|
|
|
Vojtěch SVÁTEK, Antonín ŘÍHA,
Jan PELEŠKA, Jan RAUCH |
|
EuroMISE Centre – Cardio
University of Economics, Prague
Computer Science Institute of the Czech Academy of Science |
|
|
|
|
|
Guideline compliance analysis carried out |
|
traditionally by clinicians:
mostly statistical methods |
|
more recently also by computer scientists:
possible use of modern techniques, such as knowledge modelling or data
mining |
|
We speak about the latter |
|
|
|
|
|
Discovered non-compliance may lead to |
|
feeding back to guideline authors
(incomplete, inconsistent or outdated documents) |
|
feeding back to field clinicians
(errors in clinical practice) |
|
feeding back to knowledge engineers
(errors in document formalisation) |
|
Depending on interpretation… |
|
|
|
|
|
|
Our approach |
|
detect non-compliance for individual patient
records using an operational model of the guideline (in OCML or Prolog) |
|
apply a data-mining tool to |
|
determine frequent types of non-compliance |
|
find associations among the non-compliance
patterns and other patient data |
|
frequent patterns together with their associations
submitted to medical expert for interpretation |
|
|
|
|
|
|
Experiment in the area of hypertension |
|
48 patient records |
|
compared to an (ad hoc) OCML model of the WHO
hypertension guidelines |
|
the model covered 10 generic non-compliance
patterns (NCPs) |
|
61 instances of these patterns discovered in
data |
|
subsequent analysis by the LISp-Miner tool |
|
associating 10 NCPs with 39 other binary
attributes |
|
8 associations found for a reasonable parameter
setting |
|
interpretation by a physician (data-donor) |
|
mostly missing background knowledge or outdated
guidelines |
|
|
|
|
|
|
|
|
|
Step-by-step bottom-up formalisation
(cf. the talk by Růžička...) |
|
allows to proceed from original document to
‘literal’ operational model |
|
can be carried out to large extent by knowledge
engineer alone, except for addition of background knowledge |
|
data-mining-based compliance analysis then may
serve for posterior identification of missing background knowledge! |
|
|
|
|
Thank you for your attention |
|
Notes