hpe and hortonworks join forces to deliver healthcare transformation
TRANSCRIPT
1 © Hortonworks Inc. 2011 – 2016. All Rights Reserved
HPE and Hortonworks join Forces to Deliver Healthcare TransformationRichard ProctorGM Healthcare, HortonworksJohn SansonAlliance Manager, HPE
2 © Hortonworks Inc. 2011 – 2016. All Rights Reserved
Agenda• Introductions• Healthcare - Current state• The Problem: Current data architecture• The Solution: Modern data Architecture with Hadoop• Healthcare Use Cases• HPE Hadoop Reference Architecture• Q&A
3 © Hortonworks Inc. 2011 – 2016. All Rights Reserved
Data Driven Organizations-Shifting the Data Paradigm
o Regulatory-centrico Manual data review/collectiono Repressive data silo’so What data to store?o Expensive storage w/limited options
All data types Organization & Patient centric Store everything! Inexpensive storage with lots of options
Data as an independent business process(Silo’s of data)
Reactive reporting
Data as a byproduct of patient care Prospective analysis
Primary audience – Healthcare organization
Secondary audience
Secondary audience
Primary audience – regulatory agencies
Current data process with latent architecture
Modern Data Architecture with Hadoop
4 © Hortonworks Inc. 2011 – 2016. All Rights Reserved
The problem: Current data architecture under pressureAP
PLIC
ATIO
NS
DATA
SYS
TEM
REPOSITORIES
SOU
RCES Existing Sources
ADT, Patient Accounting, GL, Payroll, Physician entry, Core measures, Patient Sat, AHRQ, External
benchmarks,, Clinical Systems (ED, Radiology, PACS), Other Sources (Clinics, home health, Ambulatory,
LTAC’s)
RDBMS EDW MPP
Business Analytics Custom Applications Packaged
ApplicationsOLTP, ERP, CRM Systems
Unstructured documents, emails
Clickstream
Server logs
Sentiment, Web Data
Sensor. Machine Data
Geolocation
Value in New data sources
• Limited Application interaction• Costly to Scale storage• Silos of Data• Inability to manage new data sources• Schema on Write vs. Read
5 © Hortonworks Inc. 2011 – 2016. All Rights Reserved
New Data Paradigm- Winners will be Proactive not Reactive!
2.8 zettabytesin 2012
44 zettabytesin 2020
N E W
1 zettabyte (ZB) = 1 million petabytes (PB); Sources: IDC, IDG Enterprise, and AMR Research
Clickstream
PROD, LOE, SHE
Web & social
Geolocation
Internet of Things
Server logs
Files, emails
Transform every industry via full fidelity of data and analytics
Opportunity
T R A D I T I O N A L
LAGGARDS
LEADERS
Ability to Consume Data
Enterprise Blind Spot
90% of all information created by humans originated in the last 2 years
6 © Hortonworks Inc. 2011 – 2016. All Rights Reserved
7 © Hortonworks Inc. 2011 – 2016. All Rights Reserved
Thank you for the diagram, Robert Wood Johnson Foundation, 2014
7
Comprehensive Health Management
80% of healthcare determinants lie outsidethe US healthcare delivery system
Can healthcare systems expand into these other areas, and become true public health systems?
8 © Hortonworks Inc. 2011 – 2016. All Rights Reserved
Imagine if your physician could say this to you…
“I can make a health optimization recommendation to you based on…
The latest clinical trials Your genomic make up and family history The local, regional, national & International data about patients just like you The real-world health outcomes over time of every patient like you the level of your interest and ability to engage in your own care
I can then tell you within a specified range of confidence, which treatment has the greatest chance of success for you.”
Thank you Dale Sanders at Health Catalyst for the quote
9 © Hortonworks Inc. 2011 – 2016. All Rights Reserved
Learning from other industries!Step 1:
Retail Approach-• Build a Central Data Repository to store and process all longitudinal patient history “Data Lake”• Analyze information across multiple touch points and episodes of care within the Healthcare
system or across Providers, Clinics, pharmacies, Home Health, and specialty providers
Step 2:
Predictive Modeling/Analytics-• 360 degree Patient view used to build Predictive Models around high cost/high quality patients
e.g. long-term chronic disease management• Data gathered on patients after they leave the care setting enables more timely visibility into
patient behavioral/clinical changes.
10 © Hortonworks Inc. 2011 – 2016. All Rights Reserved
Original 24 architects, developers, operators of Hadoop from Yahoo!
