Transcript Information Query INAHQ Presentation_Drew Richardson

Better Outcomes. Delivered.
37th Annual InAHQ Quality In Healthcare Conference
Information Query: Tools to Successful Population Health Management- Director of Population Health - Drew Richardson
May 13th, 2016
Copyright © 2015 Indiana Health Information Exchange, Inc
• Overview of IHIE
• Partnership with Regenstrief
• How does a HIE work?
• Definition of Population Health
• Why is a HIE involved in Population Health
• IQ
• Demo
• Ndepth
• Demo
• Case Studies - Our Experience
Agenda
www.ihie.org
IHIE Background
• Nation’s largest HIE
• >100 hospitals (38 health systems)
• >25,000 clinicians
• Payors
• Labs and Imaging Centers
• Public Health
• Long-term/Post-acute care (LTPAC)
• Founded in February 2004
• 501c3 not-for-profit organization
• Regenstrief Institute partnership
• Mission: Through information exchange we improve
health and healthcare
www.ihie.org
HIE Infographic
http://www.himss.org/library/health-information-exchange/improving-patient-care?navItemNumber=47236
Copyright © 2010 Indiana Health Information Exchange, Inc
www.ihie.org
Population Health
What is the definition of Population Health?
My simplistic definition
Health status and health conditions of an aggregate population.
This allows organizations, communities, providers, and public health resources to
allocate resources to overcome the problems that drive poor health conditions in
the population.
www.ihie.org
Information Query
• Population health query solution that enables healthcare administrators, quality
teams, and providers to improve outcomes across safety, quality, clinical and
operational domains.
• Rapidly search health quality, treatment information, and limitless other data
points in real-time. Identify specific population groups faster, more reliably, and
less costly than existing solutions.
Copyright © 2010 Indiana Health Information Exchange, Inc
www.ihie.org
Labs
Structured Data
ICD-9/10
Meds
CPTs
Echo’s
Op Reports
Radiology
Unstructured Data
Pathology
Provider notes
Endoscopy
Finding patients is complicated and time consuming
Due to limited data visibility and informatics
resources
Requires engaging data managers because the data
is complicated to retrieve
Request is added to the queue and often requires
refining in order to achieve
www.ihie.org
Complex patient queries can take hours or days to complete
Yesterday’s technology and infrastructure
Complex patient queries can take hours or days to complete
can’t effectively handle the deep analysis of
structured patient data – and many current
EMR’s provide a patient view, not a
population view.
Queries are filtered using discrete values
such as: medications, lab/test results,
diagnosis and procedure codes, and
demographics
www.ihie.org
Complex patient queries can take hours or days to complete
Yesterday’s technology and
infrastructure can’t effectively
handle the deep analysis of
structured patient data – and many
current EMR’s provide a patient
view, not a population view.
These complex queries require
extensive time and processing
power to retrieve, sometimes taking
hours or days instead of minutes.
www.ihie.org
Empower non-technical users to find patients quickly and easily
IQ is population health query
solution that enables clinicians,
researchers, and quality teams to
quickly search, identify, and
engage groups of patients to
transform population health.
www.ihie.org
Empower non-technical users to find patients quickly and easily
Access population statistics anytime, without
waiting for a data request, or learning complex
search patterns
Create actionable lists for faster patient
engagement
Easily identify opportunities for patient
intervention and disease management
Share and reuse validated population queries
Customize and export specific patient related
information
www.ihie.org
Let’s go find some patients….
• Building custom queries
• Using the Library
• Nested queries
• Taking action
www.ihie.org
IQ DEMO
Copyright © 2010 Indiana Health Information Exchange, Inc
www.ihie.org
80% of patient data is locked away in free text documents
Unlocking data buried within electronic health
records and other clinical documents is the
key to quality improvement, research and
outcomes analysis.
Provider notes
Operative notes
Admission and discharge summaries
Caths and Echos
Radiology and Pathology reports
Electronic, PDF, and Word documents
Copyright © 2010 Indiana Health Information Exchange, Inc
www.ihie.org
Traditional
Analytics
nDepth
Labs
Structured Data
ICD-9/10
Meds
CPTs
Echo’s
Op Reports
Radiology
Unstructured Data
Pathology
Provider notes
Endoscopy
Getting this data is time consuming and expensive
Highly qualified clinical staff
manually extract discrete
information from patient charts for
quality measures, research, clinical
improvements and outcomes
analysis.
http://www.healthcareitnews.com/infographic/infographic-clinical-claims-data-what-lies-beneath
www.ihie.org
Use nDepth to tap new sources of data
nDepth harvests discrete
information including hard-to-find
patient characteristics such as
social behaviors, symptoms and
family history.
www.ihie.org
Use nDepth to tap new sources of data
Enable deep population exploration
by trained medical staff as well as
non-technical subject matter
experts, bringing rich clinical
content to your fingertips.
www.ihie.org
How does nDepth work?
nDepth searches vast collections of data for
indicators hidden in free-text. These indicators,
called phenotypes, are a set of characteristics
that identify a specific condition or population.
