`
`M a r k e t Watc h
`Estimating The Cost Of New Drug Development:
`Is It Really $802 Million?
`Variations in cost estimates suggest that policymakers should not use
`a single number to characterize drug costs.
`
`by Christopher P. Adams and Van V. Brantner
`
`ABSTRACT: This paper replicates the drug development cost estimates of Joseph DiMasi
`and colleagues (“The Price of Innovation”), using their published cost estimates along with
`information on success rates and durations from a publicly available data set. For drugs en-
`tering human clinical trials for the first time between 1989 and 2002, the paper estimated
`the cost per new drug to be $868 million. However, our estimates vary from around $500
`million to more than $2,000 million, depending on the therapy or the developing firm.
`[Health Affairs 25, no. 2 (2006): 420–428; 10.1377/hlthaff.25.2.420]
`
`Th e e x p e c t e d c o s t of developing an
`
`average drug was recently estimated by
`Joseph DiMasi and colleagues at $802
`million per new molecular entity (in 2000
`dollars).1 The enormous cost of drug develop-
`ment is a key component of the current de-
`bates over prescription drug prices, importa-
`tion of drugs from Canada, Food and Drug
`Administration (FDA) review policies, and
`barriers to generic entry. Given the central
`role of the $802 million estimate in these de-
`bates, it is important to ask two questions.
`First, is this number an accurate estimate of
`the expected cost of developing an average
`drug? Second, even if it is accurate, what does
`the estimate mean?
`This paper independently verifies DiMasi
`and colleagues’ estimate, in “The Price of Inno-
`vation” (hereafter, DHG), using a publicly
`available data set on drug development. Our
`analysis also raises several issues that must be
`accounted for in interpreting the $802 million
`as a meaningful measure of actual drug devel-
`
`opment costs: the meaning of “average drug,”
`the impact of firms’ strategic decisions, and
`regulatory policies’ effects on development
`costs.
`Study Methods
`n DHG methodology. DiMasi and col-
`leagues took three steps to reach their $802
`million estimate. First, they randomly selected
`sixty-eight drugs from the proprietary Tufts
`Center for the Study of Drug Development
`(CSDD) database of investigational com-
`pounds for ten multinational pharmaceutical
`firms participating in a confidential survey.
`These survey data provide the average cost of
`taking a drug through each step of the drug
`development process. This is the actual money
`that the drug companies spent on the process.
`Second, they used the CSDD database to
`calculate the probability that the average drug
`will get to each phase. By multiplying the esti-
`mated average amount spent in each phase by
`the probability of getting to the phase, they
`calculated the expected cost of developing a
`
`Christopher Adams (cadams@ftc.gov) is an economist and Van Brantner is a research analyst at the Bureau of
`Economics, Federal Trade Commission, in Washington, D.C.
`
`4 2 0
`
`M a r c h / A p r i l 2 0 0 6
`
`DOI 10.1377/hlthaff.25.2.420 ©2006 Project HOPE–The People-to-People Health Foundation, Inc.
`
`Petitioner Mylan Pharmaceuticals Inc. - Exhibit 1067 - Page 1
`
`
`
`drug for market. The authors then used the
`CSDD database to estimate the probability
`that a drug in Phase I would be approved and
`used this number to calculate the expected
`cost per approved drug.
`Third, the authors used the CSDD database
`to estimate the average duration for each stage
`in the drug development process. These dura-
`tions were then used to estimate the time cost
`or opportunity cost of developing a drug.
`n Our methodology. We estimated the
`expected cost of developing an approved drug
`in the same way. However, instead of using es-
`timates from the proprietary CSDD database,
`we used estimates from the publicly available
`Pharmaprojects database. This allows others
`to verify our results. An important concern is
`that the data are likely to be less accurate than
`the survey data used to compile the CSDD da-
`tabase. The Pharmaprojects data are collected
`by the vendor (PJB Publications) based on
`press releases, academic presentations, and
`other public information about drugs in devel-
`opment. Because of this collection process, the
`data do not always include information on
`drugs in the earlier stages of human clinical
`trials. Although we have some concern about
`accuracy, we have no reason to believe that the
`data are biased.
