Transcript Document

Genetics as a determinant of health:
new challenges for epidemiology
Julian Higgins
Senior Investigator Scientist, MRC Biostatistics Unit
and
Senior Epidemiologist, PHGU
Cambridge
Most diseases have a genetic component
Heart
disease
PKU
Schizophrenia
Cancer
Cystic
fibrosis
Duchenne
muscular
dystrophy
Totally
Genetic
Fragile X
Multiple
Diabetes sclerosis
Asthma
Motor
vehicle
accident
Alzheimers
TB
Struck
by
Meningococcus
lightning
Autism
Obesity
Rheumatoid
arthritis
Totally
Environmental
Genetics in epidemiology
• Understanding genetic components gives clues to
biological mechanisms
• Predicting disease
– population relevance
– targeting preventive strategies
• More effective therapy
– biological understanding to develop new interventions
– individualising treatments
– early diagnosis
• Information now available, post Human Genome Project
– Genotyping technology
Outline
•
•
•
•
•
•
Human genome epidemiology
Some obstacles
Overcoming the obstacles
HuGENetTM and the road ahead
Example: bladder cancer
Concluding remarks
Human genome epidemiology
Gene-disease
association
C
G
NAT2
carrier
NAT2
non-carr
Total
Cases
Controls
37
89
74
118
111
207
OR = 1.51
(95% CI
0.93 to 2.44)
Human genome epidemiology
Gene-disease
association
C
C
G
smoker NAT2
carrier
yes
non
carrier
no
non
G
×
Gene-environment
interaction
Total
Cases Control
29
63
37
74
73
76
35
56
175
268
Human genome epidemiology
Gene-disease
association
C
G
Gene-gene
interaction
C
×
A
C
G
Gene
prevalence
×
Gene-environment
interaction
T
G
Human genome epidemiology
Functional effect
• Gene-based analysis
• Mendelian deconfounding
– use genetic determinants of biomarkers to determine the
effects of the biomarkers
Human genome epidemiology
The full picture
Some obstacles to the new epidemiology
• Sample size
– small effects are expected
– variants may be uncommon
– interactions require very large samples
• Too many exposures
–
–
–
–
how to choose candidate genes?
many negative findings (less likely to be published)
spurious positive findings (more likely to be published)
major reporting biases
small effects are expected
• Relative risks tend to be small
• less than 1.5
• Pro12Ala polymorphism in PPARg2 gene has RR = 1.23
for type 2 diabetes
• (an often-quoted ‘established association’)
• 20 genes with common
variants can explain 50% of
common disease burden,
even if RR = 1.2-1.5
Yang et al (2005)
Some obstacles to the new epidemiology
• Sample size
– small effects are expected
– variants may be uncommon
– interactions require very large samples
• Too many exposures
–
–
–
–
how to choose candidate genes?
many negative findings (less likely to be published)
spurious positive findings (more likely to be published)
major reporting biases
interactions require very large samples
•
•
•
•
5% prevalence of gene variant
20% prevalence of environmental factor
RR = 1.5 for gene variant (generous)
RR = 2 for environmental factor
• Sample size to detect interaction RR = 2
in case-control study with 80% power:
2742 cases and 2742 controls
Some obstacles to the new epidemiology
• Sample size
– small effects are expected
– variants may be uncommon
– interactions require very large samples
• Too many exposures
–
–
–
–
how to choose candidate genes?
many negative findings (less likely to be published)
spurious positive findings (more likely to be published)
major reporting biases
major reporting biases
>10,000,000
 >1000
 >10
 >10
 >5
 >10
Gene variants
Diseases
Outcomes
Subgroups
Genetic contrasts
Investigators
= > 5 trillion candidate analyses!
