As I argued it at this group before and as we described it elsewhere (for
details see Cancer Control 1998;5:394-405; also at
http://www.moffitt.usf.edu/pubs/ccj/v5n5/article2.html  ,  Med Decision
Making 1998;18:464
; Comp Biomed Res; or, see
http://www.hsc.usf.edu/~bdjulbeg/Programs/BR/br-java.htm  ), normatively
highest (maximum) NNT at which treatment is worth administering in
prophylactic setting is equal to:
		NNT<NNH or NNT<1/ R   
where R is risk of the treatment .
For example, for a toxicity of 2%, the treatment NNT has to be at most 50 to
be worth administering the treatment, for a toxicity of 4%, the NNT has to
be at most 25, etc.

If this condition is not satisfied, no NNT is "good NNT".

In the setting of selection of Rx1 vs Rx2, in order to even consider the
treatment Rx1 as opposed to the alternative treatment Rx2, the following
inequality must be satisfied:

NNT1 <= 1/(R1 - R2 + 1/NNT2)


This is not only normatively correct, but also makes sense intuitively: we
should not consider measures of treatment benefits without consideration of
other side of therapeutic coin (that is, treatment harm).Again, as I wrote
before, it is important to realize that same units of benefits and harms are
used when these calculations are performed.

I am not aware of any normative model that successfully integrated patients
values with NNT, NNH or with any other EBM summary measures of therapeutic
effects (although, I should say that we are currently working on one such
model)

Intertwined with this issue is the question of "action threshold". I will
try to illustrate it from perspective of ongoing discussion of CVD risk and
New Zealand tables for CVD. The tables use data from Framingham Heart study.
These tables suggest threshold for therapeutic action if risk of CVD is
10-15% at 5yrs (NNT=25 for 5 years) (see:
http://cebm.jr2.ox.ac.uk/docs/prognosis.html ) However, this treatment
threshold appears to be quite arbitrary. As I argued before, normatively
this action threshold (AT) is equal to 

AT= NNT/NNH or AT=risk of treatment*NNT=treatment
risk/(RRR*morbidity/mortality without Rx) (for details see references
provided above). 

We should treat if estimated risk of clinical event (in this case, CVD) is
above AT; conversely, we should withhold Rx if risk of CVD is below AT. For
example, package insert for pravastatin indicated that at median treatment
of 4.8 years, 0.10% of patients treated with pravastatin developed elevation
of AST as opposed to 0.03% of placebo treated patients. This would amount
roughly to NNH=1428. Using the action threshold equation this means that we
can consider treatment with statins if risk of CVD exceeds >1.7% at 5 years
(AT=25/1428). Even when I vary risk in sensitivity analysis, I still obtain
threshold for action, which is below recommended 2% or 3% per year. It is
not clear to me how current recommendations (for treatment actions) have
been derived. Can somebody care to comment? (Cardiovascular medicine is not
my field, and I am perhaps missing something here).

 (Incidentally, I developed a simple BASIC program that integrates this
threshold model with Framingham risk equations. If anybody is interested in
the program, I will be happy to send it to him/her.)



hope this clarifies this issue

ben d

Benjamin Djulbegovic, MD
Associate Professor of Medicine
H. Lee Moffitt Cancer Center & Research Institute
at the University of South Florida
Division of Blood and Bone Marrow Transplant
12902 Magnolia Drive
Tampa, FL 33612

e-mail:[log in to unmask]
http://www.hsc.usf.edu/~bdjulbeg/
phone:(813)979-7202
fax:(813)979-3071

> -----Original Message-----
> From:	Takeo Saio [SMTP:[log in to unmask]]
> Sent:	Friday, March 05, 1999 5:32 PM
> To:	Amit Ghosh
> Cc:	[log in to unmask]
> Subject:	RE: What is a good number for NNT?
> 
> Hello! That's very good question.
> 
> I think the NNTs of 2-4 are for therapeutic usage of drugs,and the NNTs
>  over 20 are for preventitive usage of drugs.
> i.e. for acute condition,NNTs of 2-4:for chronic condition,NNTs of over
> 20.
> 
> 
> > A question I have , is what is a good number for NNT? Bandolier 12
> >indicated that a NNT of 2-4 is suggestive of good NNT  ( i.e., ARR of
> >0.25-0.50). However most the the RCT reported in journal s NEJM, Annals
> >etc have statistical significance however their NNT are usually over 20,
> >ie, finasteride for BPH( NNT= 30). I reckon one of the disadvantages of
> >NNT maybe that despite the NNT being over 20 , these patients may have
> >significant improvement in quality of life and other subjective issues.
> >Or is the fact that NNT of 2-4 an over enthusiastic expectation when it
> >comes to a clinical response.I guess in the case of chronic diseases
> >with outcomes occurring over decades a NNT of 50 would be acceptable,
> >however in other more acute conditions one would consider a smaller
> >number for  NNT as desirable. Could the members throw some  additional
> >light in the defining what is a good number for NNT.
> **********************************************
> Saio,Takeo
> ///////////////CHIAKI hospital,JAPAN/////////
> e-mail;[log in to unmask]
> **********************************************
> 
> 
> 


%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%