NBER WORKING PAPER SERIES SURPLUS APPROPRIATION FROM R&D AND HEALTH CARE TECHNOLOGY ASSESSMENT PROCEDURES. Tomas J. Philipson Anupam B.

Transcription

1 NBER WORKING PAPER SERIES SURPLUS APPROPRIATION FROM R&D AND HEALTH CARE TECHNOLOGY ASSESSMENT PROCEDURES Toas J. Philipson Anupa B. Jena Working Paper NATIONAL BUREAU OF ECONOMIC RESEARCH 050 Massahusetts Avenue Cabridge, MA 0238 January 2006 Corresponding author: We are thankful for oents fro seinar partiipants at The University of Chiago, Prineton, Berkeley, SUNY Buffalo, the Kiel Institute Conferene on New Tehnology and National Health Systes, and the NIH-Diretor onferene Bioedial Innovation and the Eonoy. Philipson is thankful for finanial support fro The Milken Institute, Santa Monia, CA. Jena reeived fellowship support fro the NIH through the University of Chiago Medial Sientist Training Progra. We thank Gary Beker, Angus Deaton, Isaah Ehrlih, and Casey Mulligan for oents, Lisa Lee and Mihael Capsith CDC for providing data, and Jennifer Kates Kaiser Faily Foundation and Ruigang Song CDC for helpful advie. The views expressed herein are those of the authors and do not neessarily reflet the views of the National Bureau of Eonoi Researh by Toas J. Philipson and Anupa B. Jena. All rights reserved. Short setions of text, not to exeed two paragraphs, ay be uoted without expliit perission provided that full redit, inluding notie, is given to the soure.

2 Surplus Appropriation fro R&D and Health Care Tehnology Assessent Proedures Toas J. Philipson and Anupa B. Jena NBER Working Paper No. 206 January 2006 JEL No. I ABSTRACT Given the rapid growth in health are spending that is often attributed to tehnologial hange, any private and publi institutions are grappling with how to best assess and adopt new health are tehnologies. The leading tehnology adoption riteria proposed in theory and used in pratie involve so alled "ost-effetiveness" easures. However, little is known about the dynai effiieny ipliations of suh riteria, in partiular how they influene the R&D investents that ake tehnologies available in the first plae. We argue that suh riteria ipliitly onern axiizing onsuer surplus, whih any ties is onsistent with axiizing stati effiieny after an innovation has been developed. Dynai effiieny, however, onerns aligning the soial osts and benefits of R&D and is therefore deterined by how uh of the soial surplus fro the new tehnology is appropriated as produer surplus. We analyze the relationship between osteffetiveness easures and the degree of surplus appropriation by innovators driving dynai effiieny. We illustrate how to estiate the two for the new HIV/AIDS therapies that entered the arket after the late 980's and find that only 5% of the soial surplus is appropriated by innovators. We show how this finding an be generalized to other existing ost-effetiveness estiates by deriving how those estiates identify innovator appropriation for a set of studies of over 200 drugs. We find that these studies ipliitly support a low degree of appropriation as well. Despite the high annual ost of drugs to patients, very low shares of soial surplus ay go to innovators, whih ay iply that ost-effetiveness is too high in a dynai effiieny sense. Thoas J. Philipson Graduate Shool of Publi Poliy Studies The University of Chiago 55 E 60th Street Chiago, IL and NBER t-philipson@uhiago.edu Anupa B. Jena Departent of Eonois The University of Chiago 26 E. 59 th Street Chiago, IL ajena@uhiago.edu

3 I. Introdution Tehnologial hange is often argued to be a entral fore behind the growth in health are spending. 2 Given this rapid growth, riteria used by private and publi institutions to value the inrease in health are spending therefore reuires a ethodology to easure the value of new health are tehnologies brought about by R&D investents. There is a long-standing and vast health eonois literature that attepts to assess the value of new tehnologies by use of so alled ost-effetiveness, ost-utility, or ost-benefit analysis, hereafter referred to olletively as CE analysis. 3 This type of CE analysis has been the ajor ethod proposed to evaluate new inventions and has been argued to be entral in anaging new tehnologies, their adoptions, and their ipat on long ter health are spending. Although not expliitly stated as suh, we argue that CE riteria are ipliitly onerned with estiating the observed arket level of onsuer surplus assoiated with a given tehnology. In partiular, any tehnology assessents attept to uantify the health ipats of new tehnologies for patients or health plans by oparing patient benefits with spending at observed arket pries. Exaples inlude ost-effetiveness using spending per uality- or disability adjusted life years, as is oon by publi buyers outside the US, or ost-benefit analysis onetizing ortality redutions through value-of-life estiates, as is oon in studies assessing the gains of inreased health are spending. The entral thee of suh standard CE assessents perfored in pratie sees to be to easure onsuer surplus or net onsuer benefits. As is the nor in CE pratie, tehnologies are deeed ore valuable the larger is the patient- or health plan benefits above what is spent on the. However, when new tehnologies are brought to life fro ostly R&D, onsuer surplus is a very poor guide to induing optial seond-best R&D investents. Rather, the degree to whih produer surplus aptures soial surplus, often at the expense of onsuer surplus, beoes the entral issue that deterines dynai effiieny. This, of ourse, is the rationale for the patent syste, whih substitutes produer surplus for onsuer surplus in order to stiulate ore effiient R&D investent. Therefore, we argue that for the sae reason that patents are preferred even though they lower onsuer surplus after tehnologies are disovered, tehnology adoption riteria are preferred that do not only fous on onsuer surplus. Put differently, even though easured levels of CE would be higher without the patent syste, sine patients or health plans would spend less to get the sae tehnology, dynai effiieny would learly be lowered. An illustrative ase of the dangers of CE riteria ay be vaines, whih any ties have been estiated to be extreely ost-effetive but for that and other reasons lak any appreiable R&D investents. 4 As onsuer surplus or ost-effetiveness deterines stati effiieny and innovator appropriation deterines dynai effiieny, we analyze the relationship between the two. In the ase of onopoly R&D, we arrive at the stark ipliation that any ties both dynai effiieny and patient health are axiized when CE is iniized. We show how this ipliation is altered under publi R&D subsidies suh as those by the National Institutes of Health NIH in the US and opetition in R&D leading to patent raing that ay dupliate R&D efforts. 2 See e.g. Newhouse The literature is vast, but for exaples, see Weinstein and Stason 977, Johanneson and Weinstein 993, Gold et al. 996, Meltzer 997, Druond et al. 997, Garber and Phelps 997, Garber 2000, Cutler and MClellan 200, and Cutler A ajor onern here has, of ourse, been produt liability issues. See e.g. Manning