`To estimate the cost of developing drugs
`with different characteristics, we assumed
`that the average actual cost is the same across
`different drug characteristics. That is to say,
`the estimated variation in costs across drugs
`with different characteristics is attributable to
`differences in the estimated probability of suc-
`cess and in the estimated duration. It is impor-
`tant to also be aware that different drug types
`might have substantially different actual costs
`of clinical trials. Therefore, the estimated vari-
`ation in drug costs could be higher or lower,
`depending on whether the correlation be-
`tween actual costs, success probabilities, and
`durations is positive or negative. As discussed
`below, recent work suggests that HIV/AIDS
`drugs have high clinical costs, which may off-
`set cost reductions reported in this paper.2
`There is some controversy over how
`DiMasi and colleagues calculated their cost
`
`M a r k e t W a t c h
`
`numbers, including the use of before-tax in-
`come and different discount rates. (See the au-
`thors’ discussion of the issues and the refer-
`ences therein for more detail.) For this paper,
`we followed the DHG calculations.
`Study Data
`The data used in our study contain infor-
`mation updated monthly on drugs in a late
`stage of development, covering 1989 to the
`present, and include drugs now in develop-
`ment and those that have been discontinued or
`withdrawn from the process.3 The recorded in-
`formation includes the drug’s current status,
`the original materials, the primary therapy, the
`primary indication and other indications,
`route of administration, and the name of the
`developing firm. It also includes major event
`dates in the life of the drug, such as entry dates
`in each of the phases, as well as exit and regis-
`tration dates, when applicable. For this study,
`we limited our attention to all drugs that went
`into human clinical trials for the first time be-
`tween 1989 and 2002 and for which we have an
`entry date and at least one additional piece of
`information after entry.
`n Concern about dates. There is some
`concern about the dates available from the
`Pharmaprojects database. In particular, the
`date is often only accurate to a particular
`month. We have discussed these issues with
`the vendor, and we are confident that every ef-
`fort has been made to publish accurate dates.
`We know of no evidence that suggests that
`these dates are systematically misreported. In
`fact, we have found that statistics based on
`this database are consistent with other pub-
`licly reported statistics from other databases.
`n CSDD versus Pharmaprojects. Al-
`though both the CSDD and Pharmaprojects
`databases purport to include detailed informa-
`tion about each drug’s development mile-
`stones, there are important differences.4 The
`drugs used in the DHG analysis are all new
`molecular entities (NMEs). To obtain a sample
`of drugs that is closer to that used in the DHG
`analysis, we dropped drugs that were indi-
`cated in the database as being new formula-
`tions of previously approved drugs. The CSDD
`
`H E A LT H A F F A I R S ~ Vo l u m e 2 5 , N u m b e r 2
`
`4 2 1
`
`Petitioner Mylan Pharmaceuticals Inc. - Exhibit 1067 - Page 2
`
`
`
`H e a l t h T r a c k i n g
`
`sample is limited to self-originated drugs; un-
`fortunately, the information in Pharmaprojects
`is not detailed enough to make the same re-
`striction. The drugs used in the DHG analysis
`are drugs that first entered human clinical tri-
`als somewhere in the world after 1983. Again,
`unfortunately, the information in Pharmaproj-
`ects does not allow us to select on this crite-
`rion. The data set we used includes drugs that
`first entered one of the phases of human clini-
`cal development somewhere in the world after
`1989—the first year for which Pharmaprojects
`provides detailed and easily accessible infor-
`mation on drug histories. The data selected for
`the DHG study were all first tested in humans
`prior to 1994. Because of the limitations of our
`data, we included drugs that entered any one
`of the three stages by 2002.
`Using these criteria, our data set is much
`larger than the one selected from the CSDD
`data. Our sample includes information on 3,181
`compounds, while the DHG sample has infor-
`mation on 538 compounds. It is not clear to us
`exactly which of these differences accounts for
`the discrepancy in sample sizes. Despite these
`apparent differences, the results presented
`here show that the two data sets provide a
`similar picture of success rates and durations
`for the average drug.
`
`Replicating The DHG Results
`n Development costs. Success rates cal-
`culated from the two data sets give somewhat
`similar results (Exhibit 1). Note that the suc-
`cess rates for long-term animal testing are
`taken from the DHG study. The expected cost
`is the money that the firm expects to spend on
`the drug when it enters Phase I human clinical
`trials. This is calculated by multiplying the av-
`erage amount spent on a drug in each phase by
`the probability that the drug enters that phase.
