Andrew,
Sorry that my use of terms was unclear. Relative risk requires an estimate
of prevalence in the total population (to use your term). Odds ratios can
be calculated using only study population prevalence. Odds ratios
approximate relative risk only if the prevalence in the total population is
very low.
Relative risk can be determined if a sufficiently large sample is taken from
the total population, no matter what the true prevalence is.
Odds ratios were originally calculated when analyzing retrospective data
(case-control studies), whereas relative risk can be calculated from
prospective data (RCTs) if the sample is sufficiently representative of the
population. Since causality cannot be demonstrated with retrospective data,
the OR is a less accurate predictor of future events than is RR. This is
the usual way I distinguish between the two.
I apologize to any reader who feels I've made the destinctions too
simplistic.
______________________________________________
Eric Harvey, PharmD, MBA, BCPS
Child Health Institute, Evidence-Based Medicine Project
146 North Canal Street, Suite 300
Seattle, WA 98103-8652
University of Washington Campus Mailbox: 358853
office: (206) 616-1203
fax: (206) 543-5318
e-mail: [log in to unmask]
____________________________________________
-----Original Message-----
From: Andrew Jull <[log in to unmask]>
To: [log in to unmask]
<[log in to unmask]>
Date: Wednesday, February 24, 1999 6:03 PM
Subject: RE: Odds Ratios vs Relative Risk
>Hi Eric
>
>I'm not sure I understand this - would you not have access to the study's
>population prevalence (ie you need to know both the number of events and
the
>number without events to calculate the odds) when calculating the OR? Or
are
>you referring to knowledge of prevalence in the total population, not just
>the study population? If this former should be the case, then should not
>most studies be using ORs?
>
>regards
>Andrew
>
>> -----Original Message-----
>> From: Eric Harvey [SMTP:[log in to unmask]]
>> Sent: Thursday, 25 February 1999 09:29
>> To: [log in to unmask]
>> Subject: Re: Odds Ratios vs Relative Risk
>>
>> As I understand it, RR can only be used when the population prevalence of
>> the measured outcome is known. In many cases, population prevalence is
>> not
>> known, so OR can be properly used to approximate RR when quantifying rare
>> events. As events become more common, OR no longer accurately
approximate
>> RR. In these cases, OR will overestimate the benefits and harms of
>> treatment (RR).
>>
>> See:
>> Sackett DL. Down with odds ratios! Evidence-Based Medicine. 1996
>> Sept/Oct;1:(got from ACP website so I don't have page numbers).
>>
>> -----Original Message-----
>> From: Andrew Jull <[log in to unmask]>
>> To: [log in to unmask]
>> <[log in to unmask]>
>> Date: Wednesday, February 24, 1999 11:41 AM
>> Subject: Odds Ratios vs Relative Risk
>>
>>
>> >Dear All
>> >
>> >I was recently conversing with a colleague and the question of why use
>> ORs
>> >instead RRs came up. My naive response was that the choice seemed to be
>> >based on the individual's preference and that I had not read anything
>> that
>> >suggested the use of one was more informative than another (and indeed
>> have
>> >read some material that suggests ORs are misleading when the OR is
high -
>> >but I don't want to get onto that issue).
>> >
>> >Can anyone help me with why odds ratios might be used in preference to
>> >relative risk or vice versa.
>> >
>> >regards
>> >Andrew Jull
>> >Clinical Nurse Consultant
>> >Auckland Hospital
>> >NEW ZEALAND
>> >
>
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|