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

« Bright Lights and Applause? | Main | Lights, Camera, Pharma! »

November 6, 2007

Andy Grove: Rich, Famous, Smart and Wrong

Email This Entry

Posted by Derek

So I see that Andy Grove, ex-Intel, is telling everyone that the drug industry could use some of that Moore's Law magic. I've noticed that people who spend a lot of time in the computer business often have an. . .interesting perspective on what constitutes progress in other fields, and we might as well appoint Grove the spokesman for their worldview:

Q: In what way does the semiconductor industry offer lessons to pharma?
A: I picked the semiconductor industry because it's the one I know; I spent 40 years in it, during which it became the foundation for all of electronics. It has done a bunch of unbelievable things, powering computers of increasing power and speed. But in the treatment of Parkinson's, we have gone from levodopa to levodopa. ALS [Lou Gehrig's disease] has no good treatment; Alzheimer's has none.

To me, the first sentence of that answer is the key one. As for the rest of it, hey, it's all true. Perhaps one explanation for the difference between the two fields is that they're driven by fundamentally different processes? Nah, that can't be right:

Q: Why is the speed of progress so different in semiconductor research and drug development?
A: The fundamental tenet that drives us all in the semiconductor industry is a deeply felt conviction that what matters is time to market, or time to money. But you never hear an executive from a pharmaceutical company say, "Before the end of the year I'm going to have xyz drug," the way Steve Jobs said the iPhone would be out on schedule. The heart of every high-tech executive has been, get the product into customers' hands and ramp up production. That drive is just not present in pharma; the drive to get sufficient understanding and go for it is missing.

Well. Where to begin? Let's start with a minor fact, and work our way up. I've been in this industry for eighteen years, and I cannot count the number of year-end goals I've had to deal with. Number of new targets identified, number of new projects started, number of compounds recommended for development, number of compounds progressed to Phase II, number taken to the FDA. It never ends. If Andy Grove hasn't heard a pharma executive talk about all the wonderful things that are going to be done by a given timeline, he needs to listen harder.

But here's the rough part: although drug company people talk like this, they're full of manure when they do. These year-end goals, in my experience, do very little good and in some cases do a fair amount of harm. I'll bet some of my readers have sat in a few meetings - I sure have - and looked up at the screen thinking "Why on earth are we recommending this drug to go on?", only to have the answer be "Because it's early November". More idiotic things may get done in the name of meeting year-end numerical goals than for any other reason in this industry, so thanks, but I'll try to ignore the recommendation to do them some more, but good and hard this time.

Mr. Grove, here's the short form: medical research is different than semiconductor research. It's harder. Ever seen one of those huge blow-ups of a chip's architecture? It's awe-inspiring, the amount of detail that's crammed into such a small space. And guess what - it's nothing, it's the instructions on the back of a shampoo bottle compared to the complexity of a living system.

That's partly because we didn't build them. Making the things from the ground up is a real advantage when it comes to understanding them, but we started studying life after it had a few billion years head start. What's more, Intel chips are (presumably) actively designed to be comprehensible and efficient, whereas living systems - sorry, Intelligent Design people - have been glued together by relentless random tinkering. Mr. Grove, you can print out the technical specs for your chips. We don't have them for cells.

And believe me, there are a lot more different types of cells than there are chips. Think of the untold number of different bacteria, all mutating and evolving while you look at them. Move on to all the so-called simple organisms, your roundworms and fruit flies, which have occupied generations of scientists and still not given up their biggest and most important mysteries. Keep on until you hit the lower mammals, the rats and mice that we run our efficacy and tox models in. Notice how many different kinds there are, and reflect on how much we really know about how they differ from each other and from us. Now you're ready for human patients, in all their huge, insane variety. Genetically we're a mighty hodgepodge, and when you add environment to that it's a wonder that any drug works at all.

Andy Grove has had prostate cancer, and now suffers from Parkinson's, so it's no wonder that he's taken aback at how poorly we understand each of those diseases - not to mention all the rest of them. But his experience in the technology world has warped his worldview. We are not suffering from a lack of urgency over here - talk to anyone who's working for a small company shoveling its cash into the furnace quarter by quarter, or for a large one watching its most lucrative patents inexorably melt away. And we don't suffer from a lack of hard-charging modern management techniques, that's for sure.

What we suffer from is working on some of the hardest scientific problems in the history of the species. Mr. Grove, the rest of your recommendations don't betray much familiarity with the industry, either, so there may be only one way to make you really understand this. If you really, really believe in your ideas, please: start your own company. You've got the seed money; you can raise plenty more just by waving your hand. Start your own small pharma, your own biotech. Hire a bunch of bright no-nonsense researchers and show us all how it's done. Tell them that you're going to have a drug for Parkinson's by the end of the year, if that's what you think is lacking. Prove me and the rest of the industry wrong.

Comments (86) + TrackBacks (0) | Category: Drug Development | Drug Industry History


1. Bryan on November 6, 2007 9:24 AM writes...

A few minutes reading Gary Pisano's Science Business might enlighten Mr. Grove. Although I dont know how well received Pisano's later conjectures are in the book, he sums up the comparison between semiconductors and drug development quite succinctly. Changing the socket on a motherboard is a lot easier than altering a gene to modify a kinase's active site to fit your drug.

Permalink to Comment

2. PharmaProphet on November 6, 2007 9:54 AM writes...

Actually, I think one of the major problems facing the industry is that upper management is thinking and behaving *exactly* like Mr. Grove. Too much concern for the almighty $, and far too little understanding of how scientific progress really occurs.

Permalink to Comment

3. matt on November 6, 2007 9:56 AM writes...

Wow, great post. Hit the nail on the head. And by the way, what is up with acomplished old guys (i.e. Grove and Watson) saying dumb stuff lately?

Permalink to Comment

4. John Novak on November 6, 2007 9:56 AM writes...

Grove also doesn't have the FDA and medical ethics boards to contend with. (Not that he should, or the pharma industry shouldn't! But the semiconductor market, while ruthless in its own way, will not send you to jail for a poorly thought-out experiment.)

Permalink to Comment

5. matt on November 6, 2007 10:01 AM writes...

Wow, great post. Hit the nail on the head. And by the way, what is up with accomplished old guys (i.e. Grove and Watson) saying dumb stuff lately?

And also, what's up with the Neuroscience meeting? That's where Grove had the platform to say all this stuff. I hope people laughed him off the stage or at least gave him some hard zingers after his talk. And two years ago they invited the Dali Lama. Why do they even waste their money? These events sure do generate a lot of publicity, but it's already one of the largest meetings in the world. Not like they need to advertise.

