Tom Palmer, John Thompson and Martin Tobin · Tom Palmer, John Thompson and Martin Tobin Department...
Transcript of Tom Palmer, John Thompson and Martin Tobin · Tom Palmer, John Thompson and Martin Tobin Department...
![Page 1: Tom Palmer, John Thompson and Martin Tobin · Tom Palmer, John Thompson and Martin Tobin Department of Health Sciences, University of Leicester, UK ISCB 28, 1 August 2007 Tom Palmer](https://reader034.fdocuments.us/reader034/viewer/2022042710/5f68aa53d28cb25bd54d2a35/html5/thumbnails/1.jpg)
Meta-analysis of Mendelian randomization studies
Tom Palmer, John Thompson and Martin Tobin
Department of Health Sciences,University of Leicester, UK
ISCB 28,1 August 2007
Tom Palmer (Leicester) Meta-analysis of MR studies ISCB 28, 1 August 2007 1 / 14
![Page 2: Tom Palmer, John Thompson and Martin Tobin · Tom Palmer, John Thompson and Martin Tobin Department of Health Sciences, University of Leicester, UK ISCB 28, 1 August 2007 Tom Palmer](https://reader034.fdocuments.us/reader034/viewer/2022042710/5f68aa53d28cb25bd54d2a35/html5/thumbnails/2.jpg)
Outline
1 Brief introduction to Mendelian randomization
2 Concepts in the meta-analysis model
3 Description of the example and the meta-analysis model
4 An extension to the meta-analysis model
5 Summary and discussion
Tom Palmer (Leicester) Meta-analysis of MR studies ISCB 28, 1 August 2007 2 / 14
![Page 3: Tom Palmer, John Thompson and Martin Tobin · Tom Palmer, John Thompson and Martin Tobin Department of Health Sciences, University of Leicester, UK ISCB 28, 1 August 2007 Tom Palmer](https://reader034.fdocuments.us/reader034/viewer/2022042710/5f68aa53d28cb25bd54d2a35/html5/thumbnails/3.jpg)
Brief introduction to Mendelian randomization
Dates back to [Katan, 1986], recent revival [Katan, 2004]
Association between biological phenotype and a disease
Problems of confounding and reverse causation
Mendel’s 2nd law: individual randomized to a genotype at conception
Phenotype on pathway between gene and disease
Tom Palmer (Leicester) Meta-analysis of MR studies ISCB 28, 1 August 2007 3 / 14
![Page 4: Tom Palmer, John Thompson and Martin Tobin · Tom Palmer, John Thompson and Martin Tobin Department of Health Sciences, University of Leicester, UK ISCB 28, 1 August 2007 Tom Palmer](https://reader034.fdocuments.us/reader034/viewer/2022042710/5f68aa53d28cb25bd54d2a35/html5/thumbnails/4.jpg)
Brief introduction to Mendelian randomization
Dates back to [Katan, 1986], recent revival [Katan, 2004]
Association between biological phenotype and a disease
Problems of confounding and reverse causation
Mendel’s 2nd law: individual randomized to a genotype at conception
Phenotype on pathway between gene and disease
Tom Palmer (Leicester) Meta-analysis of MR studies ISCB 28, 1 August 2007 3 / 14
![Page 5: Tom Palmer, John Thompson and Martin Tobin · Tom Palmer, John Thompson and Martin Tobin Department of Health Sciences, University of Leicester, UK ISCB 28, 1 August 2007 Tom Palmer](https://reader034.fdocuments.us/reader034/viewer/2022042710/5f68aa53d28cb25bd54d2a35/html5/thumbnails/5.jpg)
Brief introduction to Mendelian randomization
Dates back to [Katan, 1986], recent revival [Katan, 2004]
Association between biological phenotype and a disease
Problems of confounding and reverse causation
Mendel’s 2nd law: individual randomized to a genotype at conception
Phenotype on pathway between gene and disease
Tom Palmer (Leicester) Meta-analysis of MR studies ISCB 28, 1 August 2007 3 / 14
![Page 6: Tom Palmer, John Thompson and Martin Tobin · Tom Palmer, John Thompson and Martin Tobin Department of Health Sciences, University of Leicester, UK ISCB 28, 1 August 2007 Tom Palmer](https://reader034.fdocuments.us/reader034/viewer/2022042710/5f68aa53d28cb25bd54d2a35/html5/thumbnails/6.jpg)
Brief introduction to Mendelian randomization
Dates back to [Katan, 1986], recent revival [Katan, 2004]
Association between biological phenotype and a disease
Problems of confounding and reverse causation
Mendel’s 2nd law: individual randomized to a genotype at conception
Phenotype on pathway between gene and disease
Tom Palmer (Leicester) Meta-analysis of MR studies ISCB 28, 1 August 2007 3 / 14
