Researchers and care providers wanted to have access to all of the patients` vitals signs (temperature, blood pressure, heart rate, and respiratory rate) but most of this data wasn?t recorded, only a few readings a day were posted to the patients Electronic Medical Record (EMR). The EMR isn`t meant to store such volume of data, let alone to perform any data mining on it. This session will describe the architecture of the solution that was implemented to collect these vital signs automatically from Bedside Medical Devices (BDMI), and store them into a temporary storage, then load them into a Hadoop cluster. The session will also cover how the team married this vital signs data in the HDFS (Hadoop File System) with the rest of the EMR data for our Principles Investigators (PI) in our research institute to search for correlations between administered medications, diagnosis, and vital signs readings. The session will describe the reasons behind the design decisions that were made, such as using a Cloud Hadoop cluster versus on-premises while maintaining HIPAA.
Designing IA for AI - Information Architecture Conference 2024
Elmallah june27 11am_room230_a
1. Using Hadoop for Vital Signs and EMR data
in Healthcare Research and Patient Care
Mohamed Elmallah, melmallah@chla.usc.edu
@melmallah
2. Session Plan
• Objectives
• CHLA
• Speaker
• The match
• Environment
• Why now
• Research use case
• The challenges
• What needs to happen
• Takeaways
• Q&A
3. Objectives
• Explain why Healthcare is a perfect domain for Hadoop
• Describe some of the many challenges
• Describe use cases in research and patient care
• Go over lessons learned and next steps
• How Hadoop vendors and Bigdata community can help
4. This Session will not
• dive into technical details, it will stop at the highlevel
architecture
• assume that you are a healthcare expert
• show you a production Hadoop with 40k node
• talk about petabytes of data (just a few TBs)
5. CHLA - History
• Founded in 1901, oldest children’s hospital in CA
• ~12K inpatient visits / year
• ~320K outpatient visits / year
• ~65K ED visits / year
• ~16K pediatric surgeries / year
6. CHLA - Clinical
• U.S. News and World Report’s Honor Roll
• Ranked in all 10 pediatric subspecialties
• ~5,000 employees and ~600 medical staff
• 365 active licensed beds, 85% private
• 80+ intensive care beds
7. CHLA - Clinical
• Affiliated with the Keck School of Medicine, USC
– http://www.chla.org
• The Saban Research Institute, ~100 researchers and
physicians (and data scientists :)
– http://www.chla.org/saban
8. Speaker
• Ex-developer (e.g. Manager) of Enterprise Apps and
Architecture team
• 2+ Year with CHLA
• Worked for Cedars Sinai and Kaiser Permanente
• Ex- DBA, Support Engineer, ERP Implementor: Oracle,
Qualcomm
9. Hadoop and Healthcare
• Healthcare is so different
– HealthIT for many years has been lagging
– HIPAA: Health Insurance Portability and Accountability Act
– PHI: Protected Health Information
– Public Cloud: Not there yet
– In many technology areas: A few niche players
– Some organizations are powered by research and academia
10. Hadoop and Healthcare
• Healthcare is NOT so different
– Data is growing
– HealthIT is trying to save money
– There is continued spending in HealthIT
– EMR/EDW is not a feasible solution to retain and deliver information
– Care providers are dependent on hosted applications
– Care providers are paying more attention to their social network
footprints.
– Other great candidate domains for bigdata are lagging too
11. Environment
• EMR: Cerner Millennium: PowerChart, FirstNet, SurgiNet
• Data Mart/Reporting: PowerInsight
• Patient Registration: McKesson Star
• DataMart/Dept BI/Reporting: HBI, HPM, SpotFire
• ERP: Peoplesoft Fin/HR/Materials/Contracts
• Bed Management: Teletracking
• Placement/Transfer/Scheduling: Central Logic
13. What is Changing?
• ACA, MU
– Patient Portals
– Diagnosis-Related Group (classification of cases)
• Mobile (more access, more data)
• EMR system is becoming more open
– API to access data
16. What is Changing?
• Cloud-based solutions (SaaS)
• More integration demand with other data sources
– BMDI
– Lab Outreach
– Using non-CHLA data (e.g. NIH - National Institutes of Health
data)
• Increased demand for dashboards, trends, sparkline
19. What are we doing?
• Expanding our in-house development team, not only on
middle tier and UI, but data and BI.
