R help - Nov 2012 - R: net reclassification index after Cox survival analysis

Dear all,

I am interested to evaluate reclassification using net
reclassification improvement and Integrated Discrimination Index IDI after
survival analysis (Cox proportional hazards using stcox). I search a R
package or a R code that specifically addresses the categorical NRI for
time-to-event data in the presence of censored observation and, if
possible, at different follow-up time points.
I know that the ?PredictABEL? Package contains functions for NRI and IDI
calculation but it is unclear for me if it allows censored observation.
Package ?survIDINRI? calculates only continuous NRI and the function of
Package ?Hmisc?[#rcorrp.cens(x1, x2, y) ##] is only for no censored
observations.

Many thanks.

Sincerely,

Mario Petretta
Dpt. Internal Medicine, Cardiology and Heart Surgery
Naples University Federico II - Italy

On Nov 7, 2012, at 6:54 AM, petretta at unina.it wrote:
> Dear all,
> 
> I am interested to evaluate reclassification using net
> reclassification improvement and Integrated Discrimination Index IDI after
> survival analysis (Cox proportional hazards using stcox). I search a R
> package or a R code that specifically addresses the categorical NRI for
> time-to-event data in the presence of censored observation and, if
> possible, at different follow-up time points.
> I know that the ?PredictABEL? Package contains functions for NRI and IDI
> calculation but it is unclear for me if it allows censored observation.
> Package ?survIDINRI? calculates only continuous NRI and the function of
> Package ?Hmisc?[#rcorrp.cens(x1, x2, y) ##] is only for no censored
> observations.
???. Doesn't its name , 'rcorrp.cens' suggest otherwise? Not to
mention its description int the Hmisc Index: "Rank Correlation for Paired
Predictors with a Possibly Censored Response, and Integrated Discrimination
Index". rcoop.cens is a fairly recent addition to Hmisc and I am looking at
Hmisc version 3.10-1. If you are looking at a version that is a couple of years
old, you may be seeing something different. The argument list you list looks
like the one for improveProb(), which does not appear to handle censoring. The
rcorrp.cens argument list is:

rcorrp.cens(x1, x2, S, outx=FALSE, method=1)

And the "S" object is a Surv-object.

 
> Many thanks.
> 
> Sincerely,
> 
> Mario Petretta
> Dpt. Internal Medicine, Cardiology and Heart Surgery
> Naples University Federico II - Italy
> 
> ______________________________________________
> 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.
David Winsemius, MD
Alameda, CA, USA

Many thanks to Prof. Frank Harrell for the comments and for the relevant
considerations.

In the meantime, I find an R code to estimate the net reclassification
improvement (NRI) in the context of censored survival outcomes published
by Benjamin French, Paramita Saha-Chaudhuri, Bonnie Ky, Thomas P Cappola
and Patrick J Heagerty. The link is:
https://dbe.med.upenn.edu/biostat-research/sites/files/facultyfiles/french_supplement.pdf

However, whit this code the confidence interval of NRI is not calculated
(I think there is the need for bootstrapping data)and also IDI is not
calculated.

In my opinion, a standardization (or a consensus statement or a position
paper)for this technique seems highly desirable as the NRI is currently
used in an increasing number of medical research paper dealing with
time-to-event data and censored observation.

Sincerely

Mario Petretta
Dpt Internal Medicine, Cardiology and Heart Surgery
University of Naples Federico II - Italy

_______________________________

Date: Thu, 8 Nov 2012 06:59:49 -0800 (PST)
From: Frank Harrell <f.harrell at vanderbilt.edu>
To: r-help at r-project.org
Subject: Re: [R] R: net reclassification index after Cox survival
        analysis
Message-ID: <1352386789294-4648895.post at n4.nabble.com>
Content-Type: text/plain; charset=UTF-8

And just to add a thought: Any method that requires binning of continuous
variables is suspect unless you get unanimous agreement on the category
boundaries across all subjects.  And even then continuous measures have
many advantages.
Frank