There is a fine-tuned a library of complex and
highly accurate phenotypes using the Indiana
Network for Patient Care. This library of reusable
queries is included with nDepth and continuously
updated.
www.ihie.org
Find more patients with higher accuracy using nDepth
In a 2015 study on peripheral artery
disease, nDepth was able to find
400% more patients with greater
accuracy than using either lab reports
or structured electronic health record
(EHR) data. Specificity of the
detection algorithms was a combined
96%.
Copyright © 2010 Indiana Health Information Exchange, Inc
www.ihie.org
Why nDepth is Different
Complex NLP phenotypes rendered as simple recipes
Streamlined integration between structured and unstructured data
Unparalleled user experience
Integrated validation platform
Integrated machine learning
Real-time surveillance
Deliver to downstream data warehouse
Trained on world’s largest HIE
www.ihie.org
Easily identify opportunities to improve population health
Spend less time finding patients, including
hard-to-find characteristics such as social
behaviors, symptoms, and family history
Identify larger populations by using
unstructured clinical documents
User friendly interface to support data
exploration by non-technical users
Enable health systems, health information
exchanges, and other entities to get the most
value from patient data
www.ihie.org
Example nDepth Projects
Extract LVEF values for MSSP-ACO-33
Find patients with metastatic melanoma
Identify atypical hip fractures
Capture hypoglycemic events
Identify patients w/ family history of lung cancer
Map patient trajectory following cancer treatment
Detect treatment failure in insomnia
Find reasons for refusal of osteoporosis medications
Identify ‘triple negative’ breast cancers
www.ihie.org
Lets take a closer look….
Find patients using recipes
Refine using terminology search and filter
by report/institution/date/etc.
Visualize results
Define cohort/Perform Deep Text Analytics
and advanced information extraction
Compare and explore cohorts
Validate Results
Add to library
Copyright © 2010 Indiana Health Information Exchange, Inc
www.ihie.org
Find patients using recipes
Simple and intuitive “google” interface
Immediate access to complex queries
Automatically performs deep analysis and
extraction so you don’t have to
www.ihie.org
Refine using terminology search and filters
Easily find more patients faster
Enable deep exploration using
recommended synonyms
Supports standard terminologies;
SNOMED, MEDDRA, and LOINC
www.ihie.org
Fast and powerful text retrieval
Search hundreds of millions of records
at light speed
Query deep below the surface of
electronic health data
Identify, extract, and analyze data and
relationships
www.ihie.org
www.ihie.org
Visualize results
www.ihie.org
Perform deep text analytics
Tuned for human error detection
Misspellings, Punctuation, grammar
Sentence detection
Context
www.ihie.org
Advanced information extraction
Family history
Negation
Value extraction
Cancer and biomarker staging
Karnofsky scores
www.ihie.org
Integrated machine learning
Ngram analysis of terms in
Lung Cancer discovery
Analysis of most common 3-word sets
(trigrams) from 153,000 visit notes for
patients with an ICD-9 diagnosis of lung
cancer
www.ihie.org
Compare and explore cohorts
Explore intersection of
cohorts
Save and export new
cohorts
www.ihie.org
Built-in validation
Build studies to review results
of query
Assign to team members to
review results
Randomly selects records to
represent study
Highlights key words for easy
chart review
www.ihie.org
Reusable query library
When you get it just right…
Rinse and reuse proven queries
Share with teams or other
institutions
Speeds workflow of quality,
safety, research, and clinical
teams.
www.ihie.org
Finalize and Apply
Analytics
Final
Phenotype
Real-Time
Surveillance
Write observation back to EDW
Clinical Decision
Support
Epidemiology
Observational Research
Trial Recruitment
Prospective Studies
Patient Care
Quality Metrics
EDW
www.ihie.org
nDepth Demo
Copyright © 2010 Indiana Health Information Exchange, Inc
www.ihie.org
Save hundreds of hours extracting data for quality measurements.
nDepth Case Study
The collection and use of data to drive healthcare improvements has accelerated in the past few years, and along with it the
need for quality measurements. Compiling measurements for programs such as Medicare Shared Savings Program (MSSP)
and Hospital Acquired Conditions (HAC) is very labor intensive, requiring detailed review of clinical charts by experienced
staff.
The Challenge
In order to qualify for Medicare reimbursements, health systems and ACOs provide data for a variety of patient experience,
safety, preventive health and at-risk measurements for a randomly selected group of over 600 patients. Each year, clinicians
spend significant time carefully sorting through patient information in order meet these criteria. In the case of
measurement ACO-33, abstractors typically need 45 minutes or more on average per patient in order to extract the LVEF
value. Left Ventricle Ejection Fraction (LVEF) refers to the fraction or percentage of blood that is pumped out by the left
ventricles of the heart. This measurement provides an assessment of cardiovascular limitations and indicators for health
failure.
The Solution
A local partner chose to pilot nDepth™ to more efficiently report quality measures. nDepth extracted hard-to-find data from
unstructured documents quickly and accurately. This automated extraction improved quality workflow allowing valuable
resources to focus on validation instead of manual chart abstraction. Using nDepth, we have demonstrated the ability to
extract the LVEF values per CMS-MSSP measure ACO-33 at the scale of an entire health system in minutes.
Copyright © 2010 Indiana Health Information Exchange, Inc
www.ihie.org