Re: [BIONT-DSE] Inclusion versus exclusion criteria

Kashyap, Vipul wrote:
> I guess the issue then becomes for which data items/decision  
> criteria is
> negation explicitly asserted (MRSA) vs it needs to be inferred (drugs)
>
> Also, is it the case that one can make this statement about all  
> labs without
> loss of generality? Or can this be said only in a contextual  
> manner, i.e.,
> Negative labs are explicitly asserted only for a given set of lab  
> results
> as reported by a given set of diagnostic laboratories?

We will have results only for the labs that were ordered (relevant to  
the patient condition). So I guess its a choice one has to make based  
on the application - in eligibility determination application, we  
assume that relevant lab results are available.

> So, in the open world assumption case, you would probably need to  
> compute
> (Patients satisfying Inclusion Criteria) DIFFERENCE
> (All Patients DIFFERENCE Patients Not satisfying the Exclusion  
> Criteria)

We perform a more simplistic set difference wherein we first find all  
patients that satisfy the inclusion criteria and then exclude  
patients that satisfy the non-negated exclusion crtieria. Sorry for  
the terse explanation, we describe our methodology in detail in this  
draft appearing in the proc. of ISWC 07 ( http:// 
domino.watson.ibm.com/library/cyberdig.nsf/papers/ 
BDCB732D647E2547852572EB0054C912/$File/rc24265.pdf )

Daniel Rubin wrote:
> Likewise, making any generalized statements about when open- and  
> closed-world reasoning are appropriate in health care is dangerous-- 
> one must be very specific about the context and specific case to  
> assess whether the assumption is valid.

I agree,  knowing the context is critical -- here are 2 scenarios  
that come to mind, which would require different approaches:
1. Given that we are using inpatient data where we know all pharmacy  
orders are made through a single CPOE system or aggregated from the  
pharmacy systems in to a single repository and we are not missing any  
medication information from other sources (such as clinical notes)  
then we can make an application specific choice to use closed world  
reasoning over the inpatient medical records.
2. However, if the patient repository in the institution is linked to  
a broader "web-like" system such as RHIOs wherein patients can freely  
move across the healthcare system then one has to potentially resort  
to open world view of things.

In our patient eligibility application, we were dealing with  
scenario#1, which allowed us to use a closed world assumption for the  
pharmacy data.

Thanks,
Chintan

Received on Wednesday, 12 September 2007 20:58:43 UTC