We are living in the greatest time in human history! People are living their lives on smartphones and apps, measuring themselves with wearable devices like the Apple Watch, and improving their health and care with advanced analytic algorithms. Healthcare is adopting AI, Machine Learning, and Deep Learning at an accelerated pace. “Healthcare is very important for people. We are democratizing it. We are taking what has been with the institutions, and empowering the individual to manage their health.
And we’re just getting started!” - Apple CEO Tim Cook, Jan 2019
2. @docyale
What shall we Discuss?
“Healthcare is very important for people. We are democratizing it.
We are taking what has been with the institutions, and empowering
the individual to manage their health.
And we’re just at the front end of this!”
- Tim Cook, CEO, Apple, January 8, 2019
3. @docyale
1. Strategy, People, Process, Technology
“Predictive, Preventive, Personalized, Participatory”*
DATA SCIENCE AND THE FUTURE OF HEALTH AND CARE
*P4 Medicine Institute (http://www.p4mi.org)
Deep Medicine, Eric Topol, 2019
2. Applied Advanced Analytics in Health and Care
3. The Future: ”Personal Health Management” vs “Population Health Management”
aka “the average patient does not exist”
4. @docyale
Sophistication
Gain
Standard Reports What happened?
Source: Competing on Analytics, Davenport/Harris, 2007
Ad hoc reports How many, how often, where?
Query/drill down What exactly is the problem?
Alerts What actions are needed?
Statistical analysis Why is this happening?
Forecast/extrapolation What if these trends continue?
Predictive modeling What will happen next?
Optimization What’s the best outcome?
Predictive
and
Prescriptive
Analytics
Descriptive
Analytics
DATA SCIENCE MATURITY
6. @docyale
HEALTH SYSTEM ANALYTICS STRATEGY
Source: “Shifting Into High Gear: Health systems have a growing strategic focus on analytics
today for the future”, Deloitte Insights, March 28, 2019 (http://bit.ly/2K7WNNA)
7. @docyaleSource: Preparing the Healthcare Workforce to Deliver the Digital Future, NHS, UK https://topol.hee.nhs.uk/
DATA SCIENCE WORKERS
8. @docyale
Dimensions Capabilities
Advanced Analytics Maturity Stages
1 Limited 2 Beginning 3 Momentum 4 Maturing 5 Visionary
Structure
Governance
Policies
Processes & Controls
Security & Privacy
Leadership
Organization
Culture
Impact
Strategy
Business Need & Use Cases
Analytics
Tools & Techniques
Deployment/Delivery Approach
Management/Talent
Variety, Volume, Velocity
Data
Data Access
Data Integration
Data Architecture/MDM
Funding
Resources
Talent and Skills
Roles & Responsibilities
Training
PROCESSES: ADVANCED ANALYTICS ASSESSMENT
Source: “TDWI Advanced Analytics Maturity Model Guide,” The Data Warehouse Institute, 2018 (http://bit.ly/2K8b64G)
10. @docyale
KEY BARRIERS TO DATA SCIENCE IN HEALTHCARE
https://www.datasciencecentral.com/profiles/blogs/seeing-the-ai-ml-future-in-healthcare-through-the-eyes-of-physici
11. @docyale
KEY BARRIERS TO DATA SCIENCE IN HEALTHCARE
Source: McKinsey & Company, May 2019, https://mck.co/2ExMb6T
12. @docyale
KEY BREAKTHROUGHS IN HEALTHCARE
Source: Oliver Wyman, in Clinically Integrated Care and Future of Population Health, Yale K, Miner G, HIMSS Annual Meeting 2015
Provider Value Evolution
- Population Health Management
- Descriptive Analytics
- Clinical and Claims Data
From To
Volume, patient turnover Value, patient health
Physician-Centered Patient-Centered
Transactional, episodic Coordinated Care
Sick care Wellness and prevention
Inaccessible Convenient, 24/7
Unwarranted variation Evidence based protocol
Consumer Retail Revolution
- Personal Health Management
- Predictive Analytics
- Clinical, Claims, and Context Data
From To
Uninformed Informed/Share Decisions
Limited engagement Patient Empowered
