clinical looking glass - amazon...
TRANSCRIPT
Clinical Looking Glass
Vignettes
• Summary
• What is CLG – Core Concepts
• Use Case – Diabetes
• Use Case – Pharma Analytics – Research
• Use Case – Time in Range Cost Savings Epo
Use
• API and Exporting Capabilities
Use Case – Predictive Analytics
• Use Case – PCMH Certification
• Architecture
• Q & A
“Longitudinal Analytics
for any Market”
Summary - Definition
Clinical Looking Glass is a powerful
analytic tool providing:
• On-demand healthcare analytics with
• Longitudinal and statistical capabilities
• “Bolt-On” analytical tool is EMR
agnostic – can sit on any data
warehouse
Summary - About CLG
• Developed and used by clinicians, administrators and
researchers
• Able to guide point of care clinical decision making, inform and
evaluate effects of interventions and treatments and drive
population health management
• Promotes academic and clinical rigor and innovation
• Scaleable , leverageable and valuable tool for the entire
healthcare universe: pharma, providers, payers and research
• A key differentiator in era of outcomes based performance and
greater demand for improved healthcare quality and reduced
cost (value)
Summary - CLG overview
• Started in 2002 with 3 FTE’s as part of Emerging Health
• 2013 – 22 FTE’s plus Emerging Health support
• Approx. $5.1 million annual budget and $40 million spent over past 10 years
• October 2013 Commercialization rights granted to Streamline Health and the product is known as – Looking Glass
• This is a review of the Montefiore Experience
Summary - Montefiore
Large academic medical center (Bronx, New York), 11th Largest
provider in the U.S.
4 hospitals, 1491 beds
94,000 discharges
280,000 emergency department visits
21 clinics , 2 million clinic visits
2010 - In 2010, 499,148 unique patients / 1.3M Bronx
Population = 38%
225,000 covered lives through IPA
The only Pioneer Accountable Care Organization in New York
State 2012
University Hospital of the Albert Einstein College of Medicine
Summary - CLG Development History
• Version 1.x (2002)
– Group comparison for 1 subject
(LOS)
– Embedded statistics engine
• Version 2.x (2004)
– Cohort Builder
– Time to Event for analysis
• Version 3.x (2008)
– Event Canvas with data mapping
– List any data element
– On-demand OLAP data cubes
• Version 4.x (2011-2013)
– Study Designer with iPad Viewer
App
– Performance
• 700 users trained – 5,000 analyses run per month
• Quality Improvement – Utilized in diabetes, asthma, LOS, hospital workflow, QI studies
• Research – Over 40 peer reviewed journal articles
– $18 million of Montefiore/AECOM research grants enabled by CLG analytics
• Education – Mandatory part of resident training, Albert Einstein College of
Medicine
– Utilized for CME at Montefiore
• Supports transformation of Montefiore into an ACO for 2012 and beyond
Summary - Utilization at Montefiore
What is CLG?
Core Concepts “Longitudinal Analytics
for any Market”
Story time
The Geriatrician and the Surgeons
“Use local anesthesia instead of general for elderly patients needing gall
bladder procedures to reduce complications.”
Physician decision-making did not require IT or analyst team
Creative power unleashed in real-time
Analysis model was
patient-centric
Story themes
Information must be “Patient-Centric”
Business-Centric “Patient-Centric”
How many admissions in 2011? How many unique individuals admitted in 2011?
How many visits for hypertension? How long does it take us to control hypertension?
How much do we spend on drug X? How many improved when on drug X?
