Contents
Introduction ....................................................................................................................................2
Sample Selection Bias .............................................................................................................3
Costs of Pre-Clinical Trial......................................................................................................4
Costs of Clinical Trial .............................................................................................................5
Sample Size ..............................................................................................................................6
Phase-Specific Development Times .......................................................................................7
Opportunity Cost Estimates .................................................................................................8
Tax Treatments .......................................................................................................................9
Conclusion ....................................................................................................................................10
Introduction
At the time of publishing of the paper from DiMasi et al, DiMasi had been a researcher in the
field for 10 years. This paper is the most cited research document relating to pharmaceutical
R&D expenses, and shows that DiMasi is considered a world-leading expert in the field.
Furthermore, the Tufts Center for the Study of Drug Development (CSDD) is supported by the
largest companies in the pharmaceutical industry and has unparalleled access to industry specific
data. Clearly, they are in the best position to produce high quality, unbiased research relating to
pharmaceutical R&D costs.
However, the Tufts Center is industry sponsored. One could argue that since DiMasi et al are
supported by the pharmaceutical industry, there is pressure to produce research that would help
support a continued relationship. This working relationship produces three fundamental
problems in evaluating their research findings:
1) The supported economists are unlikely to publish findings that would damage the
relationship with their sponsors.
2) The sponsoring companies are unlikely to provide data or information that would hinder
their economic goals. Automatically we assume that any provided data would be biased
in whatever direction serves their economic purpose.
3) Any data provided is only available to the Tufts center economists, so any comparative
tests to validate or oppose the results are not possible.
In the paper by DiMasi et al, the selected sample of drugs are representative of the
therapeutic classes. This means that the results of the study do not over-weight the most costly
classes of drugs to develop. Moreover, pharmaceutical R&D expenditure growth rates for the
survey firms were similar to the reported growth rates for all PhRMA member firms 1. Therefore,
the firms who provided data are not expected to have higher than average development costs.
However, according to Light and Warburton (L/W), DiMasi et al did not use a randomly selected
set of companies to participate in the study. Out of the 24 firms invited to participate in the study,
only 10 supplied data2. Additionally, these 10 firms selected which of their own drugs to provide
R&D costs for. It is possible that both the firms who chose to participate, and the drugs selected
by these firms, represent the most costly drugs to develop.
Another issue that the authors raise is that DiMasi et al do not attempt to verify the data
submitted by the firms, nor do they attempt to clean it. L/W state that, one does not know how
companies calculated their R&D costs or what they included 2. Companies follow inconsistent
accounting practices where sometimes large costs are added to R&D that are not directly related
to the production of a new drug. For example, land, legal, and general administrative costs are
often included in the estimate.
Dimasi, Joseph A., Ronald W. Hansen, and Henry G. Grabowski. "The Price of Innovation: New Estimates of Drug
Light, Donald W., and Rebecca Warburton. "Demythologizing the High Costs of Pharmaceutical Research."
Additionally, DiMasi et al only use data for drugs that were both produced in-house and are
considered new chemical entities. Drugs that are developed in-house from the ground up
represent the costliest 22% of all newly developed drugs2.
Warburton regarding DiMasi et als paper. According to the L/W, the cost of discovery is
unknown and highly variable because:
1) Costs can vary greatly between products and could take as little as 3 months to as much
as 30 years; thus, applying a single number to represent the preclinical cost for
developing new drugs is difficult.
2) Pre-clinical trials often lead to the development of several new drugs and hence applying
the total pre-clinical estimate to a single drug would over-estimate the cost of discovery.
3) The majority of funding during the pre-clinical stage comes from public sources and not
from private pharmaceutical companies. Adding the cost of discovery to the companys
R&D costs overestimates the pre-clinical cost.
Sapienza, Alice M. "Setting the Record Straight: The Office of Technology Assessment Report?pharmaceutical
R&D Costs, Risks, and Rewards." Drug Development Research 30.2 (1993): 56-61. Web.
are charged to projects based on a predetermined allocation formula. Such differences reduce the
comparability between costs reported by different companies 3.
L/W also rely on a US FDA study which states that average trial sizes for testing new molecules
from 1995 1999 averaged 2667 subjects.
reported an average cost per trial participant of $3,861 4. These figures are much lower than the
DiMasi estimates of 5,303 and $23,572 respectively.
Sample Size
DiMasi examined 68 new drugs in their study. This sample size is large enough to reliably
measure costs in each phase. Table 1 highlights some statistical tests that measure the reliability
of the mean cost estimates of each phase of development. The probabilities that the true mean is
within 10% of the estimated mean are all over 60%, even for the cost of phase 3 where the
variation is the highest of the four phases. Using statistical tests alone, the estimated mean costs
are relatively meaningful.