Permalink to Comment

6. GA on November 6, 2007 10:19 AM writes...

Derek, I'm glad you took this on - when I first read this piece (and it seems to be getting some headlines), all I could do was shake my head in amazement. I'm sure that Andy got the "process" right at Intel, so that chip after chip gave the same performance. However what he fails to understand is that the reason why drug discovery and development fails more often than not is not because the "process" is sub-optimal or the "drive" is missing. It's because of the inherent complexity of the systems we deal with.

Permalink to Comment

7. TW Andrews on November 6, 2007 10:35 AM writes...

Wow. It's not often that you hear that the problem with drug development is that the pharmaceutical industry isn't interested enough in money.

In any case, I think the root of Grove's misunderstanding is that he confuses engineering challenges like the ones that the semi-conductor industry typically faces with the scientific challenges inherent in drug development.

When Intel started putting transistors on chips, it was well understood how electrons moved through conductive materials, and how boolean gates operated. That wasn't something that they had to develop from scratch, let alone for each chip. I guess I don't need to go into detail here on how that's different from drug development where eternal principles are few and far between.

Permalink to Comment

8. RY on November 6, 2007 10:59 AM writes...

As computer scientist who re-tooled into a geneticist/biologist a few years back, I understand exactly where Grove is coming from... and why he is so wrong. The problem with bio-medical science is that all the mess and complexity gets suppressed in the popular press (i.e. NYTimes, Economist, WSJ) and luminaries like Grove get a totally warped sense of what we know and don't know. Even the neat little diagrams of pathways we sometimes make give the illusion of knowledge and simplicity which is utterly misleading. I think, we, as a community, need to do a better job of educating other professionals and the public... (though I fear it may not be possible)

Permalink to Comment

9. Bryan on November 6, 2007 11:05 AM writes...

Just a quick question, but what do you guys think about his comments and criticism of academia and peer review? (bottom page 2

Permalink to Comment

10. Nick K on November 6, 2007 11:45 AM writes...

Message to Andy Grove:

I'll listen to your ignorant, ill-informed comments about drig discovery if you listen to my equally ignorant, ill-informed comments about chip design.

Permalink to Comment

11. Nick K on November 6, 2007 11:46 AM writes...

Message to Andy Grove:

I'll listen to your ignorant, ill-informed comments about drug discovery if you listen to my equally ignorant, ill-informed comments about chip design.

Permalink to Comment

12. paiute on November 6, 2007 12:00 PM writes...

You forgot the most important difference:

Crashed computers = no big deal (patch it)

Crashed humans = multimillion dollar judgement

Permalink to Comment

13. excimer on November 6, 2007 12:27 PM writes...

Excellent post. I, too, shook my head in amazement at this guy's comments. There really is little to compare between the methods in the semiconductor industry and the pharma industry.

Permalink to Comment

14. TFox on November 6, 2007 12:46 PM writes...

In 40 years, Parkinson's has gone from levodopa to levodopa. The chip industry, on the other hand, has gone from transistors in silicon to transistors in silicon. *All* advancement in electronics is what pharma types call scale-up, doing the same thing you did last year a little cheaper. Scale-up, needless to say, is not the hard part of new drug development...

Permalink to Comment

15. DLIB on November 6, 2007 1:10 PM writes...

I'm a pharmacologist that works in the semiconductor industry ( across the street from Intel in Santa Clara ) the company I work for ( Brion technologies/ASML ) allows chips to be designed smaller and smaller -- 16nm lines are on our roadmap. It's true, there's no comparison in the complexity of the problems associated with the respective industries. We do computer modeling of the optical systems used to expose resist. Our models are good enough to resolve better than .5nm differences in edge placement. Just fine when printing an IC. Not good enough for Docking/scoring. The search algorithms bare some resemblance. Ours are physically based. We can fairly represent the complete system. You guys with much bigger and faster computation can't represent the system entropy change very accurately at all ( you need a real calorimeter for that ). There is cross fertilization that's possible but probably more the pharma industry could teach the Semiconductor industry.

Permalink to Comment

16. Kay on November 6, 2007 1:10 PM writes...

What if the electronics industry had Rules that did not work? What if they generated data but did not really believe the results? What if they chose to continue to use the Rules and results? What if the workers chose not to admit these faults to management?

Permalink to Comment

17. haywarmi on November 6, 2007 1:15 PM writes...

Hey Bryan, I'm glad someone else took note of that comment. I'd agree that the peer review system pressures conformity but who's going to weed out the "wild ducks" from the lame ducks? Especially tough when you're handed 20 50 page grants to review (and get paid nothing except the privilege of doing it)?

My impression is that this already exists, to a limited degree, with the HHMI and merit grant system. That is, those who have already proven themselves to be forward thinking and creative are given some license to explore. More of this, though, takes away from the deserving young investigators who may be more creative and technologically innovative.

Those who are so quick to point out that the peer-review process stifles creativity aren't always aware that the funding levels are nearing the single digits, so how do you make that decision? Its not like we have a lot of error to play with, which is exactly what you need to find (or encourage) those wild ducks. You use a shotgun to hunt ducks for a reason and right now the NIH is using a BB-gun (I'll stop with the analogies now but I didn't start it, "wild duck" came from Grove)

Permalink to Comment

18. Wavefunction on November 6, 2007 1:34 PM writes...

Actually a drug is just like a chip. You outsource its production to a third world country, you get all kinds of crap put into it, then it works for some time, develops a defect, kills its consumer in one way or the other and finally becomes obsolete.

Permalink to Comment

19. Chemgeek on November 6, 2007 1:36 PM writes...

A more blatant and ridiculous example of comparing apples and oranges I have never seen. (although, apples and oranges are more similar than pharma and chip design).

Permalink to Comment

20. SRC on November 6, 2007 1:39 PM writes...

I'm a bit shocked that Grove does not grasp the difference between science and technology.

Designing a chip is a matter of engineering, the underlying scientific principles having long since been worked out. It's more akin to the space program than to pharmaceutical research. Progress in scientific research happily ignores timelines that engineering development observes religiously, because caprice plays no role.

The better parallel would be to liken biomedical research to natives in Borneo attempting to build a computer.

Permalink to Comment

21. Molecular Geek on November 6, 2007 1:57 PM writes...

SfN has speakers like Grove or the Dali Lama in addition to the more traditional plenary speakers because part of the mission of a major meeting like this is to include discussion of the larger context of the science that the attendees are doing. They also have public lectures during the meeting to engage the communuity. As someone here at the meeting, I can say that his remarks were listed on being about "Ending the R01 Culture in research". I won't defend his position. Others have correctly called him to task for his misunderstanding of the relationship between chip design and drug design. I'm sure that in his mind, if every academic were to refocus their energies away from trying to get NIH or NSF support, and find a way to become a biotech/pharmaceutical entrepeneur, he wouldn't have to fear that prostate cancer or the onset of parkinsons. I won't go any further with this tired old canard. We see it often enough in critiques of the industry.

