Morteza Naghavi, M.D.
Founder and Executive Chairman of SHAPE
The 2nd Machine Learning
Vulnerable Patient Symposium
Towards Developing an Artificial Intelligence-Based Forecast System
for
Predicting Short-Term CVD Events
A Satellite Event in Conjunction with 2017 Annual Scientific Sessions of American Heart Association
Let’s Begin with
the End
Goal:
Eradicate Heart
Attacks(Unpredicted CVD Events)
Now Let’s See
Where We Are
Last year2,626,418 people died
in the US
614,348 (23.4%) of them
died due to heart disease and
133,033 (5.1%) due to stroke
Total 747,381 (28.5%)
Unlike Cancer
• Unlike in cancer where oncologists deal
with hundreds of different pathologies and
specific therapeutic strategies, in CVD over
50% of cases we deal with one pathology:
Atherosclerosis
(The number 1 killer of mankind)
Unpredicted
In >50% of victims,
the first symptom of
asymptomatic
atherosclerosis is a
sudden cardiac
death or acute MI.
Men
Women
0 10 20 30 40 50 60 70
Patients Diagnosed with CHD (%)
Murabito et al
Circulation 1993
Sudden Cardiac Death or Acute MI
as Initial Presentation of CHD
62%
42%
Add 10yrs to Life Expectancy of
Mankind
Early detection and treatment of
atherosclerosis to prevent acute CVD
events is likely to increase life
expectancy in excess of 10 years.
That’s HUGE!!!
A Vaccination Type Impact on Public Health
How Do We Get There?
A heart-attack free
future
Let’s Draw an Analogy
Heart Attack
vs. Hurricane
Imagine if the weatherman says there
is a 7.5% chance of a category 5 hurricane in
the next 10 years. Do you think people would
take immediate preventive actions like boarding
up their windows, buying hurricane supplies, or
even changing their daily routines?
Imagineif heart attack and stroke were
predicted similar to hurricanes Harvey and Irma
with sufficient short-term alerts to at-risk people
to take preventive actions.
Heart Attack vs. Hurricane
THE GALVESTON HORROR
Heart Attack vs. Hurricane
THE GALVESTON HORROR
Heart Attack vs. Hurricane
THE GALVESTON HORROR
Heart Attack vs. Hurricane
THE GALVESTON HORROR
Heart Attack vs. Hurricane
THE GALVESTON HORROR
Heart Attack vs. Hurricane
THE GALVESTON HORROR
Heart Attack vs. Hurricane
THE GALVESTON HORROR
Heart Attack vs. Hurricane
10-year Risk Prediction vs. 10-day Risk Prediction
What has SHAPE done?
Naghavi et. al. Circulation Journal
The Vulnerable Patient Consensus Statement
Naghavi et. al. Circulation Journal
The Vulnerable Patient Consensus Statement
SHAPE Task Force Meeting
SHAPE Guidelines Published
Coronary Artery Calcium Score
32
The Writing Sub-Committee of the SHAPE Task Force (left to right): Drs Budoff, Falk, Rumberger, Naghavi,
Fayad, Hecht, and Berman
Atherosclerosis Test
Very Low Risk3
Negative Test
• CACS =0
• CIMT <50th percentile
Lower
Risk
Moderate
Risk
Positive Test
• CACS ≥1
• CIMT 50th percentile or Carotid Plaque
Moderately
High Risk
High
Risk
Very
High Risk
No Risk Factors5 + Risk Factors • CACS <100 & <75th%
• CIMT <1mm & <75th%
& no Carotid Plaque
• Coronary Artery Calcium Score (CACS)
or
• Carotid IMT (CIMT) & Carotid Plaque4
• CACS 100-399 or >75th%
• CIMT 1mm or >75th%
or <50% Stenotic Plaque
• CACS >100 & >90th%
or CACS 400
• 50% Stenotic Plaque6
LDL
Target
<160 mg/dl <130 mg/dl <130 mg/dl
<100 Optional
<100 mg/dl
<70 Optional
<70 mg/dl
Re-test Interval 5-10 years 5-10 years Individualized Individualized Individualized
All >75y receive unconditional treatment2
Apparently Healthy Population Men>45y Women>55y1
ExitExit
Myocardial
IschemiaTest
NoAngiography
Follow Existing
Guidelines
Yes
The 1st SHAPE Guidelines
Step 1
Step 2
Step 3
Optional
CRP>4mg
ABI<0.9
1: No history of angina, heart attack, stroke, or peripheral arterial disease.