Patient isolated Patient socially connected
Limited Consequences Financial reward/incented
Bricks, office hours Virtual, anytime/anywhere
Physician opinion Evidence based facts
Health System Devolution
- Precision Health Management
- Prescriptive Analytics
- Comprehensive, Cradle-to-Grave Data
From To
Basic management Comprehensive life plan
Symptomatic treatment Continuous monitoring
One-size-fits-all Individual treatment
Average patient diagnosis Individual patient diagnosis
Normal range Specific results
Medical competencies Empathy, presence,
humankindness
13. @docyale
KEY BREAKTHROUGHS FOR DATA SCIENCE IN HEALTHCARE
1) Digital Inputs = Rapid Growth in
Sources of Digital Health Data
2) Data Accumulation = Proliferation of
Digitally-Native Data Sets
3) Data Insight = Generated by Data Science
using Accumulated & Integrated Data
4) Translation = Impact on
Therapeutics & Healthcare Delivery
5) Personal Health
Management =
Consumer Retail Care
Source: Adapted from Internet Trends 2017, by Mary Meeker, https://www.kleinerperkins.com/perspectives/internet-trends-report-2017
See Also: Handbook of Statistical Analysis and Data Mining Applications, Nisbet R., Miner G., Yale K., Elsevier/Academic
Press, 2017. (http://a.co/d/ihMBBKs)
6) Outcomes =
Measure Outcomes &
Iterate…
Innovation Cycle Times
Compressing
Ripe for disruption
14. @docyale
Source: PBS, Propeller Health, TechCrunch, Livongo, Ayasdi, Flatiron, Xconomy, Kinsa, Omada
Patient Empowerment &
Health Management
Propeller Health + Bluetooth
Inhaler Sensor = Improved
Medication Adherence + Insights
Livongo + Connected Glucose
Meter = Personalized Coaching
+ $100/Month Savings for
Payers
Improvements to
Clinical Pathways /
Protocol
Ayasdi AI + Mercy Health System
Patient Data = Clinical Anomaly
Detection + Improved Clinical
Pathway Development
Flatiron + Foundation Med (FMI)
= 20,000 Linked Cancer Patients
Records + Personalized Medicine
Preventative Health
Kinsa + Crowdsourced
Temperature Data = Local Flu
Predictions + Proactive
Treatments for Populations
Omada + Preventative Program
= 4-5% Body Weight Reduction
+ Reduced Risk for Stroke and
Heart Disease
KEY BREAKTHROUGHS: DIGITAL HEALTH EXAMPLES
Source: Internet Trends 2017, by Mary Meeker, https://www.kleinerperkins.com/perspectives/internet-trends-report-2017
20. @docyale
* Lynch et al. Documenting Participation in a DM Program. JOEM 2006; 48(5)
1 Frazee et al. Leveraging the Trusted Clinician: Documenting Disease Management Program
Enrollment. Disease Mgmt 2007; 10:16-29
EXAMPLE: CONNECTING WITH CONSUMERS
30%
15%
7%
3.5%
0
0.2
0.4
0.6
0.8
1
1.2
Eligible Contacted Participant Behavior
Improvement
21. Prepare
Data
Segment
Develop
Personas
- Health Insurance and
Hospital Data
- Aggregated,
Organized, Enhanced
- ”Exogenous” Data:
behaviors and lifestyles
Gather Internal
Data
Gather External
Data
Combine
Data Discovery
- Aggregate
- Integrate
- Normalize
- Standardize
Recognize
Patterns
- Machine
Learning
- Cluster Analysis
- Classification and
Regression
- Retail Marketing
“Personas”
- “Market of One”
- Optimize Service
Example of population micro-segmentation: groups with uniform behaviors, attitudes and lifestyles
SOLUTION: SOCIAL DETERMINANTS MICRO-SEGMENTATION
MicroSegmentation Source: Wiese K., 2014, Aetna Member Experience and Communications Segmentation Pilot, North Carolina State Health Plan
https://shp.nctreasurer.com/Board%20of%20Trustees%20Meeting%20Documents/BOT_4a_Segment_Pilot-8-1-2014.pdf
Wall Street Journal, April 29, 2019, online: https://www.wsj.com/articles/health-firms-are-looking-at-personal-data-11556589780
22. @docyale
EXAMPLE: LOW PREDICTABILITY
Study: Society of Actuaries (2007) A Comparative Analysis of Claims-Based Tools for Health Risk Assessment.
Clinical/Financial Model From: Wei H. “Prediction vs. Intervention (2014).
