Case-Control Study for a
Genetic Association Question

RESEARCH QUESTION

What the Association Study Set Out to Ask

This case-control study example follows a genetics research group that wants to know whether any of about 400 prespecified candidate variants is associated with dilated cardiomyopathy, in cases from specialist clinics compared with controls from a volunteer cohort. The estimate for each variant is the odds ratio per copy of the variant allele, comparing cases with controls, with its 95% confidence interval.

Cases and controls were recruited separately, so a difference in ancestry or genotyping between the two groups can look like a genetic association. The plan fixes the quality control rules, the adjustment for ancestry, the multiple-testing threshold and the replication sample before any association is tested, and it treats first-sample estimates as provisional.

STUDY DESIGN AND DATA

Cases, Controls and Prespecified Variants

A case-control sample of about 500 cases and 1,000 controls genotyped on an array, with the association tests restricted to about 400 prespecified candidate variants and the ancestry components estimated from the wider array. A second independent sample is held back for replication. Each record holds genotypes, age, sex, recruitment site and genotyping plate. Many candidate variants are rare, so some have only a handful of carriers, and ordinary estimation can fail for a variant with no carriers in one group.

METHOD AND ASSUMPTIONS

Why Quality Control Comes Before Any Test

Before any variant is tested, genotypes are filtered by call rate, by Hardy-Weinberg equilibrium checked in controls, and by differences in missingness between cases and controls and between plates. Related individuals and duplicates are identified from kinship estimates and removed or modeled. Ancestry components enter every model as covariates, because ancestry that differs between cases and controls would otherwise produce associations that reflect where people come from, not the disease. Each variant is tested with a logistic model for case status, additive coding, and adjustment for the ancestry components, age and sex.

For variants with few carriers, ordinary maximum likelihood can fail or return a very large odds ratio, so the plan names Firth's penalized likelihood for those variants, with profile likelihood intervals, and sets a minimum carrier count below which variants are collapsed into a prespecified gene-level group. A wide interval for a rare variant reflects sparse data, not an absence of effect. The significance threshold accounts for the number of variants and gene-level groups tested, and associations that pass it are tested again in the second sample, because first-sample estimates of the strongest associations tend to overstate the true effect.

OUTPUT AND INTERPRETATION

How the Results Are Read

The output has a flow diagram from the recruited sample to the analyzed sample after quality control, per-variant odds ratios with 95% confidence intervals, carrier counts in cases and controls, and a quantile-quantile plot that shows whether residual ancestry or plate effects inflate the test statistics. Where penalized and ordinary estimates both exist, they are shown side by side.

Replication results are reported beside the first-sample results, with the same coding and covariates. The interpretation is worded as an association between a variant and disease status. It states that a case-control design estimates odds ratios, not absolute risks, because the proportion of cases is set by sampling, and that an associated variant does not by itself identify the causal variant or its mechanism.

REPORTING

What the Researcher Received

A written report structured to STREGA, the extension of STROBE for genetic association studies, with statistical wording for the methods on quality control, ancestry adjustment and replication, matched to the output, for the team to check and incorporate. The research question, the biological interpretation and the conclusions remain with the research team.