2: Population over age 75y is considered high risk and must receive therapy without testing for
atherosclerosis.
3: Must not have any of the following: Chol>200 mg/dl, blood pressure >120/80 mmHg, diabetes,
smoking, family history, metabolic syndrome.
4: Pending the development of standard practice guidelines.
5: High cholesterol, high blood pressure, diabetes, smoking, family history, metabolic syndrome.
6: For stroke prevention, follow existing guidelines.
Existing Guidelines (Status Quo):
• Screen for Risk Factors of Atherosclerosis
• Treat Risk Factors of Atherosclerosis
The SHAPE Guidelines:
• Screen for Atherosclerosis (the Disease)
Regardless of Risk Factors
• Treat based on the Severity of the Disease
and its Risk Factors
SHAPE v.s. Status Quo
Number
(per year)
Estimated Impact
of SHAPE
(Sensitivity
Analysis Range)
Estimated
Change in
Cost
CVD Deaths 910,600 ↓10%
(5%-25%)
($1.2 b)
MI (prevalence) 7,200,000 ↓ 25%
(5%-35%)
($18.0 b)
Chest Pain Symptoms (ER visits) 6,500,000 ↓ 5%
(2.5%-25%)
($4.1 b)
Hospital Discharge for Primary Diagnosis of CVD 6,373,000 ↑ 10%
(5%-25%)
$3.8 b
Hospital Discharge for Primary Diagnosis of CHD 970,000 ↓ 10%
(5%-25%)
($9.9 b)
Cholesterol Lowering Therapy ↑ 50 %
(50%-65%)
8.00 b
CV Imaging 8,700,000 ↑ 10%
(5%-25%)
$358 m
Angiography 6,800,000 ↑ 15% - CTA
(2.5%-25%)
$600 m
PCI (percutaneous coronary interventions per year) 657,000 ↓ 10%
(5%-50%)
($580 m)
CABS (coronary artery bypass surgeries per year) 515,000 ↓ 5%
(2.5%-50%)
($672 m)
Total Δ in Cost ($21.5 b)
Cost Effectiveness of the SHAPE Guidelines
Heart Attack vs. Hurricane
10-year Risk Prediction vs. 10-day Risk Prediction
Long term predictions do not
trigger immediate preventive
actions.
Preventive cardiology needs a
short-term predictor.
Heart Attack vs. Hurricane
Machine Learning Vulnerable
Patient Project.
http://shapesociety.org/videos-2/
http://shapesociety.org/videos/
The Big Idea:
Developing an Artificial Intelligence-based Forecast System for
Prediction of Heart Attacks within 12 Months
Use machine learning to create new algorithms to detect who will experience
a CHD event within a year (The Vulnerable Patient). Algorithms will be
based on banked biospecimen and information collected days up to 12
months prior to the event. We will utilize existing cohorts such as MESA,
Heinz Nixdorf Recall Study, Framingham Heart Study, BioImage Study and
the Dallas Heart Study. External validation to test for discrimination and
calibration will be conducted using other longitudinal observational studies
that provide adjudicated cardiovascular event information such as the
MiHeart, JHS, DANRISK and ROBINSCA. Additionally, we will use machine
learning to characterize individuals who, despite high conventional risk, have
lived over 80 years with no CHD events (The Invulnerable). We expect to
discover new targets for drug and possibly vaccine development. We will
make the algorithms available as an open source tool to collect additional data
over time and increase its predictive value.
What a great idea, what are you
waiting for?
Funding!
Will Super Intelligent Computers Replace
Physicians?
Will Super Intelligent Computers
Replace Physicians?
Absolutely Yes
When and in What Areas?
