Statistical Analysis in Case-Control studies

Liu Tian Genome Institute of Singapore

liut2@gis.a-star.edu.sg

Summer International Workshop Aug, 09, Beijing

Outline
Introduction Basic Statistical Methods of Case-control Study GWAS A Novel Epistasis-testing Procedure

Aim of Genetic Studies

Dramatic variation do exist within a same spice Almost every biological phenomenon involves a genetic component There is always a keen need for us to seek the genetic variation relates to complex traits.

Different Design Strategies

Intervention studies
Clinic trials

Observational studies
1. Case-control studies 2. Cohort studies

Cohort Studies
A cohort study is a study where a group of individuals are followed. Cohort studies can be either prospective or retrospective
exposed

population

non-exposed
Disease +/-

Case-Control Studies
Case-control studies are used to identify factors that may contribute to a medical condition by comparing subjects who have that condition (the cases) with patients who do not have the condition but are otherwise similar (the controls) Case-control studies are retrospective and nonrandomized

Case-Control Studies
exposed
Disease population

non-exposed

exposed
Disease +

non-exposed

Selection of Cases
Population-based cases: include all subjects or a random sample of all subjects with the disease at a single point or during a given period of time in the defined population.

Hospital-based cases: All patients in a hospital department at a given time

Selection of Controls
Principles of Control Selection: Study base: Controls can be used to characterise the distribution of exposure Comparable-accuracy: Equal reliability in the information obtained from cases and controls (to avoid systematic misclassification) Overcome confounding: Elimination of confounding through control selection (matching or stratified sampling)

Selection of Controls
General population controls: registries, households, telephone sampling costly and time consuming recall bias eventually high non-response rate Hospitalised controls: Patients at the same hospital as the cases Easy to identify; less recall bias; higher response rate

Case-Control Studies vs. Cohort Studies

Cohort study Rare exposure Examine multiple effects of a single exposure Minimizes bias in the in exposure determination Direct measurements of incidence of the disease Case-control study Quick, inexpensive Well-suited to the evaluation of diseases with long latency period Rare diseases Examine multiple etiologic factors for a single disease

Case-Control Studies vs. Cohort Studies

Cohort study
Not rare diseases Prospective: Expensive and time consuming Retrospective: in adequate records Validity can be affected by losses to follow-up

Case-control study
Not rare exposure Incidence rates cannot be estimated unless the study is population based retrospective, nonrandomized nature limits the conclusions that can be drawn from them.

Data Structure of Casecontrol studies

individual 1 2 affection 1 1 gender F M SNP 1 2 2 SNP 2 1 2 SNP n 2 1

3
4

0
1

F
F

1
1

2
1

2
2

-9

sample id

case/control

genotypes

Outline
Introduction Basic Statistical Methods of Case-control Study GWAS A Novel Epistasis-testing Procedure

Population-Based Case-Control Study

Individuals are unrelated To test if marker genotypes distribute differently between the cases and controls By comparing within cases and controls, we identify those genetic factors correlated with a pre-defined phenotype

Coding Genotypes
For one marker with two alleles, there can be three possible genotypes:
Genotype AA Aa aa Coding 2 1 0

Genetic Models and Underlining Hypotheses

Genotypic Model
Genotype Genotypic Value
Genotypic value is the expected phenotypic value of a particular genotype

AA
Aa aa

AA
Aa aa

Hypothesis: all 3 different genotypes have different effects

AA vs. Aa vs. aa

Genetic Models and Underlining Hypotheses

Dominant Model
Genotype Genotypic Value

AA Aa aa

AAaa

Hypothesis: the genetic effects of AA and Aa are the same (assuming A is the minor allele)
AA and Aa vs. aa

Genetic Models and Underlining Hypotheses

Recessive Model
Genotype Genotypic Value

AA Aa aa

Aaaa

Hypothesis: the genetic effects of Aa and aa are the same (A is the minor allele)
AA vs. Aa and aa

Genetic Models and Underlining Hypotheses

Allelic Model
Genotype AA Aa aa Genotypic Value 2A A+ a 2a

Hypothesis: the genetic effects of allele A and allele a are different

A vs. a

Pearsons Chi-squared Test

Genotypic Model: Null Hypothesis: Independence

H 0 : ij i. . j
AA nAA mAA Aa nAa mAa
df = 2

cases controls

aa naa maa

Pearsons Chi-squared Test

Dominant Model: Null Hypothesis: Independence

H 0 : ij i. . j
AA+Aa nAA + nAa mAA + mAa
df = 1

cases controls

aa naa maa

Pearsons Chi-squared Test

Recessive Model: Null Hypothesis: Independence

H 0 : ij i. . j
AA nAA mAA Aa +aa nAa + naa mAa + maa
df = 1

cases controls

Pearsons Chi-squared Test

Allelic Model: Null Hypothesis: Independence

H 0 : ij i. . j
A 2nAA + nAa 2mAA + mAa a nAa + 2naa mAa +2 maa
df = 1

cases controls

Test Statistic
Chi-squared Test Statistic:

