David R. Henderson  

Hooper's Law of Drug Development

I warned the Swiss that this w... What's in Your Bag?...
Moore's Law is optimistic and reflects the ability of humans to "chip" away at a problem, making sequential, cumulative advances. Much of technology fits this pattern. One glaring exception, tragically, is the drug development conducted by pharmaceutical companies. It is hugely expensive and has gotten more so each year. If costs continue to grow at 7.5 percent per year, real costs will more than double every 10 years. The pharmaceutical industry seems to be operating under a reverse-Moore's Law. I call it Hooper's Law. Here's the short version: Drug development costs double every decade. Why? Simple: the U.S. Food and Drug Administration is steadily increasing the cost per clinical trial participant and the number of required participants per clinical trial.
This is from Charles L. Hooper, "Hooper's Law of Drug Development," one of the two Econlib Feature Articles for August.

In the piece, Charley explains why this has happened.

Comments and Sharing

COMMENTS (14 to date)
Jon Murphy writes:

Another great article by Charley Hooper (I make a point of reading anything that I see his name attached to).

I have a couple of thoughts after reading this, one statistical and one economic:


Is there some reason the Law of Large Numbers doesn't apply to medical research? I know the larger the sample size, the closer one gets to the "true" mean, but if I recall my statistics correctly (and, dear reader, keep in mind I am an economist and not a statistician), you tend to get pretty accurate answers very close to the "true" mean with fairly small samples. In other words, does the cost of increasing sample sizes from a few thousand (as in the case of the Merck study) to tens/hundreds of thousands (as in the case of the Orexigen study) provide enough extra accuracy to justify the costs? If anyone has inside knowledge of the statistical philosophy of the FDA, I'd love to hear an answer.


Price theory teaches us that, as costs rise, marginally profitable firms (that is, firms operating at the price level just enough to cover average total cost), will close down. This would, ceteris paribus, give the remaining firms more market concentration. Ordinarily, I'd say this wouldn't matter because the threat of competition would keep markets competitive; firms would be unable to fully exploit monopoly power. However, as the evidence suggested the Orexigen case shows, the regulatory burden is so high that it effectively prevents new entrants into the market. SO, it seems to me there is another way the FDA is forcing medical care costs up aside from the ways Hooper discusses in the article: implicit monopoly protection. Furthermore, since firms that are protected monopolies tend to be less innovative than firms facing competition, this may generate more of a case for gov't funding of medical research, which in turn raises taxes and effectively subsidizes the existing firms, making entry into the market harder, increasing monopoly power, reducing innovation, etc etc. In short, we might have a vicious cycle here.

I like Hooper's article. It's very thought-provoking

Alex writes:

Well said.
The FDA may have the best intentions but, in its current form at least, causes a lot of damage.
The best solution would be to legalize all drugs of course. Any company can sell anything it wants. Drug sales should be controlled, yes, but legal. If you want to buy cocaine, you should be able to.
There would be much less crime, much less prison population and much less illegal immigration.
This will not happen anytime soon but maybe one day. Gay marriage was unthinkable 30 years ago and now its the law.

Todd Kreider writes:

Jon Murphy wrote:

Is there some reason the Law of Large Numbers doesn't apply to medical research?... .... If anyone has inside knowledge of the statistical philosophy of the FDA, I'd love to hear an answer.

It looks like there is a "statistical philosophy of the FDA" based on their ridiculous request for a trial of 10,000 to 60,000 people. How many at the FDA know basic statistics? My guess is that those voices were easily drowned out by lawyers somewhere along the line.

Hooper's Law is interesting but pales compared to the mighty Moore's Law, which has held for at least 57 years so far. That will likely end in just a few years but the overall acceleration of computational power is very likely to continue at least another decade and because of this Hooper's Law is unlikely to continue out to 2027 as simulated trials become feasible.

Bill Joy estimated over ten years ago that computers will be able to mimic biology at the physics level by 2030, and I wouldn't be surprised to see very inexpensive simulated "trials" a few years before that.

Charley Hooper writes:


Thank you for the compliment.

Regarding your statistical comment, what the FDA is trying to do is look for (1) rare events and (2) collect enough data so that it can be analyzed by patient segment: by sex, by age, by stage of diagnosis, by severity, by risk factor, by concomitant condition, etc. Once we start slicing and dicing, we can generate a large number of patient segments and they each need a substantial N.

For rare events, consider the Vioxx situation where the FDA-approved Vioxx was later found to increase the risk of heart attacks by 80 percent after 18 months of therapy. There was no increase before 18 months and the risk of a patient having a cardiovascular event related to Vioxx was “very small,” according to Acting FDA Commissioner Dr. Lester M. Crawford. To measure twice a small risk we need a bunch of clinical trial participants.

Are the larger sample sizes worth it, from an economical/societal viewpoint? I suspect not. From the FDA's (i.e., public choice) perspective, they are worth it: the FDA does not pay for the expensive trials and the FDA is protected to some degree by the large volume of data collected. A larger N helps prevent another Vioxx debacle.

JB writes:

Any regime change in Pharma regulation needs to think carefully about attracting top talent. Every ambitious young man or woman wants to an invent a new drug. Those who can't make it, work for the FDA. The quality of FDA clinical reasoning is terrible in many cases, probably not adequately explained only by static public choices incentives....We need competition between different regulatory agencies which translates into monetary incentives for the regulatory staff. Better pay = better results in competitive markets.

