ArticlePDF Available

Estimating the Cost of New Drug Development: Is It Really $802 Million?


Abstract and Figures

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 entering human clinical trials for the first time between 1989 and 2002, the paper estimated the cost per new drug to be 868 million dollars. However, our estimates vary from around 500 million dollars to more than 2,000 million dollars, depending on the therapy or the developing firm.
No caption available
Content may be subject to copyright.
Health Econ. (2009)
Published online in Wiley InterScience ( DOI: 10.1002/hec.1454
Bureau of Economics, Federal Trade Commission, Washington, DC, USA
Global Consumer & Small Business Banking, Bank of America, Charlotte, NC, USA
This paper replicates DiMasi et al.(J. Health Econ. 2003; 22: 151–185; Drug Inf. J. 2004; 38: 211–223) estimates of
expenditure on new drug development using publicly available data. The paper estimates that average expenditure
on drugs in human clinical trials is around $27m per year, with $17m per year on drugs in Phase I, $34m on drugs in
Phase II and $27m per year on drugs in Phase III of the human clinical trials. The paper’s estimated expenditure on
new drug development is somewhat greater than suggested by the survey results presented in DiMasi et al .
(J. Health Econ. 2003; 22: 151–185; Drug Inf. J. 2004; 38: 211–223). The paper combines a 12-year panel of research
and development expenditure for 183 publicly traded firms in the pharmaceutical industry with panel of drugs in
human clinical trials for each firm over the same period. The paper estimates drug expenditure by estimating the
relationship between research and development expenditure and the number of drugs in development for 1682
company/years (183 firms multiplied by the number of years for which we have financial and drug development
information). The paper also estimates expenditure on drugs in various therapeutic categories. Copyright r 2009
John Wiley & Sons, Ltd.
Received 14 May 2007; Revised 24 November 2008; Accepted 5 January 2009
KEY WORDS: pharmaceuticals; drug development
DiMasi et al. (2003, 2004) estimate the cost of new drug development for all drugs and for drugs in
certain therapeutic categories, respectively. The authors estimate the average cost of new drug
development to be $802m per new drug. This number has become a central part of the policy debates on
numerous issues regarding the pharmaceutic al industry including the Medicare Prescription Drug Act,
drug importation, generic entry and vaccine development. Drug companies argue the high cost of drug
development justifies the high prices paid by governments, insurers and customers. Given the
importance of the $802m number to the debate it is important to know whether it is correct and what it
DiMasi et al. (2003) calcul ate the cost of new drug development with data from two sources. The
authors survey 10 large pharmaceutical firms and ask those firms to report the expenditure in human
clinical trials for 68 drugs chosen at random from the Tuft’s drug development database called the
CSDD. The authors then use information on average success rates and success ful durations from the
CSDD data to calculate the cost of bringing a new drug to market. Recent ly, Light and Warburton
(2005) point out numerous problem s with DiMasi et al. (2003). In particular, because ‘cost data used
The authors are not aware of any potential conflicts that may bias their work. As far as the authors are aware, the study raises no
ethical issues.
*Correspondence to: Bureau of Economics, Federal Trade Commission, 601 New Jersey Avenue NW, Washington, DC 20580,
USA. E-mail:
Copyright r 2009 John Wiley & Sons, Ltd.
was proprietary and confidential, readers cannot know how each company collected its data, or what
was counted as research costs, and no independent verific ation of the accuracy of the information is
possible’ (p. 1031). This paper provides an independent verification of the survey cost data by using an
alternative publicly available data source on research an d development expenditure. Adams and
Brantner (2006) verify the second part of DiMasi et al. (2003) paper by using publicly available data to
estimate success rates and average successful durations.
By comparing aggregate annual expenditure on research and development across firms and over time
to the number of drugs in human clinical trials for each firm and each year, we can determine the
‘marginal expenditure’ on an additional drug in development. If Drug Firm A spends an additional
$50m in 1992 relative to 1991 but in 1992 Drug Firm A has two additional drugs in development we
argue this provides an estimate of average annual expenditure by Drug Firm A, i.e. $25m per drug per
year. Similarly, if Drug Firm B spends $100m more than Drug Firm A in 1992 but Drug Firm B has an
additional four drugs in development in 1992, then we estimate drug expenditure to be $25m per drug
per year. Note that this is an estimate of the correlation between expenditure and the number of drugs in
development. We are not attempting to estimate the impact of an additional dollar of expenditure on
the number of drugs in development or the impac t of additional drug on the amount of expenditure.
There are a number of advantages to this approach. First, we are using publicly available data so
our results can be verified by other researchers. Second, we are using data from 183 publicly traded
firms rather than 10 firms selected by the study’s authors. Our selection crit eria is that the firms
have research and development expenditure information in the CompuStat data base, be in the
pharmaceutical industry (see Danzon et al., 2004) and have drugs in the Pharmaprojects data set
(see Adams and Brantner, 2006). These firms range in size from 100 employees to almost 180 000
employees with sales ranging from $2m annually to almos t $45b annually. Third, we are using
contemporaneous reports of research and development expenditure where the report s are scrutinized by
both the market and the SEC. In their comment on DiMasi et al. (2003), Light and Warburton (2005)
argue that
considering the clear interest of pharmaceutical companies in higher (rather than lower) estimates of
drug development costs, and sampled firms’ likely awareness of the intended use of the survey data, it
is not unlikely that companies would deliberately and systematically overstate costs in their survey
responses (p. 1031).
We argue that such biases are less likely he re given the large number of firms and the checks on the
reports including audits.
Of course there are also serious concerns about the approach we use here. First, the data are
aggregate research and developm ent expenditure. Those not only include expen diture on drugs in
human clinical trials but also include development expenditure on drugs yet to reach trials. To identify
the amount spent in human clinical trials we must infer the information from cross sectional and time-
series variation in expenditure that is associated with variation in the number of drugs in development.
Such variation may lead to spurious estimates. For example, if one firm specializes in anti-infec tive
drugs and we compare the specialty firm’s expenditure on anti-infective drugs to that of a firm that has
just one or two anti-infective drugs, we may estimate that expenditure on the extra drug as being small.
This low estimate may be due savings from specialization rather than an accurate measure of the cost of
adding another anti-infective drug.
Second, we are estimating changes for the ‘marginal drug’, which may be more expensive than the
average drug.
The relationship between expenditure on the marginal drug and expenditure on the
average drug depends on what assumption the reader is willing to make regarding how expenditure per
Thanks to Eric Durbin for pointing this out.
Copyright r 2009 John Wiley & Sons, Ltd. Health Econ. (2009)
DOI: 10.1002/hec
drug changes with the number of drugs. If expenditure per drug is constant then the marginal and the
average are the same. On the other han d, if expenditure per drug is increa sing with the number of drugs
in development then marginal expenditure will be higher than average expenditure. A number of papers
suggest that there may be economies of scale or scope in drug development (Cockburn and Henderson,
1996, 2001; Danzon et al., 2004). If there are economies of scale then we would expect marginal
expenditure to be less than average expenditure.