4 As the ability of innovators to appropriate the surplus of their innovations is entral to dynai effiieny, we estiate the degree of appropriation in an illustrative anner for a ajor breakthrough in ediine the new drugs to treat HIV/AIDS that entered the arket fro the late 980 s onwards. 5 Our ajor finding is that innovators aptured only 5% of the nearly $.4 trillion worth of soial surplus arising fro these new tehnologies. 6 Thus, despite the high pries of any therapies suh as the new HIV drugs, the low degree of appropriation raises onerns about whether ost-effetiveness is too high in that it indues dynai ineffiieny. We generalize this finding by deriving why and how the CE results of over 200 studies on drugs an be ipliitly viewed as identifying the degree of innovator appropriation. We derive onditions under whih the easured level of CE of a tehnology ay be used to identify the share of soial surplus appropriated by produers of that tehnology; the CE of a given tehnology reveals inforation about the ost or deand paraeters. When suh identifiation is feasible, the existing and vast CE literature infors us about the degree of innovator appropriation. We find that 25% of the interventions onsidered have estiated levels of appropriations of less than 7%, while 75% have appropriations less than 25%. Our illustrative finding for HIV/AIDS drugs suggests their appropriation of soial surplus is at the twentieth perentile of the 200 tehnologies onsidered. As opleentary evidene, we show that these estiated levels of innovator appropriation are also onsistent with alternative ethods of alulating appropriation based on prie redutions after patent expiration and the identifying assuption of profit-axiization. The paper ay be briefly outlined as follows. Setion 2 disusses the relationship between CE easures and dynai effiieny. Setion 3 presents estiates of the share of soial surplus appropriated by produers of the new HIV/AIDS drugs. Setion 4 generalizes these findings to traditional CE easures reported in the literature. Lastly, setion 5 onludes. II. Tehnology Assessent and Dynai versus Stati Effiieny In order to disuss how CE analysis relates to stati and dynai effiieny, for a given output level denote the ex-post soial surplus of a new tehnology by w. This soial surplus an be divided into a onsuer surplus, z, and produer surplus variable profits, π, as in: w z+ π For exaple, a oonly analyzed ase is when prie-disriination is infeasible, in whih ase a given output level indues both profits and onsuer surplus aording to: 5 This finding builds on and extends work of Philipson and Jena 2005 who argue that HIV/AIDS is an iportant ase to onsider in and of itself, partly beause it is perhaps the ajor disease targeted by publi setor R&D in the US. Publi R&D on HIV/AIDS was roughly $2 billion in Health, in general, is aong the three leading industries into whih the governent alloates its R&D, the other two being defense and aero-spae. The National Institutes of Health NIH is responsible for alloating the vast ajority of the publi R&D dollar in 999, NIH funding aounted for nearly 8% of publi spending on health R&D. Of the $3.9 billion that the NIH spent on researh in that year, nearly $.8 billion 3% was spent on HIV/AIDS Health, United States, Our findings relate to an existing literature on the general inability of innovators to apture the soial value of their inventions, see e.g., Mansfield et al. 977, Mansfield 985, Levin et al. 987, Hall 996, and Nordhaus

5 π p 2 z [ p y p ] dy g p 3 0 where p is the inverse deand funtion, is the variable ost funtion whih exludes the fixed ost of R&D, and g is the gross onsuer benefit. In the disussion that follows, we refer to the observed surplus as the surplus whih results at the arket uantity. The potential surplus obtains at the uantity that axiizes w in the absene of ex-post arket iperfetions, this is true at the opetitive output. A. Cost-Effetiveness Criteria In this standard fraework, we argue that typial CE tehnology evaluation has ipliitly entered on onsuer surplus, by fousing on how uh patients benefit beyond what is spent on the tehnology after it has been developed. Despite the any fors of suh riteria developed to date, their basi goal sees to be to deterine whether inreased health are spending on new tehnologies is justified by soietal, health plan, or patient benefits in ters of iproved health. Absent fro the disussion has been the effet of suh riteria on the behavior of innovators who ake the tehnologies available in the first plae. Although stati effiieny is often enhaned with inreases in CE, as it ipliitly onerns onsuer surplus, these riteria are less understood in ters of how they relate to dynai effiieny when the observed level of CE is the result of rational behavior by arket partiipants. Coon easures of CE ratios relate the here onetized patient benefits to observed spending levels. In the traditional fraework, this an be expressed by: g z z R + 4 p p This easure z R expresses onsuer benefits as a ratio to spending, siilar to the standard onsuer surplus easure z that expresses it as a differene between the two. Ratios are often estiated through spending per uality- or disability-adjusted life years or through onetized versions of health benefits, in whih the value of life is opared to observed spending levels. These attepts, however, are ipliitly related to the size of onsuer surplus, sine they opare onsuer benefits to observed spending levels. 7 In partiular, stati tehnology assessents in health are oonly rely upon the use of ost-benefit, ost-utility, or osteffetiveness riteria to deterine under what irustanes the value whose units depend on the easure of a given tehnology exeeds what is spent on it. Although it is true that CE 7 The ipliit onsuer surplus estiation of CE analysis differs fro traditional eonoi analysis the latter typially attepts to assess onsuer surplus by estiation of deand shedules, by observing hanges in deand during supply-indued prie hanges. Iportantly, the deand urve for a good suarizes the value to onsuers of both its observed and unobserved attributes. On the ontrary, estiates of onsuer surplus based on ost-effetiveness or ost-benefit analysis are typially fored indiretly by onetizing observable onsuer benefits, e.g. by use of value of life estiates to estiate the gross onsuer benefit fro ortality redutions. 4

