2. Intel Health & Life Sciences | Make it Personal Intel Confidential – Do Not Forward
Today: Many disparate
data types, streams…
Genomics/Analytics
Genomics
Clinical
Claims &
transactions
Meds &
labs
Patient
experience
Personal
data
BigDataistheFoundationofPrecisionMedicine
Future: Integrated
computing and
integrated data
Leading to better decisions
Improved patient experience
Healthier population outcomes
Reduced costs
Accelerate transition to
personalized medicine
2
3. Intel Confidential – Do Not ForwardIntel Health & Life Sciences | Make it Personal
Analyticsinaction:
PennMedicine
3
OBJECTIVE
Predict heart failure patients who are
at risk of hospital re-admission within
30 or 90 days of discharge
CHALLENGE
Analyze large amounts of
unstructured data in patient records
across multiple hospitals in a
network
4. Intel Health & Life Sciences | Make it Personal Intel Confidential – Do Not Forward 4
Predicting Heart Failure with Machine Learning
42,358 Raw
Medications
From EMRs
allopurinol, clindamycin, coumadin,
dextrose, docusate, fluconazo, gabapentin,
glargine, heparin, hydrocortisone, insulin,
lansoprazo, lantus, levothroid,
levothyroxine, lovenox, morphine,
neurontin, omeprazo, oxycodone,
pneumococcal, senna, sertraline,
subcutaneous, testosterone, therapy, valp,
warfarin, zolof …….
23,663
Standardized
Medication
Names
Pain Management, Heart Disease,
Diabetes, Liver Failure, Respiratory,
……
20 Derived
Indicators
Apply text processing & regular
expressions
Apply “LDA” machine learning
More At-risk Patients, Identified Early On, Enables Better Care
Build Model of Indicators Predict Individual Risk Using
Indicators & Machine Learning
0,9
0,95
1
1,05
1,1
1,15
1,2
Patient E H R Only
(baseline)
E H R + Meds
Before Admit
E H R + Meds
Before Discharge
15% Relative Predictive
Model Performance Improvement
5. Intel Health & Life Sciences | Make it Personal Intel Confidential – Do Not Forward
demo— —
demo
6. Intel Health & Life Sciences | Make it Personal Intel Confidential – Do Not Forward 6
Intel-led open source project
Accelerates the collaborative
creation of cloud-native
applications driven by Big Data
Analytics
Eases the development of
analytic models by data scientists
and their use by developers
Optimized for performance and
security
Trusted Analytics Platform (TAP)
Powers the journey from data’s potential to value www.trustedanalytics.org
7. Intel Health & Life Sciences | Make it Personal Intel Confidential – Do Not ForwardFrom 12 WGS in 35 hours, to 96 WGS in 11 hours
8. Intel Health & Life Sciences | Make it Personal Intel Confidential – Do Not Forward
IntelCollaborativeCancerCloud(CCC)
Q
Q
Lab #2
OHSU
Lab #1
Learn more about our work with OHSU:
OHSU’s Exacloud
Collaborative Analytics for Personalized Cancer Care
Learn more about precision medicine and genomic research:
www.intel.com/healthcare/optimizecode
https://www.whitehouse.gov/precision-medicine
9. Intel Health & Life Sciences | Make it Personal Intel Confidential – Do Not Forward
Big data Unicancer ConSoRe:
analysis of clinical records for cancer care
*Other names and brands may be claimed as the property of others.
9
Query the UNICANCER EMRs, as they’ve
accumulated records over many years with
extensive doctor annotations.
Use natural language processing and the power
of big data analytics for this purpose.
Propose a user-friendly interface for physicians.
Add potential other sources of data (next steps).
Obtain instant cohorts of real patients
allowing better decisions and easier clinical trial
recruitment.
By definition, there is not very much literature
on rare diseases.
With precision medicine, many more rare diseases
will be discovered.
Actual patient cases are few, and clinical trials must be
built over various healthcare sites. This takes time and is
costly.
Rare cancers demand rapid treatment decisions
and cannot wait for lengthy clinical trials.
Fast response to media health threats is difficult.
SolutionGeneralchallenge
10. Intel Health & Life Sciences | Make it Personal Intel Confidential – Do Not Forward
ConSoRe: using natural language processing
to identify patients for clinical trials
*Other names and brands may be claimed as the property of others.
10
We recently had a metastatic breast
cancer research project where it took
30 people reviewing patient records for
six months to assemble a cohort of
patients who had been treated in one
of the 20 French cancer centers. We
believe ConSoRe will help us do that
within a matter of hours or days.
Pierre Heudel
Oncologist
Centre Léon Bérard
Query the UNICANCER EMRs.
Use natural language processing and the power of big data
analytics.
Obtain instant cohorts of real patients
allowing better decisions and better response to media
health threats.
32%oftrialcostsareattributedto
recruitingparticipants