Rethinking and Retooling Brain Health and Mental Health

©2014 Mindstrong. All Rights Reserved. Proprietary and Confidential.
Rethinking and Retooling
Brain Health and Mental Health
Tom Insel, MD
Co-founder and President, Mindstrong Health
May 7, 2019

What is the Problem We Need to Solve?

What is the Problem We Need to Solve?
No change in morbidity or mortality
25
100
75
50
125
Peak
(1965 - 1995)
Current
(2010 - 2018)
Suicide
Stroke
Heart Disease
AIDS
Childhood Leukemia
US Burden of Disease Collaborators, JAMA, 2013. https://www.cdc.gov/vitalsigns/suicide/index.html

Why have we failed to bend the curve?
60% not receiving careLack of Engagement
Quality Fragmented, episodic, delayed
Imprecise Dx Lack of biological validity
Lack of
Measurement
We don’t manage what we
don’t measure

DSM approach:
Clinical consensus
Categorical
Non-etiologic
Non-therapeutic
“Built for billing”
265 categories – symptom based
Imprecise Dx

DSM approach:
Clinical consensus
Categorical
Non-etiologic
Non-therapeutic
“Built for billing”
R-DoC approach:
Data driven (social, cognitive, neural, genomic)
Dimensional
Biological-cognitive foundation
Validated with clinical response
“Built for research”
265 categories – symptom based 5 domains: (neg affect, pos affect, cognition, social, arousal)
Imprecise Dx

DIAGNOSTIC PRECISION IN MENTAL HEALTH
Depression
Anxiety
PTSD
Bipolar
Historical Approach
Grouping patients by symptoms;
resulting heterogeneity
New Temporal Insights
Understanding patients based
on dynamic patterns in cognitive,
neural and clinical constructs
Default
Mode
Attention
Negative
Affec t
Cognitive
Control
Salienc e
Rewa rd
Processing
Mindstrong Approach
Digital stratification into
homogenous response subgroups
within MDD
within
PTSD
within Bipolar
Assay target
engagement
with continuous
digital
measures
Executive
Function
Attention
Working
Memory
Impulsivity
Information
Proc essing
Verbal
Fluenc y
Anxiety
Anhedonia
Mood
Guilt
Suic idality
Insomnia
Lethargy
Psyc homotor
COGNITIVE NEURAL CLINICAL
Grisanzio et al, JAMA Psychiatry, 2017

SOURCES: NSDUH (2013); Kessler, Chiu, Demler, & Walters (2005); Wang, Lane, Olfson, Pincus, Wells,
Kessler (2005); Merikangas , He, Burstein, Swendsen, Avenevoli, Case, Georgiades, Heaton, Swanson,
Olfson (2011), SSA Publication 13-11827 (2014)
~44 million people in the U.S. with any
disorder; ~10 million “serious”
Lack of Engagement

Receive
Services
Lack of Engagement

Underserved
Receive Minimally
Acceptable Care
Receive
Services
Lack of Engagement

Underserved
Receive Minimally
Acceptable Care
No
Benefit
Some
Benefit
Full
Benefit
Receive
Services
The 40-40-30 Rule
Lack of Engagement

Fragmented
and
Episodic
Psychological
Care
Medical
Care
Social
Supports
Family
Support
Quality of Care

Childhood Adolescence Early Adulthood Adulthood
Psychosocial
Functioning
Psychotic
Symptoms
Psychosis
onset
Prodromal
period
Delay
Quality of Care

Childhood Adolescence Early Adulthood Adulthood
Psychosocial
Functioning
Psychotic
Symptoms
Psychosis
onset
Prodromal
period
Duration of Untreated Psychosis = 74 weeks
Addington et al, Psychiatric Services, 2015
Delay
Quality of Care
Treatment
onset

WHAT WE DO TODAY
• Subjective
• Episodic
• Clinic-based
• High burden
MEASURING MOOD, COGNITION, AND BEHAVIOR

