This document discusses using big data analytics for operational and clinical decision support in healthcare. It outlines how analytics can help optimize decisions for patients, administrators, providers and policy makers by analyzing structured and unstructured data from various sources. The document proposes creating an operational decision support center and clinical decision support center to help coordinate patient care, anticipate needs, detect bottlenecks and support clinical decisions with data-driven insights. The goal is to move from rule-based systems to more precise, predictive and transparent decision making approaches.
6. Current
Vs.
Desired
Decision
Support
6
Current
Approach
Desired
Approach
Rule
based
system
High
rate
of
false
alarms
Missed
opportuni<es
Precise
and
Context
Sensi<ve
Workflow
Interrup<on
Automated
Data
Collec<ons
Mainly
structured
EMR
or
Claims
data
Structured
and
Unstructured
data
from
EMR,
sensors,
wearable,
behavior,
and
environmental
data,
and
condi<on
focused
social
network
data.
One
<me
measurements
of
physiological
sta<s<cs
Con<nuous
measurements
and
pathway
oriented
measurements
Low
transparency
and
accountability
Transparent
and
Accountable
to
pa<ents
and
care
team
7. Building
a
“Learning
and
Improvement
Engine”
7
Sympathy-‐
Man,
We
could
do
be`er
Empathy
–
I
feel
your
pain
Compassion-‐
Let
me
help
you
Source:
HCA
8. Big
Data
Analy<cs
must
be:
Valid:
hold
on
new
data
with
some
certainty
Useful:
Should
be
ac<onable
Unexpected:
non-‐obvious
to
consumers
Understandable:
humans
should
be
able
to
interpret
Measurement
is
useful
if
it
facilitates
ac,on.
Measure
what
is
important
to
customer.
9. Examples:
Opera<onal
Decision
Support
9
• Scheduling
and
Staffing
Assistance
• Predic<ng
and
alloca<ng
service
area
and
unit
capacity
• Predic<ng
bed/room
requests,
LOS
targets,
transport
services
• Op<mizing
Asset/Inventory
U<liza<on
• Reducing
claims
processing
cost,
error,
and
<me
• Reducing
Fraud,
Waste,
and
Abuse
10. Opera<ons
Decision
Support
Center:
Air
Traffic
Control
System
for
Opera<onal
Decisions
• Modeled
afer
your
security
opera<ons
center
and
IT
opera<ons
center
• Track
pa<ent
movements
and
oversee
opera<ons
and
throughput
• Proac<vely
an<cipate
needs
for
services
• Coordinate
staffing
and
scheduling
• Coordinate
admissions,
transfers,
discharge
planning
and
execu<on
• Reduce
cross
departmental
hand-‐off
issues
11. Informa<on
Flow
in
Care
Delivery:
Spagheh
h`p://www.ncbi.nlm.nih.gov/pmc/ar<cles/PMC3002133/
12. From
Spagheh
To
Lasagna:
Reduce
Unwarranted
Varia<ons
§ Source:
processmining.org
16. Examples:
Clinical
Decision
Support
16
• Be`er
decisions
using
con<nuous
physiological
streaming
data
• Op<mize
alarm
and
alert
sehngs
in
devices
and
applica<ons
• Care
Coordina<on
for
complex
co-‐morbid
condi<ons
• Hyper-‐personalized
engagement
• Crea<ng
checklists
based
on
predic<on
of
cri<cal
events,
early
warning
signs.
17. Clinical
Decision
Support
Center:
Air
Traffic
Control
System
for
Clinical
Decisions
• Think
of
this
is
like
a
department
like
Radiology
• Helps
with
near
real
<me
evidence
findings,
implementa<ons,
and
valida<ons
• Provide
data
driven
opinions
when
no
established
guidelines
exists
• Help
validate
output
of
analy<cs
with
exis<ng
guidelines
and
evidence
from
clinical
trials.