Note that marginal expenditure may be a more useful
measure for determining the incentive effects of policy changes.
Third, we use Pharmaprojects’ definition of a ‘drug development project’ and assign the drug to the
‘originator’. In general, this definition corresponds to a new patented molecular entity. In the main
part of the analysis we drop drugs that are new formulations of existing drugs (i.e. an extended
release version of an existing drug). The analysis does not account for the fact that the drug
development project is part of a joint venture (and thus expenditure is spread across multiple firms) or is
being developed by an altogether different firm (and our method is assigning the drug project to the
wrong firm).
Such mis-mea surement may bias our estimates downward. It should be noted that our
counts of drugs in the different phases are measuring the development associated with the originating
In order to have a number that is comparable to DiMasi et al.’s (2003) average expenditure over the
sample period, we control for differences between firms and differences over time. We attempt to
control for some cross-sectional variation by conditioning on net sales. If for example, larger firms
spend more on drug development projects than smaller firms then net sales should control for this
variation. Similarly, if firms are spending more on drug development projects at the end of the period
than at the beginning then our control s for time will provide a better sense of the average expenditure
per project during the period. Note that identification of spending per drugs is coming to some extent
from the fact that larger firms have more drugs and that there are more drugs over time in the database.
The controls attempt to separat ely identify the effect of having another drug in human clinical trials
from the effect of being large or later in time.
DiMasi et al. (2003) uses a similar approach to verify their own estimates. The authors use firm level
R&D expenditure reported by PhRMA and estimate lagged expenditure on fir m level counts of
approved drugs. The authors estimate average expenditure per approved drug to be between $354m and
$558m. These numbers are similar to their estimate of $403m using the survey data. Other researchers
have simply divided aggregate R&D expenditure by the total number of approvals per year. The
concern with these approaches is that less than one in four drugs in human clinical trials actually make it
to the market and the process can take between 6 and 12 years with substantial variation across drugs
(Adams and Brantner, 2003).
The rest of the paper proceeds as follows. Section 2 discusses the data used in this study and provides
some background information on new drug development. Section 3 presents the results. Section 4
This paper combines data from two data sources. Information on each firm’s research and development
expenditure comes from the Standard Poor’s CompuStat Industrial file and Global Vantage Industrial
Commercial file used by Danzon et al. (2004).
This data set provides financial information on publicly
traded drug companie s including net sales, employment and expenditure on research and development.
To the extent one is concerned that large firms may have lower (or higher) expenditure per drug than smaller firms, some of this
variation is accounted for in the analysis through conditioning on sales revenue.
Danzon et al. (2005) analyze joint ventures.
All monetary values are in 1999 dollars using the domestic manufacturing Producer Price Index.
Copyright r 2009 John Wiley & Sons, Ltd. Health Econ. (2009)
DOI: 10.1002/hec
Information on drugs in development comes from a Pharmaprojects data set used by Adams and
Brantner (2006) and Abrantes-Metz et al. (2005). This data set uses public information to track drugs
through the development process, providing information on the length of time in different phase as well
as when and if drugs completed a development phase. The two data sets overlap for the years
1989–2001. The data sets are matched using the name of the pharmaceutical firm.
updates its information on the firms developing each drug after a merger, so we used text searches of the
database and searches of a related data set called the Manufacturing Index to determine the ownership
of drugs over time.
According to Danzon et al. (2004) there are 383 firms in their original data. Once we match these
firms to firms in the Pharmaprojects data we are left with 183 firms. It is not clear exactly why there are
firms that do not matc h. The two data sets do not exactly overlap in time and that may explain some of
it. Another explanation is that the Pharmaprojects does not capture name changes or mergers among
smaller firms (see footnote 7). Table I presents some basic summary statistics for this sample of firm/
year combinat ions. Table I shows there are an average of four drugs in development for each firm for
each year 1989–2001. Note this measure is not a very good measure of the stock of drugs in development
because we only observe drugs entering one of the stages of human clinical trials after 1989. In the
average firm/ year $264m is spent on research and development, $2355m is made in sales and there are
11 000 employees. Note that medians are substantially lower than the means suggesting that the
distributions are all skewed toward zero.
Figures 1–3 present the distribution of the number of drugs in human clinical trials per firm/year, the
amount of R&D expenditure per firm/ year, and a scatter plot of the two, respectively. The first two
figures show that the distributions of drugs and expenditures are heavily skewed to zero. The third
figure seems to show a positive correlation between the amount of R&D expenditure per firm per year
and the number of drugs in development per firm per year.
Figure 4 presents a summary of the research and development process for new drugs. The first
stage of drug discovery is commonly called ‘preclinical development’. In this stage pharmaceutical
firms analyze thousands of drugs to determine whether one may have an affect on a disease or
condition. As candidates are discovered these dru gs are tested on animals to determine whether the
drug may be safe and effective in human beings. It is estimated that drugs spend over 4 years in
preclinical testing. DiMasi et al. (2003) do not ha ve direct survey information on preclinical expenditure
because pharmaceutical firms do not track preclinical expenditure by particular drug candidates.
Given this and given that the Pharmaprojects data are based on public information and are not very
reliable regarding drugs in preclinical development, we do not estimate expenditure on preclinical
After preclinical development the sponsoring firm applies for an investigation new drug application
(IND) with the FDA in order to test the candidate in humans.
There are three steps to human clinical
Table I. Firm/year summary statistics
Variable Obs Mean Median Std. Dev. Max
Number of drugs 2245 4 2 6 45
R&D expenditure ($m) 1682 264 37 551 4678
Net sales ($m) 1701 2355 110 5438 44 611
Employees (’000) 1537 11 1 25 179
This matching was done by hand in order for it to be as accurate as possible.
This was done for all mergers involving firms in the Forbes’ top 20 of pharmaceutical industry over the period as well as any other
major mergers in the pharmaceutical industry.
If the firm wants to eventually market the drug in the US the firm must apply for an IND prior to undertaking human trials. That
said, there are exceptions.
Copyright r 2009 John Wiley & Sons, Ltd. Health Econ. (2009)
DOI: 10.1002/hec
trials. In Phase I, the drug is tested for safety on a small group (e.g. 20) of healthy volunteers. Phase II
tests concentrate on safety but the test is on a larger group of patie nts with the condition (e.g. 200).
