Re: Medical Research


"Herman Rubin" <hrubin@xxxxxxxxxxxxxxxxxxxx> wrote in message
news:e4d7t0$5pkg@xxxxxxxxxxxxxxxxxxxxxxx

You're too focused on stats.

Statistics is the science of decision making under
uncertainty. Is that not what patients are supposed
to be doing with the advice of the physician.

Certain uses of statistics have been called the "religion
of medicine". This includes the belief in "statistical
significance" and p-values.


Herman, only if the differences are very small, right? You are talking
about 3 out of 3 being cured. Maybe that would not significant, but 8 out
of 8 would be. I was involved with the evaluation of a noise lawsuit with
that kind of 'pretrail' determination by retired judges. It was settled at
that point without more suits. Both sides knew what was going to happen.



Statistics can only change the odds of the
various types of possibilities. The odds can become so
overwhelming that we may say something is proved.

In any case, the standard use of statistical significance
cannot even be translated directly into odds. That null
hypothesis is always false; how much water one drinks has
an effect on diabetes or cancer, which is probably small.
Statistical significance says absolutely nothing about the
magnitude of the effect, nor does its lack.

Whether a treatment is good is not changed by collecting
data, and intelligent decision making will cause the use of
many treatments before that much information is present.

Suppose you had a disease which was about 50% lethal, and
you gave a million people each a placebo or a treatment.
Suppose the difference of the survival rates was 1% in
favor of the treatment. That is extremely significant, so
much so that one could say it was proved that the treatment
had an effect. I suggest that an experimental treatment
which has cured three out of three is a better bet,
although this is not statistically significant.

However you number crunchers do it is your gig. If I do a study with
1000
patients and on treatment X 12 patients get better and on treatment Y 200
patients get better with no difference in side effects and all other
things
being equal, that is a practice changing study. That will most certainly
result in your P value being "significant" thought that, to me, is a
secondary concern.

I have even seen example cited where that is not the case.
See examples of Simpson's paradox.

and can
definitely not prove that a treatment has no effect.
One can get information to better assess the risks and
benefits.

While I don't treat breast cancer, I do deal much with prostate
cancer.
There are plenty of comparisons. We treat many men witth hormonal
therapy.
Many of the studies out there are old before newer diagnostic and
treatment
techniques were available. Those tests showed that early hormonal
treatment
made no difference vs. later hormonal treatment, so we might as well
wait
for a patient to become symptomatic before giving a treatment with
significant side effects. Newer studies have shown, however, that
early
hormone treatment probably does help in a selected patient
population.
We'll probably NEVER be able to prove this because those older
studies,
done
in the '70's, would be deemed unethical today. So we have studies
that
suggest a difference but can't prove it. What do we do? Most
oncologists
who treat prostate cancer would start early hormone therapy in those
patients who will probably benefit from it. We're now waiting on a
large
European study which, though it won't be exactly what we need, will
probably
help settle some of the debate.

This is the problem with MUCH of medicine. There are
ways to assess the current state of the information, and
act on it. Of course, mistakes will be made, as they are
being made now.

I know you don't like much of medicine, but if you have a better way
to
get
at the truth in such matters we'd love to hear it.

You cannot get at the "truth", you can only approximate it.

Things move slowly
in
medicine, and for a good reason. For every new treatment with a
small
study
that shows that it works... and goes on to be a good treatment, there
will
be 10 that eventually end in failure or worse. We can't just jump at
every
fad.

This is the case; things like this were the reason why
one cannot just look at a study and see what else comes
out. However, one can do much more than is being done
now, but nowhere near what we would like.

We need better studies, and more data collected on each
study.

Absolutely.

<> <> Continuous variables should be used as such, and
<> <> models based on biology used more often. All of the
<> <> studies I have seen on the use of statins are horribly
<> <> flawed; they do not take into account the actual
<> <> concentrations of the various lipids. As of this time,
<> <> there is not a single reasonable study showing that the
<> <> use of a low fat diet is beneficial.

<> <> I do not know how much followup can be done on drugs. A
<> <> physician is supposed to report adverse events reported
<> <> to him; how much is done? Also, what is adverse? I doubt
<> <> that the posters on this newsgroup will agree on which
<> <> actions of a particular drug are adverse. Some only occur
<> <> when the drug is started, or when dosage is changed, and
<> <> go away. I have seen little of this in the PDR material.

<> <> What can be easily done about these? One, more data can,
<> <> and should, be collected, and not just on drug trials.
<> <> The problem is that this can be very difficult, and the
<> <> opinions of physicians is VERY difficult to quantify.
<> <> If we had to do tests for all of this, the cost would be
<> <> very high.

<> <> Everyone needs to understand the concepts of probability
<> <> and statistical decision theory, but they do not need to
<> <> know how to carry out the calculations. Problems need to
<> <> be formulated first.

You are right Herman. As genetics move faster, we are going to have to
mine old studies for new results because the old hypotheses are going to
seem quaint, but the records of experiments will remain.


.

Relevant Pages

  • Re: Medical Research
    ... which does not understand how to use statistics. ... benefits of any available treatment, ... the individual patient. ... hormone treatment probably does help in a selected patient population. ...
    (talk.politics.medicine)
  • Re: Medical Research
    ... which does not understand how to use statistics. ... of a treatment, and especially so if qualitative results ... the individual patient. ... In actual practice, patients most ...
    (talk.politics.medicine)
  • Re: Medical Research
    ... which does not understand how to use statistics. ... of a treatment, and especially so if qualitative results ... the individual patient. ... In actual practice, patients most ...
    (talk.politics.medicine)
  • Re: Medical Research-Evidence
    ... probabilities of the various results for each type of ... and let the patient make the decision. ... preferences for various outcomes, decide on the treatment. ... Herman Rubin, Department of Statistics, Purdue University ...
    (talk.politics.medicine)
  • Medical Research
    ... the individual patient. ... One cannot "prove" anything with statistics, ... definitely not prove that a treatment has no effect. ... made no difference vs. later hormonal treatment, so we might as well wait ...
    (talk.politics.medicine)