after Ioannidis (2003)
Some obstacles to the new epidemiology
• Other biases
– genotyping errors
– choice of controls (population stratification)
– other standard biases
• Variation and poor replicability
– reporting and other biases
– sample size
– markers with different degrees of linkage to functional
variant
– other population characteristics
Overcoming the obstacles
• Lots of data
– large cohort and case-control studies, e.g. EPIC, NHANES
– UK Biobank
• Honest publication
– negative results
– web databases
• Collaboration and synthesis
– consortia of investigators
– meta-analyses
• Methods development
– integrating multiple exposures
Human Genome Epidemiology Network
(HuGENetTM)
• A global collaboration of individuals and organizations
committed to
– the assessment of the impact of human genome
variation on population health
– how genetic information can be used to improve
health and prevent disease
• Undertaking systematic reviews and meta-analyses
• Collating evidence to inform policy, practice and
research
The roadmap
SINGLE TEAMS
SINGLE STUDIES
Feedback
FIELD-WIDE
SYNOPSES
Reporting
HuGENet
Network of
Networks
PUBLISHED AND
UNPUBLISHED DATA
Grading
Synthesis
SYSTEMATIC REVIEWS
META-ANALYSES
Ioannidis et al (2006)
A systematic review
• Joint effects of NAT1, NAT2 and smoking on bladder
cancer risk
• NAT2 gene: ‘rapid’ acetylator version metabolises
aromatic amines in tobacco smoke quicker
• NAT1 gene: believed to activate aromatic amines
(so ‘slow’ version would be better)
Rather disappointing
Gene-gene-environment joint effects in
bladder cancer: a single study
Smoking
NAT1
Slow
No
Rapid
Slow
Yes
Rapid
NAT2
Cases
Controls
OR
Rapid
Slow
Rapid
Slow
6
16
8
6
13
31
16
10
1
1.12 (0.36, 3.5)
1.08 (0.30, 3.9)
1.30 (0.32, 5.3)
Rapid
Slow
Rapid
42
61
41
32
51
26
2.84 (0.97, 8.3)
2.59 (0.92, 7.3)
3.42 (1.2, 10.1)
Slow
35
12
6.32 (2.0, 20.3)
Taylor et al (1998)
Complex evidence synthesis
• It turns out we can learn about the joint effects using
studies of bladder cancer and…
–
–
–
–
–
–
NAT1
NAT2
NAT1 and NAT2
NAT1 and smoking
NAT2 and smoking
smoking
– with assumptions
More exciting
Single study vs synthesis of 28 studies
Smoking
NAT1
Slow
No
Rapid
Slow
Yes
Rapid
NAT2
OR-Taylor
OR-synthesis
Rapid
1
1
Slow
1.12 (0.36, 3.5)
0.98 (0.52, 1.8)
Rapid
1.08 (0.30, 3.9)
0.70 (0.28, 1.6)
Slow
1.30 (0.32, 5.3)
1.12 (0.52, 2.2)
Rapid
2.84 (0.97, 8.3)
1.53 (0.75, 3.0)
Slow
2.59 (0.92, 7.3)
2.23 (1.3, 3.8)
Rapid
3.42 (1.15, 10.1)
1.37 (0.74, 2.4)
Slow
6.32 (2.0, 20.3)
2.88 (1.6, 5.0)
Concluding remarks:
New challenges for epidemiology
• Large amounts of data
– big studies
– collaborative, coordinated research
• Investigating vast numbers of exposures
– methods to home in on the truth
– integrating multiple genes and environmental factors
• Evaluating the technologies
– in partnership with UK GTN etc
Warfarin
• This week’s BMJ:
– Warfarin is underprescribed to patients with atrial
fibrillation
– Physicians are less likely to prescribe warfarin after one of
their patients has a major adverse bleeding event associated
with warfarin
• Can human genome epidemiology help?
Choudhry et al (2006)
CYP2C9 variants and warfarin
metabolism
Mean difference in daily dose: 2C9*2 carriers versus non-carriers
Aithal
Taube
Furuya
Margaglione
Loebstein
Tabrizi
Scordo
Higashi
–0.85 (– 1.11, – 0.60)
Meta-analysis
-2
-1
0
1
2
Difference in mean warfarin dose (mg per day)
Lower dose for carriers
Sanderson et al (2005)