That is one tear-jerker of an article. I found myself really resenting how much I was being manipulated by the interwoven narrative. But it is a very interesting story.
I found the article to be totally obnoxious. It focused too much on the tearjerker aspects, got a bunch of quotes from doctors essentially saying "I don't trust science", and overall totally misses the point about experimental design by implying a false dichotomy (either we need people to die as controls or we don't need studies at all). There's an important point in there (as the linked post indicates), but most people won't get it from the article.
I feel like the linked post kind of glosses over the importance of getting good control data; I mean, yes, if you had excellent (or even quite good) record keeping of every cancer patient you could use an algorithm like the one they linked to get pretty good inference, but insuring that record keeping for all patients -- even those who aren't enrolled in trials -- seems like a huge change in the role and responsibility of the FDA. I mean, I see what the guy is saying, and the algorithm seems like it would be good (there are probably a number of algorithms that would do as good a job as the standard control group/experimental group paradigm, given the right data), but the hill to climb to get every single doctor (as well as large portions of the federal bureaucracy) to grasp that? That's a big hill.
But yes, in general if ML-influenced statistical techniques were more widely known in the medical/scientific community the world would be a better place. (Arguably Cosma Shalizi would spend more time annoyed at people who didn't actually understand what was happening with these inferential algorithms, but that's better than cancer. Geez, Cosma.)
I'd still worry about lack of blinding. Because placebo effects are very real and very big.
5: was there blinding in the example case? If the patients were pushing the doctors to switch them between groups it doesn't seem like it.
6: No, there wasn't blinding in that case. I meant in general I worry about it.
But yes, in general if ML-influenced statistical techniques were more widely known in the medical/scientific community the world would be a better place.
Lazy ML or OCaml?
Also, who broke gmail?
Mine's working fine. Perhaps you've been blinded and that's why you can't see it.
A control isn't a control if it was organized by different people in a different place with different demographics.
Also, who broke PLoS Pathogens?
11: but if you had accurate records of those demographics, and enough samples of enough treatments across a wide enough range, you should be able to control for that in your analysis even if there's no group that exactly matches the demographics.
the hill to climb to get every single doctor (as well as large portions of the federal bureaucracy) to grasp that
Mission creep for Obamacare. Not a bug, a feature!
Your jokes are totally evading me today, TLL.
Once the entire population of the US is covered by Obamacare, the reporting requirements to any US govt agency that wants data will go up. The MDs will resist, but they will lose. It may end up helping research because of the large data sets.
It will undoubtedly help research a huge amount. Will it hurt anything?
I feel there's something objectionable - though I'm not sure how to articulate it - in the prevalence of the sentiment "This person has to get experimental treatment ABC or they'll diiiieeee," which was also much in evidence in that recent New Yorker piece on end-of-life care. There's a reason these things are experimental, and having a terminal disease doesn't obviate that. I think TV-doctor-style magical thinking ("Dammit, at least it's a chance!") is partly at fault.
16: My memory is that everyone who specifically denounced CER was either in the pocket of industry or ideology; to the extent it may have spread further, it was probably just as a means of confirming prior beliefs that PPACA* was a government takeover.
*I like calling health reform "puh-packa."
Sorry, KR, no black helicopters here. Although your 19 last is almost sure to happen. Along with special exemptions for Members of Congress and their designees.
12: Unless there is some important factor you don't know to control for.
Apparently, in the US it is common (for ethical reasons) to end the experiment early to give everybody the new miracle drug while in England it is common (for ethical reasons) to end the experiment early to make sure no one is given the dangerous new drug.
In Germany, for what they claim are statistical reasons, all medications for clinical trials must be administered rectally. Which is why German pharmacies don't carry eye drops.
Arrowsmith Chapter 34
"Young man, if I were commanding a division at the front, with a dud show, an awful show, going on, and a War Office clerk asked me to risk the whole thing to try out some precious little invention of his own, can you imagine what I'd answer? There isn't much I can do now -- these doctor Johnnies have taken everything out of my hands -- but as far as possible I shall certainly prevent you Yankee vivisectionists from coming in and using us as a lot of sanguinary -- sorry, Evelyn -- sanguinary corpses. Good night, sir!"
||
Bonus OT Arrowsmith quote from a bit further on:
"You know. It's that renovated old part of Brooklyn where writers and economists and all those people, some of them almost as good as the very best, consort with people who are almost as smart as the very smartest. You know. Where they dress for dinner but all of them have heard about James Joyce."|>
Where they dress for dinner but all of them have heard about James Joyce.
Mouseover text?
Jeezum Crow! Way to kill the blog, TLL.
OT: The GOP's recently announced "Pledge to America" -- its platform, essentially.
According to John Cole,
it looks like the prescription for the future is tax cuts, missile defense, gay-bashing and fetus worship, and investigatingACORNthe White House.
Why anyone is not freaking the fuck out at this point is beyond me.
Along with special exemptions for Members of Congress and their designees.
Nah, they'll get special treatment the same way anyone else can in America, by paying for it. Not even the most zealous single-payer advocates that I know (including myself) believe that people should be prevented from paying for more treatment than the government insurance is willing to cover.
Floors on quality of life, not ceilings. Progressive politics is kind of like a nice patio in that way. Oh, and in being kind of green.
There are hundreds of compounds which are both efficacious for a genotypable subset of the population with a disease, often above 90%, and also harmful or fatal to the remaining subpopulation. These could save lives now at minimal cost, people have a right to get the medicine now, etc, etc.