`All results use the same spending information
`(column 2), but the Pharmaprojects data set
`has higher probabilities of drugs entering
`Phase III and thus higher expected costs ($74
`million, compared with $61 million). A drug’s
`out-of-pocket expense is the amount of money
`that a company would expect to spend to get a
`drug approved for market. This number is cal-
`culated by dividing the expected cost by the
`probability that a drug in Phase I gets ap-
`proved. Our estimated out-of-pocket costs are
`higher than those of DiMasi and colleagues—
`$310 million, compared with $282 million.
`This difference is attributable to the higher es-
`timated expected costs.
`There are a few things to note about our es-
`timates. First, our phase transition probabili-
`ties were calculated by taking the drugs in
`
`EXHIBIT 1
`Average Out-Of-Pocket Clinical Costs For Investigational Compounds
`
`Survey
`
`Mean
`costa
`
`$15
`24
`86
`
`5
`
`N
`
`66
`53
`33
`
`20
`
`Testing
`phase
`
`Phase I
`Phase II
`Phase III
`
`Animal
`Preclinical
`Total
`
`Entry probability
`
`Expected costa
`
`Totala
`
`DHG
`
`100%
`71
`31
`
`31
`
`22
`
`Pharma-
`projects
`
`100%
`74
`46
`
`31
`
`24
`
`DHG
`
`$15
`17
`27
`
`2
`
`61
`
`Pharma-
`projects
`
`DHG
`
`Pharma-
`projects
`
`$15
`17
`40
`
`2
`
`74
`
`$121
`282
`
`$133
`310
`
`SOURCES: J.A. DiMasi, R.W. Hansen, and H.G. Grabowski, “The Price of Innovation: New Estimates of Drug Development
`Costs,” Journal of Health Economics 22, no. 2 (2003): 151–185 (DHG); and authors’ calculations based on Pharmaprojects
`data.
`NOTES: All survey costs were deflated using the gross domestic product (GDP) Implicit Price Deflator, and weighted values
`were used in calculating the survey means. Preclinical costs are calculated using DHG’s preclinical to total research and
`development (R&D) expenditure ratio of 30 percent.
`a Millions of 2000 dollars.
`
`4 2 2
`
`M a r c h / A p r i l 2 0 0 6
`
`Petitioner Mylan Pharmaceuticals Inc. - Exhibit 1067 - Page 3
`
`
`
`M a r k e t W a t c h
`
`Phase II, for example, that successfully moved
`to Phase III and dividing that number by the
`same number plus the number of drugs in
`Phase II for which development was discon-
`tinued. We assumed that currently active
`drugs will experience the same probabilities of
`success and duration as drug candidates
`whose projects are completed.
`Second, our estimate for successfully mov-
`ing from Phase I to approval was calculated by
`simply multiplying the phase transition proba-
`bilities together. We did it this way because
`the data set has very few drugs with complete
`information for all three phases. This proce-
`dure is less efficient than using a duration
`model to estimate the success rates of these
`drugs (the approach taken by the DHG study).
`That approach relied on the assumption that
`the censored drugs will have the same proba-
`bility of success, conditional on time in devel-
`opment, as the uncensored drugs. The ap-
`proach we used in this paper does not rely on
`this assumption; however, the estimate could
`be biased if drugs with longer durations are
`more likely to either succeed or fail.5
`n Opportunity costs. Exhibit 2 presents a
`comparison of the capitalized expected costs
`from the two data sets. The capitalized cost is
`the opportunity cost of the money used to de-
`
`velop these drugs. It is calculated by taking the
`expected costs from the previous exhibit and
`spreading the spending uniformly over the
`length of the particular phase and then assum-
`ing that the money is all “paid back” when the
`drug is approved. Note that we followed the
`DHG approach and used an 11 percent dis-
`count rate.6 The estimate for the capitalized
`expected phase costs from the Pharmaprojects
`data is higher than the CSDD estimate, around
`$116 million rather than $100 million.
