DEMONSTRATING CLOUD-BASED CLINICAL DECISION SUPPORT AT
SCALE:
THE CLINICAL DECISION SUPPORT CONSORTIUM
Brian E. Dixon, MPA, PhD, FHIMSS Marilyn D. Paterno, MBI
Outline – Part 1
• Introductions – Overview of the CDSC and its Aims
• Theoretical Framework
– Models to accelerate knowledge to practice – Unified theory of CDS
• Knowledge Representation
– The Four layers – Authoring tools – The KM portal
Outline – Part 2
• CDS in the Cloud – The web services approach – CDSC Implementation Sites – Demo of the Regenstrief system
• CDS Dashboard
– Assessing CDS effectiveness
• Future Directions • Acknowledgements
Spectrum of CDS Systems
Clinical Reminder
Clinical Alert
Corollary Order
Population Health
CDS Grand Challenges
• Summarize patient-level information • Prioritize recommendations to users • Combine recommendations for patients with co-morbidities • Improve the human-computer interface • Use free text information in clinical decision support • Manage large clinical knowledge databases • Create a internet-accessible, clinical decision support repository • Prioritize CDS content development and implementation • Disseminate best practices • Create an architecture for sharing executable CDS modules • Mine large clinical databases to create new CDS
Sittig et al., JBI, 2008
CDS Consortium: Goal and Significance
• Goal: To assess, define, demonstrate, and evaluate best practices for knowledge management and CDS in health care IT at scale – across multiple ambulatory care settings and EHR technology platforms
• Significance: The CDS Consortium will carry out a variety of activities to improve knowledge about decision support, with the ultimate goal of supporting and enabling widespread sharing and adoption of CDS.
1. Knowledge Management Life Cycle
2. Knowledge Specification
3. Knowledge Portal and Repository
4. CDS Public Services and Content
5. Evaluation Process for each CDS Assessment and Research Area
6. Dissemination Process for each Assessment and Research Area
5-Year Timeline
2008 • October
2009
2010
2011
2012
2013 • July
Content, KM, KT Development Analysis of Best Practices
Pilots, Demonstrations Expand CDSC Membership Expand CDSC Content Evaluation
ARRA/HITECH Passed
MU Stage 2 Final
Three Models to Accelerate Knowledge -> Practice
•Current paper-based approach
•Knowledge artifact import into EMR •Cloud-based clinical decision support services
EMR Guideline
Computer Interpretable Guideline
Web Services CCD/VMR Patient Data Object
Decision Support
Clinical Knowledge
Structured Knowledge
Implementable/ Executable Knowledge
Service
EHR EHR
EHR
CDSC “L2”
GEM
CDSC “L4”
CDSC “Action – Recommendation”
CDSC “L3”
CDSC “CCD+”
The Unified Theory for CDS
Knowledge Translation and Specification: Four-Layer Model
Initial evaluation results: Structured recommendation (L3) was considered more
implementable than the semi-structured recommendation (L2).
derived from derived from
Level 1 Unstructured Format : . jpeg , . html , . doc , . xl
Level 2 Semi -
structured Format : xml
Level 3 Structured Format : xml
Level 4 Machine
Execution Format : any
derived from
+ metadata + metadata + metadata + metadata
Boxwala, A.A., et al. A multi-layered framework for disseminating knowledge for computer-based decision support. JAMIA 2011.doi:10.1136/amiajnl-2011-000334
Published Guideline
Semi-structured Recommendation
Structured Recommendation Executable Rules
Order Sets in CPOE system
Narrative Guideline Screening for High Blood Pressure Reaffirmation Recommendation Statement U.S. Preventive Services Task Force (USPSTF) The U.S. Preventive Services Task Force (USPSTF) recommends screening for high blood pressure in adults aged 18 and older. (This is a grade "A" recommendation)
Semi-Structured Recommendation Meta data Title: Screening for High Blood Pressure Reaffirmation Recommendation Statement Developer: U.S. Preventive Services Task Force (USPSTF) Strength of recommendation: Grade A Clinical Scenario: Patient age ≥18 years Blood pressure not obtained in the last year Clinical Action: Obtain and record blood pressure
Structured Recommendation Meta data Title: Screening for High Blood Pressure Developer: CDS Consortium Derived from: USPSTF BP Screening Semistructured Rec. Applicable Scenario Data Mapping: BPRecordedInLastYear: Observation = VitalSign-> select(code.equals(BPLoincCode) and vsDataTime.within(12, months)) Logical Condition: BPRecordedInLastYear->notEmpty() Recommended Action: VitalSign(code: BPLoincCode)
Arden Syntax Rule knowledge evoke: … data: BPRecordedInLastYear := read last{table=‘RES’, code=‘12345-0’} PCPemail := read {…}; Adult := …; logic: if (adult is false) then conclude false; if (BPRecordInLastYear is null) then conclude true; action: Write ‘Patient has not had a blood pressure screening in the last year’ at PCPemail;
The Four Layers Illustrated
L3 Knowledge Module
• Single knowledge representation approach for different CDS modalities – Order sets, reminders, alerts, documentation
templates – Modality features tend to mix-and-match
• Single representation for different modalities – Unified framework for tools development – Enables consistency checking
Knowledge Module Structure
Knowledge Module
Action
Behavior
Presentation
Metadata
Patient Data
Knowledge Authoring Tool
Knowledge Authoring Tool
Knowledge Authoring Tool
Knowledge Management Portal
http://cdsportal.partners.org/cdscsearch.aspx
KM Portal
Search Results for ‘diabetes’
CDS IN THE CLOUD
Web Services Approach: Hypothesis & Goal
• CDSC CDS Services team hypothesis – A service-oriented architecture (SOA) approach to decision
support is feasible and will provide benefits in interoperability, reliability, and reusability of knowledge content used in clinical decision support across multiple sites.