Patient-centric cohort is key concept
1/1/2010 1/1/2011 1/1/2012
0
0 = index date
(start therapy)
0
0
0
0
0
0
0
0
0
1
2
3
4
5
6
7
8
9
10
Patient #
What % of new diabetic patients were controlled in the year 2010? 4 / 10 = 40%
Diabetes Control
= outcome
(achieve lab value)
0 = patient experience
Cohort concept (cont)
Enrollment 1 Year 2 Years
0
0 = index date
= outcome
(start therapy)
(achieve lab value)
0 = patient experience
0
0
0
0
0
0
0
0
0
1
2
3
4
5
6
7
8
9
10
Patient #
What % of new diabetic patients were controlled within 1 year? 5 / 10 = 50%
Diabetes Control
(same data, re-sorted)
Definition
Laboratory results, in/out patient admissions, pharmacy,
radiologic data and all other types of events collected in
the Electronic Medical Records must be:
• chronologically (temporally) related and spun into a
• defined patient group (cohort paradigm) whose
• time period to outcome (trajectory) defines the…
Bottom Line: Quality of their Care
Clopidogrel example
Heart
Attack (In-Patient)
Clopidogrel (Out-
Patient
Prescription)
Prilosec (PPI) Death Rate
No
Prilosec (PPI)
Death Rate
0 30 900 Time to Event Time to Event Time to Event 365
Outcome
Temporal events leading to outcome
0 0
CLG study components
Groups I Analysis
CLG study = groups + outcome analytics
= “Index Date” I
Male diabetics I Death in 1 year
Female diabetics I Readmission 1 year
Use Case:
Diabetes Management “Longitudinal Analytics
for any Market”
Diabetes example: temporal map
New
Diabetic
A1C > 9
Hospitalization
Rate
Hospitalization
Rate
0 181 365 Time to Event Time to Event Time to Event 365
Outcome
Blackout Period Good Control
180
A1c <= 7.0
0
1) Build Cohorts 2) Study Outcome
Cohort 1 Index: A1C < 7
Cohort 2 Index: A1C > 9
Diabetes demo
Done: go to next
Diabetes study
Diabetics HgbA1c >9.5
Age >=65 I Achieve HgbA1c <7
(180 – 365 days)
Diabetes example: temporal map
New
Diabetic
A1C > 9
Hospitalization
Rate
Hospitalization
Rate
0 181 365 Time to Event Time to Event Time to Event 365
Outcome
Blackout Period Good Control
180
A1c <= 7.0
0
1) Build Cohorts 2) Study Outcome
Cohort 1 Index: A1C < 7
Cohort 2 Index: A1C > 9
Diabetics achieving
HbA1c of <=7 I
Diabetics achieving
HbA1c of >=9.0 I
Hospital Admission
Achieved a HgbA1c < = 7
Significant
difference
Financial impact of “good” diabetes mgmt
Average Annual Increased Difference in Inpatient Cost:
(Patients Under Bad Control vs. Good Control)
$760 per patient
1,086 (Patients Under Bad Control) * $760 = $825,360/year
Impact of “good” diabetes mgmt
A reduction of almost 1/3 in hospitalizations
Only ½ the story
More than a study…
It’s a tool for targeted remediation.
Remediation tool
Remediation list of patients
Implications of cohort view
335 (31%) Missing
Unlimited temporal complexity (US Patent#: 7917376 )
• Logical subgroups
• AND/OR/NOT
• Temporal rules b/w events
• Anchor events in calendar time • Group definition
• Outcome definition
Use Case:
Pharma Analytics –
Outcomes Research “Longitudinal
Analytics for any Market”
Recapitulate research in the Bronx
JAMA, March 4, 2009—Vol 301, No. 9 937
CLG in action: research scenario
Clopidogrel (Plavix):
• reduces platelet stickiness and clotting
• reduces risk of repeat myocardial infarction
• increases risk of GI bleed
Proton Pump Inhibitor (PPI - Prilosec):
• reduces risk of GI bleed
• possibly reduces efficacy of Clopidogrel
Clopidogrel example: temporal map
Heart
Attack (In-Patient)
Clopidogrel (Out-
Patient
Prescription)
Prilosec (PPI) Death Rate
No
Prilosec (PPI)
Death Rate
0 30 90 Time to Event Time to Event Time to Event 365
Outcome
Temporal Events Leading to Outcome
0 0
Clopidogrel demo
Done: go to next
Pts w MI
on clopidogrel
no PPI I
Pts w MI
0n clopidogrel
on PPI
I
Death
Repeat MI
Mortality rate (PPI vs. No PPI)
Mortality (PPI vs. No PPI)
Readmit with MI (PPI vs. No PPI)
Readmit with MI (PPI vs. No PPI)
Use Case:
Time in Range Cost Savings in Epo Use
“Longitudinal Analytics
for any Market”
Manage RBC with Epo
• Care Goal
– Manage the red blood cell count (RBC) for
patients with kidney failure in a safe range
• Therapy
– Erythropoetin stimulates production of RBCs
– Keep hematocrit (Hct) value in 29 to 32 range
• Concerns
– Hct > 32 associated with greater mortality
– Epo is expensive, don’t over prescribe
Clinical management question:
Was patient X’s hematocrit values well managed
over one year of Epo therapy?