![Page 7: Tom Palmer, John Thompson and Martin Tobin · Tom Palmer, John Thompson and Martin Tobin Department of Health Sciences, University of Leicester, UK ISCB 28, 1 August 2007 Tom Palmer](https://reader034.fdocuments.us/reader034/viewer/2022042710/5f68aa53d28cb25bd54d2a35/html5/thumbnails/7.jpg)
Brief introduction to Mendelian randomization
Dates back to [Katan, 1986], recent revival [Katan, 2004]
Association between biological phenotype and a disease
Problems of confounding and reverse causation
Mendel’s 2nd law: individual randomized to a genotype at conception
Phenotype on pathway between gene and disease
Tom Palmer (Leicester) Meta-analysis of MR studies ISCB 28, 1 August 2007 3 / 14
![Page 8: Tom Palmer, John Thompson and Martin Tobin · Tom Palmer, John Thompson and Martin Tobin Department of Health Sciences, University of Leicester, UK ISCB 28, 1 August 2007 Tom Palmer](https://reader034.fdocuments.us/reader034/viewer/2022042710/5f68aa53d28cb25bd54d2a35/html5/thumbnails/8.jpg)
θ
Genotype Phenotype Disease
�������
HHHHHHH
-δ -η
θ: Gene-Disease log odds-ratio, δ: difference in mean phenotypes,η: Phenotype-Disease log odds-ratio
Instrumental-variable (IV) methods for continuous outcome measures
IV estimates using a binary disease outcome are approximations
Ratio of coefficients approach,
η[k] ≈kθ
δ
Tom Palmer (Leicester) Meta-analysis of MR studies ISCB 28, 1 August 2007 4 / 14
![Page 9: Tom Palmer, John Thompson and Martin Tobin · Tom Palmer, John Thompson and Martin Tobin Department of Health Sciences, University of Leicester, UK ISCB 28, 1 August 2007 Tom Palmer](https://reader034.fdocuments.us/reader034/viewer/2022042710/5f68aa53d28cb25bd54d2a35/html5/thumbnails/9.jpg)
θ
Genotype Phenotype Disease
�������
HHHHHHH
-δ -η
θ: Gene-Disease log odds-ratio, δ: difference in mean phenotypes,η: Phenotype-Disease log odds-ratio
Instrumental-variable (IV) methods for continuous outcome measures
IV estimates using a binary disease outcome are approximations
Ratio of coefficients approach,
η[k] ≈kθ
δ
Tom Palmer (Leicester) Meta-analysis of MR studies ISCB 28, 1 August 2007 4 / 14
![Page 10: Tom Palmer, John Thompson and Martin Tobin · Tom Palmer, John Thompson and Martin Tobin Department of Health Sciences, University of Leicester, UK ISCB 28, 1 August 2007 Tom Palmer](https://reader034.fdocuments.us/reader034/viewer/2022042710/5f68aa53d28cb25bd54d2a35/html5/thumbnails/10.jpg)
θ
Genotype Phenotype Disease
�������
HHHHHHH
-δ -η
θ: Gene-Disease log odds-ratio, δ: difference in mean phenotypes,η: Phenotype-Disease log odds-ratio
Instrumental-variable (IV) methods for continuous outcome measures
IV estimates using a binary disease outcome are approximations
Ratio of coefficients approach,
η[k] ≈kθ
δ
Tom Palmer (Leicester) Meta-analysis of MR studies ISCB 28, 1 August 2007 4 / 14
![Page 11: Tom Palmer, John Thompson and Martin Tobin · Tom Palmer, John Thompson and Martin Tobin Department of Health Sciences, University of Leicester, UK ISCB 28, 1 August 2007 Tom Palmer](https://reader034.fdocuments.us/reader034/viewer/2022042710/5f68aa53d28cb25bd54d2a35/html5/thumbnails/11.jpg)
θ
Genotype Phenotype Disease
�������
HHHHHHH
-δ -η
θ: Gene-Disease log odds-ratio, δ: difference in mean phenotypes,η: Phenotype-Disease log odds-ratio
Instrumental-variable (IV) methods for continuous outcome measures
IV estimates using a binary disease outcome are approximations
Ratio of coefficients approach,
η[k] ≈kθ
δ
Tom Palmer (Leicester) Meta-analysis of MR studies ISCB 28, 1 August 2007 4 / 14
![Page 12: Tom Palmer, John Thompson and Martin Tobin · Tom Palmer, John Thompson and Martin Tobin Department of Health Sciences, University of Leicester, UK ISCB 28, 1 August 2007 Tom Palmer](https://reader034.fdocuments.us/reader034/viewer/2022042710/5f68aa53d28cb25bd54d2a35/html5/thumbnails/12.jpg)
θ
Genotype Phenotype Disease
�������
HHHHHHH
-δ -η
θ: Gene-Disease log odds-ratio, δ: difference in mean phenotypes,η: Phenotype-Disease log odds-ratio
Instrumental-variable (IV) methods for continuous outcome measures
IV estimates using a binary disease outcome are approximations
Ratio of coefficients approach,
η[k] ≈kθ
δ
Tom Palmer (Leicester) Meta-analysis of MR studies ISCB 28, 1 August 2007 4 / 14
