paxil CR 1.35 1.56 2.2 2.57 1 1.89 1 2.75 1 26HYDROCHLORIDE

Note: The values in this chart are calculated as the ratio of hazard rate of the patients who take drugs with respect to the hazard rate of the patientswho take placebo. AbnEja=Abnormal Ejaculation.

46

Table 12: Estimated Coefficients by Conditional Logit Model

Constant Generic Out of Pocket Age Age Male Adult Log(Income Perceived Education Having Medical Medicare MedicaidRatio Squared Per Capita) Health Insurance

Price -5.28*** 2.1*** 5.01*** -7.71*** 3.69*** -0.06*** -0.15** -0.04** 0.01 0.11**(0.44) (0.17) (0.06) (1.30) (0.62) (0.02) (0.07) (0.01) (0.01) (0.05)

Agemole 0.71*** -0.07*** -0.1*** -0.19*** 0.09(0.08) (0.02) (0.02) (0.04) (0.06)

Generic -3*** -1.57 0.99 0.02 -0.03 -0.29*** 0.18***(0.32) (1.74) (0.81) (0.02) (0.02) (0.05) (0.05)

Pediatric 0.18 0.2 0.13** 0.04 -0.19 -0.29 0 -0.12(0.17) (0.17) (0.07) (0.06) (0.2) (0.23) (0.57) (0.12)

Outside -0.1 -0.28*** -0.39*** -0.67*** -0.72*** 0.84***(0.54) (0.06) (0.05) (0.04) (0.11) (0.14)

OCD -0.17*** -0.35(0.03) (3.11)

PMDD 0.21**(0.09)

Headache -7.5*** 1.24(0.66) (7.17)

Asthenia -0.75(2.18)

Nausea 11.96*** 0.79(0.18) (1.14)

Diarrhea 2.74*** -1.17(0.12) (1.14)

Anorexia 0.37*** 0.25(0.03) (0.76)

Insomnia -6.4***(0.15)

Note: Standard errors in parenthesis.

47

Table 13: Generic Entry by Year(Source:Drug@FDA)

Trade name Manufacturers Entry YearCELEXA FOREST LABS 1999CITALOPRAM HYDROBROMIDE ACTAVIS ELIZABETH 2004CITALOPRAM HYDROBROMIDE ALPHAPHARM 2004CITALOPRAM HYDROBROMIDE APOTEX INC 2004CITALOPRAM HYDROBROMIDE AUROBINDO 2004CITALOPRAM HYDROBROMIDE CARACO 2004CITALOPRAM HYDROBROMIDE COREPHARMA 2004CITALOPRAM HYDROBROMIDE DR REDDYS LABS LTD 2004CITALOPRAM HYDROBROMIDE IVAX SUB TEVA PHARMS 2004CITALOPRAM HYDROBROMIDE MYLAN 2004CITALOPRAM HYDROBROMIDE ROXANE 2004CITALOPRAM HYDROBROMIDE SANDOZ 2004CITALOPRAM HYDROBROMIDE WATSON LABS 2004CITALOPRAM HYDROBROMIDE BIOVAIL LABS INTL 2005CITALOPRAM HYDROBROMIDE EPIC PHARMA 2005CITALOPRAM HYDROBROMIDE PLIVA 2005CITALOPRAM HYDROBROMIDE AMNEAL PHARMS NY 2006CITALOPRAM HYDROBROMIDE AUROBINDO PHARMA LTD 2006CITALOPRAM HYDROBROMIDE INVAGEN PHARMS 2006CITALOPRAM HYDROBROMIDE MUTUAL PHARM 2006CITALOPRAM HYDROBROMIDE SILARX 2006CITALOPRAM HYDROBROMIDE TARO 2006CITALOPRAM HYDROBROMIDE TEVA PHARMS 2006CITALOPRAM HYDROBROMIDE TORRENT PHARMS 2007CITALOPRAM HYDROBROMIDE NATCO PHARMA LTD 2008CITALOPRAM HYDROBROMIDE GLENMARK GENERICS 2009PAXIL GLAXOSMITHKLINE 1992PAROXETINE HYDROCHLORIDE APOTEX 2003PAROXETINE HYDROCHLORIDE ALPHAPHARM 2004PAROXETINE HYDROCHLORIDE SANDOZ 2004PAROXETINE HYDROCHLORIDE TEVA 2005PAROXETINE HYDROCHLORIDE APOTEX INC 2006PAROXETINE HYDROCHLORIDE AUROBINDO PHARMA 2007PAROXETINE HYDROCHLORIDE CARACO 2007PAROXETINE HYDROCHLORIDE MYLAN 2007PAROXETINE HYDROCHLORIDE ROXANE 2007PAROXETINE HYDROCHLORIDE TEVA PHARMS 2007PAROXETINE HYDROCHLORIDE ZYDUS PHARMS USA 2007PAROXETINE HYDROCHLORIDE PROSAM LABS 2010PROZAC LILLY 1987FLUOXETINE HYDROCHLORIDE BARR 2001FLUOXETINE HYDROCHLORIDE MYLAN 2001FLUOXETINE HYDROCHLORIDE SANDOZ 2001

