About this Author
DBL%20Hendrix%20small.png College chemistry, 1983

Derek Lowe The 2002 Model

Dbl%20new%20portrait%20B%26W.png After 10 years of blogging. . .

Derek Lowe, an Arkansan by birth, got his BA from Hendrix College and his PhD in organic chemistry from Duke before spending time in Germany on a Humboldt Fellowship on his post-doc. He's worked for several major pharmaceutical companies since 1989 on drug discovery projects against schizophrenia, Alzheimer's, diabetes, osteoporosis and other diseases. To contact Derek email him directly: Twitter: Dereklowe

Chemistry and Drug Data: Drugbank
Chempedia Lab
Synthetic Pages
Organic Chemistry Portal
Not Voodoo

Chemistry and Pharma Blogs:
Org Prep Daily
The Haystack
A New Merck, Reviewed
Liberal Arts Chemistry
Electron Pusher
All Things Metathesis
C&E News Blogs
Chemiotics II
Chemical Space
Noel O'Blog
In Vivo Blog
Terra Sigilatta
BBSRC/Douglas Kell
Realizations in Biostatistics
ChemSpider Blog
Organic Chem - Education & Industry
Pharma Strategy Blog
No Name No Slogan
Practical Fragments
The Curious Wavefunction
Natural Product Man
Fragment Literature
Chemistry World Blog
Synthetic Nature
Chemistry Blog
Synthesizing Ideas
Eye on FDA
Chemical Forums
Symyx Blog
Sceptical Chymist
Lamentations on Chemistry
Computational Organic Chemistry
Mining Drugs
Henry Rzepa

Science Blogs and News:
Bad Science
The Loom
Uncertain Principles
Fierce Biotech
Blogs for Industry
Omics! Omics!
Young Female Scientist
Notional Slurry
Nobel Intent
SciTech Daily
Science Blog
Gene Expression (I)
Gene Expression (II)
Adventures in Ethics and Science
Transterrestrial Musings
Slashdot Science
Cosmic Variance
Biology News Net

Medical Blogs
DB's Medical Rants
Science-Based Medicine
Respectful Insolence
Diabetes Mine

Economics and Business
Marginal Revolution
The Volokh Conspiracy
Knowledge Problem

Politics / Current Events
Virginia Postrel
Belmont Club
Mickey Kaus

Belles Lettres
Uncouth Reflections
Arts and Letters Daily
In the Pipeline: Don't miss Derek Lowe's excellent commentary on drug discovery and the pharma industry in general at In the Pipeline

In the Pipeline

« Animal Models: How High to Set the Bar? | Main | Fearful Symmetry? »

October 17, 2008

Down The Chute in Phase III

Email This Entry

Posted by Derek

Here's a good article over at the In Vivo Blog on this year's crop of expensive Phase III failures. They've mostly been biotech drugs (vaccines and the like), but it's a problem everywhere. As In Vivo's Chris Morrison puts it:

Look, drugs fail. That happens because drug development is very difficult. Even Phase III drugs fail, probably more than they used to, thanks to stiffer endpoints and attempts to tackle trickier diseases. Lilly Research Laboratory president Steve Paul lamented at our recent PSA meeting that Phase III is "still pretty lousy," in terms of attrition rates -- around 50%. And not always for the reasons you'd expect. "You shouldn't be losing Phase III molecules for lack of efficacy," he said, but it's happening throughout the industry.

Ah, but efficacy has come up in the world as a reason for failure. Failures due to pharmacokinetics have been going down over the years as we do a better job in the preclinical phase (and as we come up with more formulation options). Tox failures are probably running at their usual horrifying levels; I don't think that those have changed, because we don't understand toxicology much better (or worse) than we ever did.

But as we push into new mechanisms, we're pushing into territory that we don't understand very well. And many of these things don't work the way that we think that they do. And since we don't have good animal models - see yesterday's post - we're only going to find out about these things later on in the clinic. Phase II is where you'd expect a lot of these things to happen, but it's possible to cherry-pick things in that stage to get good enough numbers to continue. So on you go to Phase III, where you spend the serious money to find out that you've been wrong the whole time.

So we get efficacy failures (and we've been getting them for some time - see this piece from 2004). And we're getting them in Phase III because we're now smart and resourceful enough to worm our way through Phase II too often. The cure? To understand more biology. That's not a short-term fix - but it's the only one that's sure to work. . .

Comments (16) + TrackBacks (0) | Category: Clinical Trials | Drug Development | Drug Industry History | Pharmacokinetics | Toxicology


1. Ty on October 17, 2008 8:41 AM writes...

How far are we from pharmacogenomics? Are we stockpiling the genotype data from clinical trials? Is there a tangible push for 'personalized medicine'? Or is (was) it just a buzz word and people don't really do anything about it?

Permalink to Comment

2. petros on October 17, 2008 8:49 AM writes...

And of course another factor that oth you and Chris don't mention is how much some of these studies cost.
That depends on indication, duration and size but can run into the $100m+ range. And if you've got a phase III study running at high cost you've probably also committed to the plant scale build up, a big problem is specific facilities have to be developed.

