Interoperability and Analytics
February 29, 2016
Matthew Hoffman MD, CMIO Utah Health Information Network
Conflict of Interest
Matthew Hoffman, MD
Agenda
• Electronic Secure Data
– Supplementary Clinical Data
– Patient Overlap Between Health Systems – Diabetes Star Ratings Pilot
• Population Management
– Chronic Disease Populations – Case Management Analyses – Geomapping Tools
• Savings
– Encounter Notification Service and FHIR – Risk Adjustment Analyses
Learning Objectives
• Describe how standards define use cases and parameters for health data analytics
• Identify the ways in which data analytics can help to support the business value of health information exchange
• Evaluate how new dimensions in standard definitions will shape data analysis applications and use cases
The STEPS ™
Framework:
Electronic Secure Data
The Case for Interoperability
-Supplementary Clinical/Hospital Data for Clinicians
- Star Ratings Analysis
- Shared Identities Numbers
Supplementary Data from Health
Information Exchange
EMR DW EMR DW EMR DW EMR DW
SMFM 23,535 100,315 105,849 236,460 - 501,652 109,608 705,304 6.46 CUC 713,480 1,990,063 585,269 5,369,244 1,301,722 13,425,165 2,240,747 28,182,616 10.11
DATA GAIN
MPI RECORDS LABS ADTS
Patient Overlap Between Health
Systems
0%
5%
10%
15%
20%
25%
30%
HS 1 HS 2 HS 3 HS 4 HS 5 HS 6 HS 722%
19%
29%
20%
15%
30%
13%
Diabetes Star Ratings Analysis
[VALUE]
[VALUE]
Data Related to Diabetic Measures
HIE Data
The STEPS ™
Framework:
Population Management
Analytic Tools for the Front Line
- The Importance of Standards - Chronic Disease Populations - Case Management Analyses - Geomapping Tools
Using Standards to Combine Data
Cost to Value of Types of Analytics
Models
Clinician Created Analyses
Custom Filtering
Filter Settings
A1C
- Gender: (female, male)
- Age: (0 <= Age <= 124)
- A1C_Val: (1.30 <= A1C_Val <=
21.99)
MicroAlbumin
- Mic_Val: (0.00 <= Mic_Val <=
11.35)
BP BMI Cholesterol
- BP: (82 <= BP <= 205)
- BMI: (0.21 <= BMI <= 73.02)
- Cholesterol: (0 <= Cholesterol <=
251
)Care Management Tools:
Diabetic Population
Care Management Tools: Diabetic
Population
Geomapping
The STEPS ™
Framework:
Savings
For the CFO in All of Us
- Encounter Notification Service and FHIR - Risk Adjustment Analyses
FHIR ENS Solution Diagram
Results
• Anecdotal
– Post Surgery Re-admit reduction
– Decreased Asthma Hospital Admissions • Documented
– 3 month pilot (Coaction):
• 54% reduction in hospitalization days • 33% reduction in ED visits
• $178,547 estimated savings – Medicare Patient Population (NY)
• 2.9% reduction in likelihood of 30 day readmission • $1.24 million savings in admissions
Risk Adjustment Analysis
Architecture
ICD-X from Clinical Note
STEP Summary
Exchange
• 21% Avg Overlap Between Systems • 17% Increase in Star Ratings Data
Population Health
• Care Managers • Physicians
• Public Health
Savings
• Alerts: Decrease in Hospital Days, Readmissions and ED Visits
• Risk Assessment:
Questions
• Matthew Hoffman, MD
• CMIO, Utah Health Information
Network
Interoperability and Analytics
February 29, 2016
Daniella Meeker, PhD
Conflict of Interest
Funding
Overview
• Multi-Institutional Learning Health Systems • Data Standards and Standardization Process • Computation Standards for Analysis
• Result Standards for Dissemination and Application
Learning Objectives
• Describe how standards define use cases and parameters for health data analytics
• Identify the ways in which data analytics can help to support the business value of health information exchange
• Evaluate how new dimensions in standard definitions will shape data analysis applications and use cases
Standards for Data Exchange and
Persistence are Necessary but Not
Sufficient to Obtain Value from
Multisite Data Infrastructure.
Products of a Learning Health System
• Analysis models for causal inference and program evaluation – Program X is 15% more effective at preventing readmission
than Program Y
– Drug A has a greater risk for adverse events than Drug B • Quality measurement reports
– Practice N is achieving 80% of quality indicators – Practice L is achieving 60% of quality indicators
• Predictive models for patient-centered medicine
– For patients like John, Drug C is safer and more effective than Drug D.