Average?
Latest?
% of days in target range
What are the relevant hematocrit
ranges?
High
Low
Target
Time
Hem
ato
cri
t
32 -
29 -
X
X
X
X
X
interpolation carry forward
missing missing
How do you estimate Hematocrit
between blood collections?
Time
X
X
X
X
X
interpolation carry forward
missing missing
Low
Target
High
Hem
ato
cri
t
32 -
29 -
Time in Range Interpolation
CLG Time in Range method
• Longitudinal perspective
• Evaluates quality by counting time spent in
appropriate range of Hematocrit values
• Saves money while protecting patient’s health
by focusing on the relevant longitudinal metric
API and Exporting Capabilities
“Longitudinal
Analytics for any Market”
Exporting Temporally Enriched Data for
External Analysis
• Allow for Import of Temporally Enriched Data in:
– SAS
– SPSS
– STATA
– R
– Excel
• Advantage of this Technology
– temporally enriched data can be exported without requiring post
processing which may compromise patient privacy
– export results (counts of hospitalization) not the dates of service to
create intervals. This assures privacy throughout the analysis.
Analysis in your Favorite
Statistical Package
Niche BI Solution • Registry • DM System • Scorecard • Dashboard
CLG
API
CLG SDK
cohort method
STUDY
Application Programming Interface (API)
Platform for all clinical analytics solutions
Application
Programming
Interface (API)
enables:
• Longitudinal
perspective
• Statistical
sophistication
• Clinician
engagement
• Rapid BI
deployment
ACO Partner
ACO Partner
ACO Partner
Clinical & Financial Applications
Eagle MOC/MOTCentricity
EnterpriseSAP
Centricity
EMR
TREKS
PICIS
RIS-ICHR
ManagementKRONOS
EPF Ultra Triple G
Enterprise Data Warehouse
Apollo Peri-NatalOTTR
(Transplant)Wellsoft
CLG
Cohorts:
· Diabetes
· Cancer
· CHF
· Frequent Flyers, etc...
Events:
· Admissions
· Visits
· Transfers
· Claims
Measures:
· Smoking
· Diagnoses
· Procedures
· BP, Labs, Findings
Outcomes:
· Mortality
· Readmission
· Relapse
· Wellness
Clinical Rules
Governance
Performance
Dashboard
Diabetes
Registry
LOS
DashboardPredictive
Solutions
Report
Cards
Enterprise Objects
“What If” Analysis &
Metric Refinement
CLG Vision: Platform for All Clinical Analytics
Use Case:
Predictive Analytics CLG in Operations
“Longitudinal Analytics
for any Market”
Predicting 30 day readmission
• Predict likelihood of readmission at discharge
• Local model discovery using local clinical
intuition
• Agility in model refinement
Readmission Study Cohort
Readmit Risk?