I would also point out that Grove was here on a panel discussing funding woes in biomedical science. As a followup to that, Newt Gingrich showed up for a plenary yesterday arguing that the refusal of the current administration to keep funding the NIH budget on its previous trajectory is shortsighted, dangerous, and wrong. (Who would have ever thought he would come out for more government spending?).

Just for context, Neuroscience 2007 is huge. There are over 32,000 attendees here, and the poster hall runs for the entire length of the convention center exhibit hall (at least 500 meters, end to end) with topics ranging from molecular mechanisms of synatogenesis and ligand design through clinical behavior studies and discussions on the physiological nature of conciousness. (The latter is the topic that brought the Dali Lama to speak a couple of years ago. I wasn't there for it, but my better half said that it was a very powerful talk, and it was SRO in the 2 overflow rooms where they simulcast the presentation). They don't have sections like the ACS meetings do, so trying to decide which talks and which posters to attend is like drinking from a hydrant. They attract good speakers that can draw interest across a very broad spectrum of attendees. If anyone else gets a chance to hear Sebastian Seung from MIT speak, do it. He spoke on Monday night, and he gave one of the best lectures I have heard. It was at a Scientific American level to make sure that it was accessible to the entire audience, but his lab had posters on the details yesterday afternoon as well.


Permalink to Comment

22. CMC guy on November 6, 2007 2:02 PM writes...

This is a stimulating subject: Attempting to apply semiconductor model to pharmaceuticals does seem to be the preverbal round peg. As noted in multiple comments the complexities are vastly different. Even when we gain knowledge in areas such as mol bio and genetics the translation into acceptable drugs is a difficult pathway that isnt so straight (and paiute nails risks). I dont know how many research programs (and expense) have gone into treatments for Parkinsons and Alzhemiers since the 1950/60s (timeframe in Grove article?) but would suspect numerous approaches explored and perhaps even a few made to clinical trials.

Although at times it seems Pharma has bought into fundamental “time to money” principle at the core think most people doing the work so that will benefit sick people and have to delicately balance efficacy with ill effects. Unfortunately what works in animals often proves unsuitable for people so crossing gulf between demo a cure for a cancer in a mouse and the advancement to human the bridge collapses. Grove mentions biomarkers lack of emphasis but how many times have we seen compounds give excellent responses based on such but still fail to do the job against the disease (see PSA for Prostate Cancer for example).

Grove does have interesting comments about “conformity of thoughts and valves” (targeted academia mainly) which I do see as a problem in Pharma with too few companies willing to move off save territory. More innovations/wild ducks are needed to solve illnesses but that will not come unless the funding opens to enable these novel explorations and the short-term ROI view that now dominates investors drives to only immediate high return results.

Permalink to Comment

23. Ian Ameline on November 6, 2007 2:09 PM writes...

All the comments here are quite good, and I agree with them, but another component of why chips are "easier" than cells is that the designers of each generation of processors are using the previous generation as tools to make the next. There is a positive feedback loop in there that just doesn't exist in Derek's world.

This is not necessarily true of the lithography equipment used to manufacture the chips, or the materials science that goes into making the precise compounds that form the transistors -- but on the whole, it is just so many orders of magnitude more predictable and understandable than the processes that take place in a cell that I'm quite sure I don't grasp even a part of why Derek's field is so much harder than mine...

Keep at it Derek and colleagues -- we're all getting older, and sooner or later we're all going to depend on the fruits of your labor for our continued survival.

Permalink to Comment

24. Ai yi yi on November 6, 2007 3:07 PM writes...

There seems to be a pretty unanimous opinion here, and I voice my agreement as well. The refinements in chip design are more akin to developing a new formulation of an existing drug - perhaps a syrup for children, or a controlled release version, etc. Those are the types of refinements that the pharma industry can realistically achieve within a set time period, and accurately forecast the effort and costs involved, and they are more on a par withthe incremental advances in chip design. So did he give a timeline for when the electronics industry will finish building that Star Trek transporter? At least that would make those trips to the doctor more convenient............

Permalink to Comment

25. MTK on November 6, 2007 3:44 PM writes...

OK, there's a concensus here, so let me throw a couple of things out there. One intentionally provocative, and tongue in cheek, and the other a bona fide question.

a) If one looks at total R&D spending vs. R&D spending as a % of sales, the electronics industry leads the former, while the pharma industry leads the latter. So if Andy Grove wants the pharma industry to become as incentivized, and as fast, as the semi-conductor industry, one conclusion is that mandatory government price controls, floors not ceilings, be instituted. That would do it, right? Companies would make damn sure that stuff got done if there was guaranteed greater than market value return.

b) If Pharma can't learn from the semi-conductor industry, what industries can it learn from? I find it highly self-patronizing to think that we're so special and so difficult that we can't apply some principles from other successful industries, countries, or segments. I'm just not sure what they may be.

Permalink to Comment

26. qetzal on November 6, 2007 3:57 PM writes...

I can easily forgive Grove for not understanding why pharma is fundamentally different from the semiconducter business. I can even forgive him for not understanding that pharma is fundamentally different.

But if he honestly thinks pharma's problem is not enough drive to bring product to market, he's being an idiot. You don't have to understand pharma to know better than that. You just have to understand the tiniest bit about business.

Permalink to Comment

27. Jose on November 6, 2007 4:18 PM writes...

MTK- I think that realizing pharma *is* fundamentally different from any other industry is not self-aggrandizing or indulgent, it is the *reality." Pharmacueticals are not normal consumer products!

Permalink to Comment

28. SRC on November 6, 2007 5:47 PM writes...

Jose, actually, if I had to choose a similar industry based solely upon business model, it would probably be wildcatting for oil (albeit without the FDA, or personal injury lawyers).

Permalink to Comment

29. Great Molecular Crapshoot on November 6, 2007 6:26 PM writes...

There are some who say that pharmaceutical research needs to learn from semiconductor MANUFACTURING. Lots of talk about lean stuff with lots of beancounting and statistics, communicated with religious zeal. Nobody seems to notice that the pharmaceutical industry might be a little more regulated than the semiconductor industry. Let's also remember when the secretary of defense was recruited from General Motors and look where that ended up.

Permalink to Comment

30. Great Molecular Crapshoot on November 6, 2007 6:33 PM writes...

OK so it was Ford and not GM; hope everybody spotted the mistake.

Permalink to Comment

31. Ian Musgrave on November 6, 2007 6:39 PM writes...

Bravo, a truly excellent post Derek.!

But in the treatment of Parkinson's, we have gone from levodopa to levodopa.
(coughs politely) Well, in Australia it's levodopa plus dopadecarboxylase inhibitors (as this greatly reduces side effects), no one prescribes levodopa alone. Not to mention amantidine (okay, so it doesn't work so well, but has fewer side effects) and bromocriptine. In those 40 years we have tuned doses, added drug combinations (levodopa, MAO inhibitors and DDC inhibitor combinations work best), and tried out a heap of things that just didn't work.