Phase III are the large efficacy trials with upwards of 3,000 patients participating. Once the trials are
completed the results of all three stages are presented to the FDA in the form of a new drug application
Table II presents some basic summary statistics on the drugs owned by the firms in the sample. The
first set of three rows show the mean length in months of successful durations. The second set of three
Figure 1. Drugs in development
Figure 2. Annual R&D expenditure
Copyright r 2009 John Wiley & Sons, Ltd. Health Econ. (2009)
DOI: 10.1002/hec
Figure 3. R&D expenditure by drugs in development
Figure 4. CDER chart of the development process
Table II. Summary statistics for drugs
Obs Mean
Duration (months)
Phase I 235 16.58
Phase II 144 30.65
Phase III 130 27.15
Success (frequency)
Phase I 314 0.75
Phase II 302 0.48
Phase III 184 0.71
Copyright r 2009 John Wiley & Sons, Ltd. Health Econ. (2009)
DOI: 10.1002/hec
rows shows the frequency with which drugs successfully complete the phase. The table shows that these
drugs seem to be fairly representative (see Adams and Brantner, 2006; DiMasi et al., 2003). Successful
durations vary by a month or two and success rates vary by a few percentage points of those reported in
Adams and Brantner (2006).
3.1. Mean expendi ture estimates
Table III presents regression results for the amount of research and development expenditure on the
number of drugs in human clinical trials. There are six regressions reported in the table. First are the
basic regressions on the number of drugs in human clini cal trials then on the number of drugs in each of
the three phases of development. These regressions are then repeated adding measures of time and firm
characteristics. All results report robust standard errors clustering on firm name. The number of drugs
is the number of drugs in development for each firm/year combination. Note that ‘new formulations’ of
existing drugs are not included in the count variable.
This is done in order to make the estimates closer
to DiMasi et al. (2003) estimate for new molecular entities.
The variable time is simply the number of
years from 1988. The variable ‘sales’ is the amount of net sales for each firm/year. The time and sales
variables allow the analysis to capture change s in expenditure over the time period and across fir ms,
where ‘sales’ is probably best thought of a measure of firm size.
Table III shows that average expenditure per drug in human clinical trials is between $74m and $27m
per year.
Once we include controls for time and firm characteristics, the results suggest that the
average expenditure on drugs across all three phases of development is approximately $27m per year.
This estimate is quite precise and is statistically different from zero at traditional levels. If sales are not
accounted for then Phase I expen diture is estimated to be $81m per year, $68m for drugs in Phase II and
$77m for drugs in Phase III. Once time and sales are accounted for, these estimates fall to $16m, $34m
and $27m respectively.
The Phase II and III expenditures are estimated precisely and are statistically
different from zero. The Phase I estimate is less precisely estimated and 0 lies within the traditional
confidence interval.
How do these results compare to the estimates of expenditure in DiMasi et al. (2003)? We estimate
the average annual expenditure on drugs in all three phases of human clinical trials is $27m. If we take
DiMasi et al. (2003) estimates of expenditure for each phase of $15m, $24m and $86m for Phases I, II
and III, respectively, and weight them proportionally to the time spent in development and the
probability of being in each of the phases then we have the appropriate co mparison.
transformation gives an estimate of annual expenditure of $21m.
Our estimate is higher than this
transformed estimate from DiMasi et al. (2003) althoug h $21m lies within the 95% confidence
To compare the expenditure by phase it is necessary to do another transformation. The
numbers presented in DiMasi et al. (2003) are for the average drug over the length of the phase, while
A new formulation may, for example, be an extended release version of an existing approved drug.
Thanks to an anonymous referee for this suggestion. The estimates of expenditure per drug including formulations are lower
than the estimates presented here. The last section suggests that this occurs because expenditure on new formulations is
substantially lower than for other drugs.
Most of the decrease seems to come from including the sales variable.
Note that these results are most properly thought of as correlations between the number of drugs and the amount of expenditure.
There has been no effort made to account for endogeniety in the joint decisions to increase expenditure and take more drugs into
clinical trials.
From DiMasi et al. (2003) the average durations are 12, 26 and 34 months, respectively.
The average expenditure is (12
86)/(12126134) 5 (18014431906)/72 5 21.
$21m does not lie within the 90% confidence interval. Although, this does not account for sampling error with the original
DiMasi et al. (2003) estimate.
Copyright r 2009 John Wiley & Sons, Ltd. Health Econ. (2009)
DOI: 10.1002/hec
we have estimated expenditure for 1 year. If we use the phase durations presented in Table II we can
estimate expenditure for the whole phase. This procedure gives 1 .38
17m 5 $24m for Phase I, which is
more than DiMasi et al. (2003) estimate of $15m for Phase I. For Phase II the same method produces an
estimate of $86m, which is much higher than the DiMasi et al. (2003) estimate of $24m. Finally, for
Phase III this method gives an estimate of $61m, which is less than the DiMasi et al. (2003) estimate of
$86m. For Phases I and III, the DiMasi et al. (2003) estimates lie within the 95% confidence interval
around our estimates. However, there is no overlap between the confidence interval around the DiMasi
et al. (2003) estimate of Phase II expenditure and the confidence interval around our estimate.
It is not clear what explains such a large discrepancy between our estimate of Phase II expenditure
and DiMasi et al. (2003) estimate. One possibility and a more general concern is that our method may
be misallocating expenditure to drugs in different stages of development. This may occur for two
reasons. First, we assume that if a drug moves into a new phase in a particular year then the drug has
been in that phase for the whole year. Still, given the expected difference in Phases II and III expenditure
this assumption is more likely to lead to an underestimate of Phase III expenditure than an overestimate
of Phase II expenditure. Second, the relationship between financial years and the years assigned in the
data. Again, this may reduce the accuracy of the estimates but it is unlikely to bias the estimates.
Another possibility is that there is under reporting of drugs in human clinical trials particularly in the
earlier phases.
As other work ha s shown, expenditure on research and development is increasing at a substantial
rate. In fact, here we have it increasing at a parabolic rate although this is due to the particular
functional form that is used in the estimation. The results also show that there is a strong relationship
between sales and research and development expenditure with every $1 in sales associated with an extra
$0.07 in R&D expenditure. Note also, adding sales to the regression substantially improves the model’s
ability to explain the data. The constant in this estimation cannot really be interpreted as we do not have
a measure of the stock of drugs in development as of 1989. We are only able to observe new
development starts in 1989 and later.
The baseline regres sion measures the average expenditure on new drugs by firm and year. It does not
account for whether that average may be driven up by the large increase in R&D expenditure observed
Table III. R&D expenses OLS (robust standard errors)
12345 6
All phases 74.31
(5.87) (5.91) (3.44)
Phase I 80.72
(18.61) (18.61) (10.35)
Phase II 68.01
(12.72) (12.67) (6.80)
Phase III 76.94
(23.90) (24.07) (11.08)
Time 62.34
(10.88) (13.08) (5.91) (6.76)
(0.83) (0.97) (0.46) (0.53)
Sales 0.07
(0.01) (0.01)
Constant 22.64 21.49 147.19
19.87 23.16
(20.88) (21.00) (31.04) (36.18) (17.20) (17.69)
Observations 1682 1682 1682 1682 1682 1682
0.59 0.59 0.60 0.60 0.89 0.89
Standard errors are clustered on firm names.
