Research

JAMA | Original Investigation

Estimated Research and Development Investment Needed

to Bring a New Medicine to Market, 2009-2018
Olivier J. Wouters, PhD; Martin McKee, MD, DSc; Jeroen Luyten, PhD

Editorial pages 826, 829, and

IMPORTANCE The mean cost of developing a new drug has been the subject of debate, 831
with recent estimates ranging from $314 million to $2.8 billion. Supplemental content

OBJECTIVE To estimate the research and development investment required to bring a new CME Quiz at
therapeutic agent to market, using publicly available data. jamacmelookup.com

DESIGN AND SETTING Data were analyzed on new therapeutic agents approved by the US Related articles at
Food and Drug Administration (FDA) between 2009 and 2018 to estimate the research and jamainternalmedicine.com
development expenditure required to bring a new medicine to market. Data were accessed
from the US Securities and Exchange Commission, Drugs@FDA database, and
ClinicalTrials.gov, alongside published data on clinical trial success rates.

EXPOSURES Conduct of preclinical and clinical studies of new therapeutic agents.

MAIN OUTCOMES AND MEASURES Median and mean research and development spending
on new therapeutic agents approved by the FDA, capitalized at a real cost of capital rate
(the required rate of return for an investor) of 10.5% per year, with bootstrapped CIs.
All amounts were reported in 2018 US dollars.

RESULTS The FDA approved 355 new drugs and biologics over the study period. Research and
development expenditures were available for 63 (18%) products, developed by 47 different
companies. After accounting for the costs of failed trials, the median capitalized research and
development investment to bring a new drug to market was estimated at $985.3 million
(95% CI, $683.6 million-$1228.9 million), and the mean investment was estimated at
$1335.9 million (95% CI, $1042.5 million-$1637.5 million) in the base case analysis.
Median estimates by therapeutic area (for areas with ⱖ5 drugs) ranged from $765.9 million Author Affiliations: Department of
(95% CI, $323.0 million-$1473.5 million) for nervous system agents to $2771.6 million (95% Health Policy, London School of
Economics and Political Science,
CI, $2051.8 million-$5366.2 million) for antineoplastic and immunomodulating agents. London, United Kingdom (Wouters);
Data were mainly accessible for smaller firms, orphan drugs, products in certain therapeutic Department of Health Services
areas, first-in-class drugs, therapeutic agents that received accelerated approval, and Research and Policy, London School
of Hygiene and Tropical Medicine,
products approved between 2014 and 2018. Results varied in sensitivity analyses using
London, United Kingdom (McKee);
different estimates of clinical trial success rates, preclinical expenditures, and cost of capital. Leuven Institute for Healthcare
Policy, Department of Public Health
CONCLUSIONS AND RELEVANCE This study provides an estimate of research and development and Primary Care, KU Leuven,
costs for new therapeutic agents based on publicly available data. Differences from previous Belgium (Luyten).
studies may reflect the spectrum of products analyzed, the restricted availability of data in Corresponding Author: Olivier J.
the public domain, and differences in underlying assumptions in the cost calculations. Wouters, PhD, Department of Health
Policy, London School of Economics
and Political Science, Houghton
JAMA. 2020;323(9):844-853. doi:10.1001/jama.2020.1166 Street, London WC2A 2AE, United
Kingdom (o.j.wouters@lse.ac.uk).

R
ising drug prices have attracted public debate in the studies in this period, most of which relied on confidential or
United States and abroad on fairness of drug pricing and proprietary data, reported figures from $314 million to $2.1 bil-
revenues.1 Central to this debate is the scale of research lion (in 2018 US dollars).5-11
and development investment by biopharmaceutical compa- In 2017, Prasad and Mailankody estimated the research and
nies that is required to bring new medicines to market.2 development costs of new cancer drugs using public data re-
The most widely cited studies of the cost of develop- ported by pharmaceutical firms to the US Securities and
ing a new drug (DiMasi et al3,4) reported a sharp increase Exchange Commission (SEC).12 They estimated the median re-
in the mean cost of developing a single new therapeutic agent search and development cost of bringing a single cancer drug
from $1.1 billion in 2003 to $2.8 billion in 2013 (in 2018 US dol- to market to be $780 million (in 2018 US dollars), capitalized
lars), based on a real cost of capital rate of 11% per year in at a real cost of capital rate of 7% per year, based on a sample
the former study3 and of 10.5% per year in the latter.4 Other of 10 drugs.12

844 JAMA March 3, 2020 Volume 323, Number 9 (Reprinted) jama.com

© 2020 American Medical Association. All rights reserved.

Downloaded From: https://jamanetwork.com/ on 12/29/2021

 Research and Development Costs of Bringing a New Medicine to Market Original Investigation Research