ON
LY 100open source
Apache Hadoop data platform
%Founded in 2011
HADOOP1STprovider to go public
IPO Fall 2014 (NASDAQ: HDP)
subscriptioncustomers800 employees across
740+
countriestechnology partners1350+ 17
TM
Hortonworks Company Profile
11 © Hortonworks Inc. 2011 – 2016. All Rights Reserved
Fastest growing Fortune 1000 customer baseCustomer Momentum• More committers to Hadoop than any other distribution• Non proprietary = FREE with no vendor lock in• Fastest software company in history to 100M revenue- 4 years (Barclays)
Largest Cluster in North America
32,000 NodesLargest Cluster in Europe
1,000 Nodes
Some notable migrations include many of the early adopters of Hadoop:
© Hortonworks Inc. 2011 – 2014. All Rights Reserved
Experience at Scale80,000 nodes under contract
Largest Known Cluster in APAC
400 Nodes
• 30+ customers migrated from other distributions
12 © Hortonworks Inc. 2011 – 2016. All Rights Reserved
Use Cases - Providers
13 © Hortonworks Inc. 2011 – 2016. All Rights Reserved
Historical and Legacy system data offloadHealthcare
Problem Legacy system licensing when moved to new EMR 22 years of data for 1.2 million patients ~ 9 million records Data on legacy system was not searchable nor retrievable Cost to move data to new system not feasible Difficulty in combining data from 2 systems
Solution Unified repository provides data to both researchers & clinicians “View only” legacy system retired, saving $500K 9 million historical records now searchable & retrievable Records stored with patient identification for clinical use, same data presented
anonymously to researchers for cohort selection
MPP
SAN
Engineered System
NAS
HADOOP
Cloud Storage
$0 $20,000 $40,000 $60,000 $80,000 $180,000
Fully-loaded Cost Per Raw TB of Data (Min–Max Cost)
Hadoop Enables Scalable Compute & Storage at a Compelling Cost Structure:
Storage Costs and licensing reduction of latent systems
$500,000/year
5 x the amount of usable storage, plus 5 x processing power, for about 30% of the cost of traditional technologies,”
14 © Hortonworks Inc. 2011 – 2016. All Rights ReservedPage 14 © Hortonworks Inc. 2011 – 2015. All Rights Reserved
A Healthcare Data PlatformFor a Single View of the Patient
Mercy
– 35 hospitals and 700+ clinics, 1 million patients annually, Multi State coverage– One of the earliest and largest EPIC deployments
Objective #1 Improved Documentation- Billing & Coding AccuracyObjective #2 Improved Clinical DocumentationObjective #3 Real-Time sensor data streaming- Virtual Care Center (VCC)
15 © Hortonworks Inc. 2011 – 2016. All Rights Reserved
Mercy, St LouisObjective #1- Coding & Billing Accuracy
Problem
• Chart queries not being answered in a timely manner which produces problems with coding and payment accuracy. MDS (Medical Documentation Specialist) wastes time asking the same query without responses. Escalating problems occur too late in the billing cycle
• MDS staff use ”gut instincts” to determine which charts to review. There is 1 MDS for every 3,000 patients
Solution
• Identify the specific patient care paths to most likely include a secondary diagnoses and set up clinical rules to identify additional care delivery e.g. certain lab orders.
• Identify physicians with the lowest probability of answering follow up queries and create business rules to rank cases to determine which ones are highest value and most likely need 2nd review.
BenefitsVisibility into undocumented
secondary diagnoses and billing omissions
Improved staff utilization and chart review accuracy
More timely and accurate billing process resulting in 1M + additional annual revenue from documented secondary diagnoses and care
Ability to predict physicians documentation omission patterns
Healthcare
16 © Hortonworks Inc. 2011 – 2016. All Rights Reserved
Mercy, St LouisImproved Clinical Documentation & Analytics
Problem
Moving Epic EMR data to Clarity EDW took 24 hours and was “never going to enable near real time analytics
Difficulty in combining non traditional (unstructured data) with structured data
SolutionNeeded to get ahead of the EMR-EDW lag time…
Mercy replicated all of its Epic data in the Hortonworks Hadoop data platform• Epic data is enriched with data from other internal systems and publicly available
data (Lawson, etc.,)• NLP and advanced OCR applications deployed to drive data exploration and
discovery
BenefitsOne query on 19,000 individuals
took two weeks on the current data architecture, but it ran in 0.5 days on HDP
Hyperlinks built to bring visibility to updates pushed back into EPIC
Current lag time of 3-5 minutes for new clinical entries
Mercy searches through terabytes of free-text lab notes, speeding clinical based insights from “never” to “seconds”
Healthcare
17 © Hortonworks Inc. 2011 – 2016. All Rights Reserved
Near Real Time Analytics with Epic
EPIC W P
OracleClarity
W P
Batch
Gov
erna
nce
& In
tegr
atio
n
Secu
rity
Ope
ratio
ns
Data Access
Data Management
Near Real-time
APPL
ICAT
ION
S
Business Analytics Custom Applications Packaged
Applications
Sqoop
The Process:Initial bulk load via sqoop from Oracle/Clarity EDW to HDP. They then capture deltas every 3-5 minutes from cache into HDP.
