Do we all have an actionable

genome? Evidence from, and implications for,

registries and electronic health records

Professor Harry Hemingway, MD, FFPH, FRCP Director Farr Institute of Health Informatics Research, London

Director Institute of Health Informatics, University College London, London, UK

REGISTER IN DER MEDIZIN: WO STEHT DIE SCHWEIZ?

1 Februar 2018, UniS Bern

“justify any research application that does not include genomic analysis”

Annual Report, Chief Medical Officer of England, Dame Sally Davies 2017

Global Challenges

“Broken models of drug development and trials ” Eroom’s law

“Informatics- the ‘clean water’ of 21st c. public

health” Dame Anne Johnson, 2016

“Mining healthcare data … save 100,000 lives next year” Larry Page, 2014

Unsustainable health/social care: Costs outstrip GDP growth

King’s Fund “The new era of data based and

more precise medicine”

White House 2015

Data

21st Century challenge for HEALTH SYSTEMS: value from data

Better health outcomes for individuals and populations

[adapted from] Jameson NEJM 2015

21st Century challenge for HEALTH SYSTEMS: value from data at scale

Near term patient benefit, with precision Diagnosis Screening Treatment Prognostics Decision support

Longer term benefit for drug development, innovation and society

All driven by an emerging clinical research paradigm: Evidence Generating Medicine at scale

Scal

e: To

tal p

opul

atio

n si

ze (m

illio

ns)

Registries Longitudinal EHR

Exomes Multi-omics

Research Imaging + Sequencing (exomes)

50,000

500,000

5,000,000

15,000,000

Genome array Res’ pheno

MONDRIAAN

UK BIOBANK ♦

INTERVAL♦

SCALE and DEPTH ecosystem of data

Depth: phenotypic and omic

Health system - Health system + DNA

Research cohort + DNA

Disease-based + DNA

Disease-based -

GWAS

UK Biobank

Geisinger

CALIBER Phenome wide

Hemingway et al Eur heart J 2017

Research cohort e.g UK Biobank

Health system e.g.Geisinger

Current N participants

500k 50k

Research measures

MRI, wearables minor

Open access

Yes ++ no

Scalable

No, fixed yes

Health system embedded

no yes

Return clin’ actionable results

no yes

RESOURCES WITH GENOME SEQUENCE + EHR PHENOME SEQUENCE

Phenotypic depth

Tota

l pop

ulat

ion

size

(mill

ions

)

Registries Longitudinal EHR Multi-omics Imaging + Sequencing (exomes)

50,000

500,000

5,000,000

15,000,000

ABUCASIS

CALIBER

UK CV registries

NICOR

ESC EOPR

Swedish CV registries

SWEDEHEART

HERMES ♦ AFGEN ♦

EPIC CVD ♦

GENIUS CHD ♦

eMERGE ♦

BioVU ♦

NIHR HIC

Hospital-based

Population-based

Disease-based

Genomics

RPGEH ♦

Proposed PMI

Cohort Initiative ♦

MVP ♦

MONDRIAAN

UCLEB ♦

UK BIOBANK ♦

INTERVAL♦

Kadoorle♦

* With genetic information

Geisinger

Each of us differs in our genome sequence

Evans et al., JAMA, 2017 100 complete ‘knockouts’ genes with Loss of Function

A lot of us are getting sequenced

estimated 200 million people by 2025

Consider 4 areas in which genome sequence may be

actionable

1. Rare monogenic diseases N=5000

1. Rare monogenic

diseases N=5,000

2. Common complex diseases

N=10,000

1. Rare monogenic

diseases N=5,000

3. Drugs N=1400

2. Common complex diseases

N=10,000

1. Rare monogenic

diseases N=5,000

3. Drugs N=1400

2. Common complex diseases

N=10,000

4. Behaviours N=umpteen

Aka Omics: as the ultimate complex intervention?

1. MONOGENIC DISEASES

One monogenic disease: Familial Hypercholesterolaemia (FH)

Autosomal recessive with many variants described in LDLR, APOB, and PCSK9 How common is FH? No previous large scale WES studies Geisinger Health System EHR + Regeneron whole exome sequencing N = 229 / 50,726 patients had one or more of 35 known and predicted pathogenic variants Prevalence – higher x 2 than previous estimates or prevalence: 1:256 Abul-Husn Geisinger-Regeneron et al Science 2016

Health System Wide Screening for FH with whole exome sequencing (WES)

reveals undertreatment n=50,276

Abul-Husn et al Science 2016

But there are 5000-7000 monogenic diseases (20 000 protein coding genes)

• What is the combined prevalence of all monogenic

diseases? 8 % estimated in Europe

• …most detectable with this same ‘test’ of whole genome sequencing

• Some of these gene variants are considered clinically actionable even if the patient has – NO family history and – NO symptoms i.e. healthy population

28 monogenic diseases are currently recommended actionable ‘secondary findings’

by American College of Medical Genetics Disease name and MIM number Gene via GTR Action

Familial hypercholesterolemia (MIM 143890)

APOB (MIM 107730) LDLR (MIM 606945)

Drugs: Statins, cascade screening

Adenomatous polyposis coli (MIM 175100)

APC (MIM 611731) Endoscopic: surveillance

Aortic aneurysm, familial thoracic 4 (MIM 132900)

MYH11 (MIM 160745) Imaging: surveillance

Breast-ovarian cancer, familial 1 (MIM 604370)

BRCA1 (MIM 113705) Surgery: Prophylactic

Long QT syndrome 2 (MIM 613688) KCNH2 (MIM 152427) Device: implantable

cardiodefibrillator …. for all 28 diseases. Hunter Genetics In Medicine 2016

HOW COMMON ARE THESE 28 MONOGENIC DISEASES CLINICALLY ACTIONABLE VARIANTS?

