[R] Sensitivy / Specificity and nulls

[David Winsemius]

On Aug 28, 2013, at 7:35 AM, Donald Catanzaro wrote:

> Hi All,
> 
> Since R is used primarily as a statistical package by individuals using
> statistics to answer questions it makes sense to ask a question such as
> this.  Oftentimes individuals have encountered similar issues as the poster
> and can quickly point them in the correct direction - that is all I am
> asking.

This is a generic and pervasive problem in medicine. Most physicians do understand that the results of tests have uncertainty of various sorts, that "upper limit of normal" is not a valid criterion for diagnosis, and the even "research" definitions of disease classifications are approximate. I think the notion that R users have some special capacity to address these epistemological problems is overly optimistic.  The question is arguably about both classification and prediction, since diagnostic classfications are basically validated (and later revised) by their ability to support a prognostic/predictive purpose. You might look at the Cluster Task View: http://cran.r-project.org/web/views/Cluster.html . I would think that the Bayesian section would be most likely to yield useful packages, since you have essentially posed a problem of priors and posteriors. 

(Your request fo help about "the NULLs" has absolutely no context however.)

-- 
David Winsemius, MD

> 
> Quite honestly I have been on the R list for several years and have seen
> *plenty* of people asking for statistical advice.  Now of course often
> times the individual is flamed (something that the R list is notorious for)
> so Bert, I personally thank you for being at least courteous to offer
> advice.
> 
> Perhaps someone on the list may have a more constructive comment.
> 
> 
> On Wed, Aug 28, 2013 at 9:25 AM, Bert Gunter <gunter.berton at gene.com> wrote:
> 
>> This is not an R question. Post elsewhere. Better yet, contact a local
>> statistical consultant.
>> 
>> -- Bert
>> 
>> On Wed, Aug 28, 2013 at 7:18 AM, Donald Catanzaro <dgcatanzaro at gmail.com>
>> wrote:
>>> Good Day All,
>>> 
>>> I am working with a diagnostic test and comparing the new test to an old
>>> test.  Normally I would be able to calculate sensitivity and specificity
>>> quite easily.
>>> 
>>> However, the 'gold standard' that I am comparing my new diagnostic with
>> is
>>> really 'gold-plated' in that sometimes the 'gold standard' fails
>> completely
>>> and I have no data from the 'gold standard' but I might have data from
>> the
>>> diagnostic test.  Of course sometimes my new diagnostic fails but I have
>>> data from my 'gold standard'
>>> 
>>> To me this really starts moving towards classification but I cannot seem
>> to
>>> find the appropriate calculations.
>>> 
>>> Can someone point me to some web resources to determine the appropriate
>>> method to be able to deal with the NULLs ?  Resources within the medical
>>> realm would be better (because the rest of the folks would understand
>> them
>>> better) but not required.
>>> 
>>> --
>>> - Don
>>> 
>>> Donald Catanzaro PhD
>>> 

David Winsemius
Alameda, CA, USA