Lecture Objectives:
1) To use examples from my research to define and introduce the ideals of precision medicine and digital health. 2) To introduce how large scale population-wide analysis of data can be used to facilitate these two ideals. 3) To introduce how freely available open data can be used to facilitate these two ideals. 4) To show how mobile technology can be used to facilitate these two ideals.
10. CAG houses the world’s largest
pediatric biobank
Ø > 1M patient visits /
year to CHOP
Ø Initial 5-year goal to
establish biobank with
an emphasis on
genomic discovery
Ø Future 5-year vision is
to translate discoveries
into tangible patient
benefit
10
11. 11
Datasets (Genomics EMR)
§ Over 75K pediatric and 150K
related adult patients GWAS
genotyped with associated
longitudinal EMR since 2006
Data Analytics
§ End to end internal Next-
Gen sequencing
capabilities
§ Integrated bioinformatics
§ Rapid identification of
novel genetic biomarkers
Biobank (BB)
§ Fully automated robotic
biorepository
Consented Patients
• 85% of the BB
patients are
consented for
longitudinal follow up
and are eligible for call
back for future studies
§ ~1.2M patient visits/year
§ 10% of all R/O disease patients in
N. America are treated at CHOP
CAG’s pediatric biobank contains a high percentage of rare genetic variants
§ Population is unique in that it represents the
most severe forms of common diseases
§ Global reach in many therapy areas
In the last 8 years CAG has had over 400 peer reviewed
publications focused on novel genetic discoveries
Highly scalable infrastructure to support translational research
The Pediatric Biobank at The Center for Applied Genomics (@CHOP)
24. NFC-1 clinical trial design for ADHD
Week 1
Day 7±2
Week 2
Day 14±2
Week 3
Day 21±2
Week 4
Day 28±2
Week 5
Day 35±2
Week 9
Day 75±2
Adverse event monitoring X X X X X phone
Laboratory Safety Tests (blood and urine)A
X X X X X
Physical Examination X X X X X
Vital Signs: BP, HR, RR X X X X X
Body Weight (all points) & Height (week 1 only) X X X X X
12-lead ECG X X X X X
Urine b-hCG test (menstruating females only) X X X X X
Contraception verification (selected females) X X X X X
Vanderbilt Parent Rating Scale X X X X X
BREIF (Parent; Self) X X X X X
QuotientâADHD test X X X X X
PERMP-Math test X X X X X
Actigraphy (continuous monitoring) X X X X X
CGI-S & CGI-I X X X X X
Dispense study drugB
X X X X
NFC-1 or placebo administration at homeC
Placebo bid 50 mg bid 100 mg bid 200 mg bid 400 mg bid
Retrieve pill bottle/pill count X X X X X
A. Blood draws for hematology (RBC, WBC with differential, platelet count) and clinical chemistry (electrolytes, albumin, ALT, AST, alkaline
phosphatase, bilirubin, BUN, creatinine, glucose,
B. Study drug for Week 1 administered at end of PK study; study drug for next week dispensed at each clinic visit
C. Dose escalations to be determined by CGI-Sand CGI-I scores at end of each week of treatment; maximum doses indicated
30 mGluR+ ADHD children have completed 5 weeks on drug (FPI 01/23/15)
25. 0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
1 2 3 4 5
CGI-I: Proportion of Responders at Each Week
for All Subjects
Week Week Week Week Week
CGI-I, Clinical Global Impression
of Symptom Improvement
Responder – Global rating of much
or very much improved
NFC-1 ADHD Study Results – Clinician Rating Scale
P < 0.001
26. 0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
Week 1 Week 2 Week 3 Week 4 Week 5
Vanderbilt Scores: Proportion of Patients Improved from Pre-study baseline
for All Patients
Improvement defined as 25% improvement in hyperactivity/inattention domains
NFC-1 ADHD Study Results – Parent Rating Scale
P < 0.001
27. Current clinical trials are expensive and
inefficient
27
No Response Response
Non-targeted efficacy
(generalized population):
20 / 100 = 20%
$$$$$$$$
28. Big data to disrupt clinical trials by
minimizing cost with maximal efficacy
28
No-pathway defect Targeted pathway defect Response
Non-targeted efficacy
(generalized population):
20 / 100 = 20%
$$$$$$$$
Targeted efficacy
(personalized population):
20 / 25 = 80%
$$
Genomics
33. Dengue is “the most important mosquito-borne viral
disease in the world” – WHO
• Dengue virus causes a flu-like illness that can progress to
fatal severe dengue
• Epidemic breakouts are a leading cause of pediatric deaths
among developing Asian and Latin American countries!
• No prognostic assays or drugs are available… treatment is
largely supportive and directed at symptoms
• Neglected tropical disease
• Spreading to the US mainland!
Aedes aegypti Aedes albopictus
3.6B people at
risk
390M estimated
infections
96M manifest
clinically
2M cases progress
to severe dengue
21K fatalities!
61. We made an app for that:
STARGEO.org
Tag samples
Gene Expression Profiling During Early Acute Febrile Stage of Dengue Infection Can
Predict The Disease Outcome: GSE18090
Background: We report the detailed development of biomarkers to predict the clinical outcome
under dengue infection. Transcriptional signatures from... More.