6 analysis onerns the ratio of gross benefit and spending, while onsuer surplus onerns their differene, both hange in the sae diretion with unilateral hanges in osts and benefits. Regarding the estiated CE agnitudes, any epirial studies estiate and douent z R ratios above unity for eployed tehnologies see e.g. referenes in Introdution. Yet, it would be extreely surprising if orretly easured z R ratios were found to be below unity, at least in a standard arket eonoy. As an illustration, onsider a private arket for health are without publi or private insurane, as ight exist for ertain eletive surgeries in the US, suh as e.g. plasti surgery. A new plasti surgery tehnology would have a z R ratio above unity if estiated orretly if individuals bought the produt only when their valuation of it exeeded the prie. This, of ourse, would always be predited under standard deand analysis. Although this expeted and basi outoe has to be ualified by the presene of private or publi insurane, it is supported by a large existing and growing epirial health eonois literature on the osteffetiveness of reent innovations. 8 More iportantly, the fat that a tehnology is ost-effetive in this way only reveals that there is a positive onsuer surplus. However, a positive onsuer surplus is onsistent with any output level, regardless of how high or low it is. Conseuently, being ost-effetive in this sense bears no relationship to either stati or dynai effiieny! Even if output is not at the opetitive or onopoly level, onsuer surplus is still positive and hene the tehnology deeed ost-effetive. The proble is that being ost-effetive is only neessary but not suffiient for stati or dynai effiieny. B. Cost-Effetiveness and Dynai Effiieny To onsider the dynai effiieny indued by oon health are assessent riteria, one ust onsider how suh riteria affet effiieny in the presene of tehnologial hange driven by endogenous R&D. Let tehnologial hange be haraterized by xr, an inreasing, differentiable, and stritly onave funtion representing the probability of disovery for a given level of R&D undertaken, r. The optial level of R&D that axiizes expeted payoffs for any hypothetial ex-post prize, k, is denoted rk and is defined by: [ x r k r] r k arg ax 5 r Our assuptions about xr iply that rk is an inreasing funtion so that R&D rises with the ex-post reward. Monopoly R&D Investents First onsider the ase of a single onopolist investing in R&D who reeives a share, a, of the soial surplus w, where 0a. Then, ra w represents the R&D undertaken when those investing in R&D axiize expeted profits. If profits drive R&D investents, the expeted soial surplus is: E a, w x[ r a w] w r a w 6 where w z + π is the soial surplus ex-post. This expression diretly highlights the well-known ipliation that dynai effiieny only ours when those undertaking the osts of R&D have 8 See, for exaple, Cutler

7 inentives that are properly aligned with soiety, whih is true when soial surplus is entirely appropriated as profits, i.e. a see e.g. Arrow 96 and Tirole 988. In other words, the key fator driving dynai ineffiieny is that profits π are less than soial surplus w. More iportantly, the size of the onsuer surplus, foused on by CE riteria, is what drives a wedge between profits and soial surplus and hene leads to under-investent in R&D. Indeed, in this setting, the dynaially effiient R&D investent is rw, whih is obtained when the entire soial surplus is appropriated as profits. More generally, for any tehnology and preferenes, the observed profits assoiated with a given level of soial surplus an be written a w. The ain issue, then, is that a <. For exaple, when prodution is haraterized by onstant returns to sale, it an be shown that onopolists faing either linear or onstant-elastiity deand earn profits that are proportional to the potential soial surplus. Speifially, a /2 in the ase of linear deand and a [ -/] under onstant elastiity of deand. 9,0 In general, if the total soial surplus assoiated with a tehnology is w, the size of the under-investent in R&D is rw-rπ rw-rw-z, whih, sine r. is an inreasing funtion, rises with the onsuer surplus foused on by CE riteria. The fat that dynai effiieny is driven by the appropriation of soial surplus to innovators iplies that substituting produer surplus for onsuer surplus often raises dynai welfare. This is analogous to the arguent that patents hurt stati effiieny but raise dynai effiieny by engaging in siilar substitution. The iportant ipliation of this is that the CE assoiated with the ex-post arket for a tehnology is not learly and onotonially related to easures of stati or dynai effiieny. Indeed, in a private arket with perfet prie disriination, dynaially effiient R&D ours beause the innovator aptures the entire soial surplus. Therefore, the dynaially optial alloation of surpluses iplies that the onsuer surplus should be iniized, as opposed to axiized under a CE riteria, to enhane dynai effiieny. In this ase, dynai effiieny ditates that a tehnology should just break even ex-post i.e., z R and that epirial studies iting ore ost-effetive tehnologies are, in fat, douenting a dynai ineffiieny! Indeed, as disussed, the underinvestent in R&D fro its soially optial level, rw - rw-z, rises with how ost-effetive a tehnology is assessed to be aording to traditional CE analysis. In this ase, the dynaially effiient iniization of CE is a diret ipliation of the lassi proble of non-appropriation by innovators leading to under-investent in R&D. Iportantly, note that iniization of CE in this ontext still axiizes patient health as full deand for the health are produt obtains though not onsuer surplus. Copetitive R&D Investents and Appropriation There are iportant instanes in whih full appropriation of soial surplus by produers ay not be dynaially optial, a priary one being opetitive R&D through so-alled patent raing. Sine opetitive R&D leads to an euilibriu level of R&D that is deterined by the average rather than arginal profit assoiated with entry, non-appropriation ay enhane effiieny by taxing the over-provision of R&D. This ay be partiularly relevant to the debate over exessive 9 The soial surplus ipliit in these results is the potential soial surplus available to innovators, i.e. the soial surplus that obtains when prie is set at its opetitive level. This differs fro the observed soial surplus available to the onopolist, whih obtains when prie and uantity are deterined by the onopolist. 0 Interestingly, profits ay even exeed the private soial surplus i.e. the gross benefit to onsuers net of osts of prodution when there are external effets in onsuption. See, for e.g., Philipson, Mehoulan, and Jena 2006 who disuss R&D under altruis in health are. 6

