Date:         Tue, 19 Sep 2006 10:41:12 +0200
Reply-To:     Marta García-Granero
              <biostatistics@terra.es>
Sender:       "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From:         Marta García-Granero
              <biostatistics@terra.es>
Organization: Asesoría Bioestadística
Subject:      Re: Correlation matrix of regression coefficients - Cox
              regression analysis
In-Reply-To:  <OF275A65E3.8C582A3B-ON852571ED.0073383E-852571ED.007349C7@hjf.org>
Content-Type: text/plain; charset=ISO-8859-1

Hi Christian

Now I understand... I think the authors have the concept wrong, they
meant to talk about the correlation matrix of predictors (not the
correlation matrix of regression coefficients). At least, the
phenomenon they discuss (..."higher proportion of young HCC patients
were male"...) is simply an association (correlation) between both
predictors (age and sex). They wrongly call that "an interaction",
when they should discuss it in terms of "multicollinearity".

Interaction in this context (Cox regression model) would mean that the
effect of age on survival of HCC patients depends on gender (and vice
versa): the HR of age would be different for males and females. Here
they are simply talking about association between predictors, and
calling it - wrongly - interaction of predictors. As a matter of fact,
those associations they talk about could be related to confounding,
and the discussion of the paper is foccused mainly on confounding
(although, since they don't know the concept, they don't call it that
way).

It astounds me that the reviewers of the paper didn't correct it
before printing (I bet that none were statisticians nor
epidemiologists...). I'm going to keep this paper for my "gallery of
horrors", and I'll discuss it with my students this year.

I hope it's all clear now, but if you need more help, I can be reached
here (at the list), I check my mail quite often (as a matter of fact,
so often that my husband thinks I'm hooked on this list...)

Monday, September 18, 2006, 10:59:10 PM, You wrote:

cho> See page 3, section

cho> Interaction Between Age, Gender and HBsAg

cho> In our previous study, we found that on analyzing the correlation matrix of
cho> regression coefficients in the multivariate
cho> analysis, there were interactions between HBsAg and gender (r=-0.205,
cho> P<0.001) and age (r=0.166, P<0.001).[6] As
cho> shown in Table 4 , a higher proportion of young HCC patients was male
cho> gender. 83.7% of HCC patients below 40
cho> years of age were male, while 79.1% of HCC patients older than 40 years of
cho> age were male (P < 0.001). In addition,
cho> young HCC patients were more likely to be HBV carriers. 89.8% of young HCC
cho> patients had positive HBsAg. In
cho> contrast, only 59.1% of HCC patients older than 40 years of age had
cho> positive HBsAg (P < 0.001).


--
Regards,
Dr. Marta García-Granero,PhD           mailto:biostatistics@terra.es
Statistician

---
"It is unwise to use a statistical procedure whose use one does
not understand. SPSS syntax guide cannot supply this knowledge, and it
is certainly no substitute for the basic understanding of statistics
and statistical thinking that is essential for the wise choice of
methods and the correct interpretation of their results".

(Adapted from WinPepi manual - I'm sure Joe Abrahmson will not mind)