But what does he expect? In Parkinsons (and Alzheimer's and ALS), brain cells are dying. You can't bring back brain cells once they are gone. We have only a limited idea of why they die, and virtually no idea of how to stop them dying. Until we know mechanisms in more detail, anything we do is palliative only (levadopa helps the few remaining brain cells make more dopamine, when they finally die off, levadopa stops working). Rushing drugs to market is pointless when we don't know the mechanism of the disease. I speak here as someone who is trying to develop drugs to unravel beta amyloid, which is most likely the major pathogenic event in Alzheimer's (but maybe it's not). With Parkinson's we are even more in the dark (alpha synuclein anyone?).

Stopping brain cells dying, or replacing them, is a seriously hard problem (remember the big hoo-ha about neural cell transplantation in the late 90's, it ended up not working).

Making transistors which fit more elements on a chip is a significant challenge, but when you make a chip you know it works almost straight away. When you get a drug working in a test tube, it will be years before you know if it works in a human (and this is doubly so for diseases like Alzheimer's where you have to wait a least a year for a knock-in animal model which doesn't exactly mimic the human disease to give you any results, let alone monitoring cognitive decline in humans).

Even if you get something that works in humans, you can't rush a drug to market in the same way you can a transistor, there's this thing called the FDA (or the TGA in Australia, and the UK and European equivalents). Also, new chips aren't likely to kill people because of a rare gene polymorphism that isn't picked up on the initial tox scan.

But still and all, if you don't know the mechanism of the disease, then useful treatments will be hard to come by. And as Derek and other have pointed out, biology is complex, finding drugaable answers is a ling hard row to hoe.

Permalink to Comment

32. srp on November 6, 2007 6:47 PM writes...

In the spirit of devil's advocacy, I'll try to come up with some sense in which Grove might have half a point. I agree with the consensus above, but Grove's critique stimulates some heretical thoughts. And let me say in advance, I'm aware of the various institutional barriers to the suggestions below; but if these suggestions have some merit, then changes in those institutions would be the next order of business.

1. At Intel, and I believe at most semiconductor companies, there is very little emphasis on developing fundamental first-principle knowledge about why things work. Instead they just try to get them to work, record what they did that got them to work, and go from there.

My understanding is that pre-"rational"-drug design, the pharma industry worked the same way and generated a lot of useful drugs with high research productivity compared to today. Is there any evidence that intentional targeting of receptors is better than blind screening (based on hunches) at turning up good drugs? And if such evidence is absent, does it make sense to cling to a research model that increases emotional comfort but reduces average research productivity?

2. A weak analogy to the uncertainty of how a drug will work in a human is provided by the uncertainty of how a new semiconductor device or process technology will work on a production fab. (I realize that the uncertainty is a whole lot less, but bear with me. The difference actually works in favor of my argument.) The semiconductor folks build pilot lines and try to test in a realistic environment as quickly as possible, because the actual fabs are extremely complex and finicky systems (as human artifacts go). Stripped-down or simpler analogues to a real production line can be misleading models for what will happen when you try to do something for real at commercial quantities.

In drug research, we have a really complex and finicky environment--the human body--but a lot of time is spent working on animal models. From previous posts by Derek, I infer that there is no systematic evidence that success in animal models is highly correlated with success in people, or even that animals are really easier to cure than people. It seems to be an article of faith rather than a scientific principle that if something doesn't work in rats, it won't work in people (we know for sure that the inverse statement is false from all the rat cures that fail in human trials). Maybe getting compounds into humans earlier, faster, and more frequently and using animals less intensively is the key to getting more actual drugs into the marketplace.

Obviously, these are not airtight arguments, just ideas that Grove's analogy stimulated. How much of the standard operating procedures of pharma research is grounded in evidence that it really improves research productivity?

Permalink to Comment

33. StW on November 6, 2007 8:49 PM writes...

Reading all these comments, one might think that our basic medical research and clinical research systems are working as well as they possibly can. An objective look, however, would not produce that conclusion.

Perhaps the success and speed of the semi-conductor industry isn't the place to find all the answers, but we can take some lessons from any scientific or technical endeavor that has acheived the kind of success that the semi-conductor industry has. Perhaps the most valuable lesson we can take from Andy Grove's industry is its acceptance of change and paradigm shifts. In fact, in the semi-conductor industry, paradigm shifts are the goal, and they happen frequently. Compare that with our stagnant, entrenched, 40-year old clinical research models, and the comparison is starkly infuriating to anyone who takes the time to fully understand it, especially if one is being directly affected by the systemic failures that frequently occur in drug development and regulation. I am not talking about whether a drug works or not, but rather how desperatley awful we are at identifying the ones that work and getting them to the people who need them.

Why do drugs that obviously work (and there definitely are some of these) for diseases like cancer take 7 to 10 years to make it through an inflexible, one-size-fits-all phased clinical trial system while thousands of people die from their disease waiting for it? Why do we insist that every single "experiment" of a clinical trial be designed, run and interpreted by biostatisticians and physician-statisticians?

There are other ways to collect and evaluate data. As a very experienced applied scientist in the environmental field (highly-regulated - highly complex), and a person with deep personal eperience and extensive knowledge regarding what works and what doesn't in medical research, the arrogance and resistance to change in the medical and clinical research fields is unprecendeted in modern science.

Instead of a knee-jerk reaction that Andy Grove doesn''t know what he is talking about reflected by the post and many of the above comments, perhaps we should all take a breath and listen to what he is saying. I suspect, and I hope, we may learn some things he knows that we didn't know. And maybe they will help a little, or maybe a lot.

No one likes criticism, but medical and clinical researchers need it as much as anyone else. Nothing guarantees failure more certainly than believing one's own baloney, and coming up with arguments to justify one's own failures. Sure medical research is tough, but that doesn't mean the incredible progress we have seen in the semi-conductor industry has been easy. there are differences in the challenges they face, but there are also differences in the way they have dealt with the challenges they do face.

There is nothing sacred about the way we do anything in medical or clinical research, and it's not like we are succeeding beyond our wildest dreams in conquering disease. in fact, we are failing at a phenomenal rate and succeeding so rarely it should make us all think long and hard about what we can and should be doing differently - outside the box of accepted approaches that are so firmly anchored in medical conventional thinking.

I don't know Andy Grove, but I know and have known a great many people like him. He is motivated by is past and current personal circumstances to stir the pot, and there is ample evidence that the pot needs stirring.

Open minds, please.

Permalink to Comment

34. milkshake on November 6, 2007 8:54 PM writes...

Drug discovery is like writing software apps for a poorly documented buggy OS designed by aliens from a giant planet Red-Mont.

Except that you are not allowed to use the actual OS for testing until the very late in the process - and you must never cause it to crash.

