RE: st: RE: Poisson Regression

["Visintainer, Paul" <Paul.Visintainer@baystatehealth.org>]

It is an issue of interpretability -- and I wouldn't think that clinicians would find the issue trivial.  In their response to letters following publication of "What's the relative risk?" (Zhang & Yu, 1998.  What's the Relative Risk, JAMA, Vol 280, No 19, pp 1690-1691), Zhang & Wu intimate that clinicians are likely to sacrifice a little precision (and perhaps validity?) for something that is simple and interpretable.  

My frustration is that when the outcome is common and logistic regression is used, there virtually no discussion of clinical relevance -- only statistical significance, (e.g., is a significant odds ratio of 2.5 clinically relevant?  Perhaps if the base risk is 2%; perhaps not if the base risk 73%.  This is what underlies the popularity of the conversion approach -- it is easy and "close enough".) 

Using -margins- is another good suggestion.  However, not everyone writing for clinical journals uses Stata (they should be, perhaps) or is aware of the approach, even if they have the latest versions, etc.  

The log-binomial and Poisson approaches meet the two requirements for most clinicians doing their own analyses: they're easy to run (as easy as logistic regression) and they provide interpretable output.  I like them because they provide valid point estimates and confidence intervals.  Seeing more of these approaches in the literature can only help.

Regards,

-p

________________________________________________
Paul F. Visintainer, PhD
Baystate Medical Center
Springfield, MA 01199


-----Original Message-----
From: owner-statalist@hsphsun2.harvard.edu [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of Brendan Halpin
Sent: Monday, February 14, 2011 2:37 PM
To: statalist@hsphsun2.harvard.edu
Subject: Re: st: RE: Poisson Regression

On Mon, Feb 14 2011, Visintainer, Paul wrote:

> There is a growing literature on alternatives to logistic regression if
> the outcome is common. I've attached some of the literature below. Just
> a quick overview:

[...]

> Spiegelman & Hertzmark (2005) in a commentary go as far as to recommend
> that logistic regression not be used for risk or prevalence ratios when
> the outcome is common.

Why?