Dewey et al DiscovEHR Geinsinger-RegeneronScience 2016

• Until recently we didn’t know –most exomes done on patients SELECTED on basis of disease • Whole exome sequencing in Geisinger health system: Patients were NOT SELECTED on the basis of having a rare disease or for having family history

49/1415 = 3.5% prevalence

Registers play a key role

In understanding whether a gene variant is truly pathogenic and actionable As whole genome sequencing becomes a clinical test this need increases

2. COMMON COMPLEX DISEASES

Actions from genomic variants for Complex Diseases

• Based on causal understanding of mechanisms of disease(s) • Including genomic approaches to drug target

validation (e.g. ezetimbie and NPC1L1 NEJM 2014)

• Based on massive abilities to predict onset and progression of many diseases (irrespective of causal relevance)

One disease: coronary disease • Primary prevention with statins is based on absolute

risk (Framingham risk scores, Q risk etc)

• Based on <20 factors: age, sex, smoking, BP, chol

• What if we used millions of pieces of information (genetic variants) instead?

• Change clinical decision making, by classifying individuals above treatment threshold (10% risk) at earlier age?

One disease: coronary disease 20% of the general population with higher GRS, exceed risk threshold at much

younger age Based on 1.7 million genetic variants

Inouye et al UK Biobank Cardiometabolic Consortium

Potential for earlier diagnosis of MANY diseases: e.g. cancer?

‘liquid biopsies’ -cell free tumour DNA (2001 genomic positions) + 8 proteins

Cohen et al Science 2018

Registers play a key role

• Guess where CancerSEEK is mounting its 50k patient trial?

• ‘DNA samples’ are collected and discarded unanalysed by the billion each day…….

• Need long term support

• Need linkage between registers

3 DRUGS

Actions from genomic variants for DRUG response

• To avoid harms (safety) and deliver benefits (efficacy) – Right patient: cystic fibrosis and ivacaftor – Right drug: Stevens Johnson Synd, abacavir trial – Right dose: bleeding, warfarin trial – Right time: early in prevention e.g. Inouye 2018

– Across a wide range of drugs taken at different stages in the lifecourse

One Drug Gentotype Interaction: randomised evidence of effectiveness?

Gage et al GIFT JAMA 2017

Five Drug-Genotype Interactions

Five Drug-Genotype Interactions How common are these variants?

>90% of ‘healthy’ population

How common is exposure to one of these drugs in a 60 year old over a 5 year period?

• (only 1400 approved drugs)

• Clinical Pharmacogenetics Consortium (CPIC) – Give recommendations in the setting of the

genetic result

• Pre-emptive pharmacogenetics: – variants already in the Register or EHR with

decision support ready if and when a drug is

130 drugs have an FDA label saying ‘consider’ genetic testing

4. Behaviours

Actions from genomic variants for health related behaviours

• To avoid harms (safety) and deliver benefits

(efficacy) – Right behaviour: – Right dose: – Right time: early in prevention

– Across wide range of behaviours (not one at a time) over the ‘5000 hover hours per year’ Asch NEJM 2012

Smoking pharmacogenetics of nicotine and nicotine replacement therapy (NRT)

e.g. does NRT effectiveness differ by genotype

1

Lerman, Am J Prev Med. 2007

NRT Bupropion Varenicline

What Diet and Exercise works best for your genes?

1. Download your 23and me file 2. Search for the best SNPs below

rockstarresearch.com

Registers play a key role

• Bringing together the ‘always on’ data about behaviours from mobile and wearables with clinical data

1. Rare monogenic

diseases N=5,000

3. Drugs N=1400

2. Common complex diseases

N=10,000

4. Behaviours N=umpteen

Preventive Genomics: Pilot Randomised Trial

• Returning results from Whole Genome

Sequencing in primary care 1. Monogenic disease carrier status n=28 diseases 2. Complex diseases risks n=8 Genetic Risk Scores 3. Drug-gene interactions n=5 drug-gene pairs 4. Behaviours n=0

Vassy JL, et al MedSeq Project. Ann Intern Med. 2017

Preventive Genomics: Pilot Randomised Trial

• 51 patients WGS vs 50 patients family history

only

Vassy JL, et al MedSeq Project. Ann Intern Med. 2017

Monogenic disease risk 22% vs 0% New clinical actions recommended 34% vs 16% Behaviours change at 6 months 41% vs 30%

UK Approach?

Conclusion: Do we all have an actionable genome?

Without large scale integrated registers and EHRs with embedded genomics we will never know!

b

Thank you