This present study estimates the research and develop-

ment investment required to bring a new therapeutic agent Key Points
to market using publicly available data for products approved
Question How much do drug companies spend on research and
by the US Food and Drug Administration (FDA) between development to bring a new medicine to market?
2009 and 2018.
Findings In this study, which included 63 of 355 new therapeutic
drugs and biologic agents approved by the US Food and Drug
Administration between 2009 and 2018, the estimated median
Methods capitalized research and development cost per product was
$985 million, counting expenditures on failed trials. Data were
Sample Identification and Characteristics mainly accessible for smaller firms, products in certain therapeutic
We identified all new therapeutic agents, ie, new drug appli- areas, orphan drugs, first-in-class drugs, therapeutic agents that
cations and biologics license applications approved by the received accelerated approval, and products approved between
2014 and 2018.
FDA between 2009 and 2018, in the Drugs@FDA database.13
For each, we extracted the date of approval, date of submis- Meaning This study provides an estimate of research and
sion of investigational new drug application, date of sub- development costs for new therapeutic agents based on publicly
mission of new drug application or biologics license applica- available data; differences from previous studies may reflect the
spectrum of products analyzed and the restricted availability of
tion, indication, type (pharmacologic or biologic), expedited
data in the public domain.
programs (priority review, accelerated approval, fast track,
or breakthrough), orphan status, route of administration
(oral, injection, intravenous, or other), and manufacturer opment expenditures on individual drug candidates. We ex-
(eTable 1 in the Supplement). To capture innovation, we cluded products developed by companies that only reported
determined whether an agent was first in class using publi- total research and development expenditures across all drug
cations by FDA officials.14,15 We checked the data for consis- candidates or across therapeutic areas.
tency with published reports.15,16 For excluded products, we searched the 10-K and 10-Q
Therapeutic areas were obtained from the anatomical forms and online press releases of manufacturers at the time
therapeutic chemical classification system database.17 For that agents were approved to see if any were developed
agents that were not yet classified, we based our decision on in collaboration with other firms via licensing deals. If so,
the approved indication. we searched for 10-K and 10-Q forms from those firms in case
For each agent, we identified start and end dates of clini- there were research and development data for the product
cal studies (phases 1, 2, and 3 for the FDA-approved indica- in question.
tion) from ClinicalTrials.gov (search conducted on April 4,
2019). If there were multiple studies in the same phase, the Inclusions
earliest start date was selected. We verified these dates with For each therapeutic agent with available data, we extracted
reports in SEC filings and used the dates from SEC filings if direct and indirect research and development expenditures in
there were discrepancies. We classified combined phase 1 each year of development. Drugs were tracked across years in
and 2 trials as phase 2 and combined phase 2 and 3 trials as SEC filings using the brand, generic, or compound names of
phase 3, consistent with other studies.18-20 Dates of submis- agents, as appropriate.
sion of investigational new drug applications were used to Direct research and development expenses included all
approximate the end of preclinical testing; these dates were resources directly allocated to a particular agent. Indirect re-
checked for consistency with filings to ensure clinical testing search and development expenses, which included person-
had not already begun outside the United States. nel and overhead costs, were sometimes reported as a lump
No data were collected from human participants, and all sum across all drug development programs. If so, we applied
data in this study were publicly available. the same percentage of direct research and development costs
attributable to a particular agent to estimate indirect costs for
Research and Development Data Extraction the same agent. The proportional allocation of personnel and
Publicly traded US companies are legally required by the overhead expenses is common practice in costing studies.22
SEC to file annual 10-K and quarterly 10-Q forms, which are Costs were tracked from the year a company started
reports of key financial performance indicators that include reporting costs for a particular drug candidate in their finan-
audited financial statements and data on research and devel- cial statements until the quarter of approval, which often
opment expenditures. For every agent in our sample, we included 1 or more years of preclinical costs. In some cases, at
searched the SEC website for reports from the firm that re- the first mention of the candidate in SEC filings, companies
ceived FDA approval for it.21 reported the costs incurred since inception of the drug devel-
opment program. Certain companies only started tracking
Exclusions costs at late stages of preclinical development or at the start
As reports for private US drug firms and foreign companies of phase 1 of development, resulting in an underreporting of
listed on non-US stock exchanges were unavailable, their prod- preclinical costs.
ucts were excluded. For firms with available reports, we Some drugs were initially developed by companies that
screened 10-K and 10-Q filings for data on research and devel- subsequently licensed out their drug candidates to other firms,

jama.com (Reprinted) JAMA March 3, 2020 Volume 323, Number 9 845

© 2020 American Medical Association. All rights reserved.

Downloaded From: https://jamanetwork.com/ on 12/29/2021

 Research Original Investigation Research and Development Costs of Bringing a New Medicine to Market

which then brought these products to market. In these cases, Wong et al reported that the percentages of FDA approv-
it was assumed that any preclinical and clinical costs in- als were 13.8% for therapeutic agents entering phase 1, 35.1%
curred during initial development was included in licensing for those entering phase 2, and 59.0% for those entering
fees and milestone payments. Hence, where these fees and pay- phase 3.18
ments were recorded as research and development expenses Wong et al18 provided success rates through phase 3. We
for the agent in question, these costs were extracted. Data on supplemented these rates with a recent estimate of the pro-
costs incurred by the originator firms were not collected. portion of biologics license applications and new drug appli-
If SEC filings were missing for 3 or fewer years since the cations that are approved by the FDA (83.2%).20
inception of the drug development program (eg, if a com-
pany was privately held during early years of development) and Costing Method
the product did not move between development phases For each agent, we estimated the expected research and de-
(ie, either from 1 to 2 or 2 to 3), we extrapolated costs from the velopment investment to bring the drug to market in 3 steps.
closest available year. Products were excluded if more than 3 First, we summed direct and indirect research and devel-
years of SEC filings were missing. opment spending on a therapeutic agent in each year. All sums
Three investigators independently extracted all research were inflation adjusted to 2018 dollars using the US con-
and development data used in this study. Discrepancies were sumer price index.
resolved through discussions. Where disagreements existed, Second, we accounted for failed projects by dividing total
we assumed the higher estimate of research and develop- research and development expenditures on a drug in a par-
ment expenditures. ticular year by the corresponding aggregate phase-specific
probability of success, similar to what was done in previous
Quality Assessments studies of costs of drug development.3-7 For example, for
Consistency and completeness of company reporting in SEC each drug, we divided phase 1 costs in each year by 0.138,
filings varied over time. Many reported detailed research and which accounted for spending on the other 6.2 phase 1 trials
development costs, which allowed us to track outlays over time that would fail, on average, for each successful development
for individual candidates. Others reported costs inconsis- program. We used phase 1 rates to adjust preclinical expendi-
tently or with missing data for some years, requiring various tures, and we used the proportion of biologics license appli-
assumptions, for example on timings of transitions between cations and new drug applications that are approved by the
phases and extrapolations when SEC filings were missing. FDA to adjust costs once these applications were submitted
To aid interpretation, we categorized each estimate as high, to the agency for regulatory approval. Licensing fees and
medium, or low quality, depending on the availability and con- milestone payments, where captured, were adjusted using
sistency of reported data. The categorization was developed the success rate for the trial phase that was ongoing when the
through discussion between all authors. payments were made. When a phase shift took place within
High-quality estimates comprised drugs discovered inter- the financial year, we allocated the cost proportionally to the
nally, allowing tracking of costs back to inception of the de- time spent in each phase. For example, if development
velopment program, and products licensed at preclinical or moved from phase 1 to phase 2 on July 1 of a given year, we
phase 1 stages with minimal up-front fees or milestone pay- divided the costs equally between each phase. Similarly, in
ments captured in SEC filings. Late commercialization deals the year of approval, we multiplied the total cost by the frac-
related to marketing of products in non-US markets were also tion of the year elapsed by the time of approval. Hence, if a
deemed high-quality estimates, as they would have had little drug was approved on July 1, we only counted 50% of the
or no effect on research and development expenses incurred costs in the year of approval since firms often incurred post-
on trials required for FDA approval. approval costs related to pharmacovigilance or testing in
Low-quality estimates comprised all acquisitions, licens- other indications.
ing deals, or other collaboration agreements in phases 2 or 3, Third, we applied a real cost of capital rate of 10.5% per
earlier deals in which it was unclear whether all costs were cap- year (ie, weighted average cost of capital in the pharmaceuti-
tured in data extraction, and estimates requiring extrapola- cal industry), as in the DiMasi et al study.4 Cost of capital is the
tion of 2 to 3 years of data. We classified estimates as medium required rate of return for an investor and encapsulates a risk-
quality when other judgment calls regarding financial report- free rate (ie, opportunity cost) and premium based on the like-
ing, as agreed upon by the authors, had to be made. lihood of business failure.24
Two investigators independently categorized the quality
of estimates and resolved discrepancies through discussions. Sensitivity and Subgroup Analyses
We ran 4 univariate sensitivity analyses. First, as the results
Costs of Failed Trials were sensitive to the choice of aggregate clinical trial success
Accurate information on costs of failures, ie, research and de- rates (by phase), we recalculated the results using aggregate
velopment outlays on candidates being developed by compa- rates reported in 2 other studies (Table 1).19,20 Next, we calcu-
nies but not ultimately approved, is essential to estimating the lated a second estimate of research and development costs
costs of drug development. We accounted for failures using data using therapeutic-area–specific rates reported by Wong et al
on aggregate clinical trial success rates from a recent study by (Table 1), instead of aggregate rates. For example, oncology
Wong et al (Table 1).18 drugs in phase 1 have a 3.4% chance of ultimately receiving