18. Prac<ce
Based
Evidences
(source:
greenbu`on.stanford.edu)
Prac<ce
Research
Applying
Evidence
Genera<ng
Evidence
19. 19
Example
App:
Care
Coordina<on
Assistance-‐
Find
gaps,
redundancies,
conflicts,
and
interac<ons
and
predict
adverse
events
20. 20
Example
App:
Find
Similar
Pa<ent
Pathway
for
Evidence
Based
Interven<ons
21. Virtual
Physical
Cloud
21
Healthcare
Data
Is
Time
Oriented
and
Diverse
EHR
Systems
Web
Services
Developers
App
Support
Telecoms
Networking
Desktops
Servers
Security
Devices
Storage
Messaging
Pa<ent
Surveys
Clickstream
HIE
Pa<ent
Networks
Healthcare
Apps
IT
Systems
and
Med
Devices
Pa,ent-‐Generated
Data
Medical
Devices
CDR
Mobile
PHI
Access
Audit
Logs
HL7
Messaging
Sensors
Departmental
and
Homegrown
Applica<ons
22. Disrup,ve
Approach
to
Diverse
Data
What
Happened?
What's
Happening?
Structured
RDBMS
SQL/Cube
Schema
at
Write
ETL
Search
Schema
at
Read
Universal
Indexing
Unstructured
Volume
|
Velocity
|
Variety
22
What
Might
Happen?
Predict/Prescribe
Opera,onalize
Machine
Learning
23. Data
Analy<cs
Infrastructure
23
DATA
SOURCES
IOT
DATA
IT
DATA
Acquiring
Enriching
(real
<me)
In
Mo<on
Data
Acquisi<on,
Analysis,
and
Engagement
(security
and
privacy
monitoring
and
audit)
Searching
Analyzing
(real
<me)
Delivering
Engaging
(real
<me)
At-‐Rest
Data
Acquisi<on,
Analysis,
Compose,
and
Deploy
(security
and
privacy
monitoring
and
audit)
APPS
DATA
Data
At-‐Rest
Historical
Data
Storage
Data
Discovery,
Explora<on,
Modeling,
Evalua<on
(At
Rest)
Compose
and
Deploy
(DevOps)
Streaming
Data
Storage
Data
In
Mo<on
24. 80%
of
healthcare
data
in
unstructured
text
High
velocity
<me
series
data
from
devices-‐
different
<me
zones,
different
<me
intervals
Variety
of
structured
formats
for
the
same
object
Unit
of
Measures
do
not
match
Data
Integra<on
and
Normaliza<on
24
Probabilis<c
Methods
to
validate
exis<ng
data
or
fill
in
missing
data
25. Data
Analy<cs
Knowledgebase
25
• Computable
Care
Plans
• Guidelines/
Rules
• Health
System
Workflow
• Data
Models
• Ontologies
• Treatment-‐
Outcome
data
26. Data
Analy<cs
Methods
26
• What
you
feed
into
the
algorithm
differen<ates
winners
from
averages.
• Sophis<cated
techniques
are
generally
worse
than
simple
methods.
Visualiza<on
Search/Explora<on
Sta<s<cs
and
Machine
Learning
Sofware
Engineering
28. Data
Analy<cs
Driven
User
Engagement
28
Task
Bo`lenecks,
Issues
Knowledge
Integra<on
User
Incen<ves,
Habits
Impacts
of
new
knowledge,
Trust
Detail
or
summary
or
Both
Responsive
Adap<ve
Managed
People
have
priori,es
beyond
just
geSng
treated.
Courtesy:
DJ
Pa,l
29. Lastly,
do
not
forget
Sofware
Engineering
prac<ces
29
• Tes<ng
• Privacy
and
Security
• Design
and
Refactoring
• Version
Control
and
Provenances
• Logs
and
Documenta<ons
• Produc<on
Deployment
Review
30.
Summary
30
Aim
of
Big
Data
Analy<cs
is
to
help
make
op<mal
decisions-‐
opera<onal
or
clinical.
Success
in
analy<cs
requires
mul<-‐disciplinary
skills.
Personalize
the
analy<cs
output
to
alter
current
behavior/habits.