`The difference is due in part to the slightly
`different method of calculating the phase du-
`rations. The CSDD data include both start and
`end dates for the phases and show that there
`are some overlaps as well as some gaps be-
`tween phases. Unfortunately, in the Pharma-
`projects data, we have only phase start dates;
`we therefore assumed that the end date is
`equal to the start date of the next phase. The
`durations in these data were calculated for
`drugs that completed each phase.7 The CSDD
`durations were calculated for self-originated
`drugs that were approved between 1992 and
`1999. We estimated that the time from a new
`drug application (NDA) to approval is 15.8
`months using data from the Orange Book
`matched to the Pharmaprojects database. This
`duration is less than the DHG estimate of 18.2
`
`EXHIBIT 2
`Average Phase Time And Clinical Capitalized Costs For Investigational Compounds
`
`Duration (months)
`
`Mean costa
`
`Expected costa
`
`Totala
`
`Testing
`phase
`
`Phase I
`Phase II
`Phase III
`
`Animal
`Preclinical
`Clinical
`
`DHG 1
`
`DHG 2
`
`12
`26
`34
`
`22
`26
`31
`
`37
`
`Pharma-
`projects
`
`19
`30
`30
`
`DHG
`
`$ 31
`42
`119
`
`10
`
`Pharma-
`projects
`
`$ 32
`40
`113
`
`10
`
`DHG
`
`$ 31
`30
`37
`
`3
`
`100
`
`Pharma-
`projects
`
`DHG
`
`Pharma-
`projects
`
`$ 32
`29
`52
`
`3
`
`116
`
`$335
`467
`
`$381
`487
`
`SOURCES: J.A. DiMasi, R.W. Hansen, and H.G. Grabowski, “The Price of Innovation: New Estimates of Drug Development
`Costs,” Journal of Health Economics 22, no. 2 (2003): 151–185 (DHG); and authors’ calculations based on Pharmaprojects
`data.
`NOTES: DHG 1 is months to phase end; DHG 2 is months to start of next phase. The DHG new drug application (NDA) approval
`phase was estimated to be 18.2 months. Costs were capitalized at an 11 percent real discount rate. Pharmaprojects
`estimates used the DHG preclinical time of 52 months. The Pharmaprojects NDA approval phase was estimated to be 15.8
`months.
`a Millions of 2000 dollars.
`
`H E A LT H A F F A I R S ~ Vo l u m e 2 5 , N u m b e r 2
`
`4 2 3
`
`Petitioner Mylan Pharmaceuticals Inc. - Exhibit 1067 - Page 4
`
`
`
`H e a l t h T r a c k i n g
`
`months.8
`n Cost comparisons. Exhibit 3 presents a
`comparison between our results and previous
`estimates of drug development costs. To the
`extent that we were able to verify the estimate
`of $802 million per approved drug using pub-
`licly available data, we did that. Indeed, our es-
`timates indicate that $802 million might be an
`underestimate. Our clinical cost estimate is
`$487 million, compared with the original esti-
`mate of $467 million. Our estimate for the total
`capitalized expected cost per approved drug is
`$868 million, which is higher than the DHG
`estimate. Note that for the preclinical cost es-
`timate, we used DiMasi and colleagues’ 2003
`estimate of fifty-two months for preclinical de-
`velopment.
`Drug Development Costs By Firm
`Exhibit 4 presents cost estimates for differ-
`ent subgroups of drugs from large pharmaceu-
`tical firms. The variation reported in this ex-
`hibit is the result of variation in measured
`success rates and durations for these firms. We
`did not observe actual differences in spending
`on drugs by firm or firm group.9 This could
`lead to an overestimation of the variation
`across firms if actual spending is correlated
`with success rates and durations.
`
`The results suggest that there is little ad-
`vantage from being large and that drug devel-
`opment costs vary greatly among large firms.
`Exhibit 4 presents results using three different
`measures of “large.” “Top 10 by 2001 income”
`are the drugs being developed by public com-
`panies whose worldwide income for 2001 was
`in the top ten for drug firms. “Top 20 by For-
`tune rank” are the drugs that were being devel-
`oped by a worldwide Fortune top twenty phar-
`maceutical firm at the start of the drug’s
`development.10 “Top 10 by drug count” are
`drugs that were in a firm ranked in the top ten
`for the largest number of drugs in development
`at the start of the drug’s development. Also,
`the drugs included for each firm (A–K) are all
`of the drugs owned by that firm as of July
`2002.