846 JAMA March 3, 2020 Volume 323, Number 9 (Reprinted) jama.com

© 2020 American Medical Association. All rights reserved.

Downloaded From: https://jamanetwork.com/ on 12/29/2021

 Research and Development Costs of Bringing a New Medicine to Market Original Investigation Research

Table 1. Clinical Trial Success Rates by Phase (on Aggregate and by Therapeutic Area)a

Source Phase 1 to Approval, %b Phase 2 to Approval, %c Phase 3 to Approval, %d FDA Submission to Approval, %e
Aggregate rates
Wong et al18 13.8 35.1 59.0 83.2
Thomas et al19 9.6 15.3 49.6 85.3
Hay et al20 10.4 16.2 50.0 83.2
Therapeutic-area–specific rates18
Oncology 3.4 6.7 35.5 81.7
Metabolism and endocrinology 19.6 24.1 51.6 80.4
Cardiovascular 25.5 32.3 62.2 84.5
Central nervous system 15.0 19.5 51.1 82.2
Autoimmune and inflammation 15.1 21.2 63.7 80.3
Ophthalmologyf 32.6 33.6 74.9 80.4
Infectious disease 25.2 35.1 75.3 84.9
Otherg 20.9 27.3 63.6 80.4
Abbreviation: FDA, US Food and Drug Administration. efficacy and safety of the dose of the therapeutic agent believed to provide
a
Rates across all indications for individual therapeutic agents (as opposed to the best risk-benefit ratio.23
e
rates for lead indications, which were higher in all phases). Only the success Indicates the proportion of new drug applications and biologics license
rates used in this analysis were reported. applications approved by the FDA. Wong et al18 reported aggregate and
b
Phase 1 trials, which usually include as many as 100 healthy volunteers and therapeutic-area–specific rates through phase 3. These data were
may take several months to conduct, are primarily used to assess the supplemented with estimates of FDA submission to approval rates from Hay
tolerability and safety of a therapeutic agent in different doses; these are et al; if a particular category from the study by Wong et al was not reported by
sometimes referred to as first-in-human trials. Hay et al, the category Other was used.20
f
c
Phase 2 trials, which can involve as many as a few hundred patients with a disease This category was applied to therapeutic agents classified as treating sensory
or condition and take several months to 2 years to complete, are typically used to organ diseases, ie, anatomical therapeutic chemical classification system code S.
g
gather data on the efficacy and safety of a therapeutic agent in different doses. Values in this category were based on the rates for “all [agents] without
d
Phase 3 trials, which can involve several thousand participants with a disease oncology” reported by Wong et al.18 These rates were applied to therapeutic
or condition and may take 1 to 4 years to run, are generally used to confirm the agents that were outside the other categories.

FDA approval, so we divided each year of phase 1 costs for We conducted a nonparametric bootstrapped resampling
these products by 0.034. Third, we performed a rerun of the with replacement (1000 iterations) to calculate 95% CIs
analyses using a real cost of capital rate of 7% (as done by around the estimated mean and median investments in
Prasad and Mailankody12) and 0% (to show noncapitalized research and development in our sample. We used χ2 tests to
outlays). Fourth, to account for potentially missing preclini- identify statistically significant differences in characteristics
cal expenditures, we adopted the same assumption around of the study sample vs therapeutic agents approved by the
preclinical costs as DiMasi et al, who reported that preclinical FDA between 2009 and 2018 that were excluded from our
costs represented 42.9% of their total research and develop- analysis. We used Kruskal-Wallis and Mann-Whitney U tests,
ment estimate.4 Thus, for each product in our sample, we as appropriate, to identify statistically significant differences
isolated clinical expenditures and imputed a preclinical cost in median estimated research and development investments
that amounted to this percentage. No imputations were per- across therapeutic areas and other drug characteristics.
formed for products acquired through purchase after clinical All statistical tests were 2-tailed and used a type I error
development had begun since it was assumed that licensing rate of 0.05. The data were analyzed using Stata version 15
fees and milestone payments reflected preclinical costs (StataCorp).
incurred by the company that sold the rights to the product.
Additionally, we ran another sensitivity analysis but with
imputations done for all products, including agents acquired
through purchase.
Results
As a subgroup analysis, we reported mean and median Between 2009 and 2018, the FDA approved 355 new drugs
amounts by therapeutic area, using area-specific rates to ad- and biologics. Research and development expenditures
just for costs of failure. from SEC government filings were available for 63 of these
products, developed by 47 different companies (Figure 1).
Statistical Analysis The sample covered 17.7% (63/355) of all new therapeutic
We estimated the mean and median research and develop- agents approved by the FDA over this 10-year period.
ment investments across our sample in the base case Twenty-three of the estimates were judged of high quality,
and sensitivity analyses. We then restricted the sample to 18 medium quality, and 22 low quality. eTable 2 in the
high-quality estimates and recalculated the mean and Supplement provides the rationale for the quality categori-
median amounts. zation of each agent.

jama.com (Reprinted) JAMA March 3, 2020 Volume 323, Number 9 847

© 2020 American Medical Association. All rights reserved.

