On Tue, 26 Sep 2006, Kashyap, Vipul wrote:
1. Lack of explanation capabilities: A key feature for clinical decision
support is that physicians like to get
explanations for the recommendations proposed by the system.
Well, I'd argue that the recommendations are the accompanying literature
that document the very measured thought process that went into setting up the model: the
choice of variables, considerations in combinations of variables,
outliers & statistical anomalies, etc. provide more contextual
recommendation than a logical proof trace you would get from a purely declarative approach. The
addition of associated probabilities with the output of the model make for
a more responsible indicator especially for aspects of a pathway that
are heavily dependent on a large number of variables - each in very
specific ways.
2. Lack of "knowledge visibility": The biggest downside is from the KM
perspective, what if one of the conditions changes? We need this to be visible
so that we can have KM processes handle these changes.
I'm not sure I follow. The models I'm speaking of are 'driven' by patient
data, a different patient would result in a different outcome scenario (with
associated confidence limits and probability). The only constants are the
weights that are very much specific to the pathway (so ofcourse it would
be irresponsible to swap these into different pathways that may not have
been part of the considerations that guided the creation of the
statistical model in the first place.
Chimezie Ogbuji
Lead Systems Analyst
Thoracic and Cardiovascular Surgery
Cleveland Clinic Foundation
9500 Euclid Avenue/ W26
Cleveland, Ohio 44195
Office: (216)444-8593
[EMAIL PROTECTED]
