i am resending the massage, was rejected.
Please find the summary below.
Thanks.
*******************************************************
I often see this in papers I referee. The solution is
to change the units in which you report this variable.
As a simple example if age was of prognostic
importance and you reported age in days, it
would give hazard ratios close to 1. If you reported
in years or even decades, you would obtain the type of
discrimination you would like to obtain, and
confidence intervals would become interpretable.
*************************************
I would say that the effect of that explanatory
variable, though statistically significant, is
negligible. With a sample size large enough any small
effect would be statistically significant but this
does not mean that the effect is important.
*******************************************************
That is a very interesting situation.
I think this is a clear situation in which the
interpretation can never been done out of the contest.
Hence I wouldn't focus too much attention on the
statistical significance as I can guess that the
significance is much due to the sample size than to
the importance of the difference in log hazard between
the two groups. Thus, if the event of interest is
very rare, I would be tempted to conclude that the two
groups differ in risk of event and that the hazard in
one group is 1 per 10.000 higher/lower than in the
other. But is the event is relatively frequent, I
would conclude that the hazard are comparable between
the two groups.
I would be happy to hear what expert have suggested.
*****************************************************
Without seeing your data or knowing what software you
use one cannot be sure, but it is important to
remember that the coefficients in the case
of a continuous predictor are log ratios per unit
change in the predictor. So if it has a range of 1000
a value of 0.9999 may be different from 1. What is the
standard error of the values?
***********************************************
It may be a matter of scale and rounding. Try dividing
your continuous variable by 1000 or more and redoing
the analysis.
********************end******************************
> How can one interprete a case, when relative risk
> (hazard ratio) estimate of a continuous explanatory
> variable in a multivariate Cox PH model being
> approximately equal to 1 i.e 0.9999 or 1.0001 etc
> given that the p-value shows significance i.e <
0.05.
> Also, the relative risk estimate for this variable
is
> approximately equal to the lower limit for its 95%
CI
> for relative risk.
>
__________________________________________________
Do you Yahoo!?
Yahoo! Platinum - Watch CBS' NCAA March Madness, live on your desktop!
http://platinum.yahoo.com
|