[R] Function call error in cph/survest (package Design)

[Frank E Harrell Jr]

Dieter Menne wrote:
> 
> Dieter Menne wrote:
>> I do not know if this a problem with me, my data or cph/survest in package
>> design. The example below works with a standard data set, but not with my
>> data, but I cannot locate the problem.
>>
> 
> Dieter Menne found out after hours, that in one case and explicit cast is
> required, but none in the other, and has no idea why.
> 
> #---------------------------------------
> library(Design)
> data(ovarian)
> dd = datadist(ovarian)
> options(datadist="dd")
> ovarian$rx = as.factor(ovarian$rx)
> cp = cph(Surv(futime,fustat)~rx,data=ovarian,surv=TRUE,x=TRUE,y=TRUE)
> summary(cp)
> survplot(cp,rx=NA)
> # works ok, without casting to a data frame
> survest(cp,newdata=levels(ovarian$rx),what="survival",times=500)

It's just an accident that it ever worked without making it a data frame.

Frank

> 
> # Small data set, 223 rows, 3650 bytes
> cc = read.table("http://www.menne-biomed.de/uni/cc.csv",header=TRUE)
> #cc = read.table("cc.csv",header=TRUE)
> dd = datadist(cc)
> cp = cph(Surv(DaysToEvent,event)~ITTGroup,data=cc,surv=TRUE,x=TRUE,y=TRUE)
> survplot(cp,ITTGroup=NA)
> summary(cp)
> 
> survest(cp,newdata=levels(cc$ITTGroup),times=200)
> #Error in survfit.cph(list(coefficients = c(0.435291247251185,
> -0.143015493753166 :
> #  NA/NaN/Inf in foreign function call (arg 13)
> #In addition: Warning message:
> #NAs introduced by coercion
> 
> # Works when explicitly converted to a data frame
> survest(cp,newdata=data.frame(ITTGroup=levels(cc$ITTGroup)),times=200)
> 
>  
> 


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