Why exactly must efficacy be proven before a compound is FDA approved? Phase III of clinical trials is far and away the most expensive and prone to unpredictable failure; it raises the aggregate cost of medicine by a lot, maybe an order of magnitude.
Why exactly must efficacy be proven before a compound is FDA approved?
Because FDA approval is taken, rightly or wrongly, as an imprimatur?
Actually, I'd like to know what 31.1 is talking about because I've never heard anything like that.
And it strikes me as good that the FDA wants proof of efficacy. If you are just going to test for safety, you may as well go get some homeopathy.
Also, I don't think it's reasonable to expect doctors to interpret testing data on their own. They get too much bad stuff direct from the drug companies as it is. You need a reliable clearinghouse.
4: it is better for the sun and moon to drop from heaven, for the earth to fail, and for all the many millions on it to die of starvation in extremest agony, as far as temporal affliction goes, than that one soul, I will not say, should be lost, but should report one single estimate without uncertainty, should draw one confounded causal inference, or should confuse one poor p-value with a posterior probability.
eg
http://www.ncbi.nlm.nih.gov/pubmed/15285699
or Vioxx, recently withdrawn though an identifiable subset of people are unaffected by the problem.
searching with
polymorphism efficacy adverse
will identify more of what's public.
34.2-- OK, but this is largely what's responsible for very expensive medication in the US, and I would bet why new drugs will come from India or possibly China by 2030. There's an excluded middle between current regime and snake oil, I think.
There's an excluded middle between current regime and snake oil, I think.
Let's be careful that the middle doesn't involve moving testing to places with weak human subjects protections. Because that would be the easiest way to save money.
I'm also worried that if FDA approval comes before efficacy is established, nobody will ever test efficacy in a rigorous way.
I suspect more efficiency in testing is possible without sacrificing core safeguards. Plus China and India will probably increase, not stand still, in regulatory rigor over the next 20 years.
India has a huge body of detailed phenotype + treatment history data now.
More efficiency in testing is not likely unless there's a cost-benefit discussion, which hasn't happened yet. My understanding, possibly flawed, is that rescuing orphan already-tested compounds with genotyping is more expensive than testing a new compound.
How rigorous is efficacy testing now? Subjects present at the start of a study who are excluded during the course of the study, how tight are those guidelines?
God, I hate the Vioxx example. Just compare Vioxx to Celebrex. Same efficacy, same problems. Merck backed down, Pfizer didn't. If a 60-year wants to knowingly* take on a small risk of heart problems to alleviate pain, let them.
*Knowingly being the key word. Full disclosure of the risks should be mandatory, like smoking cigarettes.
lw
I kind of agree, but clinical trials for polymorphism testing are even more expensive than general population testing, since you need more patients, more screening, and more control groups.
And though the CYP2xxx variation is well known, it's specific effects are still being discovered.
40.1,2: You've reached a point I can't knowledgeably debate.
40.3: I don't have a great deal of confidence in the current system, but I also don't see it as a huge crisis that new drugs aren't coming out rapidly - I think as or more important is to work on doing better with what we have. (While still moving toward another generation of innovation that better uses genetics/biologics, of course.)
Of course, 43.1 to 43.2.
Anecdata: Mrs y worked crunching numbers for a cancer research outfit in the vacations when she was a student. Apparently, if you were in a trial there was a (just barely) significant improvement in the 5 year survival rate (which is one of the standard stats for this stuff), EVEN if you were in the control group.
She's always said, if anything serious goes wrong with you, get into a trial, any trial. Then take it from there.
EVEN if you were in the control group.
Why is that, do you think? Perhaps being in a trial ensures a better standard of care overall?
45: the placebo effect is probably a big chunk of it.
45. Presumably. That was a favoured hypothesis at the time. Closer monitoring leading to more responsive care, placebo, enhanced sense of self worth as on the part of the patient because they're contributing something, some combination of the above.
Question: if I know I'm in a situation where much of the benefit I feel may be placebo, do I fail to feel the benefit because I dismiss it as placebo?
Pwned by 47.2 (I swear, I was about to ask a very similar question!). My guess is that you still feel the benefit, because much of the placebo effect has to do with "emotional" rather than "intellectual" responses to treatment (that's an awkward way to express the distinction I'm getting at, but can't think of a better one at the moment).
29: Parsi, if it makes you feel better I am freaking out. About the biscuits, of course. But also about the inevitable dumping of a giant pile of shit on our heads by the radical rightists.
I am mulling over what to do about it. Last time around I just watched. I can't stomach doing that again. I will probably just write lots of letters to my elected representatives telling them I expect them to push back, but I'm strongly tempted to get into the forgery business and flood the investigation committees with bogus documents and leads just to keep them running around in circles.
Did you know that Michelle Obama is a partner in a company that trades in endangered species? Barack Obama sold crack in law school. Joe Biden buggered a gibbon.
I'd stay away from that last allegation, just in case.
49: Joe Biden buggered a gibbon.
Only because Neil Kinnock did it first.
Why anyone is not freaking the fuck out at this point is beyond me.
Speaking for myself, it's because we've been through the exact same thing before, from 1981-1992 and 2001-2008. Having freaked out at the outset of both eras, to no effect whatsoever, I've pretty much given up. We'll get the shitty government we deserve (again), Democrats will scurry to the right in response (again), and the storyline will be established (again) that it was all because an actually conservative Democratic president tried to shove MaoTseTung Thought down everybody's throat.