18 © Hortonworks Inc. 2011 – 2016. All Rights Reserved
Monitor Patient Vitals in Real-Time with Sensor data
Problem
Managing The Volumes of System Sensor Data In a typical hospital setting, nurses do rounds and manually monitor patient vital signs. They may
visit each bed every few hours to measure and record vital signs but the patient’s condition may decline between the time of scheduled visits.
This means that caregivers often respond to problems reactively, in situations where arriving earlier may have made a huge difference in the patient’s wellbeing.
Solution
Hadoop Empowers Healthcare by Converting High Volumes of Sensor Data into a Manageable Set of Data New wireless sensors can capture and transmit patient vitals at much higher frequencies, and these
measurements can stream into a Hadoop cluster.
Caregivers can use these signals for real-time alerts to respond more promptly to unexpected changes.
Over time, this data can go into algorithms that proactively predict the likelihood of an emergency even before that could be detected with a bedside visit.
BenefitsProactively Predict Events rather
than reactivelyReal-time AlertsCapture & Transmit Patient
Vitals at Much Higher Frequencies
Improve Patient SatisfactionImproved response timesReduce adverse drug response
times
Healthcare
19 © Hortonworks Inc. 2011 – 2016. All Rights ReservedPage 19 © Hortonworks Inc. 2011 – 2015. All Rights Reserved
Real Time Predictive care
Mercy’s Journey- The Virtual Care Center (VCC)
ClinicalDocs
Vital SignMonitoring
SinglePatient Record
Lab NotesArchive
PrivacyDatabase
Medical Decision Support
DeviceData
Ingest
PreventiveCare
The path to Real Time Care
• First of its kind in the World!• 24/7/365 monitoring• $54M Investment with 330 specialized
medical professionals• 36 Medical Centers/33 ED’s• 478 Critical Care beds/28 Critical Care
units• 2,431 Acute Care beds• 15% reduction in LOS
20 © Hortonworks Inc. 2011 – 2016. All Rights Reserved
Yield, Quality & Process Optimization at Merck
ProblemMerck was seeing higher-than-usual discard rates on certain vaccines. The team was looking into the causes of the low vaccine yield rates, but the usual investigative approach involved time-consuming spreadsheet-based analyses of data collected throughout the manufacturing process.
Key Challenges • Data Silos• High Cost of Data Retention• High Cost of Testing Hypothesis in the Real World
The Solution Hadoop enables the pharmaceutical company to crunch huge amounts of data resulting in the ability to develop
and bring to vaccines to market faster and at lower cost. Through 15 billion calculations and more than 5.5 million batch-to-batch comparisons, Merck discovered that
certain characteristics in the fermentation phase of vaccine production were closely tied to yield in a final purification step.
Plans for the Future• Analyze Streaming Machine Sensor Data in Real time• Proactively minimize Yield Variability• Predictive Equipment Maintenance
21 © Hortonworks Inc. 2011 – 2016. All Rights Reserved
Other common use cases with Hadoop
• Low cost & searchable centralized storage of genomic data, PDF’s, Images, etc. • Storage of ICU device data • Storage of large data not persisted in another location or repository • RTLS – Real Time Location System Patient, doctor, and hospital staff real time location tracking within
the hospitals• EPIC/Cerner Integration • Medication Safety- Error reduction through use of data• Clinical Trial Cohort Identification• Breach/Threat Security Assessment• Store medical research forever• Theft & Abuse- “Drug Diversion”• 360 View of drug safety• Clinical Text note search (Google type search)• Population Health initiatives (re-admissions)• Wearable's Monitoring Analytics• Payment Integrity/Fraud Streaming
22 © Hortonworks Inc. 2011 – 2016. All Rights ReservedPage 22 © Hortonworks Inc. 2011 – 2015. All Rights Reserved
Join the Hortonworks Healthcare User Group!