Downloaded From: https://jamanetwork.com/ on 12/29/2021

 Research Original Investigation Research and Development Costs of Bringing a New Medicine to Market

Figure 1. Selection Process for Therapeutic Agents

84 Biologics license applications approved 271 New drug applications approved

by the FDA between 2009 and 2018 by the FDA between 2009 and 2018

355 Novel therapeutic agents approved

by the FDA over the study period

299 Therapeutics excluded

112 From foreign firms not listed
on a US stock exchange
99 From private US firmsa
88 From firms reporting only
aggregate R&D figures

56 Therapeutics included after

initial screening FDA indicates the US Food and Drug
Administration; R&D, research and
development.
7 Therapeutics (out of 299 excluded a
above) added back in after screening Includes firms that went public
for licensing deals while developing the therapeutic
agent in question but for which
US Securities and Exchange
63 Therapeutics included in Commission (SEC) filings were
quantitative synthesis
missing for more than 3 years of
the drug development period.

Table 2. Characteristics of New Therapeutic Agents Approved by the US Food and Drug Administration
Between 2009 and 2018a

No. (%)
Included Agents Full Sample
Characteristics (n = 63) (n = 355) P Value
Agent type
Pharmacologic 47 (75) 271 (76)
.72
Biologic 16 (25) 84 (24)
Therapeutic areab
Antineoplastic and immunomodulating agents 20 (32) 116 (33)
Alimentary tract and metabolism 15 (24) 44 (12)
Nervous system 8 (13) 33 (9) .02
Antiinfective agents for systemic use 5 (8) 40 (11)
Other 15 (24) 122 (34)
Orphan drug 31 (49) 145 (41) .14
a
Analyses were carried out using χ2
Drug received accelerated approval 14 (22) 41 (12) .003 tests comparing the data for
Drug benefited from any expedited development 48 (76) 234 (66) .06 included agents (n = 63) vs
or approval pathwayc excluded ones (n = 292).
Innovativeness b
Other therapeutic areas included
First in class 27 (43) 127 (36) blood and blood-forming organs,
.20 cardiovascular system,
Next in class 36 (57) 228 (64)
dermatologicals, musculoskeletal
d
Route of administration system, sensory organs,
Oral 28 (44) 187 (53) and various.
c
Injection 20 (32) 87 (25) Pathways included accelerated
.20 approval, breakthrough therapy,
Intravenous 10 (16) 41 (12)
fast track, orphan drug,
Other 5 (8) 40 (11) and priority review.
Approval dates d
Injection included intramuscular
2009-2013 17 (27) 142 (40) and subcutaneous; other routes
.02 included multiple, ophthalmic,
2014-2018 46 (73) 213 (60)
and topical.

Sample Characteristics drugs, therapeutic agents that benefited from expedited de-
Table 2 presents statistics for the 63 included therapeutic velopment or approval pathways, and first-in-class drugs com-
agents. The sample contained a larger proportion of orphan pared with all FDA-approved products between 2009 and 2018,

848 JAMA March 3, 2020 Volume 323, Number 9 (Reprinted) jama.com

© 2020 American Medical Association. All rights reserved.

Downloaded From: https://jamanetwork.com/ on 12/29/2021

 Research and Development Costs of Bringing a New Medicine to Market Original Investigation Research

although these differences were not statistically significant.

Table 3. Median Expected Research and Development Expenditure
Differences in the breakdown of products by therapeutic area, on New Therapeutic Agents Approved by the US Food and Drug
accelerated vs regular approval, and approval dates were sta- Administration (2009-2018) in Main and Sensitivity Analyses
tistically significant.
Research and Development Costs
in US$, Millions (95% CI)
Research and Development Investments All Included High-Quality
Parameter Varied Agents Sample
Without adjustments for costs of failed trials, no statistically in Sensitivity Analysis (n = 63) (n = 23)
significant differences in the median research and develop- Source of aggregate
ment investment required to bring a new drug to market were clinical trial success rates
observed across any of the drug characteristics shown in Wong et al (base case)18 985.3 1048.1
(683.6-1228.9) (796.6-1180.6)
Table 2, except median costs for biologic drugs that were higher
Hay et al20 1404.9 1620.3
than those for pharmacologic drugs (eTable 3 in the Supple- (1102.2-1773.4) (1191.8-1773.4)
ment). For the 63 agents included in the analysis, outlays for Thomas et al19 1465.8 1678.4
(1121.5-1887.1) (1259.7-1999.3)
research and development (ie, total noncapitalized direct and
Aggregate vs therapeutic-area–
indirect expenses incurred during preclinical and clinical test- specific success rates
ing) were estimated at a median of $319.3 million (95% CI, Aggregate rates (base case)18 985.3 1048.1
$236.4 million-$351.4 million), and the estimated mean out- (683.6-1228.9) (796.6-1180.6)