`n Impact of size. It has been argued that
`larger companies have economies of scale and
`scope in drug development that might be asso-
`ciated with lower development costs.11 One
`difficulty in measuring such an effect is that
`large firms might be associated with success-
`ful (and lower-cost) drugs, either because such
`drugs tend to earn substantial revenues or be-
`cause mergers and acquisitions lead to such
`drugs being in larger firms.12 The results sug-
`gest that this could be a problem. When an ex
`
`EXHIBIT 3
`Capitalized Preclinical, Clinical, And Total Cost Per New Drug, In Millions Of 2000
`Dollars
`Millions of dollars
`800
`
`DiMasi 2003
`Pharmaprojects
`
`Hansen 1979
`DiMasi 1991
`
`600
`
`400
`
`200
`
`0
`
`Preclinical cost
`
`Clinical cost
`
`Total cost
`
`SOURCES: R.W. Hansen, “The Pharmaceutical Development Process: Estimates of Current Development Costs and Times and
`the Effects of Regulatory Changes,” in
`ed. R.I. Chien (Lexington, Mass.: Lexington Books,
`Issues in Pharmaceutical Economics,
`1979), 151–187; J.A. DiMasi et al., “Cost of Innovation in the Pharmaceutical Industry,”
`10, no. 2
`Journal of Health Economics
`(1991): 107–142; J.A. DiMasi, R.W. Hansen, and H.G. Grabowski, “The Price of Innovation: New Estimates of Drug Development
`Costs,”
`22, no. 2 (2003): 151–185; and data from Pharmaprojects.
`Journal of Health Economics
`
`4 2 4
`
`M a r c h / A p r i l 2 0 0 6
`
`Petitioner Mylan Pharmaceuticals Inc. - Exhibit 1067 - Page 5
`
`
`
`M a r k e t W a t c h
`
`EXHIBIT 4
`Probability Of Market Entry, Durations, And Costs For New Drugs By Firm
`
`Firm
`
`N
`
`Phase II Phase III Approval Phase I Phase II Phase III Cost ($)
`
`Entry probability (%)
`
`Duration (months)
`
`Top 10 by 2001 income
`Top 20 by Fortune rank
`Top 10 by drug count
`
`679
`549
`1,055
`
`Firm A
`Firm B
`Firm C
`Firm D
`Firm E
`Firm F
`Firm G
`Firm H
`Firm I
`Firm J
`Firm K
`
`52
`53
`92
`60
`34
`62
`74
`83
`53
`62
`58
`
`70
`61
`61
`
`56
`64
`47
`62
`88
`76
`71
`50
`82
`81
`65
`
`54
`43
`44
`
`47
`27
`31
`53
`78
`59
`38
`43
`38
`30
`46
`
`29
`20
`19
`
`23
`16
`7
`20
`58
`32
`25
`15
`23
`16
`25
`
`17
`21
`18
`
`20
`17
`20
`24
`27
`17
`15
`31
`26
`22
`18
`
`19
`23
`27
`
`20
`29
`21
`22
`26
`31
`22
`28
`19
`39
`19
`
`25
`29
`28
`
`19
`35
`33
`21
`35
`30
`31
`31
`35
`36
`33
`
`687
`942
`992
`
`751
`1,032
`2,119
`977
`521
`734
`712
`1,260
`853
`1,240
`768
`
`SOURCE: Authors’ calculations.
`NOTES: Phases are for human clinical trials. New drug application (NDA) durations are as for the average drug. Cost is the total
`expected capitalized cost per new drug (in millions of 2000 dollars).
`
`post measure of size (Top 10 by 2001 income) is
`used, the average drug from a large firm has a
`cost much lower than the overall average.
`However, when ex ante measures of size are
`used, the cost of the average drug from a large
`firm is larger than the cost for the overall aver-
`age drug. These results do not support the
`claim that larger firms tend to produce lower-
`cost drugs. Drugs from firms that had the larg-
`est number of drugs in development had an av-
`erage capitalized cost of $992 million—some
`$124 million more than the average drug.
`n Comparisons with previous work.
`These results contrast somewhat with previ-
`ous work that found that drugs from small
`firms tend to have higher costs than drugs
`from larger firms. 13 DiMasi and Henry
`Grabowski found in 1995 that this difference
`was the result of high preclinical spending and
`longer durations for drugs from small firms.