![Page 13: Tom Palmer, John Thompson and Martin Tobin · Tom Palmer, John Thompson and Martin Tobin Department of Health Sciences, University of Leicester, UK ISCB 28, 1 August 2007 Tom Palmer](https://reader034.fdocuments.us/reader034/viewer/2022042710/5f68aa53d28cb25bd54d2a35/html5/thumbnails/13.jpg)
Information from a case-control study
A biallellic polymorphism (g,G); g: common allele, G: risk allele;3 genotypes: gg, Gg, GG; (j = 1, 2, 3)
d = 0,1: control/case,ydj : observed cases and controls,
gg Gg GG
Controls y01 y02 y03
Cases y11 y12 y13
log odds-ratios θ2 θ3
Mean phenotype levels (controls) µ1 µ2 µ3
difference in mean phenotypes δ2 δ3
Tom Palmer (Leicester) Meta-analysis of MR studies ISCB 28, 1 August 2007 5 / 14
![Page 14: Tom Palmer, John Thompson and Martin Tobin · Tom Palmer, John Thompson and Martin Tobin Department of Health Sciences, University of Leicester, UK ISCB 28, 1 August 2007 Tom Palmer](https://reader034.fdocuments.us/reader034/viewer/2022042710/5f68aa53d28cb25bd54d2a35/html5/thumbnails/14.jpg)
Information from a case-control study
A biallellic polymorphism (g,G); g: common allele, G: risk allele;3 genotypes: gg, Gg, GG; (j = 1, 2, 3)
d = 0,1: control/case,ydj : observed cases and controls,
gg Gg GG
Controls y01 y02 y03
Cases y11 y12 y13
log odds-ratios θ2 θ3
Mean phenotype levels (controls) µ1 µ2 µ3
difference in mean phenotypes δ2 δ3
Tom Palmer (Leicester) Meta-analysis of MR studies ISCB 28, 1 August 2007 5 / 14
![Page 15: Tom Palmer, John Thompson and Martin Tobin · Tom Palmer, John Thompson and Martin Tobin Department of Health Sciences, University of Leicester, UK ISCB 28, 1 August 2007 Tom Palmer](https://reader034.fdocuments.us/reader034/viewer/2022042710/5f68aa53d28cb25bd54d2a35/html5/thumbnails/15.jpg)
Information from a case-control study
A biallellic polymorphism (g,G); g: common allele, G: risk allele;3 genotypes: gg, Gg, GG; (j = 1, 2, 3)
d = 0,1: control/case,ydj : observed cases and controls,
gg Gg GG
Controls y01 y02 y03
Cases y11 y12 y13
log odds-ratios θ2 θ3
Mean phenotype levels (controls) µ1 µ2 µ3
difference in mean phenotypes δ2 δ3
Tom Palmer (Leicester) Meta-analysis of MR studies ISCB 28, 1 August 2007 5 / 14
![Page 16: Tom Palmer, John Thompson and Martin Tobin · Tom Palmer, John Thompson and Martin Tobin Department of Health Sciences, University of Leicester, UK ISCB 28, 1 August 2007 Tom Palmer](https://reader034.fdocuments.us/reader034/viewer/2022042710/5f68aa53d28cb25bd54d2a35/html5/thumbnails/16.jpg)
Description of the example
Mann et al. The Journal of Clinical Investigation, (2001);phenotype - bone mineral denisty (BMD), disease - osteoporotic
fracture, gene - COL1A1 codes for collagen
Tom Palmer (Leicester) Meta-analysis of MR studies ISCB 28, 1 August 2007 6 / 14
![Page 17: Tom Palmer, John Thompson and Martin Tobin · Tom Palmer, John Thompson and Martin Tobin Department of Health Sciences, University of Leicester, UK ISCB 28, 1 August 2007 Tom Palmer](https://reader034.fdocuments.us/reader034/viewer/2022042710/5f68aa53d28cb25bd54d2a35/html5/thumbnails/17.jpg)
Description of the example
Mann et al. The Journal of Clinical Investigation, (2001);phenotype - bone mineral denisty (BMD), disease - osteoporotic
fracture, gene - COL1A1 codes for collagen
Heegard 2000
Liden 1998
Langdahl 1996
Braga 2000
Uitterinden 1998a
Weichetova 2000
Hustmyer 1999
Keen 1999
McGuigen 2000
Grant 1996a
Roux 1998
Uitterinden 1998b
Alvarez 1999
Harris 2000
Sowers 1999
Hampson 1996
Garnero 1996
Grant 1996b
Pooled
.25 1 4 16OR
Fracture risk: Gg vs gg
−.2 −.1 0 .1 .2BMD
BMD: Gg vs gg
.25 1 4 16OR
Fracture risk: GG vs gg
−.2 −.1 0 .1 .2BMD
BMD: GG vs gg
Tom Palmer (Leicester) Meta-analysis of MR studies ISCB 28, 1 August 2007 6 / 14
![Page 18: Tom Palmer, John Thompson and Martin Tobin · Tom Palmer, John Thompson and Martin Tobin Department of Health Sciences, University of Leicester, UK ISCB 28, 1 August 2007 Tom Palmer](https://reader034.fdocuments.us/reader034/viewer/2022042710/5f68aa53d28cb25bd54d2a35/html5/thumbnails/18.jpg)
Description of the meta-analysis model
Meta-analysis model for Mendelian randomization studies for a singlegenotype comparison [Thompson et al., 2005].