8 R&D into so-alled e-too drugs in the pharaeutial area. If the total fration of surplus appropriated by suessful R&D efforts is a, where 0a, the euilibriu level of R&D r is deterined by the zero-profit ondition: x [ r] aw r 7 where x. is the probability of disovery given the total investent in R&D by all firs, r. Given this partiular for of the R&D proess, it is straightforward to show that the optial level of appropriation is haraterized by : x [' r a w] r a w a 8 x[ r a w] At the optiu, surplus appropriation euals the elastiity of R&D produtivity i.e. the perentage inrease in the probability of disovery given a perent inrease in the level of R&D. Our assuptions on x., naely onavity and zero probability of suess in the absene of R&D, iply that the optial surplus appropriation is less than one the extent to whih this ours depends on the nature of R&D produtivity, x.. 2 If suh patent raes lead R&D to be over-provided, our onlusions ephasizing underprovision of R&D under onopoly R&D ay be altered. However, there appears to be an alost universal poliy towards subsidizing as opposed to taxing R&D, suh that ost nations have deided that the fores operating towards over-provision are doinated by those operating towards under-provision. In light of this, although inentives favoring over-provision ay hange the uantitative onlusions of our analysis, the ualitative onlusion that CE riteria liit already under-provided R&D sees generally appliable to ost researh areas and ountries. Publi R&D Subsidies and Appropriation Another iportant ase in whih non-full appropriation ay be optial is when publily funded R&D oprises a signifiant portion of total R&D, as is oon in US health are through NIH. Sine the dynaially optial level of total R&D is still r rw, the presene of publily funded R&D iplies that the optial private R&D and hene, appropriation should be lowered aordingly. More preisely, onsider the expeted soial surplus in the presene of publily funded R&D, s, and surplus appropriation, a, where 0a: E s, a, w x[ s + r a w] w s r a w 9 This follows fro the zero-profit ondition and the FOC for the expeted welfare, x [raw]w. 2 An interesting ase to onsider is when R&D produtivity is haraterized by xr p r, where 0p. In this ase, one an show that the optial level of R&D is r* ln[-/wlnp]/lnp and the optial appropriation is a* -[r*p r* lnp]/-p r*. For an ex-post surplus of $ trillion whih we argue is roughly the ase for HIV/AIDS, the optial appropriation ay be as low as 5% when R&D is very produtive and the optial R&D is $40 billion or perhaps as high as 80% when the optial R&D is $270 billion. The extent to whih appropriation deviates fro one depends on the produtivity of R&D, as a lower p iplies ore produtive R&D. 7

9 The probability of disovery is deterined by the su of publi and private R&D, the latter being driven by the appropriation of ex-post surplus by produers. For a given level of subsidization the optial appropriation satisfies: w x [' s + r a w] 0 whih iplies s + raw rw. Applying the ipliit funtion theore, this iplies that an inrease in the subsidy affets the optial degree of appropriation aording to: da ds < 0 w r ' a w Hene, appropriation falls with subsidization. In partiular, appropriation is unity in the absene of subsidization and less than unity under a positive subsidy. Put differently, sine the arginal produt of private R&D is dereasing in the level of subsidized R&D, private R&D and hene appropriation optially falls as its publi ounterpart inreases. III. Surplus Appropriation for the New HIV/AIDS Drugs The previous disussion highlighted the iportane of surplus appropriation by innovators and hene low levels of CE to dynai effiieny, even if that level of appropriation was not full. As the ability of innovators to appropriate the potential surplus of their innovations is entral to dynai effiieny, we illustrate the degree of appropriation for the new drugs to treat HIV/AIDS that entered the arket fro the late 980s onwards. This analysis will then be used to illustrate how ore generally levels of innovator appropriation ay be inferred fro existing CE estiates in the literature. A. Estiates of Gross Consuer Benefits The value of life indued by new drug therapies is the value of inreased survival for all individuals who hoose treatent, relative to a benhark in whih no or worse therapies exist. In a related work, we develop a ethodology to value the inreases in survival attributable to the now standard treatents for HIV/AIDS. The thought experient behind the analysis is the following. For a hypothetial individual infeted in a given year t, we exaine how that individual s survival under treatent S t opares to a ounterfatual, baseline survival in whih no drugs are available S o. 3 We then attah a onetary value to that inreased survival and su aross all infeted individuals in that ohort. This proess is repeated for eah set of ases, ohort by ohort, sine the start of the epidei and aggregated up. This approah delivers the potential aggregate value of life indued by treatent, i.e. the value of life obtained when all infeted individuals reeive treatent. This ipliitly assues that at the opetitive output, all individuals infeted with HIV i.e. the full inidene onsue drug therapy. At the opetitive output, the ex-post soial surplus arising fro a tehnology is highest and therefore represents the potential surplus available for appropriation. This differs 3 The ideal ounterfatual survival in the absene of treatent S o is the ross-setional survival of individuals infeted at the start of the epidei, here taken to be 980. S t is the longitudinal i.e., lifetie survival of individuals infeted in year t. The use of longitudinal survival aptures the benefit to individuals infeted with HIV prior to 987 when drug therapy first beae available who survive until then and onseuently fae iproved life-expetany due to treatent. 8

