How to review genetic association studies
Lavinia Paternoster3rd year PhD student
Outline
• Traditional meta-analyses• Why are genetic studies unique?• Methods
– choosing a genetic model– Multiple testing– Overall association– Per-allele mean differences
• Other things to consider
Research Question
• Does having gene “X” increase the risk of disease/trait “Y”?
• Same as:
• Does intervention “X” increase the risk of outcome “Y”?
BUT……….
Traditional meta-analysis
Intervention Control
(or intervention 2)
Observe
Outcome 1 Outcome 2
Input variable to be tested
Outcome to test success of intervention
Traditional meta-analysis
Intervention
e.g. beta-blockers
Control
(or intervention 2)
Observe
Outcome 1
e.g. cardiovascular
disease
Outcome 2
e.g. no cardiovascular
disease
Input variable to be tested
Outcome to test success of intervention
Traditional meta-analysis
Intervention
(beta-blockers)
Control
Outcome 1
(e.g. CVD)
n n
Outcome 2
(e.g. no CVD)
n n
Calculate relative risk (or odds ratio) for each study
Pool relative risks by using weighting methods
Beta-blockers & cardiovascular disease
Traditional meta-analysis
Intervention Control
Observe
Mean value of those with intervention
Mean value of controls
Variable to be tested
Outcome to test success of intervention
Traditional meta-analysis
Intervention
e.g. exercise
Control
Observe
Mean value of those with intervention
e.g. mean fatigue scale value
Mean value of controls
e.g. mean fatigue scale value
Variable to be tested
Outcome to test success of intervention
Edmonds et al. 2004. Exercise for chronic fatigue syndrome. Cochrane
Traditional meta-analysis
Observations
(e.g. fatigue scale)
Intervention 1
(exercise)
n mean sd
control n mean sd
Calculate mean difference (and 95%CI) for each study
Pool mean differences by using weighting methods
Exercise & Fatigue
Genetic Associations
• The simplest mutation (a→b) gives 3 genotypes: aa, ab, bb
• Comparing 3 groups not 2
• Conventional meta-analysis methods not suitable
Traditional meta-analysis
Intervention
e.g. beta-blockers
Control
(or intervention 2)
Observe
Outcome 1
e.g. cardiovascular
disease
Outcome 2
e.g. no cardiovascular
disease
Input variable to be tested
Outcome to test success of intervention
Traditional meta-analysis
Genotype AA
Observe
Outcome 1
e.g. cardiovascular
disease
Outcome 2
e.g. no cardiovascular
disease
Input variable to be tested
Outcome to test success of intervention
Genotype AB Genotype BB
Traditional meta-analysis
Intervention Control
Observe
Mean value of those with intervention
Mean value of controls
Variable to be tested
Outcome to test success of intervention
Traditional meta-analysis
AA BB
Observe
Mean value of those with
genotype AA
Variable to be tested
Outcome to test success of intervention
AB
Mean value of those with
genotype AB
Mean value of those with
genotype BB
My Research
• Meta-analysis of association between Carotid intima-media thickness and several genes
• Here I’ll show MTHFR example
CC / CT / TT
Data
CC CT TT
Methods in the literature
• Collapse into 2 groups– Assume genetic model
• Dominant (tt+ct v cc)• Recessive (tt v ct+cc)
– Multiple pairwise comparisons• tt v cc, tt v ct, ct v cc• dominant and recessive
Methods in the literature
• Analyse as 3 groups– Analyse as co-dominant (per-allele difference)– Meta-ANOVA
My Method
• 3 stage approach– Meta-ANOVA
• Looks for overall association between gene and trait but does not indicate which alleles increase/decrease
– Determine genetic model use linear regression
– Estimate mean differences using chosen genetic model
Meta- ANOVA
Analyse by carrying out ANOVA using ‘genotype’ and ‘study’ as categorical variables and weighting each observation
Test whether ‘genotype’ is a significant variable
P=0.026
Which genetic model?
• Recessive– TT shows effect, CT = CC– MD1 = 0, so λ=0
• Dominant– TT = CT and both show effect– MD1 = MD2, so λ=1
• Co-dominant– CT will be half way between CC
and TT– MD1/MD2 = 0.5
λ = MD1/MD2
MD1 = CT – CC
MD2 = TT - CC
Can use a linear regression of MD1 against MD2, weighted by study to determine overall the most appropriate genetic model
-0.1
0.1
0.2
0.3
-0.1 0.1 0.2 0.3
0.201
MD1
MD2
0.2 (95%CI, 0 to 0.4)
λ = 0, so recessive
Mean differences
• For dominant and recessive genetic models combine 2 genotypes and use methods previously described– Recessive
• combine CT and CC, compare with TT– Dominant
• Combine TT and CT, compare with CC
• For co-dominant models use per-allele difference– Assumes same difference between TT & CT, and CT
& CC
Mean differences
• MTHFR was associated when analysed by meta-ANOVA (p = 0.026)
• MTHFR was recessive (λ = 0.2)
• Mean difference between TT and CT/CC is: 20μm (95%CI 10 to 30)
Summary
• Genetic association studies have at least 3 groups– Chose a model based on previous evidence– Multiple comparisons– Overall association– Novel 3 stage approach
Other issues
• Other genetic models?
• Different polymorphisms within gene
• LD between genes?
• Whole genome meta-analysis