Umm let’s discuss
Inspired by IBM Watson
Google DeepMind
49
Machine vs. Cardiologist

AHA SHAPE Symposium 2017 Dr. Naghavi Presentation

  • 1.
    Morteza Naghavi, M.D. Founderand Executive Chairman of SHAPE The 2nd Machine Learning Vulnerable Patient Symposium Towards Developing an Artificial Intelligence-Based Forecast System for Predicting Short-Term CVD Events A Satellite Event in Conjunction with 2017 Annual Scientific Sessions of American Heart Association
  • 2.
  • 3.
  • 4.
  • 5.
    Last year2,626,418 peopledied in the US 614,348 (23.4%) of them died due to heart disease and 133,033 (5.1%) due to stroke Total 747,381 (28.5%)
  • 6.
    Unlike Cancer • Unlikein cancer where oncologists deal with hundreds of different pathologies and specific therapeutic strategies, in CVD over 50% of cases we deal with one pathology: Atherosclerosis (The number 1 killer of mankind)
  • 7.
  • 8.
    In >50% ofvictims, the first symptom of asymptomatic atherosclerosis is a sudden cardiac death or acute MI.
  • 9.
    Men Women 0 10 2030 40 50 60 70 Patients Diagnosed with CHD (%) Murabito et al Circulation 1993 Sudden Cardiac Death or Acute MI as Initial Presentation of CHD 62% 42%
  • 10.
    Add 10yrs toLife Expectancy of Mankind Early detection and treatment of atherosclerosis to prevent acute CVD events is likely to increase life expectancy in excess of 10 years. That’s HUGE!!! A Vaccination Type Impact on Public Health
  • 11.
    How Do WeGet There? A heart-attack free future
  • 12.
    Let’s Draw anAnalogy Heart Attack vs. Hurricane
  • 13.
    Imagine if theweatherman says there is a 7.5% chance of a category 5 hurricane in the next 10 years. Do you think people would take immediate preventive actions like boarding up their windows, buying hurricane supplies, or even changing their daily routines?
  • 14.
    Imagineif heart attackand stroke were predicted similar to hurricanes Harvey and Irma with sufficient short-term alerts to at-risk people to take preventive actions.
  • 15.
    Heart Attack vs.Hurricane THE GALVESTON HORROR
  • 16.
    Heart Attack vs.Hurricane THE GALVESTON HORROR
  • 17.
    Heart Attack vs.Hurricane THE GALVESTON HORROR
  • 18.
    Heart Attack vs.Hurricane THE GALVESTON HORROR
  • 19.
    Heart Attack vs.Hurricane THE GALVESTON HORROR
  • 20.
    Heart Attack vs.Hurricane THE GALVESTON HORROR
  • 21.
    Heart Attack vs.Hurricane THE GALVESTON HORROR
  • 22.
    Heart Attack vs.Hurricane 10-year Risk Prediction vs. 10-day Risk Prediction
  • 25.
  • 26.
    Naghavi et. al.Circulation Journal The Vulnerable Patient Consensus Statement
  • 27.
    Naghavi et. al.Circulation Journal The Vulnerable Patient Consensus Statement
  • 28.
  • 30.
  • 32.
  • 33.
    The Writing Sub-Committeeof the SHAPE Task Force (left to right): Drs Budoff, Falk, Rumberger, Naghavi, Fayad, Hecht, and Berman
  • 34.