petretta wrote
> Many thanks for Your time.
>
> M.P.
>
>>
>> On Nov 7, 2012, at 9:59 AM,
> petretta@
>  wrote:
>>
>>> Thanks to David Winsemius for the replay. i use the latest update
of
>>> Hmisc package and I try as reported in the example:
>>>
>>> set.seed(1)
>>> library(survival)
>>> x1 <- rnorm(400)
>>> x2 <- x1 + rnorm(400)
>>> d.time <- rexp(400) + (x1 - min(x1))
>>> cens <- runif(400,.5,2)
>>> death <- d.time <= cens
>>> d.time <- pmin(d.time, cens)
>>> rcorrp.cens(x1, x2, Surv(d.time, death))
>>>
>>> but to me it appears that NRI and IDi are not reported in the
results:
>>>
>>> Dxy               S.D. x1 more concordant x2 more concordant
>>>     -8.212107e-02       1.370738e-01       4.589395e-01
>>> 5.410605e-01
>>>                 n            missing         uncensored    
Relevant
>>> Pairs
>>>      4.000000e+02       0.000000e+00       1.100000e+01
>>> 4.262000e+03
>>>         Uncertain               C X1               C X2            
Dxy
>>> X1
>>>      1.553380e+05       9.920225e-01       9.258564e-01
>>> 9.840450e-01
>>>            Dxy X2
>>>      8.517128e-01
>>
>> The NRI is not reported but an equivalent measure is. As far as getting
>> output that is labeled the way you expect it, I also looked at the
>> PredictABEL::reclassification function help page:
>>
>> "The function also computes continuous NRI, which does not require
any
>> discrete risk categories and relies on the proportions of individuals
>> with
>> outcome correctly assigned a higher probability and individuals without
>> outcome correctly assigned a lower probability by an updated model
>> compared with the initial model."
>>
>> I think that is essentiality what rcorrp.cens is providing, just not
with
>> the labels you expected.
>>
>> " The function requires predicted risks estimated by using two
separate
>> risk models. Predicted risks can be obtained using the
>> functionsfitLogRegModel and predRisk or be imported from other methods
or
>> packages."
>>
>> So it would seem that you could use results from any censored survival
>> models that had a predict method.
>>
>> --
>> David.
>>
>>
>>>
>>> but only after:
>>>
>>> #rcorrp.cens(x1, x2, y) ## no censoring
>>> set.seed(1)
>>> x1 <- runif(1000)
>>> x2 <- runif(1000)
>>> y <- sample(0:1, 1000, TRUE)
>>> rcorrp.cens(x1, x2, y)
>>> improveProb(x1, x2, y)
>>>
>>> thus censoring not allowed. Or I'm in error?
>>>
>>> Many thanks
>>>
>>> David Winsemius &lt;
> dwinsemius@
> &gt; ha scritto:
>>>
>>>>
>>>> On Nov 7, 2012, at 6:54 AM,
> petretta@
>  wrote:
>>>>
>>>>> Dear all,
>>>>>
>>>>> I am interested to evaluate reclassification using net
>>>>> reclassification improvement and Integrated Discrimination
Index IDI
>>>>> after
>>>>> survival analysis (Cox proportional hazards using stcox). I
search a R
>>>>> package or a R code that specifically addresses the
categorical NRI
>>>>> for
>>>>> time-to-event data in the presence of censored observation
and, if
>>>>> possible, at different follow-up time points.
>>>>> I know that the ???PredictABEL??? Package contains
functions for NRI
>>>>> and IDI
>>>>> calculation but it is unclear for me if it allows censored
>>>>> observation.
>>>>> Package ???survIDINRI??? calculates only continuous NRI and
the
>>>>> function of
>>>>> Package ???Hmisc???[#rcorrp.cens(x1, x2, y) ##] is only for
no
>>>>> censored
>>>>> observations.
>>>>
>>>> ???. Doesn't its name , 'rcorrp.cens' suggest
otherwise? Not to mention
>>>> its description int the Hmisc Index: "Rank Correlation for
Paired
>>>> Predictors with a Possibly Censored Response, and Integrated
>>>> Discrimination Index". rcoop.cens is a fairly recent
addition to Hmisc
>>>> and I am looking at Hmisc version 3.10-1. If you are looking at
a
>>>> version that is a couple of years old, you may be seeing
something
>>>> different. The argument list you list looks like the one for
>>>> improveProb(), which does not appear to handle censoring. The
>>>> rcorrp.cens argument list is:
>>>>
>>>> rcorrp.cens(x1, x2, S, outx=FALSE, method=1)
>>>>
>>>> And the "S" object is a Surv-object.
>>>>
>>>>
>>>>> Many thanks.
>>>>>
>>>>> Sincerely,
>>>>>
>>>>> Mario Petretta
>>>>> Dpt. Internal Medicine, Cardiology and Heart Surgery
>>>>> Naples University Federico II - Italy
>>>>>
>>>>> ______________________________________________
>>>>>
> R-help@
>  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.
>>>>
>>>> David Winsemius, MD
>>>> Alameda, CA, USA
>>>>
>>>>
>>>>
>>>
>>>
>>>
>>> Mario Petretta
>>> Dipartimento di Medicina Clinica Scienze Cardiovascolari e
Immunologiche
>>> Facolt??? di Medicina e Chirurgia
>>> Universit??? di Napoli Federico II
>>> 081 - 7462233
>>>
>>> ______________________________________________
>>>
> R-help@
>  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.
>>
>> David Winsemius, MD
>> Alameda, CA, USA
>>
>>
>>
>
> ______________________________________________
> R-help@
>  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.




-----
Frank Harrell
Department of Biostatistics, Vanderbilt University
--
View this message in context:
http://r.789695.n4.nabble.com/R-net-reclassification-index-after-Cox-survival-analysis-tp4648710p4648895.html
Sent from the R help mailing list archive at Nabble.com.