• Monthly 1-hour calls covering thought leadership topics
• Past speakers: Mercy, UC Irvine, Mayo Clinic, Cardinal Health, Zirmed, Intermountain Health, CHOLA…
• 125+ members!"If I haven’t said it loud enough, this is an awesome forum!” Paul Vee, CHOLA
“This is one of the best uses of my time.” Lonny Northrup, Intermountain Healthcare
23 © Hortonworks Inc. 2011 – 2016. All Rights Reserved
Lessons Learned…
23
1. Like it or not: your business and success runs at the speed of software so IT leadership needs to lead on Hadoop
2. No Question: The Hadoop ecosystem is revolutionizing data management and analytics and evolving faster than any one company possibly can.
3. Adoption Curve: Start learning and adopting now; be ready for full adoption soon to stay competitive
4. There is only one Hadoop licensed by the Apache software foundation…choose your distribution carefully to avoid vendor lock from proprietary solutions as you scale out!
5. Our best customers are our most involved customers!
24 © Hortonworks Inc. 2011 – 2016. All Rights Reserved
HPE Hadoop Architecture
25 © Hortonworks Inc. 2011 – 2016. All Rights Reserved
HPE’s Hadoop Architecture is the difference!!!
Rapid design and deployment
Multiple Solution Options
Built and tested for Hortonworks
Reduce cost, risk, and errors
$
26 © Hortonworks Inc. 2011 – 2016. All Rights Reserved
HPE offers three (3) Deployment Options
− Hortonworks tested− Traditional, symmetric− Standard Rack Co-location − Entry Big Data system− Small deployments (~20 nodes)
− Hortonworks tested − Purpose-built, symmetric− Mid-size to large deployments− Workload/Storage optimization− High density, lower power
− Hortonworks tested− Asymmetric architecture− Mid-size to large deployments− Multiple/Dynamic workloads− High density, lowest power
1 HPE ProLiant DL Server HPE Apollo Storage Server
2 3 Moonshot Server and Apollo Storage
27 © Hortonworks Inc. 2011 – 2016. All Rights Reserved
Apollo 4530 System Hortonworks Reference Architecture Purpose-built for Hadoop and Big Data analytics
Analytics
At scale
Versatile performance
Unleash the full value of Big Data with Hadoop
42U rack comparison
Apollo 4530 - 1 management
node- 2 head nodes - 27 worker nodes - 1.8 PB raw storage
2U rack mount server- 1 management node- 2 head nodes - 18 worker nodes - 960 TB raw storage
Apollo 4530 advantage 33% more compute 2X storage capacity
28 © Hortonworks Inc. 2011 – 2016. All Rights Reserved
New approach to address Big Data demands
Two Socket, 2U ServersYARN Apps, HDFS, ORC
Files, Parquet, Hbase,
Cassandra
Compute Optimized Servers
Storage Optimized Servers
YARN Apps
HDFS, ORC Files, Parquet,
Hbase, Cassandra
− Separate processing /storage tiers connected by Ethernet networking
− Standard Hadoop installed asymmetrically• storage components on storage servers• yarn apps on processing servers
− HPE Moonshot, HPE Apollo 4200; HPE Apollo 2000
HPE Big Data Reference Architecture
− Processing / storage always collocated− All identical servers− Data partitioned across servers on direct-
attached storage (DAS)
Traditional Big Data Approach
29 © Hortonworks Inc. 2011 – 2016. All Rights Reserved
Unique ValueHPE Apollo & Moonshot Combined Solution
HPE Big Data Reference Architecture for HadoopInnovation delivering unique value to customers and the open source community
29
Data Consolidation− Shared storage pool for multiple Big Data
environments
Maximum Elasticity− Dynamic cluster provisioning from compute
pools without repartitioning data
Flexible Scalability− Scale compute and storage independently
Breakthrough Economics− Workload optimized components for better
density, cost and power
Ethernet (RoCE)
HPE Apollo 4000
HPE Moonshot or HPE Apollo 2000
30 © Hortonworks Inc. 2011 – 2016. All Rights Reserved
Advantages* of HPE Big Data Reference ArchitectureA new standard for Big Data delivery at scale
30
HPE Big Data Reference Architecture
* Normalized on performance
Traditional Architecture
Traditional Architecture BDRA
Hadoop performance Equivalent
Density >2x more dense
Network bandwidth
40Gbit versus 10Gbit
HDFS Storage performance 2x greater
Power (watts) Half the power
31 © Hortonworks Inc. 2011 – 2016. All Rights Reserved
Healthcare Customer Stories
Visit us at: http://hortonworks.com/industry/healthcare/
32 © Hortonworks Inc. 2011 – 2016. All Rights Reserved
Thank You!