10 fro the observed gross value of life indued by treatent, whih depends on the nuber of individuals who atually reeive treatent. The aggregate potential gross onsuer benefit, g, indued by the new drug onsuption is alulated by ultiplying the size or inidene of ohort t, n t, by the onetary value of inreased survival and suing over all alendar years. Moreover, the aggregate observed onsuer benefit an be alulated by replaing n t with the nuber of individuals in that ohort who atually reeive treatent. Forally, the aggregate potential gross onsuer benefit is written as: 2000 t980 g β nt gt 2 t 980 where g t gs o, S t, or the onetary value of inreasing survival fro the baseline survival S o to the higher future survival faed by ohort t, S t. The gain in survival, g t, is alulated using the infra-arginal valuation forula of Beker et al To epirially ipleent the alulation of aggregate potential onsuer benefits, we apply these forulas to published levels of HIV inidene and estiated hanges in survival indued by HIV/AIDS drugs see Philipson and Jena 2005 for a detailed disussion of the various data soures and ethods used to estiate iproveents in survival. 5 We estiate that the average individual infeted with HIV experiened an inrease in life-expetany of roughly 5 years sine the start of the epidei, fro 9 to 34 years. 6 These iproveents in survival are due to inreases in both the tie to onset of AIDS after being infeted with HIV and the period of tie alive after a diagnosis of AIDS. As desribed earlier, the potential value of iproved survival for a given ohort is oputed by ultiplying that ohort s inidene of HIV by the value of inreased survival experiened by a single individual reeiving treatent in that ohort. Table presents this value for several ohorts. 4 The value of an infra-arginal hange in survival fro S o to S t under a yearly inoe y t is deterined by V[y t + e t, S o ] V[y t, S t ], where V is the indiret lifetie utility funtion and e t is the yearly opensation reuired to ake the hypothetial individual indifferent between the two survival frontiers. The lifetie value for the gain in survival g t is alulated by suing the yearly opensation e t over tie, disounting by the rate of interest and the new survival probability. 5 Iportantly, our estiated survival urves are weighted averages aross individuals reeiving and not reeiving treatent. Thus, for eah ohort, the reported survival is lower than survival aong only those reeiving treatent, S t. We assue our reported urves to be epirial analogs for S t and therefore underestiate the aggregate observed and potential benefits of treatents introdued to date. 6 These figures are onsistent with those in the literature see e.g. Lihtenberg, To further exaine the robustness of these estiates, we predit the nuber of individuals alive with HIV/AIDS in 2003, based on the annual reported inidene of HIV and our estiated survival urves. We then opare this to the reported nuber of individuals living with HIV/AIDS in The predited and reported figures differ by only 2,000 people out of nearly illion alive with HIV/AIDS. 9

11 Table : Value of Gains in Survival for HIV Infeted Individuals Value of Survival Gains $ Year of HIV Individual Aggregate HIV Inidene $ $ Billion Infetion ,000 7, ,000 46, , , ,000 63, , , Total Disounted Value 398 All figures are disounted to 980 and are in year 2000 dollars. The total disounted value inludes all years fro 980 to These results deonstrate that the aggregate potential value of iproved survival experiened by all individuals infeted with HIV to date has been nearly $400 billion. This, of ourse, ignores the value of inreasing survival for all individuals who have not ontrated HIV yet. To add this oponent, we foreast the value to future ohorts of HIV infeted individuals by assuing that all ohorts experiene the sae aggregate gain in survival g t as the last ohort, Assuing that the future inidene of HIV is euivalent to the last period, we alulate the disounted su of future gains for individuals infeted with HIV in the future. We then add this aount to the value to date shown above, naely $398 billion. This leads to an aggregate potential value of inreased survival for all past and future ohorts of nearly $.4 trillion. The aggregate observed value of iproved survival an be alulated fro data on the annual nuber of individuals in a given ohort who later reeive HIV/AIDS drug therapy. Speifially, if S t haraterizes the survival of soeone infeted with HIV in year t who reeives treatent, the aggregate observed onsuer benefit for that ohort euals the nuber of individuals in that ohort who reeive treatent ultiplied by the value of inreased survival, gs o, S t. We use published data fro the HIV Cost and Servies Utilization Study HCSUS to approxiate the nuber of individuals in a given ohort reeiving treatent. Aording to the study, between 60 and 85 perent of HIV infeted individuals report soe for of anti-retroviral therapy Shapiro, et al., Assuing a id-range estiate of 70%, this iplies an inidene of anti-retroviral onsuers eual to 70% of HIV inidene and a orresponding aggregate observed benefit to all past and future ohorts of $980 billion i.e., $.4 trillion*0.7. B. Produer Surplus The overall produer surplus obtained fro R&D is deterined by the present value of produer surplus to firs produing HIV/AIDS drugs. We apply existing estiates of arkups for brandnae drugs as estiated fro patent expirations to approxiate variable osts as 5% of sales. Using estiates of national spending on HIV/AIDS drugs obtained fro IMS Health and reported 7 One ould use annual data on aggregate sales and arket pries to estiate the nuber of individuals reeiving treatent. However, sine the annual arket uantity is oposed of users fro all infeted ohorts, additional assuptions ust be ade to infer the share of individuals in a given infetion-ohort who ultiately reeive drug therapy. 0

