PharmExec Blog

Talkback from Tufts: Defending R&D Costs

by Joe DiMasi, director of economic analysis, Tufts Center for the Study of Drug Development

Joe Dimasi

Joe DiMasi

William Looney’s posting for this blog, “Calculating the Cost of R&D: Defending Tufts Research” (January 11, 2012), raises a number of interesting and important points.  Both the posting and the working paper by F.M. Scherer on which the posting was based, reflect a widespread concern over the state of R&D productivity in the pharmaceutical industry.  The level of resources needed to get a new drug approved by regulatory authorities is a critical aspect of the productivity of pharmaceutical R&D.

The Tufts Center for the Study of Drug Development (CSDD) has conducted a series of studies over the decades tracking industry R&D costs related to new drug approval.  Although the study criteria were defined in terms of when investigational drugs enter clinical testing, the results are easier to understand in terms of when that development resulted in success (i.e., when drugs from the study period obtained regulatory approval for marketing).  From this perspective, the first study in the series covered development that generally resulted in approvals during the 1970s, the second study was generally applicable to 1980s approvals, while the last study covered approvals during the 1990s to the early 2000s.  Taken together, the results demonstrate a marked upward trend, over and above general price inflation, in the cost per approved drug.  Alternative approaches examined as part of the last two studies to find ballpark figures by using public data confirm the study results, as does Scherer’s alternative analyses using such data.

Given scattered evidence of increasing costs for some components of the drug development process for more recent years and, the year 2011 notwithstanding, a generally stagnant level of new drug output, Tufts CSDD has undertaken a new study in our R&D cost series.  Currently, we are in the process of gathering data.  The results will cover new drug development that yielded approvals during the first decade of the 21th century.  As in the past, the cost of failures will be taken into account, and separate estimates of the time costs associated with the lengthy drug development process will be determined.  The data will reflect the diversity of development by therapeutic class and molecule type undertaken by mid-sized to large biopharmaceutical firms. There is more to R&D productivity than output per dollar spent, but clearly such a metric is needed as a starting point for a full discussion of productivity and what can be done to improve it in such a crucial segment of the health care sector.

Editor’s Note: The Tufts CSDD data has been cited by industry, which arouses skepticism, but as Forbes pointed out last week, the mean cost estimation may fall on the conservative side, particularly when current failure rates are considered.

This entry was posted in Guest Blog, pricing, R&D and tagged , , . Bookmark the permalink. Trackbacks are closed, but you can post a comment.

2 Comments

  1. Bernard Munos
    Posted February 13, 2012 at 9:29 pm | Permalink

    Dear Dr Di Masi,

    Thank you for this post. As someone with a keen interest in pharmaceutical innovation, I look forward to the publication of you new study. I hope it addresses the question that troubles many people, i.e., where is the drug R&D money really going? PhRMA keeps affirming it only costs $1.2 billion to create a new drug. Since the industry produces roughly 25 NMEs per year, this translates into an expenditure of $30 billion. Yet, public companies spend $125 billion a year on R&D. So, the output only explains 25% of the spending at best, which raises, as you might expect, serious questions about the credibility of the PhRMA claims.

    On a more technical note, I have some concerns that I hope your new study can address, i.e.,
    1. Can it alleviate concerns that the data supplied by the industry might suffer from selection bias?
    2. Is cost per NME a useful metric since it is not amenable to verification? Wouldn’t R&D spending per NME be better since that data is widely available and not subject to “adjustments” before you get it?
    3. Since the distribution of cost-per-NME (or R&D spending per NME) across the industry is heavily skewed and certainly not Gaussian, does it make sense to talk about an “average” which does not represent the population? Wouldn’t it be better to show the entire distribution, including the outliers?

    With the current innovation crisis, there are many justified questions about R&D productivity and the metrics that purport to measure it. I would hope the industry realizes it is everybody’s interest to have quality data to inform that debate, and that it will do its bit to lift the cloak of opacity that has shrouded this topic.

    Thank you for your efforts to enlighten the rest of us about this important question.