`One difference is that we did not account for
`the mixture of drugs by therapeutic category
`among firms. Nor did we account for differ-
`ences in actual spending by firm group. An-
`other explanation is that our study is more re-
`cent, and contract research organizations
`might have leveled the playing field between
`large and small firms.14
`
`n Variation within drug groups. Exhibit
`4 also presents average costs for drugs owned
`by one of eleven large drug firms. The results
`suggest that there is much variation in devel-
`opment costs even within the group of drugs
`from large firms. For example, Firm C had
`ninety-two drugs in development during this
`period, with an average expected capitalized
`cost of $2,119 million, while Firm E had thirty-
`four drugs in development, with an average
`cost of $521 million—close to one-quarter of
`the cost of drugs developed by Firm C. Note
`that the probability that a drug from Firm C
`goes from Phase I to market is only 7 percent,
`whereas that probability for a drug from Firm
`E is 58 percent. As stated above, Firms A and B
`have almost the same number of drugs in de-
`velopment, yet their costs are $751 million and
`$1,032 million, respectively.
`n Role of strategic choice. This variation
`highlights an important issue in interpreting
`cost data. These costs are not completely exog-
`enously determined; rather, these cost esti-
`mates are based on data that are the result of
`strategic behavior by the firms themselves.
`Therefore, although some of this variation is
`the result of luck or specialization in particular
`therapeutic categories, some might be the re-
`
`H E A LT H A F F A I R S ~ Vo l u m e 2 5 , N u m b e r 2
`
`4 2 5
`
`Petitioner Mylan Pharmaceuticals Inc. - Exhibit 1067 - Page 6
`
`
`
`H e a l t h T r a c k i n g
`
`sult of strategic choice. Firms may choose a
`high-risk (high-cost)/high-return strategy or a
`low-risk (low-cost)/low-return strategy.15
`Development Costs By Therapy
`Exhibit 5 presents the capitalized cost per
`drug by primary disorder for all of the major
`disorders and for the primary indication for
`some of the major indications.16 Again, the ob-
`served difference in development costs be-
`tween disorders is attributable to observed
`differences in success rates and durations. We
`did not observe differences in actual spending
`by disorder or primary indication.17 Correla-
`tion between actual spending and observed
`success rates and durations by disorder can ei-
`ther exacerbate or reduce the variation in de-
`velopment costs across therapies.18
`The exhibit shows that there is much varia-
`tion in phase transitions and success rates.
`Note that although low transition probabili-
`ties reduce the expected cost of a drug, low
`success rates increase its cost. A little algebra
`shows that the second effect always outweighs
`
`the first.19 We see that drugs in development
`for respiratory disorders such as asthma have
`very low success rates (16 percent), whereas
`drugs in development for genitourinary disor-
`ders, which include drugs such as Viagra, have
`much higher success rates.
`Exhibit 5 also shows much variation across
`major indications. Drugs designed to treat re-
`spiratory disorders such as asthma or chronic
`obstructive pulmonary disease (COPD) have
`an expected capitalized cost per approved
`drug of $1,134 million, while drugs designed to
`treat genitourinary disorders have an expected
`capitalized cost per approved drug of $635
`million. Some of this variation could be attrib-
`utable in part to decisions by the drug firm
`based on the drug’s likely revenue. For exam-
`ple, rheumatoid arthritis drugs have a very
`high cost of development and also have been
`quite successful.20
`The results also give some indication that
`regulatory policy can help to reduce develop-
`ment costs. Exhibit 5 shows that the short
`Phase III durations for HIV/AIDS drugs are as-
`
`EXHIBIT 5
`Probability Of Market Entry, Durations, And Costs For New Drugs, By Disorder And
`Primary Indication
`
`Disorder
`
`Blood
`Cardiovascular
`Dermatological
`Genitourinary
`HIV/AIDS
`Cancer
`Musculoskeletal
`Neurological
`Antiparasitic
`Respiratory
`Sensory
`
`Primary indication
`
`Alzheimer’s disease
`Rheumatoid arthritis
`Asthma
`Breast cancer
`HIV/AIDS
`
`N
`
`163
`280
`122
`120
`108
`681
`134
`192
`20
`165
`53
`
`46
`51
`74
`54
`89
`
`Entry probability (%)
`
`Duration (months)
`
`Phase II
`
`Phase III Approval Phase I
`
`Phase II
`
`Phase III Cost ($)
`
`60
`69
`84
`92
`75
`78
`73
`73
`100
`68
`88
`
`65
`91
`81
`96
`83
`
`57
`42
`44
`58
`50
`46
`41
`47
`67
`31
`60
`
`46
`33
`36
`58
`56
`
`25
`22
`29
`37
`36
`20
`22
`22
`53
`16
`40
`
`25
`23
`26
`44
`44
`
`18
`14
`13
`21
`19
`21
`19
`20
`18
`18
`11
`
`17
`18
`18
`17
`22
`
`32
`35
`29
`28
`23
`30
`39
`39
`33
`30
`44
`
`37
`36
`33
`37
`22
`
`33
`30
`24
`25
`19
`29
`30
`32
`13
`36
`30
`
`18
`39
`31
`37
`19
`
`906
`887
`677
`635
`540
`1,042
`946
`1,016
`454
`1,134
`648
`
`903
`936
`740
`610
`479
`
`SOURCE: Authors’ calculations.