    Atherosclerosis Test Very LowRisk3 Negative Test • CACS =0 • CIMT <50th percentile Lower Risk Moderate Risk Positive Test • CACS ≥1 • CIMT 50th percentile or Carotid Plaque Moderately High Risk High Risk Very High Risk No Risk Factors5 + Risk Factors • CACS <100 & <75th% • CIMT <1mm & <75th% & no Carotid Plaque • Coronary Artery Calcium Score (CACS) or • Carotid IMT (CIMT) & Carotid Plaque4 • CACS 100-399 or >75th% • CIMT 1mm or >75th% or <50% Stenotic Plaque • CACS >100 & >90th% or CACS 400 • 50% Stenotic Plaque6 LDL Target <160 mg/dl <130 mg/dl <130 mg/dl <100 Optional <100 mg/dl <70 Optional <70 mg/dl Re-test Interval 5-10 years 5-10 years Individualized Individualized Individualized All >75y receive unconditional treatment2 Apparently Healthy Population Men>45y Women>55y1 ExitExit Myocardial IschemiaTest NoAngiography Follow Existing Guidelines Yes The 1st SHAPE Guidelines Step 1 Step 2 Step 3 Optional CRP>4mg ABI<0.9 1: No history of angina, heart attack, stroke, or peripheral arterial disease. 2: Population over age 75y is considered high risk and must receive therapy without testing for atherosclerosis. 3: Must not have any of the following: Chol>200 mg/dl, blood pressure >120/80 mmHg, diabetes, smoking, family history, metabolic syndrome. 4: Pending the development of standard practice guidelines. 5: High cholesterol, high blood pressure, diabetes, smoking, family history, metabolic syndrome. 6: For stroke prevention, follow existing guidelines.
  • 35.
    Existing Guidelines (StatusQuo): • Screen for Risk Factors of Atherosclerosis • Treat Risk Factors of Atherosclerosis The SHAPE Guidelines: • Screen for Atherosclerosis (the Disease) Regardless of Risk Factors • Treat based on the Severity of the Disease and its Risk Factors SHAPE v.s. Status Quo
  • 36.
    Number (per year) Estimated Impact ofSHAPE (Sensitivity Analysis Range) Estimated Change in Cost CVD Deaths 910,600 ↓10% (5%-25%) ($1.2 b) MI (prevalence) 7,200,000 ↓ 25% (5%-35%) ($18.0 b) Chest Pain Symptoms (ER visits) 6,500,000 ↓ 5% (2.5%-25%) ($4.1 b) Hospital Discharge for Primary Diagnosis of CVD 6,373,000 ↑ 10% (5%-25%) $3.8 b Hospital Discharge for Primary Diagnosis of CHD 970,000 ↓ 10% (5%-25%) ($9.9 b) Cholesterol Lowering Therapy ↑ 50 % (50%-65%) 8.00 b CV Imaging 8,700,000 ↑ 10% (5%-25%) $358 m Angiography 6,800,000 ↑ 15% - CTA (2.5%-25%) $600 m PCI (percutaneous coronary interventions per year) 657,000 ↓ 10% (5%-50%) ($580 m) CABS (coronary artery bypass surgeries per year) 515,000 ↓ 5% (2.5%-50%) ($672 m) Total Δ in Cost ($21.5 b) Cost Effectiveness of the SHAPE Guidelines
  • 37.
    Heart Attack vs.Hurricane 10-year Risk Prediction vs. 10-day Risk Prediction
  • 38.
    Long term predictionsdo not trigger immediate preventive actions. Preventive cardiology needs a short-term predictor.
  • 39.
  • 40.
  • 41.
  • 42.
  • 43.
    The Big Idea: Developingan Artificial Intelligence-based Forecast System for Prediction of Heart Attacks within 12 Months Use machine learning to create new algorithms to detect who will experience a CHD event within a year (The Vulnerable Patient). Algorithms will be based on banked biospecimen and information collected days up to 12 months prior to the event. We will utilize existing cohorts such as MESA, Heinz Nixdorf Recall Study, Framingham Heart Study, BioImage Study and the Dallas Heart Study. External validation to test for discrimination and calibration will be conducted using other longitudinal observational studies that provide adjudicated cardiovascular event information such as the MiHeart, JHS, DANRISK and ROBINSCA. Additionally, we will use machine learning to characterize individuals who, despite high conventional risk, have lived over 80 years with no CHD events (The Invulnerable). We expect to discover new targets for drug and possibly vaccine development. We will make the algorithms available as an open source tool to collect additional data over time and increase its predictive value.
  • 44.
    What a greatidea, what are you waiting for? Funding!
  • 45.
    Will Super IntelligentComputers Replace Physicians?
  • 47.
    Will Super IntelligentComputers Replace Physicians? Absolutely Yes When and in What Areas? Umm let’s discuss
  • 48.
  • 49.
  • 50.