I want to use an opportunity provided by Jeanne's question to highlight one
aspect of contrasting of treatment benefits and harms, which, at least in my
opinion, is not sufficiently appreciated. That relates to the fact that to
justify the administration of a treatment associated with more than one
adverse event, the benefit of a given treatment aimed at preventing a single
disease (bad) outcome, has to be larger than the AGGREGATE of all adverse
events (adjusted for a patient's values and the probability of the disease).
(please see details at
http://www.medscape.com/Medscape/GeneralMedicine/journal/2000/v02.n01/mgm011
3.djul/mgm0113.djul-01.html if you are interested in learning where this
statement came from).
regards
ben djulbegovic
> -----Original Message-----
> From: Guthrie, Dr Bruce [SMTP:[log in to unmask]]
> Sent: Wednesday, August 23, 2000 12:00 PM
> To: [log in to unmask]
> Subject: Re: When treatment harms
>
> Date sent: Tue, 22 Aug 2000 03:47:22 -0400
> Subject: When treatment harms
> From: Jeanne Lenzer <[log in to unmask]>
> To: "EBM (E-mail)"
> <[log in to unmask]>
> Send reply to: Jeanne Lenzer <[log in to unmask]>
>
> > Just an uneducated question, re: a listmember's query:
> > (snip) - "I am getting some negative RRR [with CAT snipper]. What
> > would be their interpretation?. I know that RRR = 0 is null effect
> > and RRR= 1 is cure."
> >
> > Here's my question: If - as must be presumed by the range for RRR of
> > 0-1 - the term RRR assesses only efficacy of intervention and not
> > risks of intervention - is there a number that incorporates risk of
> > treatment? (In other words a more global assessment of intervention
> > that would allow a negative outcome)?
> >
> > Without such a global value, how do we measure the many thrombolytic
> > trials in stroke, for example, that show more deaths with treatment
> > intervention than without? I am increasingly concerned by the ways
> > sponsored researchers do "spin control" and it seems to me that RRR is
> > one such term that can be so abused if the reader is not alert. For
> > example, an intervention that reduces a primary endpoint (fatal MI in
> > the pt with diabetes) but increases total death rate (from
> > pancreatitis/hepatitis whatever - [shades of Rezulin]) the outcome may
> > be spun as positive through a set of maneuvers from a.) referring to a
> > positive RRR (correctly) and then b.) combining endpoints (death and
> > disability) to a more neutral - or even positive effect (as has been
> > done in the thrombolytic trials).
> >
> > Have I misunderstood something here?
>
> No, I don't think so but I think there are two separate issues.
>
> First, manipulation to show "desired" effects (usually that a
> treatment is effective). Any of the ways of presenting treatment
> effects (relative risk RR, relative risk reduction RRR, numbers
> needed to treat NNT etc) can be manipulated like this, although
> some more easily than others (eg big RRRs presented for rare
> outcomes where NNTs are high). The main outcomes I pay
> attention to are those defined as primary and secondary endpoints
> in the methods section. If these (and other endpoints) are
> combined in peculiar ways to produce statistically significant
> results then it makes me suspicious that there's been some data
> dredging to find the "right" combination because the main
> endpoints don't show benefit.
>
> Second, I don't think there's a problem in having a negative relative
> risk reduction. It means treatment is worse than control for that
> endpoint (assuming that it's statistically significant/the confidence
> intervals don't cross zero). Take away the negative and call it a
> relative risk increase of treatment. It's usually easier to see what's
> happening with numbers need to treat/harm though.
>
> Bruce
>
>
>
> Bruce Guthrie,
> MRC Training Fellow in Health Services Research,
> Department of General Practice,
> University of Edinburgh,
> 20 West Richmond Street,
> Edinburgh EH8 9DX
> Tel 0131 650 9237
> e-mail [log in to unmask]
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|