Is statisically different from 0 at 1% level and
is at 5% level.
It is not clear exactly how such under reporting would bias the results and in what direction.
Copyright r 2009 John Wiley & Sons, Ltd. Health Econ. (2009)
DOI: 10.1002/hec
during the period or by large expenditures by the large r pharmaceutical firms. Latter regressions add a
parabolic time trend and a parameter (‘sales’) to capture variation across firms. The time trend estimates
capture the large increase in expenditure that occurred during the period. The sales coefficient suggests
that large firms, at least large firm/years, are associated with large expenditures per drug.
variation across firms is also captured to some extent via the quantile regression analysis presented in
the next section.
This analysis suggests larger firms spend more on clinical trials.
Note that the measured positive relationship between sales and expenditure may not be causal.
Larger pharmac eutical firms may have different R&D strategies than smaller firms. For example, Big
Pharma may run substantially more trials over different treatments, comparison groups, and
populations compared with smaller firms. Such trials may put the firm in a better position to sell the
drug internationally and in multiple domestic markets for different indications. Note also that these
numbers do not measure expenditure by smaller non-publicly traded firms such as those funded by
venture capital firms.
3.2. Quartile e xpenditure estimates
Table IV presents results from the 25, 50 and 75% quartile regressions. Comparing these results to the
results from columns 5 and 6 in Table III we see that expenditure per drug per year is substantially less
at the lower quartiles. At the bottom quartile, expenditure per new drug in development is around $9m
per year, with $15m at the median and $18m at the top quartile. All these numbers are estimated fairly
precisely. These numbers co mpare to $27m per year for the mean. Similarly, estimated expenditures per
phase of development are substantially lower at the 25, 50 and 75% quartiles relative to the mean. The
results suggest that the distribution of expenditure on drug development is quite skewed. These results
suggest that a number of firms spend very large sums on drug development.
As before, we can transform our median estimat es to compare with the median estimates presented in
DiMasi et al. (2003). This procedure gives 1.38
14.60 5 $20m for Phase I, which is more than DiMasi
et al. (2003) estimate of $14m for Phase I. For Phase II the same method produces an estimate of $36m,
which is much higher than the DiMasi et al. (2003) estimate of $17m. Finally, for Phase III this method
gives an estimate of $30m, which is much less than the DiMa si et al. (2003) estimate of $62m.
3.3. Implications for cost estimates
If we use the mean estimates for expenditure on drugs in development in place of the survey estimates
used by DiMasi et al. (2003) we can recalculate the over all ‘cost of drug development’ or more
accurately the net revenue needed to mak e investment in drug development profitable. Doing this
calculation using the same durations and success rates as reported in Adams and Brantner (2006) we
estimate new drug development cost to be $1214m, which is much higher than the original estimate of
$802m or even the Adams and Brantner (2006) estimate of $867m. These high estimates may be due to
measurement of expenditure on the marginal drug rather than the average drug.
However, such an
estimate may be more useful to policy makers as it is more likely to measure the impact of changes in
The baseline analysis is measured at the firm/year level. This means that we are measuring the expenditure of the average firm/
year on a drug rather than the expenditure on the average drug. If the average firm/year is large then our measure may be high
because large firms happen to spend more on drugs. By adding a coefficient for firm size the measure can be adjusted to account
for the variation in expenditure across firms. Note however, that if the average drug is developed in a large firm then these results
may need to be adjusted either by adding back in the sales coefficient multiplied by the sales of the firm, which produces the
average drug or by looking at the quartile estimates in the next section.
Note that the coefficient estimate on sales is larger for the 75% quartile compared with the 50% quartile and the 25% quartile.
That is, for drugs with larger expenditures there is a stronger relationship between the size of the firm and the size of the
Note also that these estimates are based on the very high Phase II expenditure estimates.
Copyright r 2009 John Wiley & Sons, Ltd. Health Econ. (2009)
DOI: 10.1002/hec
policy on the development of new drugs. We could interpret these estimates as stating a firm would need
expected net revenue of over $1 billion to develop one more drug for the market.
3.4. Expenditure by therapy
DiMasi et al. (2004) presents estimates of drug development costs for a small number of major
therapies. In attempt to replicate this work, Table V presents results similar to those presented in
Table III but where the drug counts are by major therapeutic category. Table V presents the marginal
cost of a drug by major therapy grouping. This number is estimated by counting the number of drugs in
human clinical trials for each of the major categories presented.
Note, that we would not expect these
numbers to be negative.
The table shows cardiovascular, dermatological, genitourinary, anticancer
and neurological drugs all have more expenditure per drug in human clinical tria ls than the average
drug. Note, however, only genitourinary drugs are estimated to be statistically different from the
average at traditional levels. New formulations of existing drugs have substantially smaller expenditure
in human clinical trials than the average drug. In fact, expenditure on new formulation is not estimated
to be statistically different from zero, but is statistically different from the average. The reader may be
surprised that biotech drugs are estimated to have less than average expenditure (although the estimate
is not statistically different from the average). This may be due to imprecise measurement, or it may be
due to the way these drugs are categorized in the data. That is, more important biological drugs may be
categorized under their indication (anticancer or musculoskeletal) rather than as a product of the
biotech industry.
Some of these resul ts can be compared with the results presented in DiMasi et al. (2004). It should be
noted, however, that in both this paper and the DiMasi et al. (2004) paper the sample sizes for
individual therapeutic categories can be quite small. There are three categories in which both papers
Table IV. R&D expenses quantiles (standard errors)
25% 25% 50% 50% 75% 75%
All phases 8.88
(0.12) (0.19) (0.15)
Phase I 7.65
(0.42) (0.55) (0.44)
Phase II 10.73
(0.36) (0.46) (0.39)
Phase III 7.84
(0.44) (0.58) (0.50)
Time 2.48
(0.69) (0.80) (1.01) (1.00) (0.80) (0.85)
(0.05) (0.05) (0.07) (0.07) (0.05) (0.06)
Sales 0.07
(0.00) (0.00) (0.00) (0.00) (0.00) (0.00)
Constant 2.15 2.16 6.82
(2.27) (2.61) (3.31) (3.26) (2.52) (2.69)
Observations 1682 1682 1682 1682 1682 1682
Pseudo R
0.60 0.60 0.70 0.70 0.80 0.80
Standard errors are in parenthesis.
Refers to statistical significance from 0 at the 1% level,
at the 5% level.
Thanks to Mark Duggan for his thoughts about the differences between marginal and average drugs.
Note that we only include the larger therapeutic categories as smaller categories do not have enough drugs in human clinical
trials for reliable estimates. Thanks to an anonymous referee for this suggestion. We have also estimated expenditure by therapy
for each phase, but we do not include these results because of concern about the reliability of the estimates given the small
number of drugs in each phase for each therapy.
