Recalibrating Informatic’s

“True North”

William W. Stead, M.D.

Associate Vice Chancellor for Health Affairs
Chief Strategy & Information Officer
McKesson Foundation Professor of Biomedical
Informatics and Medicine

Disclosures: Co-inventor of two patient medical record products — one licensed to

McKesson, Inc., and one licensed to Informatics Corporation of America (ICA) — from
which I receive royalties through Vanderbilt University. Director of HealthStream, a public
company, compensated by meeting fees & an annual option grant.
 Idea – Take 1
Old New
Standardized
Unstructured
Data
Data
Capture & re-use
standardized data
 Road Map
 Idea – take 1
 The Need for a Shift
• Gap between today’s HCIT & what we need
• Root cause
 The Opportunity
• The idea – take 2
• Shifting EHR computational paradigm
 The Idea – take 3
 Action steps
Central Conclusions of NRC Report
 Current efforts aimed at nationwide
deployment of HCIT will not be
sufficient to achieve the vision of 21st
century health care, and may even
set back the cause…
 Success will require emphasis on
providing cognitive support
(assistance for thinking about and
solving problems).
 In the near term, embrace
measurable health care quality
improvement as the driving rationale
2009 for HCIT adoption efforts.
 Information-Intensive Aspects of the
IOM’s Vision for 21st Century Health Care
 Comprehensive data on patients’ conditions, treatments &
outcomes
 Cognitive support for health care professionals & patients to
help integrate
• patient-specific data
• evidence-based practice guidelines & research results
 Tools to manage a portfolio of patients & to highlight problems
as they arise
 Rapid integration of new instrumentation, biological
knowledge, treatment modalities, and so on into a “learning”
health care system
 Accommodation of growing heterogeneity of locales for
provision of care
 Empowerment of patients and their families in effective
management of health care decisions and their implementation
Stead WW, Lin HS. 2009. Computational technology for effective health care: immediate steps and strategic
directions. Comput Sci and Telecom Board, Nat Res Council. Washington: National Academies Press.
 Site Visits
 University of Pittsburgh  Partners Healthcare
Medical Center Boston, MA
Pittsburgh, PA
 Intermountain Health Care
 Veterans Administration Salt Lake City, UT
Washington, DC
 University of California,
 HCA TriStar San Francisco
Nashville, TN San Francisco, CA
 Vanderbilt University  Palo Alto Medical Foundation
Medical Center Palo Alto, CA
Nashville, TN
 Site Visit Observations
 Patient records are fragmented.
 Clinical user interfaces mimic paper without human factors
& safety design.
 Biomedical devices are poorly integrated.
 Systems are used often to document what has been done,
after the fact, for regulatory and legal uses.
 Support for evidence-based medicine and computer-based
advice is rare.
 Clinical research activities are not well integrated into
clinical care.
 Legacy systems are predominant.
 Centralization is the predominant method of
standardization.
 Implementations timelines are long and course changes are
expensive.
 Response times are variable and long down times occur.
 Root Cause: Mismatch between Computational
Technique & Scale of Problem
Automation

Connectivity Decision
Support

Data Mining
Stead WW. Electronic Health Records. In: Rouse WB, Cortese DA, eds. Engineering the system of healthcare delivery.
Tennenbaum Institute Series on Enterprise Systems, Vol. 3. Amsterdam: IOS Press; 2009.
 Overarching Grand Challenge: Cognitive Support
PATIENT CARE RESEARCH

Where
Virtual Medical
clinicians Patient Knowledge
want to stay

Decision Support

Medical Logic
Clinical
Transactions research
Where transactions
Health IT
chains us
Raw data Raw research
data

Workflow modeling and support, usability, cognitive support,

computer-supported cooperative work (CSCW), etc.
Stead WW, Lin HS. 2009. Computational technology for effective health care: immediate steps and strategic
directions. Comput Sci and Telecom Board, Nat Res Council. Washington: National Academies Press.
 Burning Platform: Overwhelming Complexity
Sets of Facts per Decision