12 by Lihtenberg 2005, this iplies lifetie sales of HIV/AIDS drugs of roughly $74 billion with orresponding profits variable osts of $63 billion $ billion. These estiates assue future sales are euivalent to year 2000 s patent-proteted sales. If the urrent level of output is 70% of the opetitive level, the variable ost of prodution is $6 billion when all infeted individuals are treated. Using the above figures, we an deopose the total lifetie value of HIV/AIDS drugs into onsuer surplus, produer surplus profits, and prodution osts. Reall that we estiated the total potential value, g, to be nearly $.4 trillion, disounted to 980 and in year 2000 dollars. This is the value that arues when all infeted individuals reeive treatent and iplies a potential soial surplus of nearly $.38 trillion $.4 trillion - $6 billion. With lifetie profits of $62.9 billion, produers appropriate only 5% of the potential soial surplus available fro their inventions 8. We an ake siilar alulations regarding the share of observed surplus aptured by produers. Reall that the observed surplus is driven by the observed arket uantity rather than the opetitive uantity whih results if all infeted individuals reeive treatent. With observed gross benefits of $980 billion, profits of $62.9 billion, and variable osts of $. billion, onsuers apture 7% of the soial surplus available at the observed arket uantities onsuer surplus $906 billion, soial surplus $969 billion. 9 IV. Extending the Analysis to Traditional Cost-Effetiveness Studies Given the low estiated share of soial surplus appropriated by produers of HIV/AIDS drugs, this raises the uestion of whether produers of siilarly CE tehnologies appropriate oparable aounts of soial surplus. And if so, an these results be generalized to obtain appropriation estiates that vary with a tehnology s observed level of ost-effetiveness? We begin this setion by disussing onditions under whih the often estiated CE of a given tehnology ay, in fat, be used to infer the share of soial surplus appropriated by produers of that tehnology. Reall fro our earlier disussion that the observed ratio of gross benefit to spending, z R, an be written as g/[p]. Siilarly, the degree of observed surplus appropriation an be written as /w, where w is the observed soial surplus assoiated with a level of output, and is the level of profit indued by that uantity. If p/[/] is the arkup above average osts, it is straightforward to show that for a given level of output, appropriation ay be written as a funtion of the CE level as in 20 : 8 It is interesting to note that the sall estiated share of soial surplus appropriated by investors sheds iportant light on the reent growth of alternative funding ehaniss to stiulate HIV/AIDS researh, e.g. through advane purhasing ontrats of governents or private foundations. Given that there is a soial surplus above a trillion dollars that is not appropriated by R&D investors, a few billion dollars added to stiulate innovation, as these publi or private ontrats see to provide, sees to pale in oparison to interventions that would better allow innovators to apture the value of their innovations. Moreover, sine both spending and arkups are higher in the US than in the rest of the world drug sales in the US aount for ore than half of worldwide spending and prie ontrols doinate foreign arkets, estiates of appropriation based on US arkets alone over-estiate worldwide appropriation. 9 Our estiated result on appropriation is onsistent with observed expenditures on HIV/AIDS drugs. For exaple, Lihtenberg 2005 estiates annual drug spending per HIV infeted individual in 200 to be $9,75 in 200 dollars, iplying a net profit of approxiately $8,300 for that individual assuing variable osts are 5% of spending, i.e. $,400. A value of a life year of $00,000 is onsistent with existing estiates and iplies an observed surplus appropriation of roughly 8% e.g. $8,300/$00,000 - $, To see this, note that p and w g. Substituting z R g/[p] into the expression for w and siplifying the appropriation share /w yields the above result.

13 π w z R 3 This expression deonstrates that highly ost-effetive tehnologies those with high z R ipliitly support low levels of observed surplus appropriation. Moreover, when free-entry is possible and firs earn zero profits prie average ost, surplus appropriation is zero. The general point, then, is that with inforation on the degree of arket power in an industry, one an use oonly reported CE estiates to infer the degree of appropriation by produers of the relevant tehnology. For the ase of HIV/AIDS, alulating the appropriation ratio based on our estiates is straightforward. First, reall that we estiated observed gross benefits to onsuers to be nearly $980 billion with spending levels of $74 billion. This iplies a CE or z R ratio of roughly 3. Estiates of the average arkup an, in turn, be obtained fro inforation on prie redutions after patent expiration, whih suggests that average osts are as low as 5% of patented pries. Put together, the average arkup and estiated CE of HIV/AIDS drugs iply a produer appropriation of observed soial surplus of 7%, idential to our diretly estiated level of appropriation. With ore restritive ost and deand assuptions, even less inforation is needed to infer the level of appropriation fro CE estiates. Under onstant returns to sale and onstant elastiity deand, it an be shown that a tehnology s CE alone identifies its elastiity of deand, whih in turns identifies the share of surplus appropriated by the produers of that tehnology. These assuptions also allow us to distinguish between appropriation of two types of surpluses disussed earlier, observed versus potential. To be sure, the observed surplus presented in euation 3 is the surplus whih obtains at the arket uantity. For exaple, for a onopoly uantity, the appropriation of observed surplus is siply /w. Alternatively, the potential surplus is that whih results if the arket uantity is deterined opetitively and hene relates to the total potential surplus available to an innovator. Iportantly, the size of profits relative to the potential soial surplus is ost relevant to dynai poliy. For a onopoly uantity, the appropriation of potential surplus is /w. Sine there is a deadweight loss assoiated with onopoly priing, the potential surplus fro an innovation exeeds the observed surplus. Conseuently, estiates of surplus appropriation based on observed surplus will underestiate the defiieny in appropriation by produers of a given tehnology. More preisely, onsider the oon odel where variable osts exhibit onstant returns, y y, and there is a onstant elastiity deand urve p x/ /, where > 0 is the elastiity of deand with respet to prie and x is a sale fator that shifts deand outward. If and denote the opetitive and onopoly output, respetively, the Appendix shows that the ratio of gross benefit to spending i.e., z R under onopoly priing satisfies: z R g p p 4 In other words, a tehnology s CE, as desribed by the ratio of gross benefit to spending, is diretly related to the failiar perentage arkup of prie over arginal ost. In addition, the 2

