1. CHIR MRSA Project:
Clinical & Operational
Significance
Healthcare-associated infections (HAIs) leading
cause of preventable death.
VA Pittsburgh Reduced MRSA Significantly!
VHA National MRSA Reduction Initiative Effective!*
HAI Surveillance Highly Time Intensive
Staff need to Focus on Problems, Intervene Early
Need for Automated Surveillance Tools
Interdisciplinary, multisite collaborative extracting
structured and unstructured data from EHR
*Jain et al, NEJM 2011
Indianapolis, IN; Salt Lake City, UT; Palo Alto, CA; West Haven, CT; Tampa, FL
4. Annotation Goals &
Progress
• Key defining what is needed from TIU’s and
where to get it for machine learning.
• Used to create guidelines and interact (2-way)
with ontologies.
– UTI guidelines finalized (465 documents annotated).
UTI, urinary cath devices, & temp and fever
– Guidelines for general machine learning written.
~anything infection related to inform a machine learning approach
and benefit ontology.
– Guidelines for detecting CVADs (pilot).
ID when cath is present, when inserted or removed
5. Example: Electronic Capture
Medical Device Use
*A problem confounding infection surveillance
1. Identify possible methods for capturing medical
device use data for acute care patients (focus
groups)
2. Identify the pros and cons associated with
specific methods to measure medical device use
(focus groups)
3. Develop a prototype system to capture medical
device use data for acute care patients in the VA
Enhanced Metafile Record (EMR)
Partnership with OI&T Innovation - West Haven-
Martinello, Brandt, et al
6. MRSA NLP Progress
• Developed YTEX – a powerful combination NLP pipeline &
database.
-Integrated into VINCI MRSA database.
• Analyzed 30,000 notes of MRSA cases, locally and on VINCI.
• Developed lists of 2000 clinical abbreviations and 4000 concepts
(manual chart review) specifying infections.
• Developed patient-centered NLP approach to analyze notes in
temporal window related to + MRSA culture in 2 use cases:
UTI and bloodstream infections.
• Prior infections in 90% of patients.
• Sentences and document fragments with relevant terms retrieved
and highlighted.
7. MRSA NLP Findings and
Ongoing Work
• NLP successfully captures relevant clinical info. related to
MRSA infection from heterogeneous clinical notes.
• False positives a major problem.
-Some rejected with improved NLP;
-Others presented to experts for judgment
Ongoing Work to Improve NLP accuracy
• Incorporation of clinical rules into NLP process
• Integration of Ontology with NLP
• Improvement of temporality, section detection, template
detection, abbreviations and negation (*key focus of
collaboration with Information Extraction Methods group)
8. Vision of Surveillance Tool
1. To rapidly design a user-centered tool to capture
MRSA UTIs by fall of 2012
2. User-centered tool for IP/ ID/ Epi experts to:
1. Display data, results, and outcomes that
2. Support key decisions & optimal workflows, and
3. Enable appropriate next steps (or actions)
3. Incorporate Black Box surveillance as appropriate
4. Demonstrate feasibility of and need for full feature
HAI Dashboard System beyond 2012
9. Workflow Capture & Analysis
To Support Rapid, User-Centered Design
Engage Leadership: Refine Design through:
- National, Regional, and
Local- Clinical and Research
- Identify Goals, Metrics, etc. Analysis:
Observations:
- Contextual Inquiry
- Grounded Theory
- Follow-A-Thread - Artifact Analysis
- Think-Aloud
Interviews:
- IP’s, Epi, ID Physicians
- Identify & Validate Activities
- Vision for Tool & Workflow Inform, Electronic Data:
Validate Refine,
Iterate - VINCI Data as available
- Machine Learning and NLP
Rapid Ethnography: Iterate &
Models
- Environmental Walkthrough Validate
- Shadow Key Personnel
Refine
- Artifact Collection
& Iterate
Iterate Surveys:
- Needs Assessment &
Representations: Validation (as needed)
- Workflows & Process Maps
- Information & Screen Flows Key Deliverables:
- Design Narratives
- Interactive Prototypes
Notas del editor
This slide can be used as a template for CHIR F2F Presentations at the October 2011 F2F Meeting. It also outlines guidelines created by the CHIR Salt Lake City group for each presenter to consider including so members and/or guests can gain a sense of each project’s relevance/status.