Gg vs gg: Bigger sample size, smaller difference in disease risk;GG vs gg: Smaller sample size, bigger difference in disease risk
Inference at the population level; marginal distribution: combinewithin and between study distributions
θ2i
δ2i
θ3i
δ3i
∼ MVN
ηδ2
δ2
ηδ3
δ3
,Vi + B
Tom Palmer (Leicester) Meta-analysis of MR studies ISCB 28, 1 August 2007 7 / 14
![Page 19: Tom Palmer, John Thompson and Martin Tobin · Tom Palmer, John Thompson and Martin Tobin Department of Health Sciences, University of Leicester, UK ISCB 28, 1 August 2007 Tom Palmer](https://reader034.fdocuments.us/reader034/viewer/2022042710/5f68aa53d28cb25bd54d2a35/html5/thumbnails/19.jpg)
Description of the meta-analysis model
Meta-analysis model for Mendelian randomization studies for a singlegenotype comparison [Thompson et al., 2005].
Gg vs gg: Bigger sample size, smaller difference in disease risk;GG vs gg: Smaller sample size, bigger difference in disease risk
Inference at the population level; marginal distribution: combinewithin and between study distributions
θ2i
δ2i
θ3i
δ3i
∼ MVN
ηδ2
δ2
ηδ3
δ3
,Vi + B
Tom Palmer (Leicester) Meta-analysis of MR studies ISCB 28, 1 August 2007 7 / 14
![Page 20: Tom Palmer, John Thompson and Martin Tobin · Tom Palmer, John Thompson and Martin Tobin Department of Health Sciences, University of Leicester, UK ISCB 28, 1 August 2007 Tom Palmer](https://reader034.fdocuments.us/reader034/viewer/2022042710/5f68aa53d28cb25bd54d2a35/html5/thumbnails/20.jpg)
Description of the meta-analysis model
Meta-analysis model for Mendelian randomization studies for a singlegenotype comparison [Thompson et al., 2005].
Gg vs gg: Bigger sample size, smaller difference in disease risk;GG vs gg: Smaller sample size, bigger difference in disease risk
Inference at the population level; marginal distribution: combinewithin and between study distributions
θ2i
δ2i
θ3i
δ3i
∼ MVN
ηδ2
δ2
ηδ3
δ3
,Vi + B
Tom Palmer (Leicester) Meta-analysis of MR studies ISCB 28, 1 August 2007 7 / 14
![Page 21: Tom Palmer, John Thompson and Martin Tobin · Tom Palmer, John Thompson and Martin Tobin Department of Health Sciences, University of Leicester, UK ISCB 28, 1 August 2007 Tom Palmer](https://reader034.fdocuments.us/reader034/viewer/2022042710/5f68aa53d28cb25bd54d2a35/html5/thumbnails/21.jpg)
Description of the meta-analysis model
Meta-analysis model for Mendelian randomization studies for a singlegenotype comparison [Thompson et al., 2005].
Gg vs gg: Bigger sample size, smaller difference in disease risk;GG vs gg: Smaller sample size, bigger difference in disease risk
Inference at the population level; marginal distribution: combinewithin and between study distributions
θ2i
δ2i
θ3i
δ3i
∼ MVN
ηδ2
δ2
ηδ3
δ3
,Vi + B
Tom Palmer (Leicester) Meta-analysis of MR studies ISCB 28, 1 August 2007 7 / 14
![Page 22: Tom Palmer, John Thompson and Martin Tobin · Tom Palmer, John Thompson and Martin Tobin Department of Health Sciences, University of Leicester, UK ISCB 28, 1 August 2007 Tom Palmer](https://reader034.fdocuments.us/reader034/viewer/2022042710/5f68aa53d28cb25bd54d2a35/html5/thumbnails/22.jpg)
Results of the meta-analysis model
Method of estimation ORpd,0.05 95% C.I.Gg vs gg 0.57 0.42 0.77GG vs gg 0.40 0.28 0.57Combined 0.50 0.39 0.62
Given a 0.05 unit decrease in bone mineral density implies a typicalpatient is at twice the risk of osteoporotic fracture
Tom Palmer (Leicester) Meta-analysis of MR studies ISCB 28, 1 August 2007 8 / 14
![Page 23: Tom Palmer, John Thompson and Martin Tobin · Tom Palmer, John Thompson and Martin Tobin Department of Health Sciences, University of Leicester, UK ISCB 28, 1 August 2007 Tom Palmer](https://reader034.fdocuments.us/reader034/viewer/2022042710/5f68aa53d28cb25bd54d2a35/html5/thumbnails/23.jpg)