Therefore, the observed negative coefficients suggest that the model may be misspecified, although none of the negative
coefficients are statistically significantly different from zero.
Copyright r 2009 John Wiley & Sons, Ltd. Health Econ. (2009)
DOI: 10.1002/hec
report results cardiovascular, anti-infective and neurological/CNS.
For these three categories the
expenditure estimates for drugs in any of the three phases of human clinical trials are $36m, $27m and
$33m, respectively, where the estimate for cardiovascular drugs and neurological drugs are stat istically
significantly different from 0 at the 95% level and the estimate for anti-infective drugs is not. To
compare these numbers to the numbers in DiMasi et al. (2004) we can make the same transformation as
above to get estimates of $17m, $33m and $17m respectively.
While the magnitudes of the estimates do
not differ that greatly, parti cularly for anti-infectives, the DiMasi et al. (2004) estimates are also located
within the traditional confidence inter vals of the estimates presented in Table V.
These results support the argument in Adams and Brantner (2006) that there is substantial variation
in drug development costs by therapeutic category. On average annual expenditure by drug is $27m.
However, for genitourinary drugs like Viagra, the expend iture number is four times higher. It is also
higher for dermatologica l and cancer drugs. In contrast annual expenditure on new formulations of
existing drugs is only a couple of million of dollars. Do therapeutic categories with high annual
expenditures have high success rates and short durations to compensate? Not necessarily. Results
presented in Adams and Brantner (2006) suggest genitourinary drugs do have very high success rates
and relatively short durations. This means that their overall drug development costs may not be too
different from ‘the average drug’. However, cancer drugs have relatively low success rates and long
durations making them much more expensive than the average drug (Adams and Brantner, 2006).
Recent critici sm of the study by DiMasi et al. (2003) argues it is not pos sible to verify the results because
the data are confidential. Further, Light and Warburton (2005) argue the sample of expenditure
estimates may not be representative and may be biased upwards. This paper attempts to replicate
DiMasi et al. (2003) expenditure estimates using publicly available da ta. By matc hing information on
Table V. R&D expenses OLS by therapy for all phases
Coefficient Robust SE Coefficient Robust SE Coefficient Robust SE
Cardiovascular 137.65
26.28 141.48
26.05 35.85
Dermatological 254.60
106.99 261.49
104.54 99.71
Formulations 3.69 8.96 6.28 8.42 4.01 3.45
Genitourinary 58.96 50.14 52.98 48.27 108.04
Anti-infective 129.96
35.87 130.07
35.50 26.56 14.83
Anticancer 56.17
27.82 52.25
26.50 43.14
Neurological 136.01
29.25 139.13
29.33 33.32
Biotechnology 26.89 15.03 26.48 14.88 24.40
Miscellaneous 10.14 16.79 9.31 16.24 2.81 5.08
Time 63.53
11.06 22.89
0.84 1.96
Sales 0.07
Constant 7.54 22.91 168.53
34.66 25.67 17.71
Observations 1682 1682 1682
0.65 0.66 0.90
Standard errors are clustered on firm names.
Refers to statistical significance at the 1% level and
refers to the 5% level.
The last category may not overlap as the definition of CNS drugs in DiMasi et al. (2004) may not be the same as the definition of
neurological drugs used in Pharmaprojects.
The formula is the same as the one presented in footnote 14 where the expenditure estimates and the probabilities are from
DiMasi et al. (2004) and the proportion of time in each phase is as for the original formula. Note that DiMasi et al. (2004) do not
present duration by phase for individual therapeutic categories.
Copyright r 2009 John Wiley & Sons, Ltd. Health Econ. (2009)
DOI: 10.1002/hec
drugs in development with research and development expenditure over the period 1989–2001 and across
some 180 firms, we infer the additional annual expenditure on each new drug in development. Our
results suggest expend iture on Phases I and II is higher than suggested by DiMasi et al. (2003) while
expenditure on Phase III is lower. If we combine our estimates in this paper with estimates on success
rates and durations from Adam s and Brantner (2006) we find that the ‘cost of drug development’ (or the
net revenue needed to make investment in new drugs profitable) is over $1 billion and higher than the
DiMasi et al. (2003) estimate of $802m. As we are estimating expenditure on the additional drug we may
be estimating the revenue needed to invest in the marginal rather than the average drug. While this may
make our estimate higher, it may also make our estimate more useful when considering the
consequences of policy changes such as price regulation. This paper also confirms results presented in
Adams and Brantner (2006) that there is a substantial amount of variation in expenditure by therapeutic
Abrantes-Metz R, Adams C, Metz A. 2005. Pharmaceutical development phases: a duration analysis. Journal
Pharmaceutical Finance, Economics and Policy 14(4): 19–42.
Adams C, Brantner V. 2003. New drug development: estimating entry from human clinical trials. FTC Working
Paper 262, July 7, 2003.
Adams C, Brantner V. 2006. Estimating the cost of new drug development: is it really $802m? Health Affairs
March/April: 420–428.
Cockburn I, Henderson R. 1996. Scale, scope, and spillovers: determinants of research productivity in the
pharmaceutical industry. RAND Journal of Economics 27(1): 32–59.
Cockburn I, Henderson R. 2001. Scale and scope in drug development: unpacking the advantages of size in the
pharmaceutical research. Journal of Health Economics 20(6): 1033–1057.
Danzon P, Epstein A, Nicholson S. 2004. Mergers and Acquisitions in the Pharmaceutical and Biotech Industries,
The Wharton School, University of Pennsylvania, November.
Danzon P, Nicholson S, Sousa Pereira N. 2005. Productivity in biotech-pharmaceutical R&D: the role of experience
and alliances. Journal of Health Economics. 24(2): 317–339.
DiMasi J, Hansen R, Grabowski H. 2003. The price of innovation: new estimates of drug development costs.
Journal of Health Economics 22: 151–185.
DiMasi J, Hansen R, Grabowski H. 2004. R&D costs and returns by therapeutic category. Drug Information
Journal 38: 211–223.
Light D, Warburton R. 2005. Extraordinary claims require extraordinary evidence. Journal of Health Economics 24:
Copyright r 2009 John Wiley & Sons, Ltd. Health Econ. (2009)
DOI: 10.1002/hec
... When firms decide whether to commence the drug development process, they need to weigh the high costs of developing a drug candidate with the considerable but highly uncertain benefits after market approval (Adams & Brantner, 2006;Arora et al., 2009;DiMasi et al., 2003). Scientific insights are considered to be an important potential input for this decision: to increase the efficiency of the drug development process and to more accurately predict the chances of success before making large investments in clinical trials, practitioners and industry experts have emphasized the importance of science (Schuhmacher et al., 2016, p123). ...