(O E ) E all cells
2

O is the observed cell counts E is the expected cell counts, under null hypothesis of independence (row total column tot al) E N

Example
The following table summarize the genotype counts of marker M :
AA Aa aa

cases
controls

36
18

100
84

64
98

Different tests can be performed: - Allelic test - Dominant gene action - Recessive gene action - Genotypic test

Example (Dominant Gene Action)

Using

R:

dominant_table <- matrix(c(80,90,20,10), ncol = 2)

print(dominant_table )
chisq.test(dominant_table ,correct=FALSE)

Example (Recessive Gene Action)

Using

R:

recessive_table <- matrix(c(36,18,164,182), ncol = 2)

print(recessive_table)
chisq.test(recessive_table,correct=FALSE)

Example (Genotypic Test)

Using

R:

genotypic_table <- matrix(c(36,18,100,84,64,98), ncol = 3)

print(genotypic_table)
chisq.test(genotypic_table,correct=FALSE)

Example (Allelic Test)

Using

R:

allelic_table <- matrix(c(172,120,228,280), ncol = 2)

print(allelic _table)
chisq.test(allelic_table,correct=FALSE)

Logistic Regression Analysis

A General Model:
pdisease logit( pdisease ) log( ) 0 1 X 1 J X J 1 pdisease
Where: pdisease is the probability that an individual has a particular disease.

0 is the intercept
1, 2 J are the effects of genetic factors X1, X2 XJ are the dummy variables of genetic factors

Logistic Regression Analysis

Logistic regression describes the relationship between a dichotomous response variable and a set of explanatory variables. Logit model is the only model under which , the effect parameter, can be estimated in retrospective studies as same as in prospective studies. If the sampling rate for cases is 10 times that for controls, the intercept estimated is log(10) =2.3 than the one estimated with a prospective study.

Inference and Interpretation

Significant test focus on:

H 0 : i 0

(i 1, .... , J )

is the estimated odds ratio for Estimator i genetic factor i. determines whether pdisease is The sign of i increasing or decreasing when the effect of genetic factor i exists.

An Example of R output

Other Options
Fishers Exact Test:
When sample size is small, the asymptotic approximation of null distribution is no longer valid. By performing Fishers exact test, exact significance of the deviation from a null hypothesis can be calculated.

For a 2 by 2 table, the exact p-value can be calculated as:

a c

b d

Other Options
Cochram-Armitage Trend Test
-- An advantage of the Cochran-Armitage test is that it
does not assume Hardy-Weinberg equilibrium

-- Typically used to test a 2 k contingency table, when the effects of AA, Aa, and aa are thought to be ordered. -- In genome-wide association studies, the additive (or codominant) version of the test is often used.

Outline
Introduction Basic Statistical Methods of Case-control Study GWAS A Novel Epistasis-testing Procedure

Genome-wide Association Study

In genetic epidemiology, a genome-wide association study (GWAS) - also known as whole genome association study (WGA study) - is an examination of genetic variation across a given genome, designed to identify genetic associations with observable traits. In human studies, this might include traits such as blood pressure or weight, or why some people get a disease or condition.
From: http://en.wikipedia.org

Genome-wide Association Study

Technology makes it feasible -- Affymetrix: 500K; 1M chip arrives in early 2007.
(Randomly distributed) -- Illumina: 550K chip costs (gene-based)

Requires little on sample, Case-control data, case-parents trio data are enough. Good for moderate effect sizes ( odds ratio < 1.5). Particularly useful in finding genetic variations that contribute to common, complex diseases.

Genome-wide Association Study

What Is A SNP?
TTCAGTCAGATCCTAGCCC AAGTCAGTCTAGGATCGGG TTCAGTCAGATCCCAGCCC
Chromosome 2
Chromosome 1

AAGTCAGTCTAGGGTCGGG
Single Nucleotide Polymorphism

Handling GWAS
Storing and converting large amounts of genotype data Quality control Generating initial association analysis Specialized analysis

Quality Control Of SNPs

Exclude SNPs that failure the Hardy-Weinberg test -- Expected proportions of genotypes are not consistent with observed allele frequency -- HWE p-value < 10-4 to 10-6 Genotyping success rate < 95% Differential missingness in cases and controls

Quality Control Of Samples

Poor quality samples
-- Sample genotype success rate < 95 to 97.5% -- Greater proportion of heterozygous genotypes than expected

Related individuals (if independent samples)

-- Based on pair-wise comparisons of similarity of genotypes

Samples with miss specified gender

Genetic Stratification
Assess population structure Adjust both phenotypes and genotypes for possible stratification using
-- principal component analysis (Prices method) -- cluster analyses (Plink)
Genomic Control

Software Demonstration
Plink
-- Case/control, TDT, quantitative traits

-- Develop by Shaun Purcell http://pngu.mgh.harvard.edu/~purcell/plink/

Results From GWAS

Results From GWAS

Software Demonstration
Haploview:

-- LD and haplotype block analysis -- tag SNP selection algorithm -- visualization and plotting GWAS results from PLINK
http://www.broadinstitute.org/haploview/haploview