Permalink to Comment

3. Hap on October 17, 2008 9:28 AM writes...

I know I'm discounting the risk that a drug may work despite so-so Phase I/II data, but I thought people wanted drugs to fail "early" (in Phase I or II, before the big trial expenses come up). If you're massaging the data to push something into Phase III that shouldn't be there, then while you may be boosting stock price temporarily, but in the long run you're likely going to cost your company a lot of money. If this is happening, I might conclude that management incentives (always a fiesta for gaming) are counterproductive.

Permalink to Comment

4. Keith Robison on October 17, 2008 9:40 AM writes...

Pharmacogenomics is here & it's real, but it also isn't a magic bullet for clinical trial failure.

First, you really need some strong hypotheses about what variation is important & how you will use it in the trial. You can't just take the trial subjects' DNA, throw it on some chips, run it through a computer & fix the trial.

However, if you have a strong hypothesis about how genetic variation might predict responders or non-responders or how it might screen out persons liable to have adverse events, then it might help. But, you're going to need to figure out how that's going to impact recruitment & whether you've just thrown a bunch more hypotheses into your trial, necessitating some statistical penalties.

In the end, pharmacogenomics is another, rich set of covariates for trial design. But they ain't a free lunch.

Permalink to Comment

5. processchemist on October 17, 2008 11:34 AM writes...


With the "take the money and run" model, phase III failures are not so strange. How many middle-to-high level managers followed a project from tox to market in the last years?
I know few of them , but the're all in small companies.
To me clinical trials are black magic: in the last few years I seen two products going up and down in phase III. Why a study tells the product is performing well and another not to me is a mistery. But, talking with an old school scientific director, I heard that with today standards in phase III trials sertraline would not be on the market.

Permalink to Comment

6. Still Scared of Dinosaurs on October 17, 2008 1:39 PM writes...

1) KR makes very good points (I'm not sure statistical penalties would often come into play with RCT endpoints but that's a REALLY minor point). It's very hard to get genetic data collected in clinical trials unless there are very specific hypotheses attached to them in the protocol. Not just genetics, but it's such a political red flag that it's likely to get approval held up or denied at IRB's. Everything you do has to be covered by the Informed Consent. You can't retest frozen samples to prove that the magic SNP is required for the drug to work unless the IC says so...or first you go back and reconsent EACH patient so tested after getting an amended protocol reapproved at each IRB. Don't laugh, it's been done.

2) "Cherry picking" Phase 2 data doesn't have to be deliberate. Just running enough products into the clinic increases the chances of some of them looking better than they should by chance. Excpect the same results in Phase 3 and these cases regress to the mean. But guess how far it gets you to say, "You know, we'd learn a lot by repeating that Phase 2 dose-ranging study."

3) The transition from Phase 2 to Phase 3 gets greatly complicated by the increase in attention paid to the product in question by Bus Dev, Marketing, and Sr Mgt. You basically get told to produce the same results as in Phase 2 but with certain changes in the protocol because of label considerations. Considerations which are always intended to increase the population convered by the indication. Increasing the pool of patients in a trial often seems to dilute treatment effects.

4) Phase 3 puts intense pressure on the team charged with running the trial. Often a big push to CRO's as well as pushes into new countries. Accrual is the biggest potential problem, often exacerbated by difficulties getting protocols approved in new countries and at new IRB's, sites started, etc. When the patients roll in too slowly the protocols get amended to be less restrictive and the results get watered down again.

Permalink to Comment

7. Hap on October 17, 2008 2:15 PM writes...

How have the pressures on Phase III managers changed in recent history? It appears that marketing might have more influence, and the safety constraints on the outcomes are likely stronger than before, but the pressures on finding patients for trials and for statistical variation in trial outcomes should have existed before this.

Are we seeing a chance variation in Phase III trial failures, or are we just (un)lucky?

Permalink to Comment

8. Hap on October 17, 2008 2:21 PM writes...

(un)lucky in the last post meaning a real variation in trial outcomes, not chance variation.

Permalink to Comment

9. GM on October 17, 2008 5:47 PM writes...

There have been some success with pharmacogenetics/geneomocs with drugs like Herceptin (effective in patients with HER2 protein), Warfarin (dose reduction for patients with polymorphism in CYP2C9 and VKORC1), Irinotecan (neutropenia in patients with UGT1A1*28), 6 Mercaptopurine (TPMT polymorphisms), Tamoxifen (CYP2D6 genotyping to identify poor metabolisers). I think pharmacogenetics may help in identifying failures in advance or identify subset of patients in whom drug is effective. May be there will be more of this in future, but is still gonna be far from personalized medicine.

Permalink to Comment

10. Andrew Ryan on October 18, 2008 5:44 PM writes...

We try to do some "pharmacogenomics" at my company, although it really is more of a buzzword for management than anything else. The problem we run into is that a lot of SNP's are quite rare and you need to analyze many thousands of samples to obtain statistically significant results correlating haplotypes to disease or somesuch. I don't know how big a typical stage III clinical study is but some of them may not be big enough.