14 share of potential surplus appropriated as profits under optial onopoly priing euals the output expansion due to opetition. 2 That is, π g z R 5 This interesting result states that, ounter-intuitively, the ore a onopolist restrits output, as perhaps estiated by patent expirations, the less of the surplus it is appropriated. 22 Note that as the elastiity approahes unity below whih profits are infinite fro above, the profits, theselves, rise but as a share of soial surplus go to zero. 23 This ours beause the nonappropriated onsuer surplus rises faster than profits as the elastiity falls. Moreover, as arket power delines and elastiity approahes infinity, the share of soial surplus appropriated as profits tends to roughly 37%. 24 Finally, there is a diret negative relationship between osteffetiveness and innovator appropriation. Under these assuptions, a given estiated CE or z R ratio iplies a speifi elastiity of deand, whih in turn iplies the degree to whih a fir appropriates soial surplus. In the ase of HIV/AIDS, for whih z R is roughly 3, the iplied elastiity of deand is around.08, whih aording to euation 6 iplies a produer share of potential soial surplus of 6%. This an be opared to our diretly estiated share of potential surplus appropriation of 5%. 25 More generally, the above relationship between CE and surplus appropriation an be used to infer the share of potential surplus appropriated by those produers whose tehnologies are exained in existing CE studies. Figure, below, graphs the relationship between surplus appropriation, ost-effetiveness, and arket power interpreted as a redution in the elastiity of deand. As arket power dereases, the produer s share of potential soial surplus approahes slightly ore than a third, while z R approahes. As desribed earlier, z R is bounded fro below by unity sine individuals only purhase goods for whih the benefits exeed the osts. 2 It is straightforward to show that the share of observed surplus appropriated by produers is -/2-, whih is greater than the potential surplus appropriated. 22 This result ay not be uniue to this partiular deand struture. For a linear deand urve, it is well known that onopoly output is half the opetitive output and that a onopolist always appropriates half the surplus, so that the surplus ondition above holds. 23 It ay even be that deand and ost paraeters do not affet the share of surplus appropriated by the produer. This is the ase when deand is linear as often estiated and there are onstant returns to sale in prodution, in whih ase the share appropriated by produers is always two thirds. 24 Note that while -/ approahes unity as elastiity beoes infinite, [-/] does not do the sae. 25 Given that the share of observed surplus appropriated by produers is -/2-, an elastiity of.08 iplied by the observed z R of 3 for HIV/AIDS suggests an observed level of appropriation of 7%, idential to our diretly estiated level. 3

15 Figure : Elastiity of Deand and Produer Shares Iplied by CE Estiates Gross Benefit/Spending Produer Share Gross Benefit/Spending Produer Share of Soial Surplus Elastiity of Deand The above figure illustrates how one an potentially use estiates of ost-effetiveness fro the large health eonoi literature to infer the share of surplus appropriated by produers of the relevant tehnology. For exaple, onsider the tehnologies to treat HIV/AIDS. With an estiated ratio of gross benefit to spending of roughly 3, this iplies an elastiity of deand of.08 and a produer share of potential soial surplus of a twentieth. We exeplify this general identifiation strategy using estiates of ost-effetiveness fro the literature. Neuann et al review the ost-effetiveness of ore than 200 pharaeutials using the established ost-utility ethod whih fouses on osts per QALY gained and therefore onern both the prolongation and uality of life. The authors note that while no aepted standards exist for how uh benefit a tehnology ust onfer to be deeed a good value, the range between $50,000 and $00,000 per QALY has been a benhark for the US. In the ontext of our fraework, this value or range is the gross benefit to onsuers of a tehnology whih leads to an additional uality adjusted year of life. Table 2 presents the spending reuired to obtain an additional QALY for several interventions reviewed by the authors. For exaple, an intervention with a prie of $,000 that leads to an inrease in 0.2 QALYs reuires the sae spending per QALY as an intervention with a prie of $5,000 that leads to an additional QALY. While the agnitude of gross benefit differs aross the two interventions, the gross benefit per QALY is the sae naely in the range desribed above. Thus, assuing the gross benefit arising fro an additional uality adjusted year of life is between $50,000 and $00,000, we an opute estiates of the ratio of gross benefit to spending per QALY for these interventions, as well as the iplied shares of soial surplus appropriated by produers. 4

16 Table 2: Estiated Produer Share of Potential Soial Surplus for Several Cost-Effetive Tehnologies Intervention Spending per QALY $ z R Gross Benefit/Spending Produer Share of Surplus $50,000 $00,000 $50,000 $00,000 Captopril Therapy 4, Horone Replaeent 2, Therapy INH Prophylaxis 8, Hip Frature Prevention 34, Cheotherapy for Breast Caner 58, Notes: CE and produer share of surplus are presented for two, separate values of an additional uality adjusted life year. The final intervention has a gross benefit less than ost when gross benefit per QALY euals $50,000. Desription of Interventions Captopril therapy in patients with yoardial infartion, 2 Horone replaeent therapy HRT, 3 Isoniazid INH prophylaxis for tuberulosis, 4 Treatent to prevent hip frature in patients with osteoporosis, and 5 Cheotherapy for breast aner. desription, see Neuann et al For a ore detailed Table 2 deonstrates that, as illustrated by the ase of HIV/AIDS, those tehnologies deeed to be extreely ost effetive ay also result in low surplus appropriation by produers. For exaple, the highly ost effetive Captopril therapy results in only 3% - 6% of potential soial surplus going to produers. While Table 2 presents estiates of the produer share of soial surplus for only five interventions, ost-effetiveness estiates fro a large, rando saple of interventions ould be used to estiate the distribution of produer shares. We use data fro over 200 published ostutility analyses ontained in the Harvard Cost-Effetiveness Analysis CEA Registry to estiate this distribution. 26 Inluding analyses fro 976 to 200, the Registry reports the spending per QALY of various interventions opared to benhark oparator groups. This spending per QALY an in turn be used to estiate the share of potential soial surplus appropriated by the produer of that tehnology, as in Table 2 above. 27 This an be opared to estiates of the produer s atual appropriation, identified by the tehnology s CE and average ark-up as in expression 3 above. The average ark-up is based on our earlier assuption that average osts are 5% of patented pries. Figure 2 plots the distribution of observed and potential produer shares for the interventions onsidered. Beause the studies inluded in the Registry ay not be a rando saple of all tehnologies, however, we an only estiate the distribution of produer shares onditional on inlusion into the Registry. 26 The Registry is not liited to only pharaeutial interventions. More detailed inforation an be found at: 27 For these alulations, we assue the gross benefit of an additional QALY to be $00,000. Conseuently, we liit our attention to those interventions with published osts of less than $00,000 per QALY gained. 5