WHAT WE DO TODAY
• Subjective
• Episodic
• Clinic-based
• High burden
WHAT WE NEED
• Objective
• Continuous
• Ecological
• Passive

22
Smartphones
A medical tool for global health – improving diagnosis and connecting care
Over 4 billion globally and 7
billion by 2024
Over 70 daily checks
Over 2600 daily “touches”
More ubiquitous than clean
water, indoor plumbing, and
stable electricity0
1000
2000
3000
4000
5000
6000
7000
8000
2017 2018 Forecast 2024
Smartphone Penetration (in millions)
Middle East & Africa
India
South East Asia & Oceania
North East Asia
Central & Eastern Europe
Western Europe
Latin America
North America
https://www.ericsson.com/assets/local/mobility-report/documents/2018/ericsson-mobility-report-november-2018.pdf

23
SENSORS
Activity
Location
Sociality
VOICE/SPEECH
Prosody
Sentiment
Coherence
HCI - KEYBOARD
Reaction Time
Attention
Memory
Executive Function
DIGITAL PHENOTYPING
A New Kind of Biomarker

24
SENSORS
Activity
Location
Sociality
VOICE/SPEECH
Prosody
Sentiment
Coherence
HCI - KEYBOARD
Reaction Time
Attention
Memory
Executive Function
DIGITAL PHENOTYPING
Digital phenotype can also include
“digital exhaust” (social media posts,
search terms, AI personal assistants
etc.)

25
SENSORS
Activity
Location
Sociality
VOICE/SPEECH
Prosody
Sentiment
Coherence
HCI - KEYBOARD
Reaction Time
Attention
Memory
Executive Function
DIGITAL PHENOTYPING
Digital phenotype can also include
“digital exhaust” (social media posts,
search terms, AI personal assistants
etc.)
Feature Extraction
Pattern Recognition
Machine Learning
Digital Phenotype =
Cognition, Mood, Behavior

Cognitive performance
Digital biomarker
Dagum, Digital Medicine, 2018
Volunteers (n = 27) compared on
neurocognitive tests and digital
biomarkers.
Correlations across multiple
cognitive trait measures = .7 - .8
(roughly test–retest variance)
Digital Biomarkers
and Cognitive
Traits

pa
ge
detecting deterioration
to prevent crisis
I'm doing a lot better. I was
experiencing a lot of auditory
hallucinations. They made it
difficult to sleep which made
things progressively worse.
I checked myself into the
hospital. They adjusted my
medications, gave group
therapy, and monitored me. I
believe I slept for 12 hours
each night 3 days in a row.
What a relief! The
hallucinations finally subsided.

tracking brain health in a 48 year old woman
under care for bipolar disorder with psychosis

WHAT WE DO TODAY
• Subjective
• Episodic
• Clinic-based
• High Burden
WHAT WE NEED
 Objective
 Continuous
 Ecological
 Passive

Mobile
Interventions
CBT, DBT, IPT; Coaching; Peer
Support; Crisis Intervention
The Digital
Health
Landscape
Not an App
But
An Operating System

Mobile
Interventions
CBT, DBT, IPT; Coaching; Peer
Support; Crisis Intervention
The Digital
Health
Landscape
Learning Engine
Not an App
But
An Operating System

Lack of Engagement
Lack of Quality
Lack of
Measurement
Imprecise Dx Objective, continuous,
ubiquitous measures
The Digital Future for Brain Health?