Full-text available
Pharmaceutical firms are extremely selective in deciding which patented drug candidates are taken up into clinical development, given the high costs and risks involved. We argue that the scientific base of drug candidates, and who was responsible for that scientific research, are key antecedents of take-up into clinical trials and whether the patent owner (‘internal take-up’) or another firm (‘external take-up’) leads the clinical development effort. We hypothesize that patented drug candidates that refer to scientific research are more likely to be taken up in development, and that in-house conducted scientific research is predominantly associated with internal take-up due to the ease of knowledge transfer within the firm. Examining 18,360 drug candidates patented by 136 pharmaceutical firms we find support for these hypotheses. In addition, drug candidates referring to in-house scientific research exhibit a higher probability of eventual drug development success. Our findings underline the importance of a ‘rational drug design’ approach that explicitly builds on scientific research. The benefits of internal scientific research in clinical development highlight the potential downside of pervasive organizational specialization in the life sciences in either scientific research or clinical development.
... Es prácticamente un lugar común sostener que la I&D farmacéutico es costoso y con tasas de falla relevantes 10,11 , no obstante el constante debate que sigue existiendo respecto a su verdadera magnitud 12 . ...
... Drug manufacturing is an extremely expensive and time-consuming process [1] requires approximated 14 years, with a total cost of about $1 billion, for a specific drug to be available in the pharmaceutical market [2]. Nevertheless, even when being in clinical uses for a while, side effects of many drugs are still unknown to scientists and/or clinical doctors [3]. Understanding drugs' side effects is essential for drug safety and toxicity. ...
Chemical compounds (drugs) and diseases are among top searched keywords on the PubMed database of biomedical literature by biomedical researchers all over the world (according to a study in 2009). Working with PubMed is essential for researchers to get insights into drugs’ side effects (chemical-induced disease relations (CDR), which is essential for drug safety and toxicity. It is, however, a catastrophic burden for them as PubMed is a huge database of unstructured texts, growing steadily very fast (~28 millions scientific articles currently, approximately two deposited per minute). As a result, biomedical text mining has been empirically demonstrated its great implications in biomedical research communities. Biomedical text has its own distinct challenging properties, attracting much attetion from natural language processing communities. A large-scale study recently in 2018 showed that incorporating information into indenpendent multiple-input layers outperforms concatenating them into a single input layer (for biLSTM), producing better performance when compared to state-of-the-art CDR classifying models. This paper demonstrates that for a CNN it is vice-versa, in which concatenation is better for CDR classification. To this end, we develop a CNN based model with multiple input concatenated for CDR classification. Experimental results on the benchmark dataset demonstrate its outperformance over other recent state-of-the-art CDR classification models. Keywords: Chemical disease relation prediction, Convolutional neural network, Biomedical text mining References [1] Paul SM, S. Mytelka, C.T. Dunwiddie, C.C. Persinger, B.H. Munos, S.R. Lindborg, A.L. Schacht, How to improve R&D productivity: The pharmaceutical industry's grand challenge, Nat Rev Drug Discov. 9(3) (2010) 203-14. [2] J.A. DiMasi, New drug development in the United States from 1963 to 1999, Clinical pharmacology and therapeutics 69 (2001) 286-296. [3] C.P. Adams, V. Van Brantner, Estimating the cost of new drug development: Is it really $802 million? Health Affairs 25 (2006) 420-428. [4] R.I. Doğan, G.C. Murray, A. Névéol et al., "Understanding PubMed user search behavior through log analysis", Oxford Database, 2009. [5] G.K. Savova, J.J. Masanz, P.V. Ogren et al., "Mayo clinical text analysis and knowledge extraction system (cTAKES): Architecture, component evaluation and applications", Journal of the American Medical Informatics Association, 2010. [6] T.C. Wiegers, A.P. Davis, C.J. Mattingly, Collaborative biocuration-text mining development task for document prioritization for curation, Database 22 (2012) pp. bas037. [7] N. Kang, B. Singh, C. Bui et al., "Knowledge-based extraction of adverse drug events from biomedical text", BMC Bioinformatics 15, 2014. [8] A. Névéol, R.L. Doğan, Z. Lu, "Semi-automatic semantic annotation of PubMed queries: A study on quality, Efficiency, Satisfaction", Journal of Biomedical Informatics 44, 2011. [9] L. Hirschman, G.A. Burns, M. Krallinger, C. Arighi, K.B. Cohen et al., Text mining for the biocuration workflow, Database Apr 18, 2012, pp. bas020. [10] Wei et al., "Overview of the BioCreative V Chemical Disease Relation (CDR) Task", Proceedings of the Fifth BioCreative Challenge Evaluation Workshop, 2015. [11] P. Verga, E. Strubell, A. McCallum, Simultaneously Self-Attending to All Mentions for Full-Abstract Biological Relation Extraction, In Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies 1 (2018) 872-884. [12] Y. Shen, X. Huang, Attention-based convolutional neural network for semantic relation extraction, In: Proceedings of COLING 2016, the Twenty-sixth International Conference on Computational Linguistics: Technical Papers, The COLING 2016 Organizing Committee, Osaka, Japan, 2016, pp. 2526-2536. [13] Y. Peng, Z. Lu, Deep learning for extracting protein-protein interactions from biomedical literature, In: Proceedings of the BioNLP 2017 Workshop, Association for Computational Linguistics, Vancouver, Canada, 2016, pp. 29-38. [14] S. Liu, F. Shen, R. Komandur Elayavilli, Y. Wang, M. Rastegar-Mojarad, V. Chaudhary, H. Liu, Extracting chemical-protein relations using attention-based neural networks, Database, 2018. [15] H. Zhou, H. Deng, L. Chen, Y. Yang, C. Jia, D. Huang, Exploiting syntactic and semantics information for chemical-disease relation extraction, Database, 2016, pp. baw048. [16] S. Liu, B. Tang, Q. Chen et al., Drug–drug interaction extraction via convolutional neural networks, Comput, Math, Methods Med, Vol (2016) 1-8. [17] L. Wang, Z. Cao, G. De Meloet al., Relation classification via multi-level attention CNNs, In: Proceedings of the Fifty-fourth Annual Meeting of the Association for Computational Linguistics 1 (2016) 1298-1307. [18] J. Gu, F. Sun, L. Qian et al., Chemical-induced disease relation extraction via convolutional neural network, Database (2017) 1-12. [19] H.Q. Le, D.C. Can, S.T. Vu, T.H. Dang, M.T. Pilehvar, N. Collier, Large-scale Exploration of Neural Relation Classification Architectures, In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, 2018, pp. 2266-2277. [20] Y. LeCun, L. Bottou, Y. Bengio, P. Haffner, Gradient-based learning applied to document recognition, In Proceedings of the IEEE. 86(11) (1998) 2278-2324. [21] Y. Kim, Convolutional neural networks for sentence classification, ArXiv preprint arXiv:1408.5882. [22] C. Nagesh, Panyam, Karin Verspoor, Trevor Cohn and Kotagiri Ramamohanarao, Exploiting graph kernels for high performance biomedical relation extraction, Journal of biomedical semantics 9(1) (2018) 7. [23] H. Zhou, H. Deng, L. Chen, Y. Yang, C. Jia, D. Huang, Exploiting syntactic and semantics information for chemical-disease relation extraction, Database, 2016.