1000
Proteomics and other
effector molecules

100
Functional Genetics:
Gene expression
profiles

10
Structural Genetics:
Human e.g. SNPs, haplotypes
Cognitive 5
Capacity Decisions by Clinical
Phenotype

1990 2000 2010 2020

Stead WW. Beyond expert-based practice. IOM (Institute of Medicine). Evidence-based medicine and the
changing nature of health care: 2007 IOM annual meeting summary,(Introduction and Overview, p. 19).
Washington, DC: The National Academies Press 2008.
 The Idea – Take 2
 Work at multiple scales to manage complexity
 Triangulate multiple signals for robustness

Satellite

Rain Gauge

Doppler Radar
 “mechanism”
“symptom”
 Shift EHR Computational Paradigm
OLD NEW
Sets of data from multiple
One integrated set of data
sources
Capture data in standardized Capture raw signal and annotate
terminology with standard terminology.
Current interpretation of multiple
Single source of truth
related signals
Seamless transfer among Visualization of the collective
systems output of relevant systems
Clinician uses the computer to Clinician & patient work together
update the record during the with shared records and
patient visit. information.
The system provides transaction- The system provides cognitive
level data. support.
Work processes are programmed
People, process and technology
and adapt through non-
work together as a system.
systematic work around.
Stead WW. Electronic Health Records. In: Rouse WB, Cortese DA, eds. Engineering the system of healthcare delivery.
Tennenbaum Institute Series on Enterprise Systems, Vol. 3. Amsterdam: IOS Press; 2009.
 Match Computational Approach to
Complexity of Data
Automation

Evidence-
based
Work lists advisors

Decision
Connectivity Support
Disease
Aggregate
management
EHR
dashboards

Data Mining
Stead WW. Electronic Health Records. In: Rouse WB, Cortese DA, eds. Engineering the system of healthcare delivery.
Tennenbaum Institute Series on Enterprise Systems, Vol. 3. Amsterdam: IOS Press; 2009.
Use different scales to support
a systems approach to care
Use different scales to support
a systems approach to care

Outcomes

Iterative
Improvement

Visualization
of Results vs. Plan

Consistent Process

Evidence-based Medicine
Use different scales to support
a systems approach to care

Outcomes

Iterative
Improvement

Visualization
of Results vs. Plan

Consistent Process

Evidence-based Medicine
Use different scales to support
a systems approach to care

Outcomes

Iterative
Improvement

Visualization
of Results vs. Plan

Consistent Process

Evidence-based Medicine
 Systems Approach to Care
2. Impact on Results
Fiscal Year
1. Number of Ventilator 2009 3. Mortality for Vanderbilt
Acquired Pneumonia (VAP) Results c/w Ventilator Patients Compare to
2008
Cases/Year at Vanderbilt all the other Hospitals
VAPs – Best in the U.S.
Prevented 108
Deaths
Avoided 16
300
$ Saved $4.3M
Hospital
Days
250
Avoided 1055
ICU Days
Avoided 431
200
Vanderbilt now # 1
O/E Vent Mortality
O/E Length of Stay
150 O/E Cost

100
2005 2006 2007 2008 2009

Source: UHC and Vanderbilt Data

 The Idea – Take 3
Abstraction Generalization
“formal
Model relationship”

Feature “package of
related
Set attributes”

“structured
Attribute information”

Data “raw signal”

Observation One Instance

 Terminology Mining & Management
External Standards
UMLS
SNOMED
FDB
…

VUMC Terminology Map

(Natural Language) Terminology
Editors

VUMC Health Language Server

Content (Curated)
Builders

Populate Operational Tools Generic Interface Populate Enterprise

HED Data Warehouse
Engine & Core of
StarForms
RxStar Interoperable Data Metadata
CHISL
…
Text String Matching: Ace Inhibitors

Ac?upril Avapro
Ac?retic Benazepril
Aceon Benicar
Altace Cand?artan
Atacand Capoten
Avalid* Capozid*

? = wild card character

* = accept any terminal ending
Low Level Facts Combine To Provide
Information

Which fuels the statistical

prediction of hypoglycemia

Product is carbohydrate
calories and fat per hour

Nutrition
provides
Nurse charts milliliters of TPN,
calories and fat
tube feeding or dextrose
per milliliter
administered per hour
 Action Steps
 Capture data from any source in any form & archive
as “raw signal”

 Keep data from different scales of biology or levels

of abstraction separate

 Aggregate and interpret data purpose by purpose

 Use statistical approaches to separate signal from

noise & correlate multiple weak signals

 Share algorithms & knowledge sources