Lack of Engagement
Lack of Quality
Lack of
Measurement
ubiquitous measures
Anonymous, person-centered
online care

Lack of Engagement
Lack of Quality
Lack of
Measurement
ubiquitous measures
online care
Coordinated, connected care
with quality metrics

Lack of Engagement
Lack of Quality
Lack of
Measurement
ubiquitous measures
online care
Coordinated, connected care
with quality metrics
Digital smoke alarms for early
detection of recovery and
relapse

Digital Tools to Reduce Suicide
Predictive signals
HCI data
Speech/text signals
Online classifiers
Crisis intervention
Upskilling tools for volunteers
On demand support
Social networks
Postvention
Care management
Peer support
AI nurse
High Tech
+
High Touch

The Digital Mental Health Challenge
Where Are We?
Improve real world outcomes
Adopted by patients and providers
Save time and money
Value?
Does it work?

Where Are We?
Save time and money
Value?
Does it work?
Privacy – Surveillance?
Agency – To me or By me?
Data – Who? When? Where?
Trust?
Acceptance?

Where Are We?
Save time and money
Value?
Does it work?
Privacy – Surveillance?
Agency – To me or By me?
Data – Who? When? Where?
Trust?
Acceptance?
Empowering Patients + Families
with Information and Connection

©2017 Mindstrong Inc. All Rights Reserved. Proprietary and Confidential.
Transforming Brain Health
tom@mindstronghealth.com
Thank You!

How to detect problems early: Examples in
Alzheimer’s Disease, Parkinson’s, anxiety and PTSD
Chaired by: Dr. Deanna
Belsky, Associate at Dolby
Family Ventures
Jan Samzelius, Co-Founder
and Chief Scientist
of NeuraMetrix
Dr. Tony Chang, Associate
at M Ventures
Dr. Srijan Sen, Professor of
Depression and
Neuroscience at University
of Michigan

Physician Training as a Model to Identify
Predictors and Preventative Interventions
for Depression under Stress
Srijan Sen MD PhD
University of Michigan
srijan@umich.edu

)
Sample
- 12 Cohorts
- 18,340 subjects (61% participation)
- 2051 in 2018 cohort to date
- 80+ Sites
- Across Specialties
- Internal Medicine, Transitional,
Surgery, Pediatrics, OB-GYN,
Psychiatry, Emergency
Medicine, Family Medicine

0.00%
5.00%
10.00%
15.00%
20.00%
25.00%
30.00%
Before Internship 3 Months 6 Months 9 Months 12 Months
PHQDepressionRate
Intern Assessment Time

Telomere Length Change with
Internship
Biological Psychiatry – in press

Methylation Changes With
Stress
Individual Factors
Longer Duty Hours
Medical Errors
Stressful Life Events
Program Factors
Low Quality Faculty Feedback
Higher Doximity Research Ranking
Lower Ethnic Diversity
Polygenic Risk Score
High Neuroticism
Low Subjective Well-Being
Childhood Stress
History of Depression
Female Gender
Minority Ethnicity
Sleep Quality
US Medical School

Genomic Risk Score and Stress
p=5x10-12

Sample Size
2014-2016: 37
2017: 546
2018: 2051

Sleep Mood b=0.12; p<0.001
Mood Sleep b=0.05; p=0.04

Change in Sleep and Wake
Time with Internship Stress

• More day-to-day variation in time asleep and wake time associated with
higher depressive symptom score during internship
Sleep Variation and Depression
• More day-to-day variation in time asleep and wake time associated with
higher depressive symptom score during internship

Circadian Genetics
• Sleep Longer
[Sleep Duration Polygenic Score]
Hours of Asleep
Change in
Sleep Schedule
• Wake up earlier
[Chronotype Polygenic Score]
Genetic Predisposition to...