... They have tripled survival rates in acute promyelocytic and chronic myeloid leukaemia, as well as medullary thyroid cancer. 41 However, the costs of cancer treatments are increasing [42][43][44] and there are two main drivers behind this. Firstly, the rates of cancer are predicted to increase as the population ages, [45][46][47] and secondly, the innovations in cancer treatment come at an increasing cost, with cancer drug prices rising five-times faster than other classes of medicine. ...
Biologicals have revolutionised modern medicine by offering vital therapeutic options to treat or prevent complex, disabling, and life-threatening diseases. Between 2013 and 2018, seven of the top ten pharmaceuticals worldwide will be biologicals; however, growing demand, combined with historically-limited competition, will continue to strain healthcare budgets and limit patient access to these treatments. Since 2006, when the first biosimilar Omnitrope® was approved in Europe, 18 other biosimilars, including the first biosimilar monoclonal antibody (mAb), infliximab (approved in 2013), have received marketing authorisation with many others currently in development. There is now extensive clinical experience with biosimilar epoetin (EPO) and filgrastim in patients with cancer, and many studies have reported comparable efficacy with the originator products, no unexpected safety concerns, and significant economic savings. Nevertheless, misconceptions concerning biosimilars remain. This educational session discussed these issues and gave an overview of biosimilar use in hematology. Dr Joerg Windisch highlighted the particular challenges and considerations associated with the development of biosimilars while Prof Steffen Thirstrup covered the approval of biosimilars from the regulatory perspective. Dr Wojciech Jurczak gave a presentation on the development of biosimilars in hematology, with a particular focus on rituximab from a clinical perspective. Dr Paul Cornes concluded with the opportunities that the introduction of biosimilars offer in terms of health economics and improved patient access to care.
In the past decades, cancer therapies are evolved from a non-specific approach to a precise and accurate therapy. Precision (or personalized) cancer therapy, in which a patient’s own tumor information is used to help make a diagnosis, a treatment plan, or a prognosis, provides an encouraging solution to overcome the failure of current conventional treatments. Relevant preclinical animal models are critical to translate this concept into clinical application. In this chapter, we overviewed how immunodeficient mouse strains were contributed to the development of personalized immune-oncology therapy, so-called personalized I/O therapy. Patient-derived xenograft (PDX) mice, which are generated by directly engrafting tumor tissues into immunodeficient mice, can largely retain the genetics of the human tumors. PDX models are a versatile platform for the study of tumor biology, and are highly predictive of patients’ responses to many cancer therapies. The human immune-tumor interaction can be further replicated in the PDX-human immune system (HIS) mice after a transplantation of PDX mice with allogeneic or autologous immune cells. This PDX-HIS dual-humanized model represents a valuable avatar of patients in clinical decision-making for I/O therapy. Many novel strategies based on personalized I/O treatment are either under preclinical study or in clinical trials, and represent our hope for fighting the cancer in the future.KeywordsHuman immune system (HIS)Humanized mouse modelsImmune-oncology therapyPatient-derived xenograft (PDX)Personalized medicinePrecision medicine
Accurate and effective drug-target interaction (DTI) prediction can greatly shorten the drug development lifecycle and reduce the cost of drug development. In the deep-learning-based paradigm for predicting DTI, robust drug and protein feature representations and their interaction features play a key role in improving the accuracy of DTI prediction. Additionally, the class imbalance problem and the overfitting problem in the drug-target dataset can also affect the prediction accuracy, and reducing the consumption of computational resources and speeding up the training process are also critical considerations. In this paper, we propose shared-weight-based MultiheadCrossAttention, a precise and concise attention mechanism that can establish the association between target and drug, making our models more accurate and faster. Then, we use the cross-attention mechanism to construct two models: MCANet and MCANet-B. In MCANet, the cross-attention mechanism is used to extract the interaction features between drugs and proteins for improving the feature representation ability of drugs and proteins, and the PolyLoss loss function is applied to alleviate the overfitting problem and the class imbalance problem in the drug-target dataset. In MCANet-B, the robustness of the model is improved by combining multiple MCANet models and prediction accuracy further increases. We train and evaluate our proposed methods on six public drug-target datasets and achieve state-of-the-art results. In comparison with other baselines, MCANet saves considerable computational resources while maintaining accuracy in the leading position; however, MCANet-B greatly improves prediction accuracy by combining multiple models while maintaining a balance between computational resource consumption and prediction accuracy.
The COVID-19 pandemic created a large, sudden unmet public health need for rapid access to safe and effective treatments. Against this backdrop, policy makers and researchers have looked to drug repurposing-using a drug previously approved for one indication to target a new indication-as a means to accelerate the identification and development of COVID-19 treatments. Using detailed data on US clinical trials initiated during the pandemic, we examined the trajectory and sources of drug repurposing initiatives for COVID-19. We found a rapid increase in repurposing efforts at the start of the pandemic, followed by a transition to greater de novo drug development. The drugs tested for repurposing treat a wide range of indications but were typically initially approved for other infectious diseases. Finally, we documented substantial variation by trial sponsor (academic, industry, or government) and generic status: Industry sponsorship for repurposing occurred much less frequently for drugs with generic competitors already on the market. Our findings inform drug repurposing policy for both future emerging diseases and drug development in general.
Full-text available
We examine the determinants and effects of M&A activity in the pharmaceutical|biotechnology industry using SDC data on 383 firms from 1988 to 2001. For large firms, mergers are a response to expected excess capacity due to patent expirations and gaps in a firm's product pipeline. For small firms, mergers are primarily an exit strategy in response to financial trouble (low Tobin's q, few marketed products, low cash-sales ratios). In estimating effects of mergers, we use a propensity score to control for selection based on observed characteristics. Controlling for merger propensity, large firms that merged experienced a similar change in enterprise value, sales, employees, and R&D, and had slower growth in operating profit, compared with similar firms that did not merge. Thus mergers may be a response to trouble, but they are not a solution. Copyright © 2007 John Wiley & Sons, Ltd.