The drug approval and the manufacturing practices are extremely burdened by regulations that are designed to ensure better drugs. But this formal system makes the process very slow and super-expensive. Many historically succesful drugs like aspirin would not pass the approval process nowadays. I don't see much push for streamlining the process, reducing the waste of money and the bureaucracy and bad management that is so prevalent in pharma.

Permalink to Comment

35. eman on November 6, 2007 8:59 PM writes...

All that fancy=shmancy research is for nothing when your company can't figure out how to get your drug into a tablet.

Permalink to Comment

36. MTK on November 6, 2007 9:18 PM writes...

srp and STW,

That's what I was looking for. We can't honestly believe that every one of our processes represents "best practices", can we? One thing, though srp, without some understanding, or at least a plausible rationalization of mechanism, your chances of getting an IND, much less an NDA approved are slim.

And Jose, I realize that there are things that make pharma different from other industries, which is why I used the words "some principles", but are we that different from medical devices? Are there not ways of thinking that might be improvements? And yes, I do think "It's a pharma thing, you wouldn't understand." is indulgent. We should be open to considering lessons that can be learned from others.

Permalink to Comment

37. SRC on November 6, 2007 9:18 PM writes...

And enough of defense. Let's go on the offensive. I'll maintain that with a less Edisonian and more scientifically rigorous approach, more like that in pharma and academic research, the semiconductor industry would be far ahead of where it is today.

There ya go, Andy. What do you make of that one?

Permalink to Comment

38. SRC on November 6, 2007 9:23 PM writes...

Sorry, the blog deleted my (g) at the end of that comment.

Still, the fact remains, Grove's authority is based upon his success in the semiconductor industry, which arguably was a matter of being in the right place at the right time (or, more evocatively, he was playing the tuba on the day it rained gold). Would he have been as successful had he entered the field 20 years earlier or later? Probably not, and so his comments, while interesting, don't warrant chiseling into stone just yet.

Permalink to Comment

39. Another Kevin on November 6, 2007 11:08 PM writes...

OK, let me try and give a contrarian view - which is what you'd expect from me: I'm an engineer, not a chemist nor a biologist. (Nevertheless, I work with many of both - I'm in that sort of lab.)

Yuri Lazebnik, in his paper, "Can a Biologist Fix a Radio? —or, What I Learned while Studying Apoptosis" gives a jaundiced view of how biologists approach the complexity of signaling pathways. (I'd extend his argument to other biological systems, such as pharmacokinetics.) He compares the tools that biology has at its disposal to attempts to understand a radio by removing selected components and seeing what breaks.

He points out that the engineering disciplines have developed tools for handling the complexity of their systems, tools that the biologists (at least in the new field of "systems biology") would do well to learn.

He's probably just as wrong as Grove, but I'm convinced that he's at least wrong for different reasons. Dr Lowe, could you be convinced to read this paper so we can at least discuss something less obviously absurd?

Permalink to Comment

40. justapdpatient on November 6, 2007 11:51 PM writes...

Stw - breath of fresh air.

I did scratch my head a little at Mr. Grove's analogy. But the real point wasn't in the analogy. It was in the slowness of the system, the actual [in much of the PD community's opinion] "wrongful conviction" of a drug that was in the Parkinson pipeline, and the lack of drive in producing innovative treatments,. Being the businessman that he is, I can't picture him using the words compassion or urgency, people like me are suffering, So he said it "his way".

The analogy isn't the point. Money is poured into research - it's only fair to ask "where's the beef?"

Permalink to Comment

41. C3PBW on November 6, 2007 11:58 PM writes...

Gordon Moore of 'Moore's law' and Intel fame is a member of the Board of Directors of Gilead Sciences.

One thing Gilead learned from the semiconductor industry was the importance of branding ala 'Intel Inside' . So when they outlicensed Tamiflu to Roche, part of the agreement was that the Gilead logo would appear side by side with Roche in every package and press-release mentions of Tamiflu. Now besides picking up some sizeable royalty checks each quarter, they still get their name out in front for something they did a long time ago.

Permalink to Comment

42. Ian Musgrave on November 7, 2007 1:37 AM writes...

From the NewsWeek article

Like an increasing number of critics who are fed up with biomedical research ... that lifts the fog of the rodent version of Alzheimer's but not people's...

We have only had an animal model of Alzheimer's for the last couple of years. Animlas don't get Alzheimer's, we had to do decades of research until we had a modest understanding of the disease process (hard to do for a relatively rare disease in elderly people that can only be reliably diagnosed after death), and the "knock-in" revolution (part of that despised "academic research that won't cure disease") allowed us to place the putative disease causing genes into rodents. Then we had to characterise the disease to make sure it was relevant to Alzheimer's, then test the drugs (in a model where it takes at least a year to be sure your drug works).

The first few drugs that really attack the disease process have only just come off animal tests, and have been tested in humans, which is not bad given we didn't even have a disease model at all a few years ago.

Predictably, both high profile treatment modalities that came though this process failed, both for adverse reactions (we expected that for one drug, but it was a proof of concept test, and new, more patient friendly drugs are being tested now.

StW wrote

Why do drugs that obviously work (and there definitely are some of these) for diseases like cancer take 7 to 10 years to make it through an inflexible, one-size-fits-all phased clinical trial system while thousands of people die from their disease waiting for it?

If there is one thing that we have learnt from biomedical and clinical research, is that drugs that "obviously work", generally don't. Cancer research is littered with drugs that "obviously worked" but failed miserably clinically.

Permalink to Comment

43. processchemist on November 7, 2007 3:54 AM writes...

I suppose that when at Intel they have ONE new chip working there's no problem in producing 100.000 pieces.

Mr Moore knows what happens when lets say 2 grams of a promising NCE (coming out from a long process, from in silico design to lead optimization) come out from a medicinal chemistry lab?

Permalink to Comment

44. Kay on November 7, 2007 6:05 AM writes...

"does it make sense to cling to a research model that increases emotional comfort but reduces average research productivity?"

If you care about your company, then please give comment 32 a careful read.

Permalink to Comment

45. Ian Musgrave on November 7, 2007 7:46 AM writes...

Sorry, this is going to be a long one.

StW wrote:

Why do drugs that obviously work (and there definitely are some of these)…

To return to this, it depends on what you mean by “obvious�. There are heaps of things that go gangbusters in a Petri dish full of cancer cell lines, but will not work in an intact animal. There are several drugs that will work well in rodent models of cancer (either spontaneously occurring cancer models or tumour xenograft models) that just won’t work in humans. Even if you take your drug straight from the Petri dish to humans, you have years of work ahead of you. In humans, to say a cancer drug works, the cancer has to be in remission for at least two, preferably 5 years. Even if you go for surrogate markers (tumour shrinkage and one year survival rates), you will still take at least a year to recruit your subjects (more if it is a relatively rare tumour, assuming that you can recruit people for a study where there is no toxicology pre-study, so you can’t tell them if the drug will not kill them outright), at least a year to run the experiment, and then another half year of so to do the analysis (measurements of tumours on X-rays, biochemical indices, histology on tumour biopsies these things take non-negligible amount of time to do).

