Hello all, This may be a naive question but since I'm new to R/survival models, I cannot figure it out the problem myself. I have a coxph model for my data and I am trying to test if the proportional hazards assumption holds. Using cox.zph on the model I get a global score: GLOBAL NA 4.20e+02 0.00e+00 Does this mean that the proportional hazard assumption does not hold? When I plot the Schoenfeld residuals, generally the plots are across the horizontal line which makes me think that the proportional hazards assumption still holds. Could someone please clarify on this? A somewhat unrelated question: I have come across several papers which just calculate the coxph model without the diagnostics for proportional hazards assumption and interpret the results of the regression directly. Should that be acceptable? Are there other ways to show the goodness of the model? Thanks! [[alternative HTML version deleted]]
Although someone on this list may respond, AFAICS this does not seem to be an R question for R-help.I would suggest that you spend some time with a local statistician. A general observation: Statistical model assumptions neither :hold" nor "don't hold." Quoting George Box, "All models are wrong, but some are useful." Determining whether a coxph regression model is useful for your data and problem is something that I think only you and someone with statistical experience and insight into these models can decide. I do not understand how this can be done remotely. Cheers, Bert On Sat, Aug 10, 2013 at 9:14 PM, Soumitro Dey <soumitrodey1 at gmail.com> wrote:> Hello all, > > This may be a naive question but since I'm new to R/survival models, I > cannot figure it out the problem myself. > > I have a coxph model for my data and I am trying to test if the > proportional hazards assumption holds. Using cox.zph on the model I get a > global score: > > GLOBAL NA 4.20e+02 0.00e+00 > > Does this mean that the proportional hazard assumption does not hold? > When I plot the Schoenfeld residuals, generally the plots are across > the horizontal line which makes me think that the proportional hazards > assumption still holds. Could someone please clarify on this? > > A somewhat unrelated question: I have come across several papers which > just calculate the coxph model without the diagnostics for > proportional hazards assumption and interpret the results of the > regression directly. Should that be acceptable? Are there other ways > to show the goodness of the model? > > Thanks! > > [[alternative HTML version deleted]] > > ______________________________________________ > R-help at r-project.org mailing list > https://stat.ethz.ch/mailman/listinfo/r-help > PLEASE do read the posting guide http://www.R-project.org/posting-guide.html > and provide commented, minimal, self-contained, reproducible code.-- Bert Gunter Genentech Nonclinical Biostatistics Internal Contact Info: Phone: 467-7374 Website: http://pharmadevelopment.roche.com/index/pdb/pdb-functional-groups/pdb-biostatistics/pdb-ncb-home.htm
On 08/11/2013 06:14 AM, Soumitro Dey wrote:> Hello all, > > This may be a naive question but since I'm new to R/survival models, I > cannot figure it out the problem myself. > > I have a coxph model for my data and I am trying to test if the > proportional hazards assumption holds. Using cox.zph on the model I get a > global score: > > GLOBAL NA 4.20e+02 0.00e+00 > > Does this mean that the proportional hazard assumption does not hold?Yes, or, the fit is very bad (see Bert's response).> When I plot the Schoenfeld residuals, generally the plots are across > the horizontal line which makes me think that the proportional hazards > assumption still holds. Could someone please clarify on this?Did you try > plot(cox.zph(fit)) Read the help pages for cox.zph and plot.cox.zph. The raw Schoenfeld residuals plots are generally of limited value. With factor covariate(s) you could also perform a graphical inspection by plotting the estimated cumulative hazards, stratifying on these factors.> > A somewhat unrelated question: I have come across several papers which > just calculate the coxph model without the diagnostics for > proportional hazards assumption and interpret the results of the > regression directly. Should that be acceptable?No.> Are there other ways > to show the goodness of the model?Yes, see above. G?ran Thanks!> > [[alternative HTML version deleted]] > > ______________________________________________ > R-help at r-project.org mailing list > https://stat.ethz.ch/mailman/listinfo/r-help > PLEASE do read the posting guide http://www.R-project.org/posting-guide.html > and provide commented, minimal, self-contained, reproducible code. >
That's the primary reason for the plot: so that you can look and think.
The test statistic is based on whether a LS line fit to the plot has zero slope.
For
larger data sets you can sometimes have a "significant" p-value but
good agreement with
proportional hazards. It's much like an example from Lincoln Moses'
begining statistics
book (now out of print, so rephrasing from memory).
"Suppose that you flip a coin 10,000 times and get 5101 heads. What
can you say?
a. The coin is not perfectly fair (p<.05). b. But it is darn close
to perfect! "
As a referee I would be comfortable using that coin to start a football game.
The Cox model gives an average hazard ratio, averaged over time. When
proportional
hazards holds that value is a complete summary-- nothing else is needed. When
it does
not hold, the average may still be useful, or not, depending on the degree of
change over
time.
Terry Therneau
On 08/13/2013 05:00 AM, r-help-request at r-project.org
wrote:> Thanks to Bert and G?ran for your responses.
>
> To answer G?ran's comment, yes I did plot the Schoenfeld residuals
using
> plot.cox.zph and the lines look horizontal (slope = 0) to me, which makes
> me think that it contradicts the results of cox.zph.
>
> What alternatives do I have if I assume proportional assumption of coxph
> does not hold?
>
> Thanks!