Is it fair to calculate NNT and NNH from a case series data? Well, I see
two problems.
First, the data is not randomized. Typically, NNT and NNH are only
calculated from a randomized trial. While you could calculate it for
observational data, it leads to some unusual interpretations.
As a trivial example, on the Titanic, 835 of the men and 33% of the
women died. That means that you can compute the NNH for gender as
1/(0.83-0.33) = 2. So you will see one additional death on average for
every two passengers who switched from female to male.
The second problem is that there is no comparison group. If you assume
that myopia does not spontaneously resolve then it would be safe perhaps
to view the NNT as 1/(0.69-0.00) = 1.45. I think it's fine, but more
than a few people would be uncomfortable with this calculation.
NNH is even trickier. While I am willing to believe that the control
group has a zero level of efficacy in some settings, I am going to be
very skeptical that you won't see some level of harmful things happening
in the control group. It's like the bumper sticker says "(Stuff) happens."
Still there is some value is saying that one in every 77 patients on
average will experience epithelial ingrowth after LASIK.
If I were in your shoes, I would report the NNT and NNH, but would note
that interpretation of these values in a case series is difficult and
these values should be treated with caution.
Steve Simon, [log in to unmask], Standard Disclaimer.
Sign up for the Monthly Mean, the newsletter that
dares to call itself average at www.pmean.com/news
|