• Using consultants and professional services in a smarter
manner
– Insist on open data access
– Insist on open architecture (SOA: Web Services, or at least
direct database access)
• Partner with bigdata vendors (small and large)
20. NICCU/PICU/CTICU Data Acquisition
Use Case
• EMR keeps a few on-demand snapshots through proprietary
integration with devices
• Researchers wants continuous access to Vital Signs from:
– Patient monitors: Philips MP70 conventional modules (HR,
SPO2, BP, TCOM), Philips MP70 Vuelink
– Cerebral/Somatic Oximeters: INVOS
– Cerebral functional monitor (CFM): Olymbic Brainz
– Respiratory monitors: Respironics NM3, BiCore II
– Cardiac monitors: Aesculon and ICON
– Ventilators: Servo, AVEA, HiFreq Oscillatory, etc.
– Infusion pumps: Medfusion
– Dialysis machines: Prisma
21. NICCU/PICU/CTICU Data Acquisition
Use Case
• Collect data (numeric and wave form) from BMDI into a SQL
Server DB
• Data has some PHI data
• Data is augmented with metadata (dates, notes, etc.)
• Instead of each vendor connecting directly to devices, we
will have a centralized, complete, controlled and in-house
repository
– Each device has a limited number of ports
22. Data Acquisition Architecture
Short Term
SQL Server
Short Term
SQL Server
Web ServicesWeb Services
Trend ViewerTrend Viewer
Data
Access
Layer
Data
Access
Layer
Live
XML
Stream
Live
XML
Stream
Hubs + XML
Serialization
Service
Hubs + XML
Serialization
Service
PlaybackPlayback
Access &
Auditing
Access &
Auditing
PI
23. Challenges
• We have a development team, but mainly in .Net Web
Development, have been preparing them
– Data (modeling, ETL, BI, Analytics) training
– Considering Java training
• Getting the buy-in from business side
• Many of the niche healthcare vendors are small, not
Hadoop-ready
• The small but successful bigdata vendors are busy, and the
big ones are usually expensive
24. Our Next Steps
• Marry the Vital Signs to the Patient Chart and its Events
(e.g. administration of medication), and access one from
the other, keeping HIPAA in sight.
• Once clinical/research use case is proven, low hanging fruits
in social networking, and operations/finances (claims,
payroll, etc.) should be next
– Not to reinvent the wheel
– Use analytics/algorithms already proven
– Partner with other care providers
25. Our Next Steps
• Expose practical Bigdata to end-users and business
stakeholders
• Bigdata is not an IT thing
• Hadoop is an echosystem not just one product
26. What Needs to Happen
• Healthcare focus
– Bigdata vendors need to understand the domain
– Repeating the word “HIPAA” many times is not enough
• Fill in a gap
– BI
– Archiving
– Data Governance
– Enterprise Search
– Social Networking
• Agile implementations
27. What Needs to Happen
• HIPAA Ready
• Offer quick start packages
– Help Hospitals to teach their HealthIT staff Hadoop
– Teach Hadoop/MR to Oracle/.Net Developers
– Don’t forget the administrators
28. Objectives
• Explain why Healthcare is a perfect domain for Hadoop
• Describe some of the many challenges
• Describe use cases in research and patient care
• Go over lessons learned and next steps
• How Hadoop vendors and Bigdata community can help
29. For more info
• MUCMD: Meaningful Use of Complex Medical Data
Conference, August 16-17, 2013 Los Angeles
– http://mucmd.org
• VPICU: Virtual Pediatric Intensive Care Unit
– http://www.picu.net