[R] Clinical significance - Equivalence test

[Greg Snow]

I agree with the others that you should consult with a statistician,
but here are some additional things to consider:

The usual equivalence test can be done much simpler (both computation
and conception) by just calculating a confidence interval on the
difference and seeing if the entire interval is within the equivalence
region.  But you may also want to look at non-inferiority.

But the straight forward equivalence testing may not be the best
option, consider this comparison:  Lets assume that systolic blood
pressure in our population of interest has a mean of 120 with a
standard deviation of 20 (those are actually fairly close to truth for
healthy adults), my new test is to flip a fair coin and report a
systolic blood pressure of 20 if the coin is tails and a systolic
blood pressure of 220 if the coin lands heads.  These 2 distributions
have the same mean of 120, so with a large enough sample size we can
show these 2 to be equivalent (according to the mean).  But I for one
do not want my doctor using the second approach in deciding if I need
medication for high or low blood pressure.

What if I have 2 measures that have a perfect monotonic relationship,
but different means for my population.  This means that they are not
equivalent, but knowing one value would give me a perfect knowledge of
the other (just need some calculations/adjustment), I would definitely
prefer this to the coin flipping above.

I would suggest googling for "Bland Altman", they are 2 authors of
several papers, tutorials, websites, etc. that talk about many of
these issues is more depth looking at better approaches in many cases
than a simple equivalence test.

On Thu, Mar 27, 2014 at 7:53 AM, Manuel Carona <unkuiri at gmail.com> wrote:
> Hi,
>
> I have implemented a therapeutic intervention on two groups (one is a
> control group) and tested them in two moments using some assessment
> tools (with normative data). Now I want to compare the experimental
> group with the control group using clinical equivalence testing. To do
> this I need to specify a range of closeness (One for each assessment
> tool according to the specificity of this same tool) and do two
> one-tailed tests to test if the two groups are considered clinically
> equivalent in the first moment and on the end I want to compare the
> experimental group with the normative data (Here I have to add the mean
> and standard deviation of the normative sample because I don't have the
> normative sample).
>
> I know that R has a package named equivalence but I don't know how to do
> this kind of calculations with it. Is it even possible with the actual
> packages?
>
> Thanks in advance
>
> ______________________________________________
> R-help at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.



-- 
Gregory (Greg) L. Snow Ph.D.
538280 at gmail.com