So it will take you a minimum of 2.5 to 3 years just to see if the drug works in humans straight from the Petri dish (which may have been time and money wasted, as without a preliminary pharmacokinetics study you have no idea whether the dose you give your subjects, extrapolated from the Petri dish data, actually gives a high enough tissue concentration to do anything in the first place, again assuming the drug doesn’t kill your subjects outright, or makes them throw up all the time etc. because you haven’t done a preliminary tox study). So, minimum of 2.5 to 3 years from Petri dish to proof of concept in humans, then you have to convince the FDA on the basis of one years data, that your drug will cause long term cancer remission and won’t cause long term problems further down the road and justifies the inevitable higher cost of your drug on the basis of minuscule knowledge.

Now, most of the drugs that are going to successfully kill cells in a Petri dish will fail, and cancer patients go through enough as it is without testing hundreds of drugs on them that will not work. So we are obliged to do preliminary animal studies, just to make get an inkling of metabolism problems, and that the drug is not insanely toxic. And then human pharmacokinetic and toxicity studies to make sure that the right dose gets in and there is not a weird wrinkle in the human toxicology profile (cytokine storm anyone?). So of course it takes longer.

Why do we insist that every single "experiment" of a clinical trial be designed, run and interpreted by biostatisticians and physician-statisticians?

Because if you don’t, you have just wasted several million dollars, and probably trashed the chance of a decent drug to be marketed (or let through a dog of a drug).

It’s hard to get this across, but statistics isn’t a bolt on frippery to keep maths geeks happy. It is an integral part of experiment design. If you don’t do it right, then you can kiss millions of dollars good bye as you do a study that won’t actually be able to tell you if your drug has any effect.

In chemistry, if you use the same reagents and apparatus, there is a tolerable chance that you will get roughly the same yield every time you run a reaction. In biology, especially for human trials, Harvard’s Law rules with a vengeance (when two sets of animals are treated under the same conditions of lighting, heating, food intake and environmental stimulus …. They will do as they damn well please). With cells in a Petri dish, you are dealing with a genetically uniform clone with none of those nasty absorption, distribution, metabolism and excretion issues. With xenograft mice, you have a bit more complexity, but the mice are still fairly vanilla genetically, and the tumour grafts fairly uniform too.

In people, that’s a whole ‘nother ball game. They vary genetically with everything, they have a wide range of environmental influences, ingesting substances that may interfere with the already variable absorption, metabolism and excretion of your drug (some like grapefruit juice (remember Terfanidine) we can control, others we can’t), so we have a whole parcel of complicated interactions that need to be carefully sorted out it you want to see if your drug works (and the tumours are genetically heterogeneous too, and will evolve ways to beat your drug while you are testing it, cancer biology is a nightmare).

Except or a few rare drugs, like Gleevec, most drug effects on cancer are relatively small, and have to be dug out of the noise with careful experimental design. Like it or not, there is a good reason why we do clinical trials the way they are done, because the other ways we have tried just don’t work. I’ve also done epidemiology, and worked briefly as a biogeographers assistant, so I know about other approaches to experimental design and analysis. Biology is complicated and the ways we have are the ones that give us a fighting chance of getting drugs that work without harming people too much. Even then the system isn't fool proof, as organisms are just damn complicated, and nasty surprises can lurk in the wood work.

So yes, if at the end of the day you want to give cancer sufferers a relatively safe drug that will actually be of some benefit to them, rather than an expensive placebo, we will have to wait years and have statisticians on board.

Permalink to Comment

46. Anonymous BMS Researcher on November 7, 2007 8:10 AM writes...

I strongly agree with many of the points made above, in particular the stupidities that can be caused by numerical metrics and the fundamental difficulties of drug development.

On metrics, like Derek I too have seen how intense the pressure to make the numbers (which, not incidentally, are included in our bonus formulas) can become late in the year.

On biostatistics: I once was an electrical engineer before going back to graduate school and getting a doctorate in biology. Believe me, engineers also use statistics -- a lot. But the engineering world is fundamentally much easier than is the pharmaceutical world, because circuits are as close to identical as we can make them.

However, there is another fundamental difference between the engineering world and the pharma world that I have yet to see mentioned in this thread: the primary locus of value. Ford and Toyota both have large R&D departments and patent portfolios, but the reasons I'm about to drive 15 miles in a Camry instead of in a Taurus have very little to do with their makers' respective patent portfolios. The number one reason why Toyota has a much larger market cap than Ford is what happens on the factory floor. Clearly making the cars efficiently is one of the hardest parts of being a car company.

But Wall Street pays very little attention to what happens in the factories where BMS and Pfizer and the rest of us make our pills, it is basically assumed that if the drug gets approved the pills will get made. I am not saying the manufacturing end is unimportant, it is quite important and we spend a fair amount of effort on it, but making the physical pills is clearly not the hard part of our industry, as the existence of generic competition clearly proves. Our market cap is driven mainly by Wall Street's perception of our pipeline.

Permalink to Comment

47. emjeff on November 7, 2007 8:59 AM writes...

I don't have any more comments to add, but I have a suggestion; Derek, you need to work this particular blog up into an article and get it published.

Permalink to Comment

48. STW on November 7, 2007 9:01 AM writes...

In response to Ian Musgrave. The argument you make about drugs that obviously work usually don't, falls into the "believing your own baloney" category. It is the response given by statisticians who believe only in things like confidence limits and p-values. I am not being combative here, only pointing out that the cliches and sound bites so automatically thrown around by defenders of the status quo are more of the one-size-fits-all problem we have in medical and clinical research. I am not a rookie on this. I am something of an expert with many thousands of hours under my belt studying the problems and solutions, and it is in fact possible to identify some of the drugs that obviously work long before the statisticians are finally satisfied with the data from multiple, years long, double-blind, randomized, placebo-controlled clinical trials, for example, in terminal cancer patients. Good decisions have two parts, they must be sufficiently correct (but not perfect) - and they must be made in time. A correct decision made later than it should have been often makes the decision either wrong, useless or far less effective than it could have been. Since we are often talking about lives with medical treatment decisions, one would think we would be good at making them. We aren't. The art of good decision-making seems to be non-existent in clinical research, replaced by a myopic vision focused like a laser beam on an arcance set of algorithms called relative frequentist statistics. No other field of science has ever taken it to the extreme clinical reserachers and regulators have, for good reason. Strap in the scientists to a single method, and it isn't science anymore. The data is never perfect, and waiting for it to become statistically perfect, which is what the FDA is doing with increasing frequency, imposes a human cost and a stagnation of progress that far outweighs the diminishing benefit we get from waiting for that perfect statistic to finally emerge.