Tag: (Dengue hemorrhagic fever)DHF
All (26) DHF(10) Unmatched (16)
Tag Value Sample_acc sample_characteristics sample_title sample_source_name
GSM452242
gender: female| |age: 23| |days
of symptoms: 7| |igm: Pos| |igg:
Neg| |pcr/virus isolation: Pos
DF Patient 8 PBMCs from DF patient
GSM452243
gender: male| |age: 41| |days of
symptoms: 3| |igm: Neg| |igg:
Pos| |pcr/virus isolation: Pos
DHF Patient 1 PBMCs from DHF patient
GSM452244
gender: male| |age: 41| |days of
symptoms: 3| |igm: Neg| |igg:
Pos| |pcr/virus isolation: Pos
DHF Patient 2 PBMCs from DHF patient
DHF
DHF
Column Regex Saveall ▼ DHF
DHF
DHF
DHF
DHF
NCI BD2K Funded (PI: Hadley)
62. We made an app for that:
STARGEO.org
Tag samples
Gene Expression Profiling During Early Acute Febrile Stage of Dengue Infection Can
Predict The Disease Outcome: GSE18090
Background: We report the detailed development of biomarkers to predict the clinical outcome
under dengue infection. Transcriptional signatures from... More.
Tag: (Dengue hemorrhagic fever)DHF
All (26) DHF(10) Unmatched (16)
Tag Value Sample_acc sample_characteristics sample_title sample_source_name
GSM452242
gender: female| |age: 23| |days
of symptoms: 7| |igm: Pos| |igg:
Neg| |pcr/virus isolation: Pos
DF Patient 8 PBMCs from DF patient
GSM452243
gender: male| |age: 41| |days of
symptoms: 3| |igm: Neg| |igg:
Pos| |pcr/virus isolation: Pos
DHF Patient 1 PBMCs from DHF patient
GSM452244
gender: male| |age: 41| |days of
symptoms: 3| |igm: Neg| |igg:
Pos| |pcr/virus isolation: Pos
DHF Patient 2 PBMCs from DHF patient
DHF
DHF
Column Regex Saveall ▼ DHF
DHF
DHF
DHF
DHF
Search free text
attributes of human
microarray expression
11,903 Series à
465,770 Samples
Tag samples across
multiple studies to
annotate features
278 Tags à
5,798 Series annotations à
490,110 Sample annotations
Analyze genomic
signatures by meta-
analysis
1,682 microarray platforms à
28,254,323 gene probes
NCI BD2K Funded (PI: Hadley)
78. Digital Health
Digital health is the
convergence of the
digital and genomic
revolutions with health,
healthcare, living, and
society. Digital health is
empowering people to
better track, manage,
and improve their own
and their family's health,
live better, more
productive lives, and
improve society.
81. • Early accurate diagnosis improves melanoma outcomes
– Deadliest cancer among young adults with increasing incidence
– 5th most common type of cancer in America
• 73,000 new cases estimated this year
• 9,000 deaths are expected to occur
– >97% survivable with early detection
• Overdiagnosis is a problem
– Current clinical methods subjective
• Poor specificity (<60%)
– Imprecise histopathology standard
• Poor precision among pathologists (< 30% )
• 36 biopsies for every one melanoma confirmed
• Poor diagnostic precision adds an estimated $673 million in overall cost to
manage the disease
Melanoma diagnosis lacks precision…
Our goal is to develop an objective clinical-grade diagnostic!
87. Current melanoma (mis)diagnosis
• Dermatologists
– Problem: Have high sensitivity, but
low specificity
– Solution: Aggressive excision of
skin lesions
• General Practitioners
– Problem: Lack dermatology
expertise
– Solution: Quick referral to
dermatologist
• Residents
– Problem: Variable performance in
predicting melanoma
– Solution: Dermatology consult
The common method for identifying concerning moles is using
the ABCDE rule, which focuses on Asymmetry, Border, Color,
Diameter, and Evolution of a skin lesion.
The Ugly Duckling method is a newer classification model that
looks at the surrounding mole pattern to find the outliers that
might be cancerous.
Standard Mole Checkup
89. SkinDeep surveillance for Melanoma
• Aims
– Digital screening and molecular confirmation of
skin cancer with complete precision and
accuracy
– Real-time data collection platform for
multimodal surveillance and analytics of skin
lesion evolution
• Approach
– Physician prescribes smartphone app for
patient to follow their suspicions moles
– App analyzes the patient-captured image using
a DL screening algorithm and alerts physician of
results
– Physician can elect for our non-invasive
molecular diagnostic for confirmation
• Innovation
– Multimodal deep learning (DL) and predictive
algorithms
– Augmented reality capture and image analysis
in a mobile application
– Non-invasive molecular profiling of tumor
UCSF Inaugural Marcus Award for Precision Medicine (PIs: Judd, Wei, & Hadley)
90. SkinDeep Precision Diagnostics
• Self learning expert system for
precise melanoma diagnosis
– 83% accuracy with < 200 images
scraped from Google
– Converges to complete precision
with use
• Serial digital imaging currently
outperforms a general practitioner
– Expected > 90% accuracy with
enough training data
• Serial molecular profiling for
complete accuracy
– Multimodal DL algorithms
• First success led to melanoma
excision in UCSB CS prof!
Courtesy Abhishek Bhattacharya,
UCSB Undergraduate (CS/Bio, honors)
92. Summary
• We can use protected Big Data (hospital-based) to molecularly dissect disease and
personalize novel drugs and biomarker discovery
– Ex: ADHD & Autism
• But we already have a lot of open biomedical Big Data that can be used to better
characterize disease and discover novel drugs and biomarkers if structured
properly
– Ex: Severe Dengue
• Web-based tools are emerging to empower physician scientists to structure open
data and formulate genomics hypotheses about disease
– Ex: STARGEO nosology
• Emerging mobile technologies will facilitate bigger data collections and massive
recruitment facilitate digital health
– SkinIQ melanoma surveillance
We can translate the Big Data into Biomedical Innovation to DISRUPT MEDICINE!