    Bernard Munos

  2. Joseph DiMasi
    Posted February 14, 2012 at 6:08 pm | Permalink

    Dear Bernard,

    Thank you for your comments. Dividing contemporaneous industry or company R&D expenditures by the number of new drug approvals can yield an estimate that is seriously biased upward. Approvals this year are really associated with R&D expenditures made over a number of past years, and the bulk of R&D expenditures this year will be associated with future approvals. These facts, combined with a strong trend of increasing R&D expenditures over decades, even after adjusting for inflation, guarantee an upward biased estimate. Using ten or fifteen years of contemporaneous R&D spending and approvals does not eliminate the problem. You might want to argue that the number of approvals is effectively constant, but there was something of an upward trend in approvals over the periods covered by our past studies, and we really don’t know what the number of future approvals will be. No doubt, as I have repeatedly pointed out in the past, increasing aggregate R&D expenditures and a more slowly rising or stagnant number of approvals clearly indicates increasing costs per new drug, but you cannot get very precise about levels from the aggregate data. If one is very careful about using the PhRMA R&D expenditure and the new drug approvals data series, as we were for our studies in 1991 and 2003, then the best that one can do is to get a range of values for a given period. In section 8.2.3 of our 2003 paper, along with an appendix that was available to the reader upon request, we describe and present such an analysis in the way of verification of our project-specific results. Both our out-of-pocket and capitalized cost estimates sit fairly centrally located with the ranges obtained from the published aggregate industry data.

    Aggregate analyses using reported individual company R&D spending and approval counts are also fraught with problems. The same kind of lag issue (R&D expenditures and approvals separated in time) that exists for industry level data also applies at the company level. In addition some companies have multiple lines of business and do not separate out prescription drug R&D. Companies also differ in terms of the extent to which they license-in drugs. A portion of the R&D costs for the licensed-in drugs will not be captured in the financial statements of either the licensee or licensor firms. Finally, while industry level approval counts can be relatively stable over time, the same cannot be said about approvals at the company level.

    The $1.2 billion figure applies to a particular period. Our new study will obviously cover a more recent period, and, of course, the results can be different. As a point of comparison, I should note that the $1.2 billion includes time costs in addition to cash outlays. Below are some comments on your technical points.

    “1. Can it alleviate concerns that the data supplied by the industry might suffer from selection bias?”

    I am not sure what you mean by selection bias here. We select the drugs, and the selections in the past have matched closely with the broader portfolios by therapeutic class and type.

    “2. Is cost per NME a useful metric since it is not amenable to verification? Wouldn’t R&D spending per NME be better since that data is widely available and not subject to “adjustments” before you get it?”

    Our cost per new drug estimate methodology really is a way to get at, in an appropriate manner, total (pre-approval) R&D spending per new drug. I should also note that in our 2003 study we have an estimate for post-approval cost per new drug which includes R&D on new indications, required Phase IV studies, new dosage regimens and strengths, new formulations, etc. Our papers contain numerous approaches to verifying our results in total and in part. Two FTC economists in two published papers have also effectively roughly verified our results using other public data. I don’t know what you mean by “adjustments.”

    “3. Since the distribution of cost-per-NME (or R&D spending per NME) across the industry is heavily skewed and certainly not Gaussian, does it make sense to talk about an “average” which does not represent the population? Wouldn’t it be better to show the entire distribution, including the outliers?”

    I believe that you have misinterpreted our methodology and to what our results refer. Our most fundamental result, cost per approved new drug, is not the mean of a set of values for individual new drugs. The methodology incorporates the costs of failures and accounts for fixed costs that cannot be attributed to specific compounds. Thus, there is no “distribution” for total cost per approved new drug. If one drills down to components of the overall cost estimate, then one can talk about distributions. We have done that where possible and provided variability measures in our studies (e.g., for the distribution of phase II costs for investigational drugs, or the distribution of clinical period costs for just the approved drugs in the sample).

    Joseph DiMasi

Post a Comment

Your email is never published nor shared. Required fields are marked *

*
*

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <strike> <strong>

  • Categories

  • Meta