lay was $374.1 million (95 CI, $301.9 million-$464.2 million), Area-specific rates18 1385.2 1220.1
(1053.9-1971.8) (994.3-2118.3)
(eTable 4 in the Supplement). The mean (SD) number of years Cost of capital rate
of data per drug was 8.3 (2.8) years. eTable 5 in the Supplement 10.5% (base case) 985.3 1048.1
shows the dates of phase changes for clinical trials of each in- (683.6-1228.9) (796.6-1180.6)
cluded agent. 7% 848.9 930.0
(671.1-1076.6) (692.0-1018.1)
After accounting for costs of failed trials, the estimated
0% 688.2 697.9
median research and development investment required to (450.8-850.6) (567.9-850.6)
bring a new drug to market, capitalized at a rate of 10.5% per Adjustment for potential
year, was $985.3 million (95% CI, $683.6 million-$1228.9 mil- underreporting of spending
on preclinical trials
lion), and the estimated mean was $1335.9 million (95% CI, No (base case) 985.3 1048.1
$1042.5 million-$1637.5 million) (Table 3). Figure 2 shows (683.6-1228.9) (796.6-1180.6)
point estimates for each of the 63 agents, which ranged from Yes, excluding agents acquired 1228.9 1482.0
through purchase (943.8-1900.9) (1104.7-1860.9)
$143.2 million for Mytesi (crofelemer) to $6419.0 million for
Yes, including all products 1628.4 1751.1
Dupixent (dupilumab). (1196.5-2072.2) (1255.3-2246.9)
Restricting the analysis to high-quality estimates (n = 23),
the estimated median research and development investment from $1335.9 million to $2307.2 million (95% CI, $1726.9 mil-
increased from $985.3 million to $1048.1 million (95% CI, lion-$3013.0 million).
$796.6 million-$1180.6 million), while the estimated mean When the costs were capitalized at an annual rate of 7%
declined from $1335.9 million to $1143.3 million (95% CI, instead of 10.5%, the median expected investment decreased
$880.4 million-$1442.1 million). from $985.3 million to $848.9 million (95% CI, $671.1 million-
$1076.6 million), and the mean decreased from $1335.9 mil-
Sensitivity Analyses lion to $1158.6 million (95% CI, $929.3 million-$1407.2 mil-
Table 3 shows the results of univariate sensitivity analyses. lion). When costs were not capitalized, rather than capitalized
When the aggregate success rates reported by Hay et al20 were at an annual rate of 10.5%, the median expected invest-
used instead of those reported by Wong et al,18 the estimated ment decreased from $985.3 million to $688.2 million (95%
median research and development investment, capitalized CI, $450.8 million-$850.6 million), and the mean decreased
at a rate of 10.5% per year, increased from $985.3 million from $1335.9 million to $884.8 million (95% CI, $717.1 million-
to $1404.9 million (95% CI, $1102.2 million-$1773.4 million), $1084.7 million).
while the estimated mean increased from $1335.9 million to With the adjustments for potentially missing preclinical
$1976.6 million (95% CI, $1595.5 million-$2454.8 million). costs done for 33 of 63 products (ie, excluding agents acquired
When the rates from Thomas et al19 were used, the estimated through purchase), based on the DiMasi et al4 approach, the es-
median research and development investment, capitalized timated median research and development investment in-
at an annual rate of 10.5%, was $1465.8 million (95% CI, creased from $985.3 million to $1228.9 million (95% CI,
$1121.5 million-$1887.1 million), and the estimated mean $943.8 million-$1900.9 million), while the estimated mean in-
was $2059.5 million (95% CI, $1639.9 million-$2511.7 million). creased from $1335.9 million to $1800.7 million (95% CI,
When therapeutic-area–specific rates from Wong et al,18 $1396.4 million-$2268.5 million). With the adjustments for po-
rather than aggregate rates, were used to account for costs of tentially missing preclinical costs done for all 63 products, the
failed trials for each agent, the estimated median research and estimated median research and development investment in-
development investment, capitalized at a rate of 10.5% per year, creased to $1628.4 million (95% CI, $1196.5 million-$2072.2 mil-
increased from $985.3 million to $1385.2 million (95% CI, lion), while the estimated mean increased to $2214.7 million
$1053.9 million-$1971.8 million), and the estimated mean rose (95% CI, $1734.1 million-$2719.9 million).

jama.com (Reprinted) JAMA March 3, 2020 Volume 323, Number 9 849

© 2020 American Medical Association. All rights reserved.

Downloaded From: https://jamanetwork.com/ on 12/29/2021

 Research Original Investigation Research and Development Costs of Bringing a New Medicine to Market

Figure 2. Estimated Expenditures for Therapeutic Agents by Quality of the Estimates

High quality Viibryd (vilazodone hydrochloride)

Mepsevii (vestronidase alfa-vjbk)
Vitrakvi (larotrectinib sulfate)
Rhropressa (netarsudil dimesylate)
Folotyn (pralatrexate)
Elzonris (tagraxofusp-erzs)
Aristada (aripiprazole lauroxil)
Ultomiris (ravulizumab-cwvz)
Eucrisa (crisaborole)
Trulance (plecanatide)
Iclusig (ponatinib hydrochloride)
Gattex (teduglutide recombinant)
Ocaliva (obeticholic acid)
Ingrezza (valbenazine tosylate)
Brineura (cerliponase alfa)
Alunbrig (brigatinib)
Tymlos (abaloparatide)
Xerava (eravacycline dihydrochloride)
Veltassa (patiromer sorbitex calcium)
Vimizim (elosulfase alfa)
Galafold (migalastat hydrochloride)
Nuplazid (pimavanserin tartrate)
Belviq (lorcaserin hydrochloride)
Median estimate
Medium quality Juxtapid (lomitapide mesylate)
Orbactiv (oritavancin diphosphate)
Crysvita (burosumaab-twza)
Dificid (fidaxomicin)
Varubi (rolapitant)
Strensiq (asfotase alfa)
Tegsedi (inotersen sodium)
Xtandi (enzalutamide)
Kerydin (tavaborole)
Kalbitor (ecallantide)
Rubraca (rucaparib camsylate)
Viberzi (eluxadoline)
Imbruvica (ibrutinib)
Rapivab (peramivir)
Adcetris (brentuximab vedotin)
Linzess (linaclotide)
Nuzyra (omadacycline tosylate)
Palynziq (pegvaliase-pqpz)
Median estimate
Low quality Mytesi (crofelemer)
Fanapt (iloperidone)
Firdapse (amifampridine phosphate)
Beleodaq (belinostat)
Northera (droxidopa)
Idhifa (enasidenib mesylate)
Hetlioz (tasimelteon)
Kybella (deoxycholic acid)
Kyprolis (carfilzomib)
Zejula (niraparib tosylate)
Kengreal (cangrelor)
Exondys 51 (eteplirsen)
Talzenna (talazoparib tosylate)
Libtayo (cemiplimab-rwlc)
Zerbaxa (ceftolozane sulfate/tazoparib tosylate)
Tibsovo (ivosidenib)
Zaltrap (ziv-aflibercept)
Kevzara (sarilumab)
Copiktra (duvelisib) Indicates the estimated research
Eylea (aflibercept) and development investment needed
Praluent (alirocumab) for an average company to bring one
Dupixent (dupilumab) of these products to market based on
Median estimate aggregate clinical trial success rates.
Median overall Individual companies may have spent
0 1000 2000 3000 4000 5000 6000 7000 more or less on their respective
Expenditure in Millions, $ development pipelines, based on
their individual success rates.

