Kevork Hopayian writes:
>The event rates in such trials could be worked out in at
>least two different ways and they give two different
>answers. You could calculate the event rate from the
>events per patient-years as stated in the reportıs tables
>(method 1) OR you could calculate it from the number of
>events divided by the number in the group over the median
>or mean follow up (method 2). Method 1, I would say, is
>the more accurate of the two.
This is an interesting example. It seems to be related to the fact that the
average of a ratio is not equal to the ratio of the averages. For example,
to get the average of 1/2 and 3/4 you can't divide the average of the two
numerators by the average of the two denominators. That would get you a
value of 2/3, when the correct answer would be 5/8. For the most part, the
discrepancy will be small (e.g., .667 vs. .625),but if there is a lot of
variation in the denominator then the discrepancy could be quite large.
If the denominators are constant, (e.g., 4/8 and 6/8) then either approach
will give you the same answer.
Perhaps this answer is more intuitive than the one you provided, or perhaps
it is just confusing. But it does illustrate those tricky mathematical
paradoxes, like when a raise of 50% followed by a 50% salary cut leaves you
worse off than when you started.
Steve Simon, [log in to unmask], Standard Disclaimer.
STATS - Steve's Attempt to Teach Statistics: http://www.cmh.edu/stats
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|