Results of the meta-analysis model
Method of estimation ORpd,0.05 95% C.I.Gg vs gg 0.57 0.42 0.77GG vs gg 0.40 0.28 0.57Combined 0.50 0.39 0.62
Given a 0.05 unit decrease in bone mineral density implies a typicalpatient is at twice the risk of osteoporotic fracture
Tom Palmer (Leicester) Meta-analysis of MR studies ISCB 28, 1 August 2007 8 / 14
![Page 24: Tom Palmer, John Thompson and Martin Tobin · Tom Palmer, John Thompson and Martin Tobin Department of Health Sciences, University of Leicester, UK ISCB 28, 1 August 2007 Tom Palmer](https://reader034.fdocuments.us/reader034/viewer/2022042710/5f68aa53d28cb25bd54d2a35/html5/thumbnails/24.jpg)
Results of the meta-analysis model
Method of estimation ORpd,0.05 95% C.I.Gg vs gg 0.57 0.42 0.77GG vs gg 0.40 0.28 0.57Combined 0.50 0.39 0.62
Given a 0.05 unit decrease in bone mineral density implies a typicalpatient is at twice the risk of osteoporotic fracture
Tom Palmer (Leicester) Meta-analysis of MR studies ISCB 28, 1 August 2007 8 / 14
![Page 25: Tom Palmer, John Thompson and Martin Tobin · Tom Palmer, John Thompson and Martin Tobin Department of Health Sciences, University of Leicester, UK ISCB 28, 1 August 2007 Tom Palmer](https://reader034.fdocuments.us/reader034/viewer/2022042710/5f68aa53d28cb25bd54d2a35/html5/thumbnails/25.jpg)
Results of the meta-analysis modelMethod of estimation ORpd,0.05 95% C.I.Gg vs gg 0.57 0.42 0.77GG vs gg 0.40 0.28 0.57Combined 0.50 0.39 0.62
Given a 0.05 unit decrease in bone mineral density implies a typicalpatient is at twice the risk of osteoporotic fracture
−1
0
1
2
3
logO
R(g
d)
−.15 −.1 −.05 0 .05Difference in mean phenotype (g/cm²)
gg versus Gg gg versus GG
gg versus Gg gg versus GG
Combined
Tom Palmer (Leicester) Meta-analysis of MR studies ISCB 28, 1 August 2007 8 / 14
![Page 26: Tom Palmer, John Thompson and Martin Tobin · Tom Palmer, John Thompson and Martin Tobin Department of Health Sciences, University of Leicester, UK ISCB 28, 1 August 2007 Tom Palmer](https://reader034.fdocuments.us/reader034/viewer/2022042710/5f68aa53d28cb25bd54d2a35/html5/thumbnails/26.jpg)
An extension to the meta-analysis model
Genetic model-free approach - estimate λ - summary of mode ofinheritance of the risk allele [Minelli et al., 2005]
Interpretation of λ; 0 - recessive, 0.5 - co-dominant, 1 - dominant
Provided genotype is independent of the phenotype (andconfounders) [Thompson et al., 2006]
λ =θ2
θ3=
δ2
δ3
Extended meta-analysis modelθ2i
δ2i
θ3i
δ3i
∼ MVN
ηλδλδηδδ
,Vi + Σ
Tom Palmer (Leicester) Meta-analysis of MR studies ISCB 28, 1 August 2007 9 / 14
![Page 27: Tom Palmer, John Thompson and Martin Tobin · Tom Palmer, John Thompson and Martin Tobin Department of Health Sciences, University of Leicester, UK ISCB 28, 1 August 2007 Tom Palmer](https://reader034.fdocuments.us/reader034/viewer/2022042710/5f68aa53d28cb25bd54d2a35/html5/thumbnails/27.jpg)
An extension to the meta-analysis model
Genetic model-free approach - estimate λ - summary of mode ofinheritance of the risk allele [Minelli et al., 2005]
Interpretation of λ; 0 - recessive, 0.5 - co-dominant, 1 - dominant
Provided genotype is independent of the phenotype (andconfounders) [Thompson et al., 2006]
λ =θ2
θ3=
δ2
δ3
Extended meta-analysis modelθ2i
δ2i
θ3i
δ3i
∼ MVN
ηλδλδηδδ
,Vi + Σ
Tom Palmer (Leicester) Meta-analysis of MR studies ISCB 28, 1 August 2007 9 / 14
![Page 28: Tom Palmer, John Thompson and Martin Tobin · Tom Palmer, John Thompson and Martin Tobin Department of Health Sciences, University of Leicester, UK ISCB 28, 1 August 2007 Tom Palmer](https://reader034.fdocuments.us/reader034/viewer/2022042710/5f68aa53d28cb25bd54d2a35/html5/thumbnails/28.jpg)
An extension to the meta-analysis model