Restricting the sensitivity analyses to high-quality esti- ket increased in most cases (Table 3). eTable 6 in the Supplement
mates (n = 23), the estimated median and mean research and shows the estimates for each agent in the base case and sensi-
development investments required to bring a new drug to mar- tivity analyses.

850 JAMA March 3, 2020 Volume 323, Number 9 (Reprinted) jama.com

© 2020 American Medical Association. All rights reserved.

Downloaded From: https://jamanetwork.com/ on 12/29/2021

 Research and Development Costs of Bringing a New Medicine to Market Original Investigation Research

a
Table 4. Mean And Median Expected Research and Development Expenditure on New Therapeutic Agents Therapeutic areas were obtained
Approved by the US Food and Drug Administration (2009-2018) by Therapeutic Area from the anatomical therapeutic
chemical classification system
Expenditure in US$, Millions database.17 Where agents were not
(95% CI)b
Sample yet classified, the categorization
Therapeutic Areaa Size Median Mean was based on the approved
Antineoplastic and immunomodulating agents 20 2771.6 (2051.8-5366.2) 4461.2 (3114.0-6001.3) indication.
b
Alimentary tract and metabolism 15 1217.6 (613.9-1792.4) 1430.3 (920.8-2078.7) Bootstrapped CIs were not
calculated for therapeutic areas
Nervous system 8 765.9 (323.0-1473.5) 1076.9 (508.7-1847.1)
with less than 5 samples. Estimates
Antiinfectives for systemic use 5 1259.9 (265.9-2128.3) 1297.2 (672.5-1858.5) were based on therapeutic-area–
Dermatologicals 4 747.4 1998.3 specific success rates reported by
Wong et al.18
Cardiovascular system 3 339.4 1152.4
c
The product Veltassa (patiromer
Musculoskeletal system 3 1052.6 937.3
sorbitex calcium) was assigned
Blood and blood-forming organs 2 793.0 793.0 to the therapeutic area Various
Sensory organs 2 1302.8 1302.8 under the subgroup Drugs for
treatment of hyperkalemia and
Otherc 1 1121.0 1121.0
hyperphosphatemia.

Subgroup Analyses by Therapeutic Area developed by 10 large firms.4 The estimate by DiMasi et al
Median estimates by therapeutic area (for areas with ≥5 drugs), used confidential data on costs voluntarily submitted by
adjusted using area-specific rates and capitalized at 10.5% per anonymous companies without independent verification,
year, ranged from $765.9 million (95% CI, $323.0 million- making them difficult to validate.12,28-30 The higher esti-
$1473.5 million) for nervous system agents to $2771.6 million mate of DiMasi et al seems to reflect a combination of higher
(95% CI, $2051.8 million-$5366.2 million) for antineoplastic clinical costs incurred by larger drug developers, lower esti-
and immunomodulating agents. The corresponding mean es- mates of trial success for each stage of development com-
timates ranged from $1076.9 million (95% CI, $508.7 million- pared to the more recent data presented by Wong et al, and
$1847.1 million) for nervous system agents to $4461.2 million different assumptions about preclinical expenditures as
(95% CI, $3114.0 million-$6001.3 million) for antineoplastic and their data set did not permit allocating these expenditures
immunomodulating agents (Table 4). to specific agents.
The results of the present study varied widely when sub-
ject to sensitivity analyses, especially using different success
rates. The methods employed by Wong et al to handle miss-
Discussion ing data were an improvement on earlier studies of trial suc-
Based on data for 63 therapeutic agents developed by 47 com- cess rates, and their study was based on a larger sample.18
panies between 2009 and 2018, the median research and de- Wong et al also noted that the most cited studies of success
velopment investment required to bring a new drug to mar- rates 19,20,31 originated from researchers with ties to the
ket was estimated to be $985 million, and the mean was pharmaceutical industry, and elaborated that “previous esti-
estimated to be $1336 million. Estimates differed across thera- mates of drug development success rates [relied] on rela-
peutic areas, with costs of developing cancer drugs the high- tively small samples from databases curated by the pharma-
est. The results included costs of failed clinical trials and var- ceutical industry and [were] subject to potential selection
ied in sensitivity analyses using different estimates of trial biases.”18 Also, compared with these earlier studies of suc-
success, preclinical expenditures, and cost of capital. cess rates,19,20,31 the timing of the work by Wong et al18 more
These figures were higher than the median capitalized re- closely aligned with that of the present study, thereby improv-
search and development cost of $780 million (in 2018 US dol- ing its internal validity.
lars) reported by Prasad and Mailankody for oncology drugs.12 There are challenges in isolating preclinical investments
This may be because adjustments based on clinical trial suc- by drug companies. It is especially difficult to identify the
cess rates were applied in the present study to account for costs exact date from which costs should start being allocated
of failures, whereas Prasad and Mailankody restricted their to individual agents during the early stages of preclinical
analysis to companies bringing their first drug to market and research. The base case scenario in this study relied on pre-
then summed the total research and development expendi- clinical costs reported by firms in SEC filings, which were
tures of each company during the development periods of the likely underestimated since many companies did not attri-
drugs in their sample. Most of the companies included in their bute costs during the drug discovery stages to individual
study appeared to be more successful than the average candidates. DiMasi et al estimated that preclinical costs
company.25-27 Moreover, their analysis was based on data for accounted, on average, for 42.9% of total capitalized costs,
10 oncology drugs, which limits the comparability of their re- based on aggregated data on preclinical spending and
sults with the present study. assumptions around the duration of preclinical testing. 4
The mean estimate of $1.3 billion in the present study Although preclinical costs were variously estimated in the
was lower than the $2.8 billion (in 2018 US dollars) reported present study, including indirectly through license fees, pre-
by DiMasi et al, which was based on data for 106 products clinical data were directly captured for 19 products. For these

jama.com (Reprinted) JAMA March 3, 2020 Volume 323, Number 9 851

© 2020 American Medical Association. All rights reserved.

