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: Re: st: Parametric survival analysis with competing risks

From   "Seed, Paul" <>
To   "" <>
Subject   RE: Re: st: Parametric survival analysis with competing risks
Date   Thu, 26 Apr 2012 15:41:14 +0100

Dear Roger, 
Thanks for the interest.

What I want is an algorithm to give me the probability of delivery for any women before any gestation up to 37 weeks (in particular before 30, 34, 37 weeks, and within 2 & 4 weeks of the test), based on the results of two clinical tests conducted at a known gestation.  Data are censored at 37 weeks, as delivery after then is not clinically important. The probability should be derived from a formula for the Hazard that can be generally applied; so Cox's regression is out.

Numbers would be available to obstetricians, midwives and women to help decisions about clinical care (& holiday planning for the women).  

The problem is complicated by the two competing risks of spontaneous onset and iatrogenic onset (mainly elective C/S and induced labour). As you imply, the way the analysis is done affects the meaning of the answer obtained; and I am still getting to grips with the implications.  It is, I think easier to validate a competing risk model, as all events are explicitly included, and you are describing what actually happens.  Of course like any model it is only applicable if circumstances (in particular the pressure for iatrogenic delivery) remain the same.

In fact, since posting, I have read around & done some web searches & realised that survival analysis with competing risks is not so different from ordinary survival analysis, which answers my immediate problem .  The same regression methods and software can apparently be used.  It is only necessary to treat the subjects with competing events (women undergoing induction or C/S) as censored at the end of the study period.  Separate analyses are needed for each of the different risks; and the individual probabilities and hazards can then be summed.  

Best Wishes, 

> Date: Wed, 25 Apr 2012 20:57:50 +0100
> From: "Roger B. Newson" <>
> Subject: Re: st: Parametric survival analysis with competing risks
> What exactly do you mean by "The goal is the prediction of spontaneous
> preterm labour"? Did you want relative hazard rates for spontaneous
> pre-term labour, or a predictive score for spontaneous pre-term labour,
> or an estimate for the probability of spontaneous pre-term labour that
> you can quote to pregnant women, assuming that they do not opt for
> induction of labour or surgery? Or something else?
> Best wishes
> Roger
> Roger B Newson BSc MSc DPhil
> Lecturer in Medical Statistics
> Respiratory Epidemiology and Public Health Group
> National Heart and Lung Institute
> Imperial College London
> Royal Brompton Campus
> Room 33, Emmanuel Kaye Building
> 1B Manresa Road
> London SW3 6LR
> Tel: +44 (0)20 7352 8121 ext 3381
> Fax: +44 (0)20 7351 8322
> Email:
> Web page:
> Departmental Web page:
> pgenetics/reph/
> Opinions expressed are those of the author, not of the institution.
> On 25/04/2012 13:04, Paul Seed wrote:
> > Dear Statalist,
> >
> > The -st- suite contains two commands for parametric survival analysis
> > regression;
> > and for survival analysis with competing risks (using a Cox's model):
> > -streg , model()-, and -stcrreg- respectively.
> > Is there any way of combining the two features?
> >
> > I need a parametric method as I want to put the formulae into an
> > prediction probability calculator, and the problem really does
> > have competing risks.  The goal is the prediction of spontaneous
> > preterm labour, with Induction of labour or surgery as competing
> risks.
> >
> > As far as I can see, -findit competing risk- does not produce
> > any user-written command that does this, and I can see nothing
> relevant
> > in Statalist archives.
> >
> > Alternatively, is there a better way of allowing for competing risks
> in a
> > parametric model than by simply treating subjects as censored at the
> time of
> >
> > the competing event?
> >
> > Best wishes,
> >
> >
> > Paul T Seed
> > Division of Women's Health, King's College London
> > Women's Health Academic Centre KHP
> > 020 7188 3642,

*   For searches and help try:

© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index