`NOTES: Phases are for human clinical trials. New drug application (NDA) durations are as for the average drug. Cost is the total
`expected capitalized cost per new drug (in millions of 2000 dollars).
`
`4 2 6
`
`M a r c h / A p r i l 2 0 0 6
`
`Petitioner Mylan Pharmaceuticals Inc. - Exhibit 1067 - Page 7
`
`
`
`sociated with lower capitalized costs for those
`drugs.21 Almost all AIDS drugs were allowed
`to file NDAs without completing large-scale
`human clinical trials. Our results for these
`drugs contrast with the results presented in a
`recent extension of the DHG analysis.22 That
`analysis found that HIV/AIDS drugs have quite
`high clinical costs and anti-infectives (of
`which HIV/AIDS drugs are a part) have some-
`what higher-than-average expected capital-
`ized clinical costs. Another issue is that a siz-
`able proportion of HIV/AIDS dr ugs’
`development costs was moved from the preap-
`proval clinical trials to the required postap-
`proval studies.23
`Discussion
`n Variation by drug type. The results pre-
`sented here suggest that there is considerable
`variation in the estimated cost of developing
`different drugs. The estimated expected cost
`of developing an HIV/AIDS drug is $479 mil-
`lion, while the expected cost of developing a
`rheumatoid arthritis drug is $936 million.
`DiMasi and colleagues similarly found large
`variation in the estimated expected develop-
`ment costs.24 Using the same data as in their
`original 1991 study, they found that capitalized
`clinical costs per approved drug were 25 per-
`cent below the average for anti-infectives
`(such as penicillin) and 75 percent above the
`average for nonsteroidal anti-inflammatory
`drugs (NSAIDs, such as Celebrex).25 Their
`more recent work reports variations from 13
`percent above the average to 20 percent below
`the average. These estimated differences imply
`that different therapies might have different
`costs. For example, anticancer drugs have
`much higher expected durations, implying
`higher development costs.
`n Other factors affecting the esti-
`mates. Another issue is that these estimates
`are based on observed success rates and dura-
`tions of actual drugs. The concern is that these
`numbers are affected by many factors, includ-
`ing factors under the control of the firms de-
`veloping the drugs. This fact makes it difficult
`to determine the extent to which these high
`measured costs really impede new drug devel-
`
`M a r k e t W a t c h
`
`opment or reduce drug companies’ incentives
`to develop new drugs or types of new drugs.
`The results show that for one large pharma-
`ceutical firm, the expected cost of developing a
`drug is $521 million, while for another large
`firm, it is $2,119 million. This difference sug-
`gests that some of the estimated costs could be
`attributable to the strategic decisions of the
`drug firms themselves.
`n Impact of regulatory policies. The es-
`timated cost of developing HIV/AIDS drugs
`suggests that regulatory policy can also have a
`substantive effect on the cost of drug develop-
`ment. In particular, the low cost estimates of
`developing HIV/AIDS drugs seem to be in
`some part the result of the short durations for
`these drugs, which is in part attributable to
`FDA policy regarding review of these drugs.26
`However, as discussed above, there may be rea-
`sons to be cautious about this explanation.
`
`Re c e n t e s t i m at e s on the cost of
`
`drug development play an important
`role in the current debates on drug
`prices, regulatory policy, generic entry, and
`drug importation. This paper attempts to ver-
`ify the accuracy of the DHG estimate that the
`expected capitalized cost per approved drug
`is $802 million. Our estimate of $868 million
`suggests, if anything, that $802 million is an
`underestimate. However, we also found sub-
`stantial variation in estimated drug costs,
`which suggests that policymakers should
`take care in using a single number to charac-
`terize drug costs and that these cost numbers
`are determined by a series of factors including
`the strategic decision making of the drug
`firms themselves.