Love, J (2003) Evidence Regarding Research and Development Investments in Innovative and Non-Innovative
Table 1 Out-of-pocket Phase Costs For Approved Compounds (in $MM of 2000 dollars)
Testing phase
Mean Sample
Standard
95% confidence
Probability that true
Cost (in $MM of
Error
interval for mean
mean is within 10% of
2000 dollars)
estimated mean
Phase 1
$15.2
1.55
12.07
18.33
66.9
Phase 2
$23.5
2.68
18.14
28.86
61.64
Phase 3
$86.3
7.35
71.60
101.00
75.56
Long-term
animal
$5.2
0.58
4.04
6.36
62.52
However, DiMasis cost estimates are based purely on mean figures. This indicates that certain
outlying drugs with high development costs skew data toward costlier estimates. A more
meaningful R&D measure would be using the median development cost, which in DiMasis
paper equaled only 74% of the mean estimate 2. Furthermore, the cost of each phase may be
autocorrelated. In other words, a drug with high pre-clinical R&D costs would have higher
clinical costs, and without controlling for this possibility the precision of the total cost estimate is
further complicated.
drugs covered in this study. The time from the start of clinical testing to marketing approval
averaged about 90 months. When we include the 52 month estimate for the pre-clinical phase,
the total R&D time estimate equals 142 months1. Also, the survey participants do not have as
strong an incentive to misrepresent such data, so we can assume the data is generally reliable.
However, according to the US FDA in the Federal Register, trial times have decreased from 72
months to 36 during the years 1985 to 1995. The regulatory review time also decreased from 2.5
years to less than a year. These facts alone reduce the time estimate of DiMasis study by at least
3.5 years. The issue is further sullied by the fact that any over-estimate of time will also be
reflected in opportunity cost, which after 12 years of R&D (the DiMasi estimate) amounts to a
significant compounding effect2.
(stock market returns are by no means guaranteed). If the opportunity cost is used in pricing
decisions, the true risk free rate should be used as a compounding rate5.
Tax Treatments
In DiMasi et als research, they did not include the tax deduction in the estimation of R&D costs. If
tax shelters are included in the R&D estimate, then changes in tax treatments would
simultaneously indicate a change in R&D cost estimates. In order to reliably track R&D cost
trends over time, they must omit factors depending on jurisdictional subjectivity including tax
treatments. They also argue that it is hard to evaluate the quantitative impact of tax policy.
Moreover, they claimed that it is unlikely that the credit has had a substantial economic impact on
large multinational pharmaceutical firms1.
Light and Warburton argue that tax savings should be deducted from the total cost of R&D
because:
1) R&D costs are directly expensed to a companys income statement and provide an
immediate reduction to taxable income. Also, the US office of Technology
Assessment clearly states that any dollar saved in tax due to research costs must
deducted from the net cost of R&D.
2) In some jurisdictions, manufacturing firms and, more specifically, companies with
high R&D costs are eligible for special tax credits amounting to 20% of total R&D,
United States Office of Management and Budget. (2003) Guidelines and Discount Rates for Benefit-Cost
Analysis of Federal Programs. Washington DC: United States Office of Management and Budget.
Conclusion
Though we have highlighted some concerns raised with the assumptions made in DiMasi et al.s
paper, the general procedure used to calculate the total R&D cost of developing a new drug is
reasonable. Where possible, we adjust the resulting R&D cost estimate by making, in our view,
more reasonable assumptions. By doing so we reach a more conservative figure for the R&D costs
of developing new drug.
The following table lists a proposed set of assumptions:
Preclinical Trial Costs
Clinical Trial Costs
Opportunity Cost Estimate
Tax Treatments
By using the assumptions in the previous table we have built a revised model to estimate the mean
R&D costs for developing new drugs. We arrive at a value of $271.62 MM USD before the
inclusion of tax savings, and $165.69 MM after the inclusion of tax savings.
10
Revised Cost Estimates, Self-Originated New Chemical Entities ($MM US, year 2000)
Expected Cost
Preclinical Trial Costs
121
PV of Cost
207.99
13.9
Phase II
12.07
Phase III
19.47
Animal
0.97
Trail total
46.41
63.63
271.62
Tax Savings
105.93
165.69
Opportunity cost*
Preclinical duration(Yrs)*
Clinical duration (Yrs)*
Reductions due to tax savings*
*Revised Assumptions
5.40%
4.3
6
39%
11