Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: predicting survival with a semiparameteric model

From   Maarten buis <>
Subject   Re: st: predicting survival with a semiparameteric model
Date   Tue, 24 Aug 2010 19:02:31 +0000 (GMT)

--- On Tue, 24/8/10, Feiveson, Alan H. (JSC-SK311)  wrote:
> Of course, I can always get a standard error of S(t) with a
> fully parametric model using -streg-, but if possible, I'd
> like to use a PH model without having to specify a
> distribution.

The key issue with specifying the distribution that -stcox- 
tries to solve is that you don't have to specify the baseline
hazard rate. This is a strength and a weakness, you make the
model more robust, but you limit what you can say about the
distribution, as you find out.

A solution is to choose a parametric model, and add a 
parametric form for the baseline, but keep it flexible:

A common method is the piecewise constant model, but you can
go further and model it with (restricted) cubic splines, like

*------------- begin example --------------
sysuse cancer, clear
gen int id = _n
stset studytime, failure(died) id(id)
stsplit time, every(1)
mkspline sp = time, cubic knots(5 10 20 30)
streg i.drug age sp*, dist(exp) hr
*-------------- end example ---------------

I believe Paul Lambert and Patrick Royston has done and published 
stuff like this, see:

P. C. Lambert and P. Royston (2009) Further development of flexible 
parametric models for survival. The Stata Journal, 9(2):265--290.

Hope this helps,

Maarten L. Buis
Institut fuer Soziologie
Universitaet Tuebingen
Wilhelmstrasse 36
72074 Tuebingen


*   For searches and help try:

© Copyright 1996–2017 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index