Re: "Statins caused my kidney failure"

From: Herman Rubin (hrubin_at_odds.stat.purdue.edu)
Date: 07/14/04 

Date: 14 Jul 2004 15:28:19 -0500

In article <slrncf3h24.n1d.mbkennelSPAMBEGONE@lyapunov.ucsd.edu>,
Dr Chaos <mbkennelSPAMBEGONE@NOSPAMyahoo.com> wrote:
>On 11 Jul 2004 14:49:06 -0500, Herman Rubin <hrubin@odds.stat.purdue.edu> wrote:
>> In article <10f0ibdlkqejtc7@corp.supernews.com>,
>> Robert <Robert@hotmail.com> wrote:

>>>"George Conklin" <nilknoc@earthlink.net> wrote in message > > What do you
>>>mean no one noticed? How do you know about now?

>>>> We know now because 40-50 years after treatment began, the proper
>>>> research was done. I am surprised how gloriously ignorant you are.

>>>How did they know then about doing the proper research about that specific
>>>problem that we know now?
>>>That information was gained ONLY after 40-50 years of treatment experience
>>>AND our current understanding of cancer physiology. They didn't even have
>>>tumor markers then so how in the hell would they have done the proper
>>>research then? You totally don't have a clue. You are mixing up 50 years ago
>>>and today. Fifty years ago there was no HIV so why didn't they do the
>>>research concerning the bad side effects in HIV infected women as far as
>>>cervical and uterine cancer back then? Dahhh!!!!!!

>> You are quite correct; we cannot wait until the "correct
>> research" is being done. In fact, in too many cases, the
>> FDA waits too long, and it is acting under pressure. The
>> probabilistic analysis needed cannot be done except
>> approximately, but those indoctrinated in the classical
>> statistical religion are unable to understand this. Giving
>> me the information that something was significant at a
>> certain level, or the p-value, gives me some information,
>> but not the relevant information.

>Would you favor a Bayesian estimate of likely hazard ratios
>under the treatment?

As I seem to have been the first to show that
self-consistency requires Bayesian behavior, and I have
been advocating that on this group for years, I believe
that is a partial answer.

What is needed is an estimate of the probabilities of the
prognoses under the various modes of treatment. This will
also require some judgment about how the patient fits into
the categories. It will be necessary to use "sloppy"
methods; even if one was able to come up with a good model,
and I do not see that ever being done for very high
dimensional problems, which this is, it may still be
possible to get a robust prior Bayes procedure, which does
a good job of evaluating the various expected risks.

Risk has to be taken into account, not just probability.
This is now done to a small extent, but it is only the patient
who has a legitimate input into risk unless public health is
directly involved. Dynamic programming, which is what is done
in industry with known or assumed probabilities, is more than
a half century old.

A major problem is finding a way to handle the amount of data.
Medical researchers leave much out in an attempt to use what
are classical, or even simplistic Bayes, methods of analysis.
For the real problems, I do not see a possibility of using
such simple procedures, but there is a possibility in finding
ways around them.

Much progress is now being made with highly inbred laboratory
animals. This way, one can eliminate extraneous variables.
But one cannot do this with human populations. One cannot
even tell different groups of humans how much exercise to do,
so the estimated effects of exercise may totally wrong. Is it
that people are healthy because they exercise, or that they
exercise because they feel healthy? A major problem with the
statin studies is that I do not believe they have included
quite a few of the relevant factors, and the medical researchers
do not know how to carry out the analyses if they did.

>Can you elaborate on your preferred way for quantitative inference?

>(this is not a rhetorical question, I am honestly interested in the
>answer)

I believe we can use better statistical methods. Some of the
modern methods have gotten people to consider some of the worst
problems, but very much needs to be done, and in something as
complicated as biological systems, it must be realized that the
ideas of classical testing and "statistical significance" have
to be abandoned; the decision approach shows that even in simple
situations, they are wrong. One needs to have people who know
lots of theory apply it, and this is not being done. 

-- 
This address is for information only.  I do not claim that these views
are those of the Statistics Department or of Purdue University.
Herman Rubin, Department of Statistics, Purdue University
hrubin@stat.purdue.edu         Phone: (765)494-6054   FAX: (765)494-0558

Relevant Pages

  • Re: FAQ * r.g.l.faq * 01/17/2006
    ... > chance none of the numbers from the previous draw will repeat ... Of the 55% chance it's about 30% one number ... challenges show the probabilities rule. ... Although making use of these sorts of statistics hasn't ...
    (rec.gambling.lottery)
  • Re: Multiple comparison correction using p values alone
    ... asserts Pis not P*P, ... probabilities, not p values, there would be less kerfuffle. ... correlation and sufficient statistics, if we have a molecularly based ... posterior probabilities of 0.95 for each of two lineages of the split, ...
    (sci.stat.consult)
  • Re: The Philosophy of War
    ... basic understanding of how probabilities work. ... of knowledge and understanding of statistics. ... the coin analogy, and figure out why the coin is not fair, and how we ... Bet on a coin flip. ...
    (rec.martial-arts)
  • Re: ACS Law - ooops
    ... and secure a judgement 'on the balance of probabilities', ... _and_ evidence to support it. ... you are conflating 'probability' and 'statistics'. ... that statistics can be found to show that the majority of unlawful ...
    (uk.legal)