Size matters: the value of large scale epidemiology

1
Size matters the value of large scale
epidemiology

• Paul Burton
• Professor of Genetic Epidemiology
• University of Leicester
• P³G Consortium
• PHOEBE

2
A daunting task!

• Need for extensive, valid information
• Developments in biotechnology, IT
• Pre-morbid and longitudinal life-style/environment
  relevant
• Bioclinical complexity ? low statistical power!!

3
Large scale genetic epidemiology

• Focus on the aetiology of complex diseases
• Common disease common variant hypothesis
• a shift in paradigm from linkage to association
• BUT serious failure to identify associations
  that can consistently be replicated

4
Hattersley AT, McCarthy MI. Lancet
20053661315-1323 Examples of some polymorphisms
or haplotypes that have shown consistent
association with complex disease
5
Why has replicationproved to be so difficult?

• Poorly designed studies
• e.g. wrong controls, family v non-family designs
• Poorly conducted analyses and meta-analyses
• e.g. use of inefficient or inconsistent methods
  failure to take proper account of extreme
  multiple testing publication and/or reporting
  bias
• Inconsistent definitions of outcome or exposure
• e.g. what do we mean by asthma?
• Poor methods of assessment
• e.g. bad choice of SNP genotyping platform

6
Why has replicationproved to be so difficult?

• Heterogeneity
• e.g. stroke encompasses important
  subcategories phenocopies pleiotropy
• Population substructure
• Latent stratification and admixture pertaining to
  population of origin

7
Why has replicationproved to be so difficult?

• LOW STATISTICAL POWER!!
• A key feature of almost all proffered
  explanations, and/or of the approaches needed to
  correct for them
• If we need 5,000 cases to test for a given
  aetiological effect with a power of 80, and with
  a critical p-value of 0.0001, how much power
  would there be for a study with 500 cases?

8
Why has replicationproved to be so difficult?

• LOW STATISTICAL POWER!!
• A key feature of almost all proffered
  explanations, and/or of the approach needed to
  correct for them
• If we need 5,000 cases to test for a given
  aetiological effect with a power of 80, and with
  a critical p-value of 0.0001, how much power
  would there be for a study with 500 cases?

?0.008!!
9
How should we respond?

• Increase the quality of individual studies
• Limit measurement/assessment error
• Increase the size of individual studies
• Promote harmonization to enable data pooling and
  integration

10
How should we respond?

• Increase the quality of individual studies
• Limit measurement/assessment error
• Increase the size of individual studies
• Promote harmonization to enable data pooling and
  integration

? MAJOR international investment in biobanks
and biobank harmonization
11
What is a biobank?

• An organised collection of human biological
  material and associated information stored for
  one or more research purposes
• Population Biobanks Lexicon (P3G, PHOEBE)
• Types
• Disease-specific
• Exposure-focused
• Population-based

12
Justification for large-scalegenetic-epidemiology
programs
13
BIG per se

• No argument about
• Need to increase statistical power
• Benefit of constructing biobanks containing
  extensive case-series for case-control studies
• Benefit of constructing large acceptably
  representative series of controls for each nation

14
BIG cohort studies

• Studies of the joint effects of genes and
  environment/life-style
• Genotype-based studies
• The genetics of disease progression
• Direct association of genes with disease
• Population-based replication studies
• Universal controls

15
BUT how big is big?With Anna Hansell,
Imperial College
16
The statistical power ofcase-control studies

• Contemporary pre-eminence of genetic association
  studies rather than genetic linkage studies
• Covers both stand-alone case-control studies, and
  nested case-control studies in large cohorts.
  Main issue is the number of cases.
• Sample size determining in both settings

17
Simulation-based power calculations

• Work with the least powerful (common) setting
• Disease outcome and exposures all binary
• Logistic regression interactions departure
  from a multiplicative model
• Complexity (arbitrary but realistic).
• Four controls per case

18
Diabetes mellitus defined by HbA1C 97.5
percentile
19
Genetic main effects
Prevalence of at-risk genotype 0.1, 0.5
20
Lifestyle main effects
Prevalence of at-risk life-style determinant
0.5
Reliability 1.0 measured height 0.9 self
reported weight 0.7 office BP, measured
serum cholesterol 0.5 dietary recall of
many components (424 hr recalls)
21
Gene-lifestyle interactions
Prevalence of at-risk genotype 0.1
Prevalence of at-risk life-style determinant
0.5
22
Mean power ? 55
23
What is needed?

• Genetic main effects
• 2,500-10,000 cases
• Life-style main effects
• 5,000-20,000 cases
• Gene-lifestyle interactions
• Probably need at least 20,000 cases

24
How can this be achieved?

• Large disease-based biobanks
• Very large cohort-based biobanks
• But how large do these need to be?