Previous comment about IRB and consent is spot on as well, unless the samples are discarded and the people who collected them are no longer involved in the research you will need consent.

Permalink to Comment

11. Morten on October 18, 2008 5:49 PM writes...

Wouldn't an incentive model based on dividends rather than stock price be better for a company's long term health? Rather than stock options the company could lend stocks to employees for 20-25 years. As an employee I would no longer have an incentive to make the earning potential of my company appear greater than it actually is - there's no personal gain from a positive impression on stock brokers and the public.

just my 2 cents - good or bad idea?

Permalink to Comment

12. Great Molecular Crapshoot on October 19, 2008 7:53 AM writes...

I’m not sure that you can always label failures in development as being due to PK, efficacy and toxicity. Failure can be connected with a competitor’s success.

Normally you’ll have an idea of the human exposure that you’ll need for efficacy and if your phase I suggests that you’re not going to achieve this then this is a good time to take a firm grip of your ripcord and proceed to the nearest exit.

It may be that the phase I study suggests that you’ll be able to achieve adequate exposure. You go to phase II but your compound doesn’t do anything there. You knew the PK wasn’t great but provided that the necessary exposure had been estimated correctly it’d have been more than adequate. The blamestorming starts. If you’d designed a compound with better PK we’d have been able to achieve higher free blood levels. Don’t blame us, our prediction of the human PK profile was actually spot on. The pharmacologists said that we’d see an exploitable effect if we could maintain free blood level above 80nM so blame them instead.

Once you get into phase III you really shouldn’t be failing due to a complete lack of efficacy. You also should be able to link the efficacy that you observed to exposure. The phase 3 trials are where you’re more likely to see rare toxicity simply because you’re dosing more patients. If the toxicity can be linked to a particular target, you might be able to use pharmacogenomics to contain the problem but that's going to show up on the label later. If not, you might wonder what would have happened if you’d come up with something that was efficacious at a lower dose.

Paracelsus has the last laugh

Permalink to Comment

13. Still Scared of Dinosaurs on October 19, 2008 8:21 AM writes...

Hap - I think the pressures on people running the trials have always been there and a compensating factor is that it's actually a lot easier to find people who've done it before when ramping up. There's still a gazillion opportunities for mistakes but if you hire well it's still easier to run trials today than it was 15-20 years ago.

The real pressures are on companies having pipelines so they push substandard drugs into the clinic. I've heard of companies where the performance metrics were based on # drugs in the clinics, # in Phase 2, etc. Almost designed to produce later stage failures.

GM - One question for each of the drugs you mention is when in the dev process the association of treatment effect with the genetic markers was discovered. The ideal I suppose would be different gene -> different target -> new drug.
The hope for Phase 3 failure -> post-hoc assay festival -> miracle SNP is too desperate for serious consideration. I suspect that most of the drugs listed were somewhere in the middle but closer to the beginning than the end. In any case did it happen after initiation of the Phase 3 studies?

Oh, and one huge stumbling block when you do find a miracle factor is that you've just shrunk your total market down to the % of patients with that characteristic and sometimes the money types can't get past that. Explaining that locking up 20% of the market may be better than competing for 100% doesn't register. Especially to the budding empire builder who realizes that such a drug may only require 5% of the salesforce.

Permalink to Comment

14. Kay on October 20, 2008 7:42 AM writes...

"Failures due to pharmacokinetics have been going down over the years as we do a better job in the preclinical phase"

Derek: Care to offer numbers that can be or have been audited? No one reveals real attrition explanations as truth is rare in our industry. What's more, clinical failures have many parallel causes.

Permalink to Comment

15. GM on October 20, 2008 5:10 PM writes...

Still Scared of Dinosaurs - I think Herceptin was the only drug where patients expressing HER2 protein were enrolled in all phases of clinical trial. For other drugs (Irinotecan, Warfarin, tamoxifen), association of SNPs to efficacy/toxicity was made once the drug was marketed and changes in the drug label were made accordingly. I totally agree to the fact that it reduces the market size. Suppose a NCE is metabolised by CYP2D6 (highly polymorphic enzyme), wouldnt it be prudent to genotype patients and explore the association of poor metabolizers/extensive metabolizers with the efficay/safety of the drug. Atleast this would help in modifying the dosage regimen, if not prevent a drug failure.

Permalink to Comment

16. Craig on March 20, 2009 3:27 PM writes...

What are your thoughts on a Phase 3 trial that is late in getting to their first-interim analysis? Is this an indication that the drug is doing well, or better than initially anticipated by the management of the company? On the reverse, what is the trial reaches P3 first-interim analysis too quickly? - I am speaking of cancer drugs.

Thank you

Permalink to Comment


Remember Me?


Email this entry to:

Your email address:

Message (optional):

The Last Post
The GSK Layoffs Continue, By Proxy
The Move is Nigh
Another Alzheimer's IPO
Cutbacks at C&E News
Sanofi Pays to Get Back Into Oncology
An Irresponsible Statement About Curing Cancer
Oliver Sacks on Turning Back to Chemistry