`
`This paper does not necessarily represent the views of
`the Federal Trade Commission (FTC) or any
`individual commissioner. The authors thank PJB
`Publications for providing and answering questions
`about the data; Stephen Bonventre for excellent
`research assistance; and their colleagues at the FTC for
`helpful comments and suggestions. They also thank two
`anonymous reviewers and the Health Affairs
`editorial staff. In addition, they thank Rosa Abrantes-
`Metz, Ana Aizcorbe, Ernie Berndt, Joe DiMasi, Mark
`
`H E A LT H A F F A I R S ~ Vo l u m e 2 5 , N u m b e r 2
`
`4 2 7
`
`Petitioner Mylan Pharmaceuticals Inc. - Exhibit 1067 - Page 8
`
`
`
`H e a l t h T r a c k i n g
`
`Duggan, Richard Frank, Sean Nicholson, Chris Snyder,
`as well as participants at the National Bureau of
`Economic Research (NBER) Summer Institute and the
`George Washington University and Bureau of
`Economic Analysis (BEA) seminars for helpful
`suggestions. They are particularly grateful to Bill Vogt
`for his encouragement and suggestions. All errors are
`the authors’ own.
`
`2.
`
`NOTES
`1.
`J.A. DiMasi, R.W. Hansen, and H.G. Grabowski,
`“The Price of Innovation: New Estimates of Drug
`Development Costs,” Journal of Health Economics 22,
`no. 2 (2003): 151–185.
`J.A. DiMasi et al., “Research and Development
`Costs for New Drugs by Therapeutic Category:
`A Study of the U.S. Pharmaceutical Industry,”
`PharmacoEconomics 7, no. 2 (1995): 152–169; and
`J.A. DiMasi, H.G. Grabowski, and J. Vernon,
`“R&D Costs and Returns by Therapeutic Cate-
`gory,” Drug Information Journal 38, no. 3 (2004):
`211–223.
`3. See C.P. Adams and V.V. Brantner, “New Drug
`Development: Estimating Entry from Human
`Clinical Trials,” FTC Working Paper no. 262
`(Washington: Federal Trade Commission, 2003).
`4. Pharmaprojects is supplemented with data from
`the Orange Book.
`5. R.M. Abrantes-Metz, C.P. Adams, and A.D.
`Metz, “Pharmaceutical Development Phases: A
`Duration Analysis,” Journal of Pharmaceutical Fi-
`nance, Economics, and Policy (forthcoming).
`6. See DiMasi et al., “The Price of Innovation.”
`7. See Abrantes-Metz et al., “Pharmaceutical Devel-
`opment Phases.”
`8. There is evidence that time in regulatory review
`has fallen in recent years, particularly after 1995.
`Ibid. E.R. Berndt et al., “Industry Funding of the
`FDA: Effects of PDUFA on Approval Times and
`Withdrawal Rates,” Nature Reviews: Drug Discovery
`4, no. 7 (2004): 545–554, estimates this duration
`as having fallen from 24.2 months to 14.2 months
`between 1992 and 2002. However, our shorter
`estimates may be attributable to censoring bias
`toward observing completed durations for
`quicker drugs.
`J.A. DiMasi and H.G. Grabowski, “R&D Costs,
`Innovative Output, and Firm Size in the Pharma-
`ceutical Industry,” International Journal of the Eco-
`nomics of Business 2, no. 2 (1995): 201–221, presents
`variation in costs of development by groups of
`firms where the authors also measure variation in
`actual expenditure by firm.
`10. The date used is the first date we have for the
`
`9.
`
`drug’s human clinical trials and could be from
`any phase.
`11. R. Henderson and I.M. Cockburn, “Scale, Scope,
`and Spillovers: The Determinants of Research
`Productivity in the Pharmaceutical Industry,”
`RAND Journal of Economics 27, no. 1 (1996): 32–59.
`12. P.M. Danzon, A.J. Epstein, and S. Nicholson,
`“Mergers and Acquisitions in the Pharmaceutical
`and Biotech Industries,” NBER Working Paper
`no. 10536 (Cambridge, Mass.: National Bureau of
`Economic Research, 2004).
`13. DiMasi and Grabowski, “R&D Costs.”
`14. Thanks to an anonymous reviewer for pointing
`this out.
`15. See DiMasi and Grabowski, “R&D Costs.”
`16. The categorizations a