Downloaded From: https://jamanetwork.com/ on 12/29/2021

 Research Original Investigation Research and Development Costs of Bringing a New Medicine to Market

products, preclinical costs generally accounted for a lower cult to ensure perfect comparability of research and develop-
share of the total capitalized costs (ranging from 0.3% to ment figures between firms. These inconsistencies may have
50.7%; median 12.4%) than what was estimated by DiMasi been explained by differences in accounting policies. For in-
et al (eTable 4 in the Supplement). For comparison, however, stance, some firms allocated overhead and administrative costs
the 42.9% estimate was used to impute preclinical costs in to direct research and development figures, while others re-
sensitivity analyses in this study. Further validation work is ported these costs separately. Some reported preclinical re-
needed to establish the preclinical share of research and search costs as a separate line item, while others incorpo-
development estimates for individual products. rated them in overhead costs. Companies also reported costs
Greater transparency around research and development associated with licensing deals, drug acquisitions, and col-
costs is essential for analysts to check the veracity of claims laboration agreements differently, so it is likely that not all costs
by companies that the steep prices of new drugs are driven were fully reflected in some estimates.
by high development outlays. While these expenditures are Fourth, uncertainties in the analysis may have resulted in
undoubtedly high, as shown in this study, it is important for under- or overestimations of research and development ex-
policy makers, regulators, and payers to know the exact scale penditures for some products. It is difficult to attribute costs
of these investments. This knowledge can inform the design to individual drug candidates in the early stages of preclini-
of pricing policies that give adequate rewards for innovative cal development, so only the costs reported by firms in SEC
drugs that bring value to health care systems. filings were considered in the base case analysis. However,
since preclinical costs may have been underreported by some
Limitations companies, sensitivity analyses were conducted to produce an
This study has several limitations. First, data were unavail- upper-bound estimate of preclinical expenditures. Con-
able for many products approved by the FDA during the study versely, many drug firms conducted trials for a particular can-
period. No data were available for products developed by didate for multiple indications, which may have led to over-
non-US companies not listed on a US stock exchange and large estimations of costs since trial expenditures were not broken
drug firms that did not report research and development fig- down by indication but instead reported as annual lump sums
ures for individual drug candidates. Thus, there was likely an for each agent. Also, the estimates did not reflect any public
overrepresentation of smaller firms, which may have run leaner tax credits or subsidies, which may have led to further over-
operations than larger ones. This limited the generalizability estimations of costs incurred by companies.
of the results to all products.
Second, the included agents differed from other drugs ap-
proved by the FDA between 2009 and 2018, although not all
differences were statistically significant. The sample in-
Conclusions
cluded a larger proportion of orphan drugs, products in cer- This study provides an estimate of research and develop-
tain therapeutic areas, first-in-class drugs, therapeutic agents ment costs for new therapeutic agents based on publicly avail-
that received accelerated approval, and products approved be- able data. Differences from previous studies may reflect the
tween 2014 and 2018. spectrum of products analyzed, the restricted availability of
Third, there were inconsistencies in research and devel- data in the public domain, and differences in underlying as-
opment reporting between companies, which made it diffi- sumptions in the cost calculations.

ARTICLE INFORMATION extraction file is available from the corresponding 2. Morgan S, Grootendorst P, Lexchin J,
Accepted for Publication: January 28, 2020. author upon request. Cunningham C, Greyson D. The cost of drug
Additional Contributions: We thank Evelyn S. development: a systematic review. Health Policy.
Author Contributions: Dr Wouters had full access 2011;100(1):4-17. doi:10.1016/j.healthpol.2010.12.002
to all of the data in the study and takes Warner, MSc (Duff and Phelps, UK), for her
responsibility for the integrity of the data and the contribution to the study design and data 3. DiMasi JA, Hansen RW, Grabowski HG. The price
accuracy of the data analysis. collection, as well as her comments on an earlier of innovation: new estimates of drug development
Concept and design: Wouters. draft of the manuscript. We thank Jouni Kuha, PhD costs. J Health Econ. 2003;22(2):151-185. doi:10.
Acquisition, analysis, or interpretation of data: All (London School of Economics and Political Science, 1016/S0167-6296(02)00126-1
authors. UK), and Geert Verbeke, PhD (KU Leuven, 4. DiMasi JA, Grabowski HG, Hansen RW.
Drafting of the manuscript: Wouters. Belgium), for input on the statistical analysis, as well Innovation in the pharmaceutical industry: new
Critical revision of the manuscript for important as Alistair Milne, PhD (Loughborough University, estimates of R&D costs. J Health Econ. 2016;47:20-
intellectual content: All authors. UK), and Max King, BA (retired; formerly Investec 33. doi:10.1016/j.jhealeco.2016.01.012
Statistical analysis: Wouters. Asset Management, UK), for advice on costing
capital investments. None of the individuals listed 5. Jayasundara K, Hollis A, Krahn M, Mamdani M,
Administrative, technical, or material support: All Hoch JS, Grootendorst P. Estimating the clinical
authors. in the acknowledgments received compensation
for their role in the study. cost of drug development for orphan versus
Supervision: Wouters. non-orphan drugs. Orphanet J Rare Dis. 2019;14(1):
Conflict of Interest Disclosures: Dr McKee REFERENCES 12. doi:10.1186/s13023-018-0990-4
reported receipt of grants from the Wellcome Trust, 6. Mestre-Ferrandiz J, Sussex J, Towse A. The R&D
European Commission, and United Kingdom 1. Kesselheim AS, Avorn J, Sarpatwari A. The high
cost of prescription drugs in the United States: Cost of a New Medicine. London, UK: Office of Health
Research and Innovation outside the submitted Economics; 2012.
work. No other disclosures were reported. origins and prospects for reform. JAMA. 2016;316
(8):858-871. doi:10.1001/jama.2016.11237 7. Adams CP, Brantner VV. Estimating the cost of
Data Sharing Statement: All data used in this study new drug development: is it really 802 million
were in the public domain. An example data

852 JAMA March 3, 2020 Volume 323, Number 9 (Reprinted) jama.com

© 2020 American Medical Association. All rights reserved.

