Jeanne Lenzer writes:
>The National Cancer Institute just announced that "screening and/or
>treatment" is responsible for downward trends in prostate cancer deaths.
>This seems like a real stretch to me and I need help because they are a
>term and technique terms I don't understand.
>
>First, to me, it seems that a downward trend in prostate cancer morbidity
>and mortality could simply be due to a cohort effect, another possibility
>is "detections effect" in which skyrocketing increases due to better
>detection begin to taper off as more and more men have had PSA testing so
>fewer cases may be detected causing both a false initial increase in death
>rates and then a slowing of death rates.
>
>But they stumped me when they said they arrived at their conclusions using
>a "new statistical technique" called "joinpoint" analysis in which they
>look at segments of time rather than overall trends. Come again? can
>someone help me out here? This doesn't sound anything like proof of effect
>to me. Or am I nuts? Why are "segments" in time being used instead of
>overall trends? Were regions in which there is a high rate of screening
>and treatment compared with regions with four-fold variations of the same?
And Sue Kaiser writes
>In PubMed I found "Permutation tests for joinpoint regression with
>applications to cancer rates," Kim et al., Stat Med 2000 Feb
>15;19(3):335-51. I was particularly delighted to read in the abstract that
>"Each p-value is found using Monte Carlo methods." Perhaps one of the list
>subscribers gifted in explaining statistics (a couple of names spring to
>mind) would clarify all this. I'd be grateful.
I can only guess at the term "joinpoint analysis" without reading the
Statistics in Medicine article. I will try to track down this article. I do
have two comments though, for now.
First, both permutation tests and Monte Carlo methods are statistical
approaches that use brute force computer power to solve an otherwise
difficult problem.
Second, if all the problems that Jeanne Lenzer describes are real, I don't
see how brute force computer power can overcome them.
A good rule to keep in mind is that if there is a problem with the
statistics in a paper, more often than not, it is in how the data was
collected rather than in how it was analyzed.
Steve Simon, [log in to unmask], Standard Disclaimer.
STATS - Steve's Attempt to Teach Statistics: http://www.cmh.edu/stats
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|