[R] Defining reference category for a cph model summary inside of a "for" loop

[Frank E Harrell Jr]

Wells, Brian wrote:
> Frank, 
> 
> Thanks again, I didn't realize that continuous variables could be
> manipulated that way inside of the summary function. 
> 
> I realize that my code was kind of confusing. 
> 
> The variables "A"..."F" are all categorical variables. They each have
> four levels named "1st Quartile"...."4th Quartile"
> 
> I tried the code below with the same result. 
>> print(summary(f, eval(parse(text=paste(i,"='1st Quartile'", sep='')))))
> 
> In the output, the reference category is different for each of the
> variables. 
> 
> Brian 

Thanks for clarifying.  That approach will NOT provide estimates at the 
quartiles.  For example a hazard ratio for the "upper quartile category" 
to the "lower quartile category" will estimate the ratio of hazards when 
X>Q3 to when X<Q1 where outer quartiles are Q1 and Q3.  This represents 
a hazard ratio of an unknown mixture of distributions and will not 
transport to another sample with a different mixture.

In addition you will have serious residual confounding with that 
approach by not adjusting for all the information in continuous predictors.

Frank

> -----Original Message-----
> From: Frank E Harrell Jr [mailto:f.harrell at vanderbilt.edu] 
> Sent: Sunday, March 30, 2008 9:14 AM
> To: Wells, Brian
> Cc: r-help at r-project.org
> Subject: Re: [R] Defining reference category for a cph model summary
> inside of a "for" loop
> 
> Wells, Brian wrote:
>> Dr. Harrell, 
>> Thanks for you help. 
>>
>> I tried:
>>
>>> print(summary(f,parse(text=paste(i,'="1st Quartile"', sep=''))))
>> Same result. No error, the reference category simply doesn't change. 
> 
> That's good, because the default in summary is to compare the outer 
> quartiles for a continuous variable.  And as I said before the string 
> '1st Quartile' has no special meaning for R or Design.
> 
> Get what you are trying to do to work without parse (and you'll need 
> eval() with parse) first.  When you want total control over a setting, 
> say getting a hazard ratio for the .2 to the .8 quantile, do something
> like
> 
> summary(f, age=quantile(age,c(.2,.8),na.rm=TRUE))
> 
> Frank
> 
>> Brian 
>>
>> -----Original Message-----
>> From: Frank E Harrell Jr [mailto:f.harrell at vanderbilt.edu] 
>> Sent: Friday, March 28, 2008 8:34 PM
>> To: Wells, Brian
>> Cc: r-help at r-project.org
>> Subject: Re: [R] Defining reference category for a cph model summary
>> inside of a "for" loop
>>
>> Wells, Brian wrote:
>>> I have the following code. 
>>>
>>>  
>>>
>>>  
>>>
>>>> f <- cph(formula = Surv(TimeToDeath, Dead == "Yes")
>>> ~1,data=single.dat, x=T, y=T, surv=T)
>>>
>>>> for(i in c('A', 'B', 'C', 'D', 'E', 'F')){
>>>> f <-update(f,as.formula(paste('Surv(TimeToDeath, Dead ==
>>> "Yes")~',i,sep='')))
>>>
>>>> print(summary(f, paste(i,"=1st Quartile", sep='')))
>>>  
>>>
>>>  
>>>
>>> There is no error message generated in R, but R ignores the reference
>>> category defined with paste in the summary function for the cph
> model.
>>>  
>>>
>>> The output uses the "1st Quartile" as the reference category to
>>> calculate hazards for some of the variables defined by i, but not all
>> of
>>> them. 
>>
>> Your code is confusing.  What is to the right of ~ in a formula is a 
>> predictor variable name, not a value.  If your variables are named A,
> B,
>> C, ... you are OK.
>>
>> '1st Quartile' has no special meaning to R or Design, and you can't
> pass
>> a character string as a second argument to summary and expect it to
>> work.
>>
>> You will need parse(text=paste(...)) to create an appropriate
>> expression.
>>
>> But Design gives you inter-quartile range hazard ratios by default
>> anyway.
>>
>> Beware of getting hazard ratios that are not adjusted for other 
>> variables needed in the model.
>>
>> Frank Harrell
>>
>>>  
>>>
>>>  
>>>
>>> Any help would be greatly appreciated. 
>>>
>>>  
>>>
>>> thanks
>>>
>>>  
>>>
>>> Brian J. Wells, MD, MS
>>>
>>> Research Associate
>>>
>>> Quantitative Health Sciences
>>>
>>> Cleveland Clinic
>>>
> 
> 


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