Predictive Modeling Summit presentation. November 13, 2014, Washington, DC.Clinical/Real World
20% 30% 40% 50%
ACG
CDPS
Clinical Risk Group
DxCG (Verisk)
DxCG (Verisk)
Ingenix (Optum)
Medicaid Rx
Impact Pro
Ingenix ERG (Optum)
ACG Dx+Rx
DxCG UW Model
MEDai
Clin/Fin - Train
Clin/Fin - Test
Groupers
Novel Predictive Models for Metabolic Syndrome Risk: A “Big Data” Analytic Approach, Am J Manag Care. 2014 Jun 1;20(6):e221-8..
SOLUTION: STRATEGIC DATA
ACQUISITION & COMBINATION
23. @docyale
EXAMPLE: READMISSION RISK
Nearly 1 in 5 Medicare patients discharged from a hospital (approximately 2.6 million seniors)
is readmitted within 30 days, at a cost of more than $26 billion every year - CMS, 2016
Affordable Care Act
Section 3026 Community Care
Transitions Program “Test models to
improve care transitions from hospital
to other settings and reduce
readmissions for high-risk Medicare
beneficiaries.” – CMS, 2011
---
“No statistically significant impacts of
the CCTP on readmission rates or
Medicare Part A and Part B
expenditures” – Mathematica Policy
Research, Final Evaluation Report,
November 2017
Care Transitions Program
§ Medication self-management:
Patient engaged & medication
management system
§ Patient-centered record: Patient
use ”PHR” (communication/continuity)
§ PCP & Specialist Follow-up:
Patient schedules appointments
§ Knowledge of Red Flags: Patient
understands when condition is worsening
Care Transitions “Red Flags”
§ Cardiac
§ Pulmonary/COPD
§ Heart Failure
§ Diabetes
§ DVT
§ Peripheral Vascular
§ Stroke
§ Other: fever, bleeding,
confusion, pain, fatigue
24. @docyale
SOLUTION: DATA SCIENCE PREDICTIVE ANALYTICS
One in five hospital patients experienced an adverse event within three weeks of discharge;
60% were medication related and could have been avoided, $60 million loss
Risk Assessment
Medicare readmission penalty
Heart failure, CAD, dysrhythmia
Acute myocardial infarction
COPD, Pneumonia, Asthma
Joint replacements
Coronary artery bypass
Coronary stents
Stroke
Diabetes
LOS > 7 days (any diagnosis)
Home Location
Home Discharge
Right Care, Right Time & Place
Pre-discharge hospital visit
RN home visit within 24 hours:
§ Medication reconciliation/education
§ Schedule follow-up appointments
§ Educate self management, “red
flags,” doctor visits
§ “Personal Health Record”
Follow-up phone calls to reinforce
and ensure appropriate follow up
and care
Results
“Significant” readmissions decline
9% decrease in total cost of ER
visits
“Population Health Management”
or “Personal Health Managed”
“Market of One”
25. @docyale
EXAMPLE: DIAGNOSE SKIN LESIONS
• Smartphone Selfie Skin Lesion Diagnosis
• Pattern recognition
• Melanoma survival rate:
• Early detection: 99% survival
• Late stage detection: 14% survival
Deep Medicine, Eric Topol, 2019
“Dermatologist-Level Classification of Skin Cancer with Deep Neural Networks, Esteva et al. Nature vol. 542, pgs 115–118, Feb. 2, 2017
“The Final Frontier in Cancer Diagnosis,” Leachman S, Merlino G, Nature vol. 542, pgs 36-38, Feb. 2, 2017
”How Accurate Are Smart Phone Apps for Detecting Melanoma in Adults?” Chuchu, N et al. Cochrane, Dec 4, 2018 (http://bit.ly/2wCJj4w)
Risk Assessment
• “In Silico” does not equal Real World
• Not clinically validated
• Accuracy not demonstrated (6.8% to 98.1%)
• Poor Sensitivity, highly variable Specificity
• “High likelihood of missing melanomas”
(Cochran)
”Lesions Learnt”
• Classification: Benign or Malignant
• If Malignant: Melanoma or Non-Melanoma
• Convolutional Neural Network (CNN)
• Algorithm outperformed dermatologists
• Medical staff replaced by AI?
• Algorithm released to the public
29. @docyale
2. Individual Data Health & Personal Data
1. Curated Medical Knowledge
3. Patient Relationships
4. Outcomes
Predict, Prescribe, Perform
TRANSLATE OUTPUT