Genetic model-free approach - estimate λ - summary of mode ofinheritance of the risk allele [Minelli et al., 2005]
Interpretation of λ; 0 - recessive, 0.5 - co-dominant, 1 - dominant
Provided genotype is independent of the phenotype (andconfounders) [Thompson et al., 2006]
λ =θ2
θ3=
δ2
δ3
Extended meta-analysis modelθ2i
δ2i
θ3i
δ3i
∼ MVN
ηλδλδηδδ
,Vi + Σ
Tom Palmer (Leicester) Meta-analysis of MR studies ISCB 28, 1 August 2007 9 / 14
![Page 29: Tom Palmer, John Thompson and Martin Tobin · Tom Palmer, John Thompson and Martin Tobin Department of Health Sciences, University of Leicester, UK ISCB 28, 1 August 2007 Tom Palmer](https://reader034.fdocuments.us/reader034/viewer/2022042710/5f68aa53d28cb25bd54d2a35/html5/thumbnails/29.jpg)
An extension to the meta-analysis model
Genetic model-free approach - estimate λ - summary of mode ofinheritance of the risk allele [Minelli et al., 2005]
Interpretation of λ; 0 - recessive, 0.5 - co-dominant, 1 - dominant
Provided genotype is independent of the phenotype (andconfounders) [Thompson et al., 2006]
λ =θ2
θ3=
δ2
δ3
Extended meta-analysis model
θ2i
δ2i
θ3i
δ3i
∼ MVN
ηλδλδηδδ
,Vi + Σ
Tom Palmer (Leicester) Meta-analysis of MR studies ISCB 28, 1 August 2007 9 / 14
![Page 30: Tom Palmer, John Thompson and Martin Tobin · Tom Palmer, John Thompson and Martin Tobin Department of Health Sciences, University of Leicester, UK ISCB 28, 1 August 2007 Tom Palmer](https://reader034.fdocuments.us/reader034/viewer/2022042710/5f68aa53d28cb25bd54d2a35/html5/thumbnails/30.jpg)
An extension to the meta-analysis model
Genetic model-free approach - estimate λ - summary of mode ofinheritance of the risk allele [Minelli et al., 2005]
Interpretation of λ; 0 - recessive, 0.5 - co-dominant, 1 - dominant
Provided genotype is independent of the phenotype (andconfounders) [Thompson et al., 2006]
λ =θ2
θ3=
δ2
δ3
Extended meta-analysis modelθ2i
δ2i
θ3i
δ3i
∼ MVN
ηλδλδηδδ
,Vi + Σ
Tom Palmer (Leicester) Meta-analysis of MR studies ISCB 28, 1 August 2007 9 / 14
![Page 31: Tom Palmer, John Thompson and Martin Tobin · Tom Palmer, John Thompson and Martin Tobin Department of Health Sciences, University of Leicester, UK ISCB 28, 1 August 2007 Tom Palmer](https://reader034.fdocuments.us/reader034/viewer/2022042710/5f68aa53d28cb25bd54d2a35/html5/thumbnails/31.jpg)
Estimation; maximum likelihood or Bayesian using Product NormalFormulation [Spiegelhalter, 1998]
θ2i ∼ N(ηλδ, var(θ2i )), δ2i ∼ N(λδ, var(δ2i ))
θ3i ∼ N(ηδ, var(θ3i )), δ3i ∼ N(δ, var(δ3i ))
Prior distributions
δ, η ∼ N(0, 1× 106), λ ∼ Beta(0.5, 0.5)
Results
Method of estimation ORpd,0.05 95% C.I./Cr.I. λ 95% C.I./Cr.I.ML - MVN 0.42 0.28 0.61 0.33 0.19 0.47Bayesian - PNF 0.46 0.32 0.61 0.30 0.17 0.45
Tom Palmer (Leicester) Meta-analysis of MR studies ISCB 28, 1 August 2007 10 / 14
![Page 32: Tom Palmer, John Thompson and Martin Tobin · Tom Palmer, John Thompson and Martin Tobin Department of Health Sciences, University of Leicester, UK ISCB 28, 1 August 2007 Tom Palmer](https://reader034.fdocuments.us/reader034/viewer/2022042710/5f68aa53d28cb25bd54d2a35/html5/thumbnails/32.jpg)
Estimation; maximum likelihood or Bayesian using Product NormalFormulation [Spiegelhalter, 1998]
θ2i ∼ N(ηλδ, var(θ2i )), δ2i ∼ N(λδ, var(δ2i ))
θ3i ∼ N(ηδ, var(θ3i )), δ3i ∼ N(δ, var(δ3i ))
Prior distributions
δ, η ∼ N(0, 1× 106), λ ∼ Beta(0.5, 0.5)
Results
Method of estimation ORpd,0.05 95% C.I./Cr.I. λ 95% C.I./Cr.I.ML - MVN 0.42 0.28 0.61 0.33 0.19 0.47Bayesian - PNF 0.46 0.32 0.61 0.30 0.17 0.45
Tom Palmer (Leicester) Meta-analysis of MR studies ISCB 28, 1 August 2007 10 / 14
![Page 33: Tom Palmer, John Thompson and Martin Tobin · Tom Palmer, John Thompson and Martin Tobin Department of Health Sciences, University of Leicester, UK ISCB 28, 1 August 2007 Tom Palmer](https://reader034.fdocuments.us/reader034/viewer/2022042710/5f68aa53d28cb25bd54d2a35/html5/thumbnails/33.jpg)