17 Figure 2: Cuulative Distribution of Atual and Potential Surplus Appropriation Potential 0.7 Atual Probability Median Pot Median At Appropriation of Atual and Potential Surplus Sine the onstant elastiity of deand assuption predits a produer appropriation of soial surplus of no ore than 37%, all interventions onsidered in Figure 2 have estiated produer shares less than this aount. The edian intervention reuires a spending per QALY of roughly $9,000, whih orresponds to a produer share of potential atual soial surplus of nearly 3% 7%. 28 Moreover, 25% of the interventions onsidered have estiated potential appropriations of less than 7%, while 75% have appropriations less than a fourth. Moreover, 75% of the interventions have an atual appropriation of less than 40%. If the estiated distribution of produer shares generalizes to the distribution aross all health interventions i.e., not only those inluded in the Registry, our epirial finding for produers of HIV/AIDS drugs suggests their appropriation of potential soial surplus is at the twentieth perentile. Our estiates of appropriation an be opared to alternative, theory-based ethods of alulating this share. Speifially, given the previously desribed relationship between the elastiity of deand and the share of potential soial surplus appropriated by innovators, one an use inforation on prie redutions after patent expiration to estiate patent-proteted arkups Caves et al These arkups identify the elastiity of deand for the patent-proteted drugs and thus the share of surplus alloated to the produer. In partiular, the larger is the prie redution upon patent-expiration, the lower is the elastiity and the saller is the share of surplus alloated to the produer. Existing estiates suggest that prie redutions are on the agnitude of 85% perent, iplying a deand-elastiity around.7. This elastiity iplies a produer share of soial surplus of 0%, whih is highly related to our ajor finding that the share of 28 If the gross benefit of an additional QALY is assued to be $50,000 rather than $00,000, the edian intervention has an iplied produer share of soial surplus loser to 20%. 29 Caves, Whinston, and Hurwitz 99 estiate that with 20 generi opetitors, the ratio of pries between generi- and brand-drugs is roughly 7%. We use the prie of generi drugs as an upper bound of the arginal osts of prodution. 6

18 soial surplus appropriated by R&D investors in this area is, in fat, uite low. This is true even though pries for these drugs are high, presuably due to the inelasti nature of deand. V. Conluding Rearks and Future Researh We argued that popular tehnology assessent riteria in health are going under the rubri of ost-effetiveness are often ipliitly onerned with axiizing the observed level of onsuer surplus, whih is any ties onsistent with axiizing stati effiieny after an innovation has been developed. Dynai effiieny, however, aligns the soial osts and benefits of R&D and is therefore deterined by the how uh of the soial surplus fro a new tehnology is appropriated by innovators. For the ase of HIV/AIDS, our earlier estiates suggested that produers appropriated only 5% of the soial surplus arising fro new drug therapies. Given the low degree of appropriation by produers of the highly ost-effetive HIV/AIDS therapies, we showed how other CE estiates in the literature ould be related to the standard fraework our ain finding was that these CE estiates ipliitly support a low degree of surplus appropriation by produers, oparable to our diretly easured estiates for HIV/AIDS. Despite the high annual osts of these drugs to patients, the low share of soial surplus going to innovators raises onerns about advoating ost-effetiveness riteria that would further redue appropriation share, and hene further redue dynai effiieny. In addressing why produer surplus is so sall, and why the CE of therapies ay be ineffiiently high, it sees natural to suggest that this is ultiately due to a lak of profits and arket power. This is potentially due to pries being held down by: the threat of publi regulation if pharaeutial opanies raise pries, or 2 patents that are weakly enfored or too narrowly defined to allow patent-proteted onopolies to raise prie appropriately. However, even with free priing and nearly-inelasti deand, the share of soial surplus alloated to produers ay be sall. This is ost easily illustrated by the onstant elastiity ase, in whih a produer share of soial surplus of 5% is onsistent with onopoly priing under a deand urve that is alost as inelasti as it an be,.08, as an elastiity below unity, of ourse, leads to infinite profits. In fat, we showed that higher pries suh as those of HIV/AIDS drugs indued by lower elastiities of deand often lead to less surplus aptured by inventors. Put differently, even though profits, of ourse, rise as the elastiity of deand falls, any ties the share of soial surplus appropriated by the onopolist falls. This ours beause the non-appropriated onsuer surplus rises faster than profits as the elastiity falls. It is iportant to stress that arguents about the differene between stati and dynai effiieny are a different atter than whether pries used for alulating spending in CE analysis reflet osts of prodution in general, and average osts of prodution refleting R&D osts, in partiular. Under traditional CE analysis, even if one ould easure osts perfetly, and did not need to approxiate unobserved osts by observed pries, one would be onerned with the wrong easure, total ex-post surplus. This is beause the division of the surplus is what atters for dynai R&D poliy, as opposed to only the total surplus whih is relevant for stati poliy. In partiular, this holds true whether the osts represented are arginal or average osts, the latter potentially inluding fixed osts suh as R&D. In both ases, the division of soial surplus is ignored but is what drives optial R&D poliy. Several issues ay be iportant in generalizing our onlusions and are therefore suitable for future researh. The first onerns the interpretation of CE analysis in a nononopoly ontext; the field of industrial organization of tehnology adoption needs to be better understood. Another onern is the effet of altruis, whih seeingly otivates uh of publi 7