Downloaded From: https://jamanetwork.com/ on 12/29/2021

 Research and Development Costs of Bringing a New Medicine to Market Original Investigation Research

dollars? Health Aff (Millwood). 2006;25(2):420-428. 16. Darrow JJ, Kesselheim AS. Drug development clinical-research#Clinical_Research_Phase_Studies.
doi:10.1377/hlthaff.25.2.420 and FDA approval, 1938-2013. N Engl J Med. 2014; Published 2018. Accessed January 27, 2020.
8. Adams CP, Brantner VV. Spending on new drug 370(26):e39. doi:10.1056/NEJMp1402114 24. Chit A, Chit A, Papadimitropoulos M, Krahn M,
development. Health Econ. 2010;19(2):130-141. doi: 17. WHO Collaborating Centre for Drug Statistics Parker J, Grootendorst P. The opportunity cost of
10.1002/hec.1454 Methodology. ATC/DDD Index 2020. https://www. capital: development of new pharmaceuticals.
9. DiMasi JA, Grabowski HG. The cost of whocc.no/atc_ddd_index/. Published 2020. Accessed Inquiry. 2015;52:pii:004695801558464. doi:10.1177/
biopharmaceutical R&D: is biotech different? January 27, 2020. 0046958015584641
Manage Decis Econ. 2007;28(4-5):469-479. doi:10. 18. Wong CH, Siah KW, Lo AW. Estimation of 25. van der Gronde T, Pieters T. Assessing
1002/mde.1360 clinical trial success rates and related parameters. pharmaceutical research and development costs.
10. DiMasi JA, Grabowski HG, Vernon J. R&D costs Biostatistics. 2019;20(2):273-286. doi:10.1093/ JAMA Intern Med. 2018;178(4):587-588. doi:10.
and returns by therapeutic category. Drug Inf J. biostatistics/kxx069 1001/jamainternmed.2017.8706
2004;38(3):211-223. doi:10.1177/ 19. Thomas DW, Burns J, Audette J, Carroll A, 26. DiMasi JA. Assessing pharmaceutical research
009286150403800301 Dow-Hygelund C, Hay M. Clinical Development and development costs. JAMA Intern Med. 2018;178
11. Paul SM, Mytelka DS, Dunwiddie CT, et al. How Success Rates 2006-2015. Washington, DC: (4):587. doi:10.1001/jamainternmed.2017.8703
to improve R&D productivity: the pharmaceutical Biotechnology Innovation Organization, Amplion, 27. Prasad V, Mailankody S. Assessing
industry’s grand challenge. Nat Rev Drug Discov. and Biomedtracker; 2016. pharmaceutical research and development
2010;9(3):203-214. doi:10.1038/nrd3078 20. Hay M, Thomas DW, Craighead JL, Economides costs—reply. JAMA Intern Med. 2018;178(4):588-589.
12. Prasad V, Mailankody S. Research and C, Rosenthal J. Clinical development success rates doi:10.1001/jamainternmed.2017.8737
development spending to bring a single cancer drug for investigational drugs. Nat Biotechnol. 2014;32 28. Light DW, Warburton RN. Extraordinary claims
to market and revenues after approval. JAMA Intern (1):40-51. doi:10.1038/nbt.2786 require extraordinary evidence. J Health Econ.
Med. 2017;177(11):1569-1575. doi:10.1001/ 21. US Securities and Exchange Commission. 2005;24(5):1030-1033. doi:10.1016/j.jhealeco.2005.
jamainternmed.2017.3601 EDGAR company filings. https://www.sec.gov/ 07.001
13. US Food and Drug Administration. Drugs@FDA: edgar/searchedgar/companysearch.html. Published 29. Goozner M. A much-needed corrective on drug
FDA-approved drugs. https://www.accessdata.fda. 2019. Accessed April 2, 2019. development costs. JAMA Intern Med. 2017;177(11):
gov/scripts/cder/daf/. Published 2019. Accessed 22. Drummond MF, Sculpher MJ, Torrance GW, 1575-1576. doi:10.1001/jamainternmed.2017.4997
March 3, 2019. O’Brien BJ, Stoddart GL. Methods for the Economic 30. Avorn J. The $2.6 billion pill—methodologic
14. Lanthier M, Miller KL, Nardinelli C, Woodcock J. Evaluation of Health Care Programme. 3rd ed. Oxford, and policy considerations. N Engl J Med. 2015;372
An improved approach to measuring drug UK: Oxford University Press; 2005, https://pure.york. (20):1877-1879. doi:10.1056/NEJMp1500848
innovation finds steady rates of first-in-class ac.uk/portal/en/publications/methods-for-the-
economic-evaluation-of-health-care-programme- 31. DiMasi JA, Feldman L, Seckler A, Wilson A.
pharmaceuticals, 1987-2011. Health Aff (Millwood). Trends in risks associated with new drug
2013;32(8):1433-1439. doi:10.1377/hlthaff.2012.0541 third-edition(e43f24cd-099a-4d56-97e6-
6524afaa37d1)/export.html. Accessed March 2, 2019. development: success rates for investigational
15. US Food and Drug Administration. Novel drug drugs. Clin Pharmacol Ther. 2010;87(3):272-277.
approvals (2011-2018). https://www.fda.gov. 23. US Food and Drug Administration. Step 3: doi:10.1038/clpt.2009.295
Accessed March 2, 2019. Clinical research phase studies. https://www.fda.
gov/patients/drug-development-process/step-3-

jama.com (Reprinted) JAMA March 3, 2020 Volume 323, Number 9 853

© 2020 American Medical Association. All rights reserved.

Downloaded From: https://jamanetwork.com/ on 12/29/2021