Estimation; maximum likelihood or Bayesian using Product NormalFormulation [Spiegelhalter, 1998]
θ2i ∼ N(ηλδ, var(θ2i )), δ2i ∼ N(λδ, var(δ2i ))
θ3i ∼ N(ηδ, var(θ3i )), δ3i ∼ N(δ, var(δ3i ))
Prior distributions
δ, η ∼ N(0, 1× 106), λ ∼ Beta(0.5, 0.5)
Results
Method of estimation ORpd,0.05 95% C.I./Cr.I. λ 95% C.I./Cr.I.ML - MVN 0.42 0.28 0.61 0.33 0.19 0.47Bayesian - PNF 0.46 0.32 0.61 0.30 0.17 0.45
Tom Palmer (Leicester) Meta-analysis of MR studies ISCB 28, 1 August 2007 10 / 14
![Page 34: Tom Palmer, John Thompson and Martin Tobin · Tom Palmer, John Thompson and Martin Tobin Department of Health Sciences, University of Leicester, UK ISCB 28, 1 August 2007 Tom Palmer](https://reader034.fdocuments.us/reader034/viewer/2022042710/5f68aa53d28cb25bd54d2a35/html5/thumbnails/34.jpg)
Estimation; maximum likelihood or Bayesian using Product NormalFormulation [Spiegelhalter, 1998]
θ2i ∼ N(ηλδ, var(θ2i )), δ2i ∼ N(λδ, var(δ2i ))
θ3i ∼ N(ηδ, var(θ3i )), δ3i ∼ N(δ, var(δ3i ))
Prior distributions
δ, η ∼ N(0, 1× 106), λ ∼ Beta(0.5, 0.5)
Results
Method of estimation ORpd,0.05 95% C.I./Cr.I. λ 95% C.I./Cr.I.ML - MVN 0.42 0.28 0.61 0.33 0.19 0.47Bayesian - PNF 0.46 0.32 0.61 0.30 0.17 0.45
Tom Palmer (Leicester) Meta-analysis of MR studies ISCB 28, 1 August 2007 10 / 14
![Page 35: Tom Palmer, John Thompson and Martin Tobin · Tom Palmer, John Thompson and Martin Tobin Department of Health Sciences, University of Leicester, UK ISCB 28, 1 August 2007 Tom Palmer](https://reader034.fdocuments.us/reader034/viewer/2022042710/5f68aa53d28cb25bd54d2a35/html5/thumbnails/35.jpg)
Estimation; maximum likelihood or Bayesian using Product NormalFormulation [Spiegelhalter, 1998]
θ2i ∼ N(ηλδ, var(θ2i )), δ2i ∼ N(λδ, var(δ2i ))
θ3i ∼ N(ηδ, var(θ3i )), δ3i ∼ N(δ, var(δ3i ))
Prior distributions
δ, η ∼ N(0, 1× 106), λ ∼ Beta(0.5, 0.5)
Results
Method of estimation ORpd,0.05 95% C.I./Cr.I. λ 95% C.I./Cr.I.ML - MVN 0.42 0.28 0.61 0.33 0.19 0.47Bayesian - PNF 0.46 0.32 0.61 0.30 0.17 0.45
Tom Palmer (Leicester) Meta-analysis of MR studies ISCB 28, 1 August 2007 10 / 14
![Page 36: Tom Palmer, John Thompson and Martin Tobin · Tom Palmer, John Thompson and Martin Tobin Department of Health Sciences, University of Leicester, UK ISCB 28, 1 August 2007 Tom Palmer](https://reader034.fdocuments.us/reader034/viewer/2022042710/5f68aa53d28cb25bd54d2a35/html5/thumbnails/36.jpg)
Dominant
Co−dominant
Recessive
−1
0
1
2
3lo
gOR
(Gg)
−.5 0 .5 1 1.5 2logOR(GG)
Tom Palmer (Leicester) Meta-analysis of MR studies ISCB 28, 1 August 2007 11 / 14
![Page 37: Tom Palmer, John Thompson and Martin Tobin · Tom Palmer, John Thompson and Martin Tobin Department of Health Sciences, University of Leicester, UK ISCB 28, 1 August 2007 Tom Palmer](https://reader034.fdocuments.us/reader034/viewer/2022042710/5f68aa53d28cb25bd54d2a35/html5/thumbnails/37.jpg)
Dominant
Co−dominant
Recessive
−1
0
1
2
3lo
gOR
(Gg)
−.5 0 .5 1 1.5 2logOR(GG)
Dominant
Co−dominant
Recessive
−.05
0
.05
.1
.15
Del
ta IP
(Gg)
−.05 0 .05 .1 .15Delta IP(GG)
Tom Palmer (Leicester) Meta-analysis of MR studies ISCB 28, 1 August 2007 11 / 14
![Page 38: Tom Palmer, John Thompson and Martin Tobin · Tom Palmer, John Thompson and Martin Tobin Department of Health Sciences, University of Leicester, UK ISCB 28, 1 August 2007 Tom Palmer](https://reader034.fdocuments.us/reader034/viewer/2022042710/5f68aa53d28cb25bd54d2a35/html5/thumbnails/38.jpg)
Talk summary and discussion
Random allocation of an individual’s genotype allows use ofinstrumental variable theory
Meta-analysis analysis of two genotype comparisons;extended to include the genetic model-free approach