Predictive Model
Today
Model development
Historical Data, Clinical Variables, Social Factors
Future Past
MRN
Probability
Score Influencers INDEXAdmission INDEXDischarge
00001 0.97 ER6mCount;Inpat6mCount;FluidAndElectrolyteDisorders 8/27/2012 8/28/2012
00002 0.83 ER6mCount;CharlsonScore;FluidAndElectrolyteDisorders 8/25/2012 8/28/2012
00003 0.72 FluidAndElectrolyteDisorders;ER6mCount;Inpat6mCount 8/26/2012 8/28/2012
00004 0.68 ER6mCount;CharlsonScore;FluidAndElectrolyteDisorders 8/19/2012 8/28/2012
00005 0.59 CharlsonScore;Obesity;ER6mCount 8/17/2012 8/28/2012
00006 0.58 CharlsonScore;FluidAndElectrolyteDisorders;Coagulopathy 8/12/2012 8/28/2012
00007 0.45 CharlsonScore;WeightLoss;Coagulopathy 8/23/2012 8/28/2012
00008 0.45 FluidAndElectrolyteDisorders;DeficiencyAnemia;Inpat6mCount 8/19/2012 8/28/2012
00009 0.43 Obesity;CharlsonScore;DrugAbuse 8/24/2012 8/28/2012
00010 0.40 FluidAndElectrolyteDisorders;DeficiencyAnemia;OtherNeurologicalDisorders 8/19/2012 8/28/2012
00011 0.37 FluidAndElectrolyteDisorders;DeficiencyAnemia;DrugAbuse 8/24/2012 8/28/2012
00012 0.37 CharlsonScore;FluidAndElectrolyteDisorders;DeficiencyAnemia 8/14/2012 8/28/2012
00013 0.32 FluidAndElectrolyteDisorders;CharlsonScore;DeficiencyAnemia 8/25/2012 8/28/2012
00014 0.32 CharlsonScore;FluidAndElectrolyteDisorders;Coagulopathy 8/25/2012 8/28/2012
00015 0.30 CharlsonScore;Coagulopathy;DeficiencyAnemia 8/27/2012 8/28/2012
Probability score of readmission
Intervention Team Daily Workflow
CLG A P I
MRN Probability Score
00001 0.97
00002 0.83
00003 0.72
00004 0.68
• Med reconciliation
• Setup referral appts
• Needs assessments
• Log actions
Care Mgmt System
Predictive Model reclassified patients
• High risk discharges went from 3000 to 500
• Readmit rate for high risk went from 19% to 46%
Efficacy of intervention under evaluation
Care Transition Team
Use Case:
PCMH Certification “Longitudinal Analytics
for any Market”
First at Montefiore: PCMH Level 3
South Bronx Health
Center for Children
and Families
… and
CLG Power Users!
Shareable CLG Study to facilitate PCMH Level 3 certification
Additional CLG Uses at SBHC and NYCHP
CLG
Quality Indicators
Patient Worklists
Outcomes Research Grant Seeking and Reporting
Program Evaluation
Architecture “Longitudinal
Analytics for any Market”
CLG SOA Integrated Architecture
Technologies Powering CLG
• User Interface
– ASP.NET ( Including AJAX Framework)
– Silverlight 5.0
– Infragistics , Telerik and Component Art controls.
– Silverlight Toolkit.
– MVC 4.
– PRISM
– IPAD - IOS
• Authentication/Authorization
– Single Sign On.
– Membership Provider Model.
– Forms Authentication.
• Services Layer
– Windows Communication Foundation 4.0
– Windows Workflow Foundation 4.0
– MVC 4 API Controllers
• Database
– SQL Server 2008 R2/ 2012
Backend Modules/ORM – Entity Framework – MEF – TEMPORAL Engines.
Statistics – Revolution R
Server/Operating Systems – Windows Server 2008 r2
Languages – C# 4.0 – VB.NET 4.0 – Visual C++ 4.0 – R – JavaScript. – JQUERY. – T-SQL/ Ansi-SQL – IOS 5.1
Framework – Net 3.5SP1, .NET 4.0 – PRISM – MEF – MVC
Process and Tools
Process:
– Scrum agile workflow model
– 4 week cadence
– Definition of Done
– Continuous integration
– Nightly build and auto deploy
– Automated system testing
– Data mapper to facilitate implementation
Tools:
– Visual Studio 2010 , 2012
– Team Foundation Server 2010 SP1
– SQL Server 2008 R2/2012
– Expression Blend
– Visio for UML modeling
– IBM DOORS for requirement model (in promo video)
72
Questions “Longitudinal
Analytics for any Market”
http://exploreclg.montefiore.org
74
If interested in Commercial
Contact I provide the contact
Clinical Analytics
Screenshot: Login Page
Sales Contact
Daniel McDuffie Vice President, Sales
Streamline Health
1230 Peachtree Street NE | Ste. 600 | Atlanta, GA 30309
[email protected] | www.streamlinehealth.net (C) 404.771.1781 NASDAQ: STRM