Continued on next page

48

Table 13 – continuedTrade name Manufacturers Entry YearFLUOXETINE HYDROCHLORIDE TEVA 2001FLUOXETINE HYDROCHLORIDE ACTAVIS MID ATLANTIC 2002FLUOXETINE HYDROCHLORIDE ALPHAPHARM 2002FLUOXETINE HYDROCHLORIDE BEIJING DOUBLE CRANE 2002FLUOXETINE HYDROCHLORIDE CARLSBAD 2002FLUOXETINE HYDROCHLORIDE DR REDDYS LABS INC 2002FLUOXETINE HYDROCHLORIDE HI TECH PHARMA 2002FLUOXETINE HYDROCHLORIDE IVAX SUB TEVA PHARMS 2002FLUOXETINE HYDROCHLORIDE LANDELA PHARM 2002FLUOXETINE HYDROCHLORIDE MALLINCKRODT 2002FLUOXETINE MUTUAL PHARMA 2002FLUOXETINE HYDROCHLORIDE NOVEX 2002FLUOXETINE HYDROCHLORIDE PHARM ASSOC 2002FLUOXETINE HYDROCHLORIDE PLIVA 2002FLUOXETINE WATSON LABS 2002FLUOXETINE HYDROCHLORIDE WOCKHARDT 2002FLUOXETINE HYDROCHLORIDE LANNETT 2004FLUOXETINE HYDROCHLORIDE PAR PHARM 2004FLUOXETINE HYDROCHLORIDE RANBAXY 2004FLUOXETINE HYDROCHLORIDE SILARX 2007FLUOXETINE HYDROCHLORIDE AUROBINDO PHARMA 2008FLUOXETINE HYDROCHLORIDE ALEMBIC PHARMS LTD 2009FLUOXETINE HYDROCHLORIDE AUROBINDO PHARM 2009FLUOXETINE HYDROCHLORIDE DR REDDYS LABS LTD 2010FLUOXETINE HYDROCHLORIDE EDGEMONT PHARMS LLC 2011FLUOXETINE HYDROCHLORIDE HERITAGE PHARMS INC 2012ZOLOFT PFIZER 1991SERTRALINE HYDROCHLORIDE IVAX SUB TEVA PHARMS 2006SERTRALINE HYDROCHLORIDE ROXANE 2006SERTRALINE HYDROCHLORIDE TEVA 2006SERTRALINE HYDROCHLORIDE ACTAVIS ELIZABETH 2007SERTRALINE HYDROCHLORIDE APOTEX INC 2007SERTRALINE HYDROCHLORIDE AUROBINDO PHARMA 2007SERTRALINE HYDROCHLORIDE DR REDDYS LABS LTD 2007SERTRALINE HYDROCHLORIDE INVAGEN PHARMS 2007SERTRALINE HYDROCHLORIDE LUPIN 2007SERTRALINE HYDROCHLORIDE MUTUAL PHARM 2007SERTRALINE HYDROCHLORIDE MYLAN 2007SERTRALINE HYDROCHLORIDE MYLAN PHARMS INC 2007SERTRALINE HYDROCHLORIDE PLIVA HRVATSKA DOO 2007SERTRALINE HYDROCHLORIDE RANBAXY 2007SERTRALINE HYDROCHLORIDE RANBAXY LABS LTD 2007SERTRALINE HYDROCHLORIDE SANDOZ 2007SERTRALINE HYDROCHLORIDE SUN PHARM INDS (IN) 2007