– For patients with Debbie’s goals and comorbidity profile, Program Y is better than Program X
pSCANNER Network – 21M patients
• 5 University of California Medical Centers • Cedar’s Sinai Hospital
• Pacific NW Rural Health Practice-Based Research Network
• Los Angeles Department of Health Services
• 5 multi-site FQHCs
• Children’s Hospital of Los Angeles • Keck Medicine of USC
• Service Oriented Architecture for privacy-preserving analytics in distributed research networks
– Focus on map-reduce, iterative algorithms based on parallel-distributed processing algorithms
• Standardized Analytic Data Warehouse at every site
• Role based access control -> Attribute Based Access Control
• De-centralized study administration – no “coordinating center” peer-to-peer analytics.
• GUI for menu-driven queries for multiple activities
– Proposing collaboration partners, data sets, and protocols – Specifying regression analytics
– Data set extractions – Cohort discovery
Distributed Research Network
Requirements
• Methods repository for contributing and sharing analytic algorithms for classical statistics and machine learning
• Result repository for contributing and sharing analytic results –(e.g. causal models, predictive models, quality reports)
Distributed Research Network
Requirements
Need to select and implement standards for the entire analysis cycle:
• Data
• Data Processing Algorithms • Data Analysis Algorithms • Data Analysis Results
SCANNER Network Software
Health Economic Domains are
not represented in other research and quality information models,
Implementation
Aside on Data Quality & Availability
Dx R x Lab Procedures text CCD LIS Finance EHROMOP OMOP OMOP OMOP OMOP OMOP OMOP OMOP OMOP SHARED TRANSFORMATIO N PROGRAMS Quality Model Quality Model Quality Model Quality Model Quality Model Quality Model Quality Model Quality Model Quality Model CUSTOM PROGRAMMING QRDA DISTRIBUTED ANALYTICS
Aside on Data Quality & Availability
Dx R x Lab Procedures text CCD LIS Finance EHRQRDA STANDARDIZED WAREHOUSE
Why do we need standards for
computation if the data is already
standardized?
• Reproducibility
• Robustness to innovation – update parts without starting from scratch • Flexibility to local implementation
• Transparency • Comparability
Direct and Quantifiable Comparisons
dataset A dataset
A
Selecting a Standard for Computation
Specification
• Sufficiently expressive to represent data processing concepts for
transactional, time-series data (e.g. interval logic)
• Sufficiently expressive to represent data processing concepts that are specific to healthcare (e.g. time of administration, age of onset) • Sufficiently expressive to represent basic statistics and data
analysis algorithms • Supported/Adopted Quality Data Model Data Processing Semantics Predictive Model Markup Language SQL
OMOP OMOP OMOP OMOP OMOP OMOP OMOP OMOP DISTRIBUTED ANALYTICS PORTABLE DATA PROCESSING STANDARDS
Products of a Learning Health System
• Analysis models for causal inference and program evaluation – Program X is 15% more effective at preventing readmission
than Program Y
– Drug A has a greater risk for adverse events than Drug B • Quality measurement reports
– Practice N is achieving 80% of quality indicators – Practice L is achieving 60% of quality indicators
• Predictive models for patient-centered medicine
– For patients like John, Drug C is safer and more effective than Drug D.
– For patients with Debbie’s goals and comorbidity profile, Program Y is better than Program X
Disseminating Products of a Learning Health
System
OMOP Reference Data Model Extracted Data Set (“Flat File”) Data Set Extraction Program Analysis Program Estimated Predictive Modelaka “Data Processing” “Computable Phenotype” “Cohort” “Inclusion Criteria” “Measure Denominator” research Patient Record Predictive Model Computation Standardized Extracted Record Patient Centered Prediction Publish care Publish Record Processing Program CCD
Selecting Standards for Disseminating
LHS Products
• Quality Indicators can be shared & Reported with QRDA
• Predictive Models can be shared & Reported with PMML (but not part of Health IT standards)
• Causal inference & program evaluation results are (still) disseminated on paper.
Summary: Standards for Exchanging Analysis
Process and Products
Process we need to represent
Standard Rationale
Data processing rules HQMF>CQL CMS, ONC, HL7 endorsed
Part of EHR certification process New standards in trial use
Cohort definition rules HQMF>CQL 100’s of established data sets 1000s of cohort criteria
Data set description QRDA
PMML
QRDA – Quality Measurement EHR Certification & CMS PMML – Data analysis
Data Analysis Methods PMML UCSD Data Mining Group
Extensible to support model specifications Data Analysis Results
(Estimated Models, Produce Predictions)
PMML Developed to represent results
Adopted by most stats packages
What Next?
• FHIR
• Clinical Quality Framework • Data Access Framework