Disease is not caused by p-values, and no one ever calls a statistician in areal emergency, unless the energency happens to be a serious or life-threatening disease, then you are stuck with them.

Again. Open minds, and instead of repeating the conventional wisdom (which is often the "baloney" I was talking about), question the conventional wisdom. Test the comventional wisdom. Conventional wisdom is nothing more than yesterday's stale knowledge. I suspect Andy Grove would tell you that, in the semi-conductor industry, following the conventional wisdom is the quickest path to failure.

Permalink to Comment

49. Hap on November 7, 2007 12:04 PM writes...

Science isn't a library of knowledge but the method to get that knowledge. Statistics (when performed honestly) is a tool to understand what you actually know rather than what you think you know. If there's a better way to get the knowledge of what drugs work and are safe, people would be interested because they can make a whole more money and help people more effectively in doing so, but it doesn't sound like AG actually has such an idea.

The "dose and hope" school of pharmacology sort of went down the tubes when someone decided to dose 150 children with cough medicine dissolved in diethylene glycol. People are pretty risk-averse - they don't generally want to assume risks (even very small ones) for the benefit of others, and their risk-aversity decreases as they pay more for something - and this is unlikely to change. People are willing to assume larger risks either for larger benefits or when they have no choice - and (as with cancer and AIDS) the FDA already factors for those circumstances, as do the drug companies. Attempting to ignore what we know of statistics won't make the lack of knowledge or lack of pliability of human nature, human biology, and systems biology go away.

Permalink to Comment

50. Abigail Alliance on November 7, 2007 3:31 PM writes...

"Every drug for cancer and other serious life-threatening illnesses that the Abigail Alliance has pushed for earlier access to in our five and half year history is now approved by the FDA! Many lives could have been saved or extended, if there had been earlier access to these drugs!" The number of drugs is now up to 16!
(More detail in the August 14, 2007 Abigail Alliance Wall Street Journal op-ed, ‘FDA’s Deadly Track Record’)
Also go to and in the upper left hand corner click on the (short) FDA rally video button.
Frank Burroughs
President, Abigail Alliance for Better Access to Developmental Drugs

Permalink to Comment

51. Joerg Kurt Wegner on November 7, 2007 3:48 PM writes...

I was reading his article and was thinking exactly the same way you nicely summarized! I am not sure how much he really knows about drug design, but for me there is a *big* difference between a carbon and a silicon based molecular structure. Is this guy telling us that we should start making Si-based drugs? Note: We are still talking about a drug for humans, right; or are we talking about robots with an artificial bloodstream we can control?

Finally, here a statement for highlighting the differences in the industrial setup:
"If you want to understand why something happens in business, study the disk drive industry. Those companies are the closest things to fruit flies that the business world will ever see. Drug design is a process between 9 to 15 years! So, which object to study lies in-between a fruit fly and a hard-disk? A high-throughput screening, a biological assay or an 'in silico' 3D/2D/xD model of a drug?"

Permalink to Comment

52. Ian Musgrave on November 7, 2007 3:58 PM writes...

StW wrote:

The argument you make about drugs that obviously work usually don't, falls into the "believing your own baloney" category. It is the response given by statisticians who believe only in things like confidence limits and p-values.

A brief disclaimer here, as well as trying to unravel beta amyloid, I also teach biostatistics. One of the first things I teach my students about is the “bloody obvious test”, where a result is so obvious that any statistics is window dressing. The only anti-cancer drug that falls in the BOT category is Gleevec (and its variants), with a stunning 80% response rate. Gleevec was fast forwarded through the approval process. On the flip side is herceptin, which works marginally in a special subtype of cancer, which wouldn’t have been found without statistics (or we could have just given expensive herceptin placebos to women with cancer). Leading anti-cancer drugs produce effects on the order of 20% increases in remission (eg Rituximab, ot the antibody conjugated to a plant toxin, gemtuzumab ozogamicin), so you definitely need statistics to sort these out. With very few exceptions, the new anti-cancer drugs we produce are incremental improvements, and we need statistics to see these incremental improvements.

StW, could you give us an example of an "obviously working" anti-cancer agent that wasn't fast forwarded through approval?

Permalink to Comment

53. stw on November 7, 2007 5:22 PM writes...

Gleevec certainly does fall in to that category, but there have been others. And by the way, gleevec wasn't "fast-forwarded" nearly as much as the FDA wants us to think. the FDA actually dithered over that one for about two years, and was still dithering and asking for more trials after a 100 percent response rate in phase 1 and response rate above 80- percent in Phase II. there have actually been virtually no exceptionally promising drugs that weren't held up to some significant extent by the FDA. And they are still being held up. On the subject of FDA dithering and delays, I really am an expert.

The inability to see the other drugs for which this was true has its roots in the rule of statistical analysis that requires everything to be viewed in the context of populations, whether we have defined an actual controlled population or not. A drug that obviously works for a subset of patients in a trial is their "Gleevec" even if it is not "Gleevec" for everyine in the trial.

The fact that it doesn't effectively treat everyone chosen to be in the trial population is a failing of the method and the designers of the trial, not the drug. The designers of those trials failed to identify (i,e,. control) the population properly. What they did was run a trial that included patients with different diseases. This is so obvious it almost intellectually hurts. Macro control variables like prior treatments, mets or no mets, no prior surgeries or specific prior surgeries, etc. have little at all to do with whether a patient responds to a drug or doesn't for genetic/proteomic diseases. It is a construct of the ignorance that existed back in 1962 when statistics was chosen as the basis for all clinical research and all approval endpoints. The diseases we are still having the most trouble with are the ones driven at the molecular level, so in order for statistical trials to work well, we have to control for molecular differences. The easy way to think of this is that Tarceva works well for about 10 percent of lung cancer patients and hardly at all or not at all for the rest. This teaches us that lung cancer is not lung cancer, and that we didn't control the trial for that, not that the drug doesn't work. For those it does help, the results can be miraculous if unfortunately, not permanent. That doesn't mean the drug doesn't work - it means we have to figure out who it will work for, who it won't, and why. Once we do that, we begin to understand not just the effect of the drug (which is all we ever really get from statistical trials) but both the effect and the cause of that effect. Once we link the cause with the effect, we no longer need staistical clinical trials for that question because we have gotten to what is generally termed, "first-principal" science.

Oddly, just as a couple of teams working independently had managed to tease this information out for the very similar drug Iressa, FDA decided to pull Iressa off the market because a randomized trial that averaged the survival outcomes of the 90-percent of non-responders in with the 1 percent of responders fell just short of statistical significance for showing a survival benefit. So a useful drug for a few who could be fairly confidently idebntified in advance of treatment was yanked just as the information needed to optimize its use was learned. Why? because FDA doesn't beleive in anything but statistics, whether it produces a rational and medically correct decision or not.