• Current state
moderates
effect of
messages
slope = -.074 (on square root scale)
p-value = .007
slope = -.039 (on square root scale)
p-value = .01
slope = -.051; p-value = .001

Promise of Digital Tools in
Depression
• Prediction
– objective, real-time biomarkers
– elucidate pathway from biology to phenotype
• Prevention
– improved precision in customizing for the right
person at the right time

Acknowledgements
• Participating interns and program directors
• Key Personnel
– Yu Fang
– Elena Frank
– Joan Zhao
– Douglas Mata
– Yu Fang
– David Kalmbach
– Arbormoon Software
Funding: NIMH, UM Depression Center, Alfred A. Taubman Medical Institute,
American Foundation for Suicide Prevention
• Key Collaborators
– Connie Guille
– Zhenke Wu
– Ambuj Tewari
– Danny Forger
– Peter Song
– Margit Burmeister
– Todd Arnedt
– Laura Scott

ConfidentialMay 7, 2019
Detecting Brain Diseases Early

• Bill Gates: …Alzheimer’s starts damaging the
brain more than a decade before symptoms
start showing. That’s probably when we need
to start treating people to have the best shot at
an effective drug.
• 400 trial failures due to lack of early detection
Early Detection is Key to Solving Brain Diseases

• Extremely high quality data
• So sensitive, it can pick up very slight changes
• So statistically powerful that diagnostic capability
can be proven
• Basically, 0% false negatives or positives, AUC=1.0
• This means much higher bar than we are used to
– r2 of 0.9 or better
• Generates lots of variables, so fingerprints for each
disease can be generated
Early Detection: Requirements

•Can be deployed widely
• Easy on the person
• Runs in background
• No additional hardware
• Low cost
• Easy to implement
Early Detection: Requirements

•Can be done with significant risk factors
• Huntington’s (HD) 100%
• Alzheimer’s (AD) APOE4
• 1 – 20% of population – 35-40% risk
• 2 – 2% of population – 100% risk
Challenge #1: Proving Diagnostic Capability

Huntington’s Provides a Promising Model
0.1
0.2
0.3
0.4
0.5
0.5 1.5 2.5 3.5
Inconsistency by group
Controls Presymptomatic Symptomatic
Inconsistency

Huntington’s Provides a Promising Model
0.1
0.2
0.3
0.4
0.5
0.5 1.5 2.5 3.5
Inconsistency by group
Controls Presymptomatic Symptomatic
Inconsistency
DiseaseActiveHealthy

Challenge #2: Develop Disease Fingerprints
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 5 10 15 20 25 30 35
PD patients and controls
PD patients Controls

•Assemble AD APOE4 cohort or very large
random sample
•Identify genetic risk factors for other
diseases
•Continue development of fingerprints
Next Steps

Confidential
May 7, 2019
Detecting Brain Diseases Early
Jan Samzelius, CEO
Jan.Samzelius@neurametrix.com
+1415-420-6636

2019 SharpBrains Virtual Summit: The Future of Brain Health
How to detect problems early
Tony Chang
07 May 2019

76 2019 SharpBrains Virtual Summit: The Future of Brain Health | May 2019 |
Healthcare
Next generation of
drugs
 Biologics and small-
molecules
 Investment focus:
Oncology, Immunology &
Immuno-Oncology
 First-in-class
Life Sciences
Next generation life
science technologies
 Next generation tools
and services for
biotech research &
production
 Tools and laboratory
supply for the
academic research
and industrial testing
Performance
Materials
Next generation
materials
 Innovative display
materials, pigments
and functional
materials, and
applications
 High-tech materials for
electronics and
applications
Investments in new
business fields
 Cross-sector and brand
new business verticals for
the Merck Group
 Interest in multiple fields,
ranging from deep health-
tech solutions, agtech &
food tech, to deep tech
enabling technologies
New Businesses
At M Ventures we discover exciting
ideas and novel technologies by thinking
beyond our own imagination; helping our
partners create tomorrow’s greatest
ideas, building the future together.

