doing the impossible with Diver
M
ANAGING
C
LINICAL
B
IG
D
ATA
4 TIPS TO DEPLOY ACTIONABLE CONTENT TO YOUR USERS
Aaron McMaster | Sr. Data Science Professional Munson Healthcare
ABOUT
M
UNSON
H
EALTHCARE
8 Hospitals located in
Northern Michigan
871 Licensed beds
7,000 employees
160 Employed
Physicians
Cerner EMR
McKesson Star
billing/rev cycle
A
BOUT
M
E
Been with Munson Healthcare for 12
years
Functions as a conduit between IT and
clinicians
Drives the adaption of analytics and
support of process improvement efforts
Responsible for guiding multidisciplinary
teams on the development of clinical intelligence applications to support evidence based decisions
As a non-clinician having the ability to
increase the quality of care to our patients incredibly engaging and rewarding
Our goals & challenges
Tip 1. Structure for Success
Tip 2. Efficient Extraction
Tip 3. Big Data Processing
Tip 4. Speed to Users
What we learned
O
UR
CBI G
OAL
Unlock data within the EMR to facilitate getting the right information for decision making, to the right people, at the right time.
Workflow decision:
At the
bedside Not at the
bedside
DI
THE CHALLENGE
Lots of data (TB’s)
Highly normalized Complex schema
Multiple & inconsistent levels of parent/child
relationships
Inconsistent usage, integration to workflows Data integrity challenges
Many ambiguous customer requests
Challenges in request prioritization
TIP 1: STRUCTURE FOR SUCCESS
F
OUNDATION FOR
R
APID
D
EVELOPMENT
Focus on the
customers request
Get the right people
involved, cannot be driven by IT priorities
Balance between
getting value now, and making data available for the future
Drive towards rapid
development lifecycles of CBI projects projects
W
AREHOUSING
T
HEORY
Hybrid Model:
• Focus on specific customer requirement
• Acquire most accessible data when accessing a
source system
• Cleanse/validate on a project-by-project basis • Focus on reusable content
Speed-to-Value Approach Traditional Approach
TIP 2: EFFICIENT EXTRACTION
G
ET THE
D
ATA
!
No. 1 priority: do not disrupt the production
system
Stability is the key to success
Create monitoring tools
O
RACLE
: M
ATERIALIZED
V
IEWS
“An object that contains the result
of a query…”
Looks and functions like a table
Updated at some regular frequency Low/negligible impact on the
source system
Optimize by balancing the update
frequency (inverse relationship
between network traffic generated & quantity of data transferred)
O
RACLE
: M
ATERIALIZED
V
IEWS
EMR
Oracle Database: Materialized ViewsDI Servers
(stores historic data)
TIP 3: BIG DATA PROCESSING
B
IG
D
ATA
R
ESOURCE
M
ANAGEMENT
Process only the data that is
required to meet the customers needs
Leverage DI tools/objects to
facilitate this
As a developer, you need
access to tools to monitor system performance during testing:
CPU utilization
Memory utilization
E
XAMPLE
: B
ASIC
S
CRIPT
S
ELECTIVE
P
ROCESSING OF
D
ATA
1. Splitting the repository by date
2. Transform directory file list to include only specific files, create File List Input
3. Can be used to process different amounts of data depending on day-of-the-week
Transform
C
REATE
S
MALL
, E
FFICIENT
M
ODELS
1. Split adhoc into multiple files
2. Transform file list to create parameters for model building (i.e. filename, model number, etc…)
3. Pass parameters to loop & build models
Transform Data
File List Parameter File
TIP 4: SPEED TO USERS
T
WO
T
YPES OF
S
PEED
D
ESIGN FOR
S
PEED OF
U
SE
Think process improvement – what is the best use of
data/technology to meet the customers requirements
How does the solution fit into the customers most
efficient workflow?
The solutions architecture depends on customer
needs
Near real-time (NRT) Hourly
Daily
Manage increase in granularity when creating
repositories and models
I
NCREASE
P
ROJECT
I
TERATION
S
PEED
Think process improvement not report
remediation when defining customer
requirements
Invest in developing regular customers
By making data accessible from the source
system even though its not validated
Perform data validation on a project by project
basis
G
OAL
: B
ALANCE
O
BJECTIVES
Maximize
Speed
Maximize
Automation
Minimize
Complexity
RESULTS
MHC C
LINICAL
D
ATA
W
AREHOUSE
Pharm Encounter PersonTasks
Events
Order CatalogOrders
Production
Rad MBO DocOUR RESULTS
Data and reusable content is available for use in
many upcoming projects
Ease in ongoing prioritization Faster project iterations
Clinician workflow changes from “static” reports to
LESSONS LEARNED
Communicate to customers that CBI projects are
not just report remediation
The timing of initiating customer engagement is
extremely important
Make developing regular customers must be a
priority
Focus on delivering the customers content, try
not to get too distracted
However; if a task will most likely add value at
WHERE WE GO FROM HERE
Develop and publish more content faster
leveraging existing data & programs
More clinical content encompassing the
continuity of care
Regular linking of orders to results
More near-real-time content to clinicians More predictive content to clinicians