Transcript M8120_EBP_IR - Columbia University

Information Retrieval for
Evidence-based Practice
Fall 2001
Suzanne Bakken, RN, DNSc, FAAN
School of Nursing & Department of Medical Informatics
Columbia University
Definition
• Information retrieval (IR) is the
science and practice of
identification and efficient use
of recorded media
Information Retrieval and Analysis
• Indexing (using standardized
terminology and other methods)
• Query formulation
• Retrieval
• Evaluation and refinement
Relevance
• Recall - ratio of number of documents
retrieved and relevant to number of
relevant documents in database
(sensitivity)
• Precision - ratio of relevant documents
retrieved to total number of documents
retrieved (positive predictive value)
User-defined Relevance
• Information content of document
• Provision of references to other
sources of information
• Subjective qualities
• Situational aspects - clinical vs.
student
Types of Clinical Decisions
• Diagnostic reasoning - differential
diagnosis
• Decisions related to actions - involve risk:
cost, time, morbidity, mortality, discomfort
– Diagnostic tests - treatment-threshold
– Screening
– Management (including prevention) medications, surgery, education,
watchful waiting
General Search Strategy for
EBP Information Retrieval
Clinical problem
Define searchable question
Select most likely resource
Design search strategy
Summarize the evidence yield
Apply the evidence
Select another resource
Poor
General Search Strategy for
EBP Information Retrieval
Clinical problem
Define searchable question
Select most likely resource
Design search strategy
Summarize the evidence yield
Apply the evidence
Select another resource
Poor
Define Searchable Question
• Background question – general (who,
what, where, when, why, and how)
• Foreground question – patient-specific
knowledge questions
Informatics Support for
Background Questions at CPMC
• Generic queries – Medline Button/
InfoButton
– Is [laboratory test] indicated in [disease or
syndrome]?
– What is the drug of choice for [disease or
syndrome]?
• Supported through the Medical Entities
Dictionary (the MED), the conceptoriented terminology
Elements of Well-built
Foreground Questions
• Patient or problem being addressed
• “Intervention” (natural or therapeutic)
being considered
• Comparison intervention being
considered, where relevant
• Outcome of interest
Selecting a Diagnostic Test
• In patients with heart failure, how
accurate is the bedside test for postValsalva systolic overshoot in predicting
left ventricular function?
• Problem
– Intervention
– Comparison intervention
– Outcome
Determining Prognosis
• In an elderly woman with coronary
heart disease, by how much does the
appearance of heart failure shorten her
life?
– Problem
– Intervention
– Comparison intervention
– Outcome
Selecting Therapy
• In coronary patients who develop heart failure
on ibuprofen, would simple drug withdrawal
plus diuretics suffice or should they be offered
an angiotensin-converting enzyme (ACE)
inhibitor as well, in order to alleviate the heart
failure?
• Problem
– Intervention
– Comparison intervention
– Outcome
Selecting Secondary Prevention
Measures
• In patients with heart failure, would
long-term ACE inhibitor therapy prevent
recurrences and improve the quality of
life?
– Problem
– Intervention
– Comparison intervention
– Outcome
Informatics Support for
Background Questions at CPMC
• MED
• Query matching
– In the context of the particular application, what is
the closest generic question?
– Matches displayed to user and user may add more
information then query is passed to multimedia
search process
• Query refinement
– Additional information can be added (e.g., age) if
search results are not satisfactory