Todd Kreider writes:

Charley Hooper respnoded to Jon:

Regarding your statistical comment, what the FDA is trying to do is look for (1) rare events and (2) collect enough data so that it can be analyzed by patient segment: by sex, by age, by stage of diagnosis, by severity, by risk factor, by concomitant condition, etc. Once we start slicing and dicing, we can generate a large number of patient segments and they each need a substantial N.

Hooper used the Vioxx case as an example but the problem wasn't an insufficient sample size. Instead, it was Merck that pulled this very successful drug off the market because it didn't want to deal with any risk when it could have put a warning on the label about the tiny risk of heart attack increase.

Going to 60,000 to 100,000 person trials wouldn't have changed the results one bit.

Jon Murphy writes:

@Charley Hooper

Thanks for the explanation. It makes sense, especially from the FDA's standpoint: you'd want to know the risk in the tails, especially if the cost is extremely high.

Todd Kreider writes:


Keep in mind that the increase in heart attack risk from Vioxx was also later discovered other COX-2 inhibitors and NSAIDS like ibuprofen where the old clinical trials didn't catch heart issues.

Also, Merck knew the risk in the tails with the Vioxx trials but hid the results at first. Recently, there have been FDA warnings to people who take ibuprofen, etc. with heart conditions.

Jon (different one) writes:

Some years after first publishing Moore's law, Gordon gave the keynote at the first VLSI conference. In that presentation he showed a different version of Moore's law: exponentially increasing cost.


See page 12. Breaking this cost curve happened for while in the 80s and 90s because of computer aided design (automation). The last generation of computers helped us build the next. But the gains there were done maybe fifteen years ago. The cost problem returned and was slain by replication and modularization, but its come back again. So the curve is bent at best, not broken, and where a large die could be done for $25M fifteen years ago; today it will take $100M. Transistor for transistor it's cheaper, but dollar per ambition is more expensive.

The same is true searching for drugs. In the past, we found that which was readily at-hand, formalizing what we already new from hundreds or thousands of years of experience. Today we are searching further and further afield of our cultural inheritance. We aren't getting less efficient; our ambitions are actually getting bigger.

Charley Hooper writes:

@Todd Kreider,

I agree about Vioxx.

Merck should have kept Vioxx on the market and instead limited its usage to those patients with the best benefit/risk profile. The withdrawal gave ammunition to those who argued that Vioxx’s benefits weren’t significant enough to warrant any increased risks. Keeping Vioxx on the market would have helped squash the image of Vioxx being a wildly dangerous drug and it may have dissuaded some proportion of the subsequent lawsuits. We should note that there were patients who strongly preferred Vioxx over other drugs, even with the increased cardiovascular risk.

Charley Hooper writes:

@Jon (different one),

You make a good point. Moore's Law and Hooper's Law aren't an apples-to-apples comparison because Moore's Law looks at output per chip and Hooper's Law looks at inputs per drug.

The cost of designing each chip has been going up, but, because of Moore's Law, the performance per dollar of each chip has been increasing. Moore's Law reflects manufacturing improvements and the chip industry is so competitive that manufacturing costs are key.

The pharmaceutical industry is different, due to patents. Prices in the pharmaceutical industry often reflect "value," primarily, and amortized R&D costs and manufacturing costs, secondarily.

Alan Goldhammer writes:

I think Charley Hooper's article is a little too simple minded. I spent 90% of my working career in the biopharma industry working in regulatory affairs and drug safety. The key problem today is that the science has grown much more difficult as the drug targets are more complex (intrepid readers would do well to follow Derek Lowe's 'In the Pipeline' blog for a first hand account of someone involved in this daily.

Yes, the Merck withdrawal of Vioxx and Pfizer's of Bextra (that one was accompanied by a large criminal judgement against Pfizer for fraudulent marketing) left US patients with only on Cox-2 inhibitor on the market. Ironically, Pfizer did a very large safety study of Celebrex that was completed last year, showing that the drug was no worse than other NSAIDs with respect to cardiovascular mortality (the other two are worse). It didn't help Pfizer much as the drug went off patent several years before the completion of the study. it should also be noted that Celebrex is a very weak drug in terms of it's anti-inflammatory properties. Todd Kreider's points are right on the mark.

Regarding the Orexigen study, it was never clear that the combination drug was really going to work all that well (the history of drug development in this area is not all that encouraging). I don't know why Hooper writes that development was terminated. In fact the drug was approved a couple of years ago and is being marketed by Takeda under license from Orexigen. It's questionable whether this will make any dent in the obesity treatment market.

Most of the drugs that run into trouble at the FDA possess only marginal clinical efficacy or pose significant safety issues that need to be carefully managed. Research in a number of therapeutic areas is being stopped not because of FDA problems, but rather because existing drug therapies are more than adequate to treat the condition.

Companies that have good drug candidates and the data to support licensure don't have regulatory problems.

Charley Hooper writes:

@Alan Goldhammer,

Read again what I said about Orexigen. I said that development was discontinued until the company and the FDA negotiated a smaller trial size.

Regarding drug targets becoming more complex, that relates to drug discovery, an area I didn't address. I merely pointed out that clinical trials have become more expensive, which is an aspect of drug development. You have not explained why I'm wrong.

Vipul Naik writes:

There's already an Eroom's law (Moore's law reversed).

Comments for this entry have been closed
Return to top