Statistics can be a useful tool, but it cannot be the only tool we use, because it too often forces us to ask and answer questions in the wrong way, or to ask the wrong questions altogether, in which case one can't possibly get the right answers. Methods and experimental designs are just recipes for cooking up data, and if we confine ourselves to one recipe, we severely limit the quesiotns we can ask, and the answers we can get. Sorry, but defending statistics as the only valid basis for the design, conduct and analysis of experimental data in clinical research is like trying to swim with a strait jacket on (those darn analogies again).

As I said in an earlier post, once scientists get strapped into using a single approach to experimental design, or data analysis, or anything else for that matter, it isn't science anymore. It is repetition, and we should not expect much progress from it.

Science is the serach for knowledge. The methods we used to made today's discovery is often not gong to be the method we need to use to make the next discovery. Since 1962 we have gone from barely knowing what the genome was to haveing decoded the entire thing more than once, and we are approaching very rapdily the ability to decode human genomes the way we check for cholesterol levels. Over that time, our narrow clinical reserach model has not changed at all and remains firmly in the hands of reserachers who refuse to beleive anything but the results of a Kaplan-Meier curve based on a clinical trial designed to produce nothing but a Kaplan-Meier curve.

No one should be surprised we are failing a lot more than we are succeeding. If we don't start looking where the exponentially expanding knowledge of human disease biology is telling us the answers are, we don't have much chance of finding those answers.

Permalink to Comment

54. Hap on November 7, 2007 5:56 PM writes...

How do you determine whether something works in a subset of a population without statistics? Statistics isn't sufficient (you need to have an idea that some people with be more susceptible to a drug than others), but once the idea exists, the only way to test it is statistical (because human biology isn't well enough understood and because people's behavior can't be held constant to do a controlled lab experiment).

The other problem seems to be that we can't decode the genome as easily as a cholesterol test - people have been plugging away for some time even after the sequence of the basic human genome is known, because people don't know what it all does. We don't know what proteins are produced (let alone how) or what they all do. Mechanism-based logic is tough without a mechanism.

It doesn't seem as if we have the tools to make safety and efficacy testing faster, either. The FDA insists on time-consuming safety tests because their output is predictable and (mostly) accurate. If one had a different method to determine safety and efficacy, one would still have to prove to the FDA that the new tests are functionally equivalent to the old ones before the faster tests could be deployed in practice. In addition, some aspects of drug testing in people are slow because it takes time to see an effect (as indicated above).

Considering the problems with the drug industry, it's clear it can learn from someone how to do its job better - but since people can't agree on what's wrong, it's hard to understand who might understand its problems and have a better method in which to do its job.

Permalink to Comment

55. stw on November 7, 2007 6:46 PM writes...

You determine it clinically. You are confusing the observations that produce the data points used in statistical analysis with the clinical observations that produce the data points in the first place. An example. A patient with Stage IV, advanced and progressing colon cancer presents for enty into a trial and is found to be eligible. I know from very involved personal experience and lots of research that colon cancer is a progressive disease. this is cauise the natural course of the disease, and we understand this pretty well after watching these diseases kill people millions of times. The patient will have been taken off his/her last therapy because it was no longer working (i.e., the tumors were growing and/or new tumors were forming - this is how progression is measured in colon cancer along with some blood markers that are less defintive than Ct scans of tumors). The natural course of the disease is continued progression in the absence of treatment, and in colon cancer, spontaneous stable disease and/or regression are so rare as to be non-concerns. the patient is then given a drug and the tunors stop growing and spreading for an extended period, or they shrink and dtsbailize, or in rare cases, they completely disaappear (called a complete response) for a time period long enough to be measured on periodidc scans (usually every 6 weeks for colon cancer). These are all considered responses and they are the observations that the statisticians use as evidence of clinical benefit. They are the plus side of the trial data. people who whose disease continues to progress (even if it may be slowed a little) are considered non-responders. For the people that experienced the complete responses and the partial responses (generally defined quiet arbitrarily as tumor shrinkage of 50 percent or more), they certainly responded and the drug worked for them for as long as the response lasts, which is referred to as duration. This is the clinical data statisticains use from clinical trials. it is meaningful in iots own right because it shows that there are patients who respond to the drug, but statisticians don't care about that obvious fact and rejecty it as anecdotal because they deal only in populations. this is actually a profoundly unscientific concept, but they do it nonetheless.

What they then do with it is kind of like deciding whether you like a movie by reading a review written by someone else (analogy again). They calculate the average number of responses in each arm of the trial (and if it is against a placebo in advanced colon cancer, there won't be any responses in the control arm) and then they compare them. They also calculate arcane almost meninlkess metrics called 95-percent confidence and related p-values, and they decide whether the drug is better than the placebo based primarily on the p-value, which is yet another arcane and almost m,eaningless measure of whether the result could be due to chance. the problem with all of this is that is completely removed form ewhat actually happened. Some people responded and benefitted from the drug, so there is obviously a subset of patients who should be geting it. it has nothing to do with statistics. you learn it before a statistician ever touces the data. The statistician then loses and obscures the inherent knowledge imparted by the "anecdotal" data by sunmerging it in a bunch of sequential averaging calculations. So what we really learned is iognored, and the next quesiotn we should be asking - which is how to identify the patients who should get the drug, and conversely how to oidewntify those who shouldn't because it won'y work for them, is lost.

It would literally take a book to explain whty this is incredibly stupid given what we now know about the molecular causes of disease like colon cancer, but because the next step is invariably another trial desitgned by statisticains because the FDA virtually mandates that those are the only experiemntal designs they will accept, we never get around to doncutcting the kind of noin-staitsical clinical trioals that wouyld give us the answers we need. Instead the FDA either approves the drug or doesn't based on the averages, and if they approve it, doctor's will prescribe the drug to patients with no idea at all who will respond or for how long, except that if tey treat enough of them, eventually they end up somewhere around the average response and duration found in the trial, if they bother to average the results experienced by their won patients.

This is stupidity, not science. It is also what is mandated by the FDA and staunchly defended by a lot of people who can't really explain why we can't change it, except to say we can't. This string of blog posts presents a pretty good record of that kind of thinking.

Permalink to Comment

56. stw on November 7, 2007 6:59 PM writes...

Actually, some of the people engaged in developing genomic profiling systems are predicting that we will be able to decode an indidviuals genome for just a few thousand dollars as soon as about 5 years from now. With those projections coming down quickly, the reality is probably closer than that. genomic preofiling svcience is one of the few things in medical reserach that is going vertical in terms of progress - perhaps because doing it is very similar to the kind of thing intel does. It is developing a technology where we alre