Continued on next page

49

Table 13 – continuedTrade name Manufacturers Entry YearSERTRALINE HYDROCHLORIDE TORRENT PHARMS 2007SERTRALINE HYDROCHLORIDE WATSON LABS 2007SERTRALINE HYDROCHLORIDE ZYDUS PHARMS USA 2007SERTRALINE HYDROCHLORIDE MATRIX LABS LTD 2008SERTRALINE HYDROCHLORIDE WOCKHARDT 2008SERTRALINE HYDROCHLORIDE AUSTARPHARMA LLC 2009SERTRALINE HYDROCHLORIDE HIKMA PHARMS 2009SERTRALINE HYDROCHLORIDE PROSAM LABS 2010LEXAPRO FOREST LABS 2002ESCITALOPRAM OXALATE ACCORD HLTHCARE 2012ESCITALOPRAM OXALATE AMNEAL PHARMS 2012ESCITALOPRAM OXALATE APOTEX INC 2012ESCITALOPRAM OXALATE AUROBINDO PHARMA LTD 2012ESCITALOPRAM OXALATE HETERO LABS LTD III 2012ESCITALOPRAM OXALATE HIKMA PHARMS 2012ESCITALOPRAM OXALATE INVAGEN PHARMS 2012ESCITALOPRAM OXALATE IVAX SUB TEVA PHARMS 2012ESCITALOPRAM OXALATE JUBILANT ORGANOSYS 2012ESCITALOPRAM OXALATE LUPIN LTD 2012ESCITALOPRAM OXALATE MACLEODS PHARMS LTD 2012ESCITALOPRAM OXALATE STI PHARMA LLC 2012ESCITALOPRAM OXALATE TARO 2012ESCITALOPRAM OXALATE TORRENT PHARMS LTD 2012ESCITALOPRAM OXALATE ZYDUS PHARMS USA INC 2012ESCITALOPRAM OXALATE SILARX PHARMS INC 2013

Table 14: Estimated Drug-year Fixed Effects

Drugname Year Coefficients St. Err. P-value SignificanceCelexa 1998 -6.14 0.49 0.00 ***Celexa 1999 -3.22 0.32 0.00 ***Celexa 2000 -2.62 0.29 0.00 ***Celexa 2001 -1.97 0.27 0.00 ***Celexa 2002 -1.61 0.27 0.00 ***Celexa 2003 -2.05 0.26 0.00 ***Celexa 2004 -2.94 0.27 0.00 ***Celexa 2005 -3.51 0.28 0.00 ***Celexa 2006 -3.60 0.28 0.00 ***Celexa 2007 -5.17 0.30 0.00 ***Celexa 2008 -4.63 0.29 0.00 ***Celexa 2009 -5.84 0.30 0.00 ***Celexa 2010 -7.66 1.03 0.00 ***

Continued on next page

50

Table 14 – continuedDrugname Year Coefficients St. Err. P-value SignificanceCelexa 2011 -5.76 0.52 0.00 ***Generic Celexa 2004 -6.10 0.61 0.00 ***Generic Celexa 2005 -5.13 0.60 0.00 ***Generic Celexa 2006 -4.89 0.59 0.00 ***Generic Celexa 2007 -5.38 0.59 0.00 ***Generic Celexa 2008 -5.04 0.59 0.00 ***Generic Celexa 2009 -4.80 0.59 0.00 ***Generic Celexa 2010 -4.79 0.59 0.00 ***Generic Celexa 2011 -4.55 0.59 0.00 ***Generic Paxil 2004 -3.55 0.58 0.00 ***Generic Paxil 2005 -3.28 0.58 0.00 ***Generic Paxil 2006 -3.23 0.58 0.00 ***Generic Paxil 2007 -4.18 0.58 0.00 ***Generic Paxil 2008 -4.01 0.58 0.00 ***Generic Paxil 2009 -3.88 0.58 0.00 ***Generic Paxil 2010 -4.05 0.58 0.00 ***Generic Paxil 2011 -4.00 0.58 0.00 ***Generic Prozac 2001 -4.52 0.60 0.00 ***Generic Prozac 2002 -3.67 0.60 0.00 ***Generic Prozac 2003 -3.87 0.60 0.00 ***Generic Prozac 2004 -4.06 0.60 0.00 ***Generic Prozac 2005 -3.92 0.60 0.00 ***Generic Prozac 2006 -3.83 0.60 0.00 ***Generic Prozac 2007 -5.13 0.60 0.00 ***Generic Prozac 2008 -5.07 0.60 0.00 ***Generic Prozac 2009 -4.90 0.59 0.00 ***Generic Prozac 2010 -4.99 0.60 0.00 ***Generic Prozac 2011 -4.92 0.60 0.00 ***Generic Zoloft 2006 -3.43 0.58 0.00 ***Generic Zoloft 2007 -3.86 0.57 0.00 ***Generic Zoloft 2008 -3.80 0.57 0.00 ***Generic Zoloft 2009 -3.61 0.57 0.00 ***Generic Zoloft 2010 -3.73 0.57 0.00 ***Generic Zoloft 2011 -3.70 0.57 0.00 ***Lexapro 2002 -4.30 0.30 0.00 ***Lexapro 2003 -2.79 0.28 0.00 ***Lexapro 2004 -2.72 0.28 0.00 ***Lexapro 2005 -2.43 0.28 0.00 ***Lexapro 2006 -2.24 0.27 0.00 ***Lexapro 2007 -2.45 0.28 0.00 ***Lexapro 2008 -1.94 0.28 0.00 ***Lexapro 2009 -2.43 0.28 0.00 ***Lexapro 2010 -2.57 0.28 0.00 ***Lexapro 2011 -2.84 0.28 0.00 ***

