|Date: ||Thu, 15 Jul 2004 22:42:55 -0400|
|Reply-To: ||"Edmund J. Bini, M.D., M.P.H." <email@example.com>|
|Sender: ||"SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>|
|From: ||"Edmund J. Bini, M.D., M.P.H." <firstname.lastname@example.org>|
|Subject: ||Re: Odds Ratios and Proportions|
|Content-Type: ||text/plain; charset="us-ascii"|
Thanks Paul. I understand the part about OR overestimating RR when the
outcome is common. Logistic regression is often used even if the outcome is
common in order to adjust for potential confounders but people have to
accept odds ratios for what they are - odds ratios - and not try to convert
them to a RR. Do others feel the logistic regression should not be used if
the outcome is > 10%? The part I was confused about is when they asked us to
convert the OR from logistic regression to a proportion. Maybe I am reading
too deep into this and they just want us to abandon the use of logistic
regression and just present out data unadjusted as percentages.
From: email@example.com [mailto:firstname.lastname@example.org]
Sent: Thursday, July 15, 2004 10:29 PM
To: Edmund J. Bini, M.D., M.P.H.
Subject: Re: Odds Ratios and Proportions
Hi there Edmund
'However, as both the prevalence and OR increase, the error in the
quickly becomes unacceptable: if the baseline prevalence is 10% an OR of 4
equivalent to a RR of 3, a discrepancy of 25%'
> Edmund J. Bini, M.D., M.P.H. <email@example.com> wrote:
> I sent this last week but am not sure if it was delivered. I apologize
> if it
> is a duplicate e-mail. Can someone please help me address the issue
> using SPSS?
> I recently submitted a paper to a journal describing a randomized
> study we did to increase compliance with colon cancer screening and
> have a
> statistical question about one of the comments we received when the
> was reviewed. We randomly allocated 788 patients to receive an
> or standard of care and looked at compliance with screening at 6 months
> after the intervention. By 6 months, 65.9% of patients in the
> group completed the screening test compared with 51.3% in the standard
> care group (p < 0.001). We then did multivariable logistic regression
> determine if our intervention was significantly associated with
> after adjusting for several potential confounding variables. The
> relative odds of compliance was 1.9 (95% CI, 1.4 - 2.6) comparing the
> intervention to the standard of care group.
> One of the associate editors of the journal made the statement below
> and I
> am interested in your comments about this as well as any information
> how to convert odds ratios from logistic regression to proportions in
> Is this possible or logical? Any help or link to a reference about this
> would be very much appreciated.
> Given that your outcome of interest was relatively common, we are
> that this may
> violate the assumptions of logistic regression. Specifically, in order
> avoid overstating the prevalence ratio (which the odds ratio is
> intended to
> approximate), the outcome of interest
> needs to occur <=10% of the time. It is more accurate to convert the
> ratios into proportions when the outcome is more frequent than that.