Dear List
I am doing Cox-regression with a continuous exposure, x
I would like to fit a crude model with x modelled as a restricted
cubic spline. In Stata 10, I do
mkspline rcx = x, cubic nknots(7)
/* crude */
stcox rcx*
Now, I would like to draw the hazard ratio as a function of x with one
value of x as reference level. Also, I want 95% confidence limits.
For the crude model, I could try
predict hr
predict sd, stdp
But, despite the label given to "hr", this is not the relative hazard,
but rather the hazard function. I need to divide by the value at the
reference level of x, which I call xref.
replace hr=hr/xref
and a novel attempt at calculating 95% confidence limits would be
gen ll = exp(log(hr) - 2*sd)
gen ul = exp(log(hr) + 2*sd)
line hr x, sort
However, in order to draw 95% confidence limits, I have to "adjust" sd
accordingly so that the width of the confidence interval is zero at
the reference value. The plot should look like a "butterfly". Is there
a way of obtaining this?
I have tried replacing x with x - xref, but the predicted value of the
"hazard ratio" does not become 1 as expected at the value xref. I
tried looking at the mvrs package, but the fracplot does not give me a
"butterfly".
Any help is highly appreciated,
Claus Dethlefsen
Center for Cardiovascular Research
Aalborg Hospital, Aarhus University Hospital
Denmark
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