Continued on next page

51

Table 14 – continuedDrugname Year Coefficients St. Err. P-value SignificancePaxil 1996 -1.55 0.25 0.00 ***Paxil 1997 -0.80 0.24 0.00 ***Paxil 1998 -0.43 0.24 0.07 *Paxil 1999 0.06 0.24 0.80Paxil 2000 0.37 0.24 0.12Paxil 2001 0.88 0.24 0.00 ***Paxil 2002 0.87 0.24 0.00 ***Paxil 2003 0.69 0.24 0.00 ***Paxil 2004 -0.51 0.24 0.04 **Paxil 2005 -0.98 0.25 0.00 ***Paxil 2006 -0.79 0.28 0.01 ***Paxil 2007 -3.81 0.27 0.00 ***Paxil 2008 -3.42 0.27 0.00 ***Paxil 2009 -4.38 0.28 0.00 ***Paxil 2010 -4.26 0.35 0.00 ***Paxil 2011 -3.08 0.30 0.00 ***Paxil CR 2002 -0.29 0.50 0.56Paxil CR 2003 0.38 0.44 0.39Paxil CR 2004 0.05 0.44 0.90Paxil CR 2005 -0.41 0.45 0.36Paxil CR 2006 -0.23 0.45 0.61Paxil CR 2007 -2.08 0.47 0.00 ***Paxil CR 2008 -2.08 0.50 0.00 ***Paxil CR 2009 -4.70 0.72 0.00 ***Paxil CR 2010 -3.25 0.72 0.00 ***Paxil CR 2011 -3.76 0.72 0.00 ***Prozac 1996 -0.32 0.33 0.35Prozac 1997 -0.13 0.33 0.71Prozac 1998 -0.07 0.34 0.83Prozac 1999 0.05 0.34 0.89Prozac 2000 -0.17 0.28 0.54Prozac 2001 0.15 0.28 0.60Prozac 2002 0.23 0.30 0.44Prozac 2003 -0.41 0.30 0.17Prozac 2004 -0.29 0.31 0.34Prozac 2005 -0.63 0.31 0.05 **Prozac 2006 -0.51 0.32 0.11Prozac 2007 -2.28 0.29 0.00 ***Prozac 2008 -1.32 0.31 0.00 ***Prozac 2009 -5.33 0.30 0.00 ***Prozac 2010 -1.58 0.37 0.00 ***Prozac 2011 -2.39 0.43 0.00 ***Zoloft 1996 -0.41 0.29 0.16Zoloft 1997 -0.19 0.28 0.51

Continued on next page

52

Table 14 – continuedDrugname Year Coefficients St. Err. P-value SignificanceZoloft 1998 -0.05 0.28 0.87Zoloft 1999 -0.05 0.28 0.87Zoloft 2000 0.58 0.28 0.04 **Zoloft 2001 0.53 0.28 0.06 *Zoloft 2002 0.09 0.22 0.67Zoloft 2003 0.15 0.22 0.51Zoloft 2004 -0.09 0.22 0.68Zoloft 2005 0.00 0.22 1.00Zoloft 2006 -0.16 0.23 0.47Zoloft 2007 -2.44 0.23 0.00 ***Zoloft 2008 -3.31 0.24 0.00 ***Zoloft 2009 -3.70 0.25 0.00 ***Zoloft 2010 -2.25 0.28 0.00 ***Zoloft 2011 -3.34 0.27 0.00 ***

53