we seem to forget that ARR and NNT express the same thing - conceptually,
there is no difference between them, but how individuals perceive them seems
to matter a lot. steve simon's last mail exemplifies this well when he says:
>I prefer to see the NNT because it is usually a nice round number.<
Actually, it is a round number because we round it by convention. Suppose
the ARR in a trial was 11%, the inverse is 9.09 BUT BY CONVENTION we round
it to 9.
He also says, >It's also easier to do division with NNTs and NNHs< BUT only
because they have been rounded!
>Small ARRs are notoriously hard to interpret.< BUT this is not so if you
reflect that a small ARR means very little effect - reflected by the large
NNT.
there is no real contest between ARR and NNT - whether you say >you are 11%
less likely to suffer X if you do Y< or you say >9 people would have to take
Y to avoid X< you are saying the same thing. what should matter is which
gives us and patients the best handle on a choice. i would like to ask:
1. has anyone seen any evidence that NNT is superior to ARR in talking to
patients?
2. is there not something to be gained by mentioning both ARR and NNT
amongst clinicians so that we avoid the mathematical difficulties thrown up
when a wide confidence interval of ARR crosses zero?
Kevork Hopayian, Leiston, Suffolk
----------
>From: "Simon, Steve, PhD" <[log in to unmask]>
>To: [log in to unmask]
>Subject: NNTs versus ARRs
>Date: Fri, May 21, 1999, 9:06 pm
>
>I've enjoyed the discussion so far about Numbers Needed to Treat (NNTs) and
>Absolute Risk Reductions (ARRs).
>
>I prefer to see the NNT because it is usually a nice round number. For
>example, "treat 9 people to see one success on average" is a whole lot
>easier to say than "we see an additional 0.11 successes with the new therapy
>on average."
>
>It's also easier to do division with NNTs and NNHs. Suppose we see an
>additional .037 side effects, we have an NNH of 27. This implies that we
>have 3 (=27/9) successes for each side effect. Try doing that same division
>using 0.11 and 0.037.
>
>You could simplify the division by rounding, but it is not always clear how
>much you can safely round an ARR. It's usually pretty clear how much you can
>safely round an NNT.
>
>Small ARRs are notoriously hard to interpret. Suppose a screening test
>detected additional cases of disease with probability 0.00235. Isn't it a
>whole lot easier to say that we have to screen 426 people to detect one
>additional disease on average? Also those very small numbers make it easy to
>accidentally drop or add a decimal. 0.00235 and 0.000235 look very similar,
>but 426 and 4,260 look very dissimilar.
>
>Using percentages or rates per 10,000 can sometimes help with these small
>probabilities, but even then I still usually prefer the NNT.
>
>I'm sure there are examples where the ARR is a nice round number like 0.6
>and the NNT is an awkward 1.7. But in general, the NNT leads to simpler
>numbers that are easier to manipulate in your head.
>
>Steve Simon, [log in to unmask], Standard Disclaimer.
>How to Read a Medical Journal Article: http://www.cmh.edu/stats/journal.htm
>
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|