• Interoperability goal – A web service that is – External to the application and/or system – Agnostic to the technology of the calling site – Capable of being called from inside or outside its firewall – Supports / makes use of emerging standards
Web Services Approach: Service Flexibility
Service Input Formats
Web Services Approach: Flexible yet Standard Output
• Action – Recommendation Model – Utilizes HL7 Datatypes R1 standard
• Makes the decomposing task easier • Mappable to local EHRs that use coded data
– Actionable for clinical use, as in: • Create Orders – Medications, Procedures • Record Observations – Problems • Display Messages – Text-based Alerts • Provide Knowledge Assets – Patient Education Material • Recommend Encounters – Referrals
Message
Observation
Encounter
Knowledge Asset
Message
Sample
WVP Health Authority Salem,Oregon
Wishard Hospital Indianapolis,IN
RWJ Medical Group New Brunswick,NJ
PHS Host Boston, MA
Web Services Approach: Implementations
Current Status
Site - Guideline Clinics Providers
Partners - CDSC 2 48
Partners - Immunization 4 40
Regenstrief - CDSC 2 72
NextGen - CDSC 4 10
GE – CDSC (Expects to start late August) 1 10*
* Estimate
Performance over 90 days*
Site - Guideline # days Type+ Average
Calls/day Avg time
(secs)
Partners - CDSC 90 S 1,573 0.98
Partners - Immunization 80 S 526 0.82
Regenstrief - CDSC 64 A 113 1.06
NextGen - CDSC 87 A 80 1.68
+ Synchronous / Asynchronous * May 13, 2013 – August 10, 2013
Partners HealthCare @ LMR CDSC Guidelines
Reminders appear on Summary Screen
Partners HealthCare @ LMR CDSC Guidelines
And on Reminder Screen
Partners HealthCare @ LMR CDSC Guidelines
And at time of signing
Partners HealthCare @ LMR CDC Immunizations
Partners HealthCare @ LMR CDC Immunizations
Contraindication warning
Regenstrief Institute @ Wishard Hospital
DEMONSTRATION CDSC Guidelines at Wishard - Gopher
NextGen @ WVP Health Authority
GE Centricity @ Rutgers Robert Wood Johnson Medical Group
CDS DASHBOARD
Purpose
• Develop performance reporting tools and CDS dashboards to –review adherence to CDS Consortium guidelines –assess effectiveness of CDS on patient care
outcomes
• How does the dashboard help assess CDS
effectiveness?
Putting Reminders in Context
Patient becomes member of eligible population
Reminder logic becomes true
Reminder displayed
Reminder accepted
Right action documented
Clinical outcome
“Prevalence” “Logic” “Display” “Acknowledged” “Performance” “Outcome”
Measurement Period
T0 T1
Patients with Type2 DM
Overdue for A1 C Test
Reminder displayed to user
User clicks on reminder and chooses coded response
A1 C test result documented
A1 C < =7. 0
Epidemiology of CDS: the CDS/Reminder Lifecycle
“How well are the reminders working?”
• To measure the Effectiveness of CDS, dashboard uses
– Display (Counts) – Acknowledged (%) – Performance (“Right” action taken) – Contribution to Clinical Performance
• Number Needed to Remind (NNTR) – the number of reminders needed to be displayed to
a provider for that provider to take the recommended action
Dashboard – CDS Designer View
Total patients 47,782 Performing total 28,476
Patients where reminders displayed 2,757 Total count of displays 14,944
NNTR 3.87
Reminders in Context
CDS Reminders Displayed
NNTR is: 2757 patients with reminder displayed divided by 713 patients who had reminder displayed and then had aspirin added to the Med List = 3.87 (If look at total #reminder displays rather than #patients, then NNTR is 20.96)
Number Needed to Remind (NNTR)
Less effective More patients
Less effective Fewer patients
More effective Fewer patients
More effective More patients
CDS Reminder Effectiveness
Uses for Dashboard Results
• Systematically monitor and evaluate effectiveness of clinical decision support
• Prioritize which decision support guidelines or reminders to implement or enhance
FUTURE PLANS
Future Directions for the CDSC
• Current demonstrations – Complete site trials currently running – Publish multi-site trial results, analysis
• Continuing to expand – Technology development (e.g., support for DSS, SMART) – Clinical content offerings (e.g., pharmacogenomic, other
rules in development)
• Relocating Administrative Support – Vanderbilt University Medical Center
FOR MORE INFORMATION
CDSC Blackford Middleton [email protected] Four-Layer Process and Knowledge Authoring Tool Aziz Boxwala [email protected] KM Portal Tonya Hongsermeier [email protected] ECRS Web Service Howard Goldberg [email protected] Action-Recommendation Model Beatriz Rocha [email protected] CDS Dashboard Jonathan Einbinder [email protected]
THANK YOU Presenters: Brian E. Dixon, MPA, PhD, FHIMSS [email protected] Marilyn D. Paterno, MBI [email protected] Contact: Blackford Middleton, MD, MPH, MSc [email protected]