s-news
[Top] [All Lists]

Re: Estimating median survival from Cox model including categorical ind

To: "Hunsicker, Lawrence" <lawrence-hunsicker@uiowa.edu>
Subject: Re: Estimating median survival from Cox model including categorical independents
From: Frank E Harrell Jr <f.harrell@vanderbilt.edu>
Date: Tue, 11 Dec 2007 11:51:15 -0600
Cc: s-news@lists.biostat.wustl.edu
In-reply-to: <825A13A34899AB40A5039B9E1B71939F01DE2E04@HC-MAIL12.healthcare.uiowa.edu>
References: <825A13A34899AB40A5039B9E1B71939F01DE2E04@HC-MAIL12.healthcare.uiowa.edu>
User-agent: Thunderbird 2.0.0.6 (X11/20071022)
Hunsicker, Lawrence wrote:


Greetings, all:

I would like to estimate predicted survival for patients adjusted for specific values for two independent variables, but using “ average ” characteristics for all of the other independent predictors. I know that I can use the survfit( ) function to do this on the fitted Cox model , assuming that I enter the new data using the newdata=list() parameter. The issue is how to enter data for the “ average ” patient when the covariate is categorical. I suppose that I could do this entering the fraction of patients at each level of the categorical variable , after determining the parameterization by looking at contrasts(variable). But is there an easier way that uses the mean values that must have be en used by the survfit function to generate the values for the function when newdata is not given ? These values must be recorded somewhere.

Many thanks in advance for any help that you can give me.

Larry Hunsicker


Take a look at the gendata function in the Design package, in conjunction with predict.cph or predict.psm.

Frank

--
Frank E Harrell Jr   Professor and Chair           School of Medicine
                     Department of Biostatistics   Vanderbilt University

<Prev in Thread] Current Thread [Next in Thread>