25
Expected event ratesin UK BiobankWith Anna
Hansell, Imperial College
26
Taking account of

• Age range at recruitment 40-69 years
• Recruitment over 5 years
• All cause mortality
• Disease incidence (healthy cohort effect)
• Migration overseas
• Withdrawal from the study

27
(No Transcript)
28
Conclusions

• Having taken account of realistic bioclinical
  complexity, a cohort-based biobank needs to be
  very large if it is to provide a stand-alone
  infrastructure
• Anything much less than 500,000 recruits severely
  curtails the number of diseases that will be able
  to be studied based on that biobank alone
• The value of any biobank will be greatly
  augmented if it proves possible to set up a
  coherent and scientifically harmonized
  international network of biobanks

29
What is biobank harmonization?
30
Biobank harmonization

• A set of procedures that promote, both now and in
  the future, the effective interchange of valid
  information and samples between a number of
  studies or biobanks, accepting that there may be
  important differences between those studies
• With thanks to Alastair Kent

31
Biobank harmonization

• Prospective harmonization
• Aims to modify study design and conduct, ahead of
  time, in order to render subsequent data and
  sample pooling more efficient and more
  straightforward
• Retrospective harmonization
• Aims to optimize the pooling of data, samples and
  phenotypes that have already been collected,
  between studies with inevitably heterogeneous
  designs.

32
Why harmonize?

• Investigate less common (but not rare!!!)
  conditions
• UKBB Ca stomach 2,500 cases in 29 years
• 6 UKBB equivalents ? 10,000 cases in 20 years
• Investigate smaller ORs
• GME 1.5 ? 1.2 requires 2,000 ? 12,600
• 6.3 UKBB equivalents
• Analysis based on subsets homogeneous classes
  of phenotype, or e.g. by sex

33
Why harmonize?

• Earlier analyses
• UKBB Alzheimers disease, 10,000 cases in 18 yrs
• 5 UKBB equivalents ? 9 years
• Events at younger ages
• Broad range of environmental exposures
• Aim for 5-6 UKBB equivalents
• 2.5M 3M recruits

34
Some key issues

• Scientifically and politically VERY challenging
• Laboratory science, clinical science, population
  science, IT challenges, ethico-legal issues
• A need for REAL collaboration and tools that are
  ACCESSIBLE and USABLE
• Case-control and cohort studies

35
International biobankharmonization programs

• Public Population Program in Genomics (P3G)
• Tom Hudson, Bartha Knoppers, Isabel Fortier
• Population Biobanks
• FP6 Co-ordination Action (PHOEBE Promoting
  Harmonization Of Epidemiological Biobanks in
  Europe)
• Jennifer Harris, Leena Peltonen, Paul Burton
• Human Genome Epidemiology Network (HuGENet)
• Muin Khoury, Julian Little
• ESSENTIAL THAT ALL INITIATIVES WORK TOGETHER!!

36
Extra slides
37
Rarer genotypes

• Genetic main effects

38
Proposed assessment visit model
39
Taking account of

• Age range at recruitment 40-69 years
• Recruitment over 5 years
• All cause mortality
• Disease incidence (healthy cohort effect)
• Migration overseas
• Comprehensive withdrawal (max 1/500 p.a.)
• Partial withdrawal (c.f. 1958 Birth Cohort)

40
(No Transcript)
41
Necessary to contact subjects
42
Issues that are often ignored in standard power
calculations

• Multiple testing/low prior probability of
  association
• Interactions
• Unobserved frailty
• Misclassification
• Genotype
• Environmental determinant
• Case-control status
• Subgroup analyses
• Population substructure

43
Harmonisation

• Prospective
• Retrospective
• Description
• Comparison
• Harmonised synthesis

44
(No Transcript)
45
(No Transcript)
46
Recruitment and assessment

• Recruitment via centrally held list of
  individuals registered with Primary Care
  Practitioners (GPs)
• Assessment in large centres (?100 subjects per
  day)
• Assessment ? 70 minutes
• Questionnaire, physical examination, bloods

47
Assessment visit model
48
Summary

• 80 power for genotype frequency 0.1
• Genetic main effect ? 1.5, p10-4 ? 2,000 cases
• Genetic main effect ? 1.3, p10-4 ? 5,500 cases
• Genetic main effect ? 1.2, p10-4 ? 12,600 cases
• Genetic main effect ? 1.7, p10-7 ? 2,000 cases
• Genetic main effect ? 1.5, p10-7 ? 3,400 cases
• Genetic main effect ? 1.3, p10-7 ? 9,500 cases
• Genetic main effect ? 1.2, p10-7 ? 21,500
  cases
• GE interaction with environmental exp.
  prevalence
• 0.5 ? 2.0, p10-4 ? 10,000 to 30,000
  cases

49
UK Biobank

• A prospective cohort study
• 500,000 adults (40-69 years) across UK
• A population-based biobank
• Not disease or exposure based
• Recruitment via electronic GP lists
• Broad spectrum not fully representative
• Individuals not families
• MRC, Wellcome Trust, DH, Scottish Executive
• 61M

50
UK Biobank

• Initial data/sample collection and subsequent
  longitudinal health tracking
• Nested case-control studies
• Long time-horizon
• Owned by the Nation
• Central Administration Manchester
• PI Prof Rory Collins - Oxford
• 6 collaborating groups (RCCs) of university
  scientists

51
Smaller sample sizes