Meta-analysis of genetic association studies using merged genotypecomparisons [Salanti and Higgins, 2007]
Tom Palmer (Leicester) Meta-analysis of MR studies ISCB 28, 1 August 2007 12 / 14
![Page 39: Tom Palmer, John Thompson and Martin Tobin · Tom Palmer, John Thompson and Martin Tobin Department of Health Sciences, University of Leicester, UK ISCB 28, 1 August 2007 Tom Palmer](https://reader034.fdocuments.us/reader034/viewer/2022042710/5f68aa53d28cb25bd54d2a35/html5/thumbnails/39.jpg)
Talk summary and discussion
Random allocation of an individual’s genotype allows use ofinstrumental variable theory
Meta-analysis analysis of two genotype comparisons;extended to include the genetic model-free approach
Meta-analysis of genetic association studies using merged genotypecomparisons [Salanti and Higgins, 2007]
Tom Palmer (Leicester) Meta-analysis of MR studies ISCB 28, 1 August 2007 12 / 14
![Page 40: Tom Palmer, John Thompson and Martin Tobin · Tom Palmer, John Thompson and Martin Tobin Department of Health Sciences, University of Leicester, UK ISCB 28, 1 August 2007 Tom Palmer](https://reader034.fdocuments.us/reader034/viewer/2022042710/5f68aa53d28cb25bd54d2a35/html5/thumbnails/40.jpg)
Talk summary and discussion
Random allocation of an individual’s genotype allows use ofinstrumental variable theory
Meta-analysis analysis of two genotype comparisons;extended to include the genetic model-free approach
Meta-analysis of genetic association studies using merged genotypecomparisons [Salanti and Higgins, 2007]
Tom Palmer (Leicester) Meta-analysis of MR studies ISCB 28, 1 August 2007 12 / 14
![Page 41: Tom Palmer, John Thompson and Martin Tobin · Tom Palmer, John Thompson and Martin Tobin Department of Health Sciences, University of Leicester, UK ISCB 28, 1 August 2007 Tom Palmer](https://reader034.fdocuments.us/reader034/viewer/2022042710/5f68aa53d28cb25bd54d2a35/html5/thumbnails/41.jpg)
Acknowledgements
Student conference award
Medical Research Council Capacity Building PhD Studentship inGenetic Epidemiology (G0501386).
Dr. Martin Tobin is funded by a Medical Research Council ClinicianScientist Fellowship (G0501942).
Tom Palmer (Leicester) Meta-analysis of MR studies ISCB 28, 1 August 2007 13 / 14
![Page 42: Tom Palmer, John Thompson and Martin Tobin · Tom Palmer, John Thompson and Martin Tobin Department of Health Sciences, University of Leicester, UK ISCB 28, 1 August 2007 Tom Palmer](https://reader034.fdocuments.us/reader034/viewer/2022042710/5f68aa53d28cb25bd54d2a35/html5/thumbnails/42.jpg)
References
Katan, M. (1986).
Apolipoprotein e isoforms, serum cholesterol, and cancer.Lancet, 327:507–508.
Katan, M. (2004).
Commentary: Mendelian randomization, 18 years on.International Journal of Epidemiology, 33(1):10–11.
Minelli, C., Thompson, J., Abrams, K., Thakkinstian, A., and Attia, J. (2005).
The choice of a genetic model in the meta-analysis of molecular association studies.International Journal of Epidemiology, 34:1319–1328.
Salanti, G. and Higgins, J. P. T. (2007).
Meta-analysis of genetic association studies under different inheritance models using data reported as merged genotypes.Statistics in Medicine.in press.
Spiegelhalter, D. (1998).
Bayesian graphical modelling: a case-study in monitoring health outcomes.Applied Statistics, 47(1):115–133.
Thompson, J., Minelli, C., Abrams, K., Tobin, M., and Riley, R. (2005).
Meta-analysis of genetic studies using Mendelian randomization - a multivariate approach.Statistics in Medicine, 24:2241–2254.
Thompson, J. R., Minelli, C., and Abrams, K. R. (2006).
Model-free meta-analysis of genetic association studies of continuous outcomes.In ISCB 27 Abstract Book. ISCB.
Tom Palmer (Leicester) Meta-analysis of MR studies ISCB 28, 1 August 2007 14 / 14