Pushing the boundary
Benefits of earlier brain health detection is well recognized
Early symptom
At risk group
Triggers screening Clinical diagnosis Treatment
Functional decline
Earlier detection benefits:
 Medical: earlier access to intervention to improve outcome
 Financial: better disease outcome leads healthcare cost saving
 R&D: open up new time windows for treatment investigation
 Social & emotional: Enable patients to plan ahead while cognitively able to understand choices

Road blocks
Status quo hampers early brain health detection
Social
Biology Tools
Social stigma
Lack of awareness
Subjective measures
Accessibility of tools
Heterogenous causes
Inadequate biomarkers

Untapped data sources
The human body generates vast amounts of proxy data
Source: Kourtis et al., npj Digital Medicine, 2019

Turning data into action
Proxy data can be useful in different context of early detection
 Know the patient population
 High degree of control over
environment
 Know the patient population
 Low degree of control over
environment
 Don’t know the patient
population
 Low degree of control over
environment

In pursuit
Investments into digital proxies for early detection is on the rise
0
5
10
15
20
25
0
20
40
60
80
100
120
2014 2015 2016 2017 2018
No.ofdeals
USDmillion
N.A. Europe APAC Israel Deal value ($m)
Source: Pitchbook; M Ventures analysis

M Ventures investment – Sonde Health
Sonde is developing a voice-based health measurement tech platform
 Founded: 2015
 Location: Boston, MA, US
 Investments: Closed $16m Series A in Apr
2019
 IP: Exclusive license from MIT Lincoln Lab

Speech production starts in the brain – voice as a proxy
Complexity of Core Speech Network Approximate view of speech production
Speech
Concept
Sentences and words
Syllables and phonemes
Prosodics
Phonetic representation:
Position/state of articulators/
folds
Timing and coordination
of
articulators and vocal
folds
Neural signaling
Muscle activation
Differentbrainregions
Auditoryandtactileself-monitoring
Sentence/word
s from concept
Prosodics
Syllables
Phonemes
Articulator
positions
Vocal cord
source
state
Articular and cords
timing/coordination
Neural motor
signals
Auditory
feedback
Tactile
feedback
Parkinson’s
mTBI
Cognitive
Impairment

Translate vocal signatures into brain health measurement models
Vocal biomarker 1
Distinct Acoustic Feature of
voice correlated with a
with an element of health,
e.g. pitch slope
Vocal biomarker 2
e.g. Formant Frequencies
Vocal biomarker 3
e.g. Harmonic to Noise
Ratio
Health measures are the outcome of
Sonde’s machine learning scoring
models
Vocal biomarkers are distinct acoustic
features, correlated with an element of
health or disease
Health measure
• Disease screening
• Response to treatment
• Etc. etc.

Voice-based health measure on self-owned device is feasible
Proof of concept in depression
• Voice sample collected from more than 4k
subjects from U.S. and India
• Samples of only 6 seconds of free speech were
collected, no baseline
• Sample collection was performed remotely, on
self-owned smartphones (no site personnel, no
lab technologies) by means of app download
• The performance of Sonde’s technology was on
par with gold standard clinical screening
instrument
Sonde’s depression screening vocal
biomarker achieved
• 80% true positive rate (or sensitivity)
• below 10% false positive rate (or
>90% specificity)
reported PHQ-9 sensitivity=77.5% &
specificity=86.7%

Voice as a proxy
Big tech brands also showing interest to measure health through voice
Amazon is building a 'health & wellness’
team within Alexa as it aims to upend
health care – CNBC, May 2018

Key take away
New data opens up new opportunities, but also new challenges
Extending early brain disease detection into the real world
Seamlessly integrate into everyday life
Proxy data that translate into actionable insight
Scalable across geographies & conditions

2019 SharpBrains Virtual Summit: The Future of Brain Health

Access recorded talks, Q&A, and more at:
SharpBrains.com

Rethinking and Retooling Brain Health and Mental Health

Recommended

Recommended

More Related Content

What's hot

What's hot (20)

Similar to Rethinking and Retooling Brain Health and Mental Health

Similar to Rethinking and Retooling Brain Health and Mental Health (20)

More from SharpBrains

More from SharpBrains (20)

Recently uploaded

Recently uploaded (20)

Rethinking and Retooling Brain Health and Mental Health