Full-text available
The clinical period (i.e. clinical trial and long term animal testing) development costs of a random sample of new chemical entities (NCEs) were examined for differences in average cost. All of the NCEs studied were first tested in humans between 1970 and 1982, and were classified for the purposes of the study by therapeutic class. The costs of unsuccessful projects were included with those of projects that resulted in US marketing approval. Including income forgone from expending funds before returns are earned (‘time costs’), the capitalised (i.e. out-of-pocket plus time) clinical period costs per approved NCE were $US70, $US98, $USI03 and $US163 million (1993 dollars) for anti-infective, cardiovascular, neuropharmacological and nonsteroidal anti-inflammatory drugs, respectively. Combining the data for all therapeutic categories, the mean clinical period cost per approved NCE was $US93 million. Omitting costs associated with unsuccessful projects, the mean capitalised clinical period costs for approved NCEs ranged from $US7.1 million (for topical steroids) to $US66.7 million (for cardiovascular agents) [ 1993 dollars ]. The estimates of total clinical period costs per approved NCE depend on average out-of-pocket clinical phase costs, attrition rates across phases (i.e. the rates at which compounds drop out of active testing), the probability of marketing approval and deve lopment and regulatory review times. Phase attrition and approval rates are the most imponant sources of variability in total clinical period costs between therapeutic categories. Development cost estimates by therapeutic category did not correlate strongly with US sales in the fifth year of marketing. Cardiovascular NCEs had much higher than average sales revenues but clinical development costs for these drugs were only slightly above average. Conversely, nonsteroidal anti-inflammatory drugs attained average sales revenues but had much higher than average development costs.
Full-text available
We examine the relationship between firm size and research productivity in the pharmaceutical industry. Using detailed internal firm data, we find that larger research efforts are more productive, not only because they enjoy economies of scale, but also because they realize economies of scope by sustaining diverse portfolios of research projects that capture internal and external knowledge spillovers. In pharmaceuticals, economies of scope in research are important in shaping the boundaries of the firm, and it may be worth tolerating the static efficiency loss attributable to the market power of large firms in exchange for their superior innovative performance.
Objectives: This study examines the degree to which therapeutic class accounts for variability in drug development costs. It also scrutinizes how sales levels vary across the associated therapeutic classes for those drugs that have reached the marketplace. Data and Methods: A stratified random sample of 68 investigational drugs that first entered clinical testing anywhere in the world from 1983 to 1994 was selected from the pipelines of 10 pharmaceutical firms. Clinical period cost data were obtained for these compounds by phase. The sample consisted both of drugs that failed in testing and drugs that obtained marketing approval. We grouped the drugs by therapeutic category. Clinical period costs per approved new drug (inclusive of failures) were obtained for the analgesic/anesthetic, antiinfective, cardiovascular, and central nervous system (CNS) therapeutic classes. Worldwide sales profiles for new drugs approved in the United States from 1990 to 1994 over a 20-year product life cycle were computed based on IMS Health sales data. All costs and sales were expressed in year 2000 dollars. Results: Out-of-pocket clinical period cost per approved drug (inclusive of failures) for cardiovascular ($277 million) and CNS ($273 million) drugs was close to the overall average ($282 million). However, antiinfective drug costs were considerably above average ($362 million) and analgesic/anesthetic drug costs were modestly below average ($252 million). The results were qualitatively similar when the development timelines were used to determine capitalized (out-of-pocket plus time) costs. In comparison to the overall average of $466 million, the capitalized cost per approved drug was slightly lower for CNS ($464 million) and for cardiovascular ($460 million) drugs. The capitalized costs were $375 million for analgesic/anesthetic drugs and $492 million for antiinfective drugs. The mean net present values of life cycle sales for new drugs approved in the first half of the 1990s were $2434 million, $ 1080 million, $2199 million, $3668 million, and $4177 million for all drugs, analgesic/anesthetic drugs, antiinfective drugs, cardiovascular drugs, and CNS drugs, respectively. Conclusions: Development costs vary substantially from drug to drug. A drug's therapeutic class can explain some of that variability. The sales of new drugs by broad therapeutic category did not correlate well with average development costs. However, given the dynamic nature of pharmaceutical markets and changes over time in research and development (R&D) expenditure shares, the results are still consistent with a model of firm behavior that posits that R&D efforts will generally shift toward high net return, and away from low net return, therapeutic areas.
This paper estimates a duration model of late stage drug development in the pharmaceutical industry using publicly available data. The paper presents descriptive results on the estimated relationship between a particular drug's characteristics such as therapy category, route of administration and originator's size, and that drug's pathway through the three stages of human clinical trials and regulatory review. The results suggest that drugs with longer durations are less likely to succeed, drugs from larger firms are more likely to succeed and faster in the later phases of development, and that durations fell between 1995 and 2002.
This paper analyses a detailed data set on drugs in human clinical trials around the world between 1989 and 2002. The data provides information on the probabilities with which drugs successfully complete the different phases of the trials and the durations of successful completions. The paper shows that success rates and durations can vary substantially across observable characteristics of the drugs, including primary indication, originating company, route of administration and chemistry. It suggests that analysis of this type of data can help us to answer questions such as: Do AIDS drugs get to market faster? Do Biotech drugs have higher probabilities of getting to market? This paper provides some general statistics for analyzing these questions.
Drug development performance is examined using data on clinical research projects of 10 pharmaceutical companies. In contrast to previous work on the discovery phase of pharmaceutical R&D we find a strong correlation between the diversity of firms' development efforts and the success probability of individual projects, but no effect of scale per se. Large firms' superior performance in drug development appears to be driven by returns to scope rather than returns to scale. Scope is confounded with firm fixed effects, however, suggesting an important role for inter-firm differences in the organization and management of the development function.
The research and development costs of 68 randomly selected new drugs were obtained from a survey of 10 pharmaceutical firms. These data were used to estimate the average pre-tax cost of new drug development. The costs of compounds abandoned during testing were linked to the costs of compounds that obtained marketing approval. The estimated average out-of-pocket cost per new drug is 403 million US dollars (2000 dollars). Capitalizing out-of-pocket costs to the point of marketing approval at a real discount rate of 11% yields a total pre-approval cost estimate of 802 million US dollars (2000 dollars). When compared to the results of an earlier study with a similar methodology, total capitalized costs were shown to have increased at an annual rate of 7.4% above general price inflation.
Using data on over 900 firms for the period 1988-2000, we estimate the effect on phase-specific biotech and pharmaceutical R&D success rates of a firm's overall experience, its experience in the relevant therapeutic category, the diversification of its experience across categories, the industry's experience in the category, and alliances with large and small firms. We find that success probabilities vary substantially across therapeutic categories and are negatively correlated with mean sales by category, which is consistent with a model of dynamic, competitive entry. Returns to experience are statistically significant but economically small for the relatively straightforward phase 1 trials. We find evidence of large, positive and diminishing returns to a firm's overall experience (across all therapeutic categories) for the larger and more complex late-stage trials that focus on a drug's efficacy. There is some evidence that a drug is more likely to complete phase 3 if developed by firms whose experience is focused rather than broad (diseconomies of scope). There is evidence of positive knowledge spillovers across firms for phase 1. However, for phase 2 and phase 3 the estimated effects of industry-wide experience are negative, which may reflect either higher Food and Drug Administration (FDA) approval standards in crowded therapeutic categories or that firms in such categories must pursue more difficult targets. Products developed in an alliance tend to have a higher probability of success, at least for the more complex phase 2 and phase 3 trials, and particularly if the licensee is a large firm.