eduKate Singapore

VIEW THIS AS

Auto mode follows the Route Engine until you choose a viewpoint.

YOU ARE HERE

ROUTE CHECK

CONNECTED TO

WHAT NEXT

Use the canonical route for this room, or HELP if you are unsure.

Next useful routeCanonical hubHELP

The Gold Standard of Genetic Epidemiology

—

by

eduKate Asia

A headline says that researchers have found a gene for a disease. The original study may actually show that a genetic marker is more common among people with a particular recorded outcome. The marker could be linked to another variant, the sample might reflect population structure, and the result may need replication. Genetic epidemiology exists to make sense of that gap between a statistical signal and a justified health claim.

The gold standard of genetic epidemiology is a study whose population, phenotype, genetic measurements, statistical tests and causal interpretation are openly defined. This guide introduces eighteen controls, two worked fictional calculations, twelve FAQs and an interconnected pathway through genetics, genomics, epidemiology and biostatistics.

Part of The Gold Standard, this is general scientific education, not personal genetic risk assessment, clinical screening advice or a substitute for qualified health professionals.

Quick information: six conditions for credible genetic epidemiology

  • Define cases, outcomes and population selection consistently.
  • Report reliable genomic measurements, effect alleles and statistical models.
  • Control population structure and relevant study biases.
  • Account for multiple testing, uncertainty and independent replication.
  • Avoid conflating association, causation and individual disease probability.
  • Protect genomic privacy and evaluate portability fairly.

Chapter index

1. Begin with a population question, not a genetic verdict

Back to contents

Genetic epidemiology asks how genetic differences and their interplay with other factors relate to the distribution of health and disease across populations. It uses study designs and statistical methods to understand associations, their possible causes and their limits. It does not turn DNA sequences into unchangeable predictions of one person’s health.

Start with the phenotype or outcome, the population, the timing and the question of interest. A variant linked statistically to a condition in one dataset is a lead for further inquiry. The gold standard distinguishes association from causation, population results from clinical interpretation and genuine evidence from genetic determinism.

2. Define the phenotype before searching variants

Back to contents

A study’s outcome may be a laboratory measurement, clinician-defined diagnosis, survey answer or administrative code. These do not have identical specificity. If cases are defined inconsistently, apparent associations can be affected by misclassification, healthcare access or differences in record keeping.

State inclusion criteria, diagnostic standards, measurement dates and known sources of error. Separate an identified condition from an inferred risk factor or a convenient billing label. A reproducible phenotype definition is as important as a reproducible genomic pipeline.

3. Choose a study design that can support the claim

Back to contents

Case-control studies compare characteristics of people with and without a defined outcome, while cohort studies can examine outcomes over follow-up. Family-based designs and biobanks answer different questions and have distinct strengths and biases. Sampling, loss to follow-up and access to diagnosis can all affect representativeness.

Define source population, recruitment, exclusions and comparison group before analysis. A study recruited from specialist clinics may not represent a general community. Report who was eligible but not included and which conclusions cannot be transferred beyond the sampled setting.

4. Understand genotypes, variants and alleles

Back to contents

A genetic variant describes a difference at a genomic location, while an allele identifies a particular sequence form. A variant may be directly genotyped or inferred by imputation; those data have different uncertainty. Association results depend on the declared effect allele, coding model and reference genome assembly.

Never interpret a reported effect without checking which allele it describes and whether the analysis is additive, dominant, recessive or another specified model. Allele orientation and genome-build mismatches can reverse or misplace findings in combined datasets.

5. Check genotyping and sequencing quality

Back to contents

Genomic assays can contain missing calls, batch effects, contamination, platform-specific artefacts and errors in variant annotation. Quality control may include sample-level checks, call-rate filters and variant-level diagnostics appropriate to the study. No single filter threshold is correct for every platform or objective.

Preserve raw and reviewed data and document the reasons for exclusion. When errors correlate with case status or recruitment site, an artefact can resemble biological association. A large sample does not automatically compensate for systematic data problems.

6. Treat population structure as a real confounding risk

Back to contents

Allele frequencies and outcome prevalence may both differ across sampled groups for historical, social, environmental or methodological reasons. Pooling participants without accounting for relevant population structure can create spurious genetic associations. Genetic ancestry and social identity are not interchangeable concepts.

Describe recruitment, ancestry-estimation methods and population descriptors precisely. The GWAS Catalog recommends avoiding citizenship or a geographic cohort name as a substitute for genetic ancestry. Adjustment methods should be justified and checked; simplistic racial categories cannot substitute for careful genetic and social context.

7. Distinguish an effect estimate from absolute disease risk

Back to contents

Odds ratios, risk ratios and regression coefficients describe different quantities. In a case-control design, the sample’s fraction of cases is usually set by recruitment, so it cannot ordinarily be read as population disease prevalence. An odds ratio of two does not mean every carrier has a fifty-percent probability of disease.

Present the coding model, uncertainty interval, baseline outcome context and relevant covariates. A statistically significant relative association can have modest practical value when the absolute risk remains low. In individual counselling, clinical decisions require far more than a research association table.

8. Use genome-wide testing thresholds correctly

Back to contents

A genome-wide association study examines large numbers of variants and therefore faces a serious multiple-testing problem. The conventional genome-wide significance benchmark of approximately five times ten to the negative eight is much more stringent than an ordinary 0.05 level. The exact testing strategy may depend on design and the statistical method.

The GWAS Catalog’s indexing threshold for reported top associations is p below one times ten to the negative five; its documentation explicitly warns that this curation threshold is not the same as a genome-wide significance criterion. Being listed in a catalog does not by itself establish replication, causality or clinical actionability.

9. Understand linkage disequilibrium before naming a causal gene

Back to contents

Nearby variants can be correlated because they are inherited together more often than independent assortment would predict. A significant marker may tag a region containing a causal mechanism without being that mechanism itself. Assigning the nearest gene as the true causal gene can be misleading.

Fine-mapping, functional genomics, colocalisation and experimental follow-up can narrow hypotheses, but each brings its own assumptions. Report statistical signals as loci or associations until the evidence justifies a stronger biological claim. A pathway story created after seeing an attractive result is not proof of causation.

10. Replicate findings in an appropriate new sample

Back to contents

A result discovered and tested in the same data is vulnerable to overfitting and the winner’s curse, in which initial estimated effects may be exaggerated. Independent replication asks whether the association remains credible in distinct evidence under a comparable phenotype and compatible method.

Specify whether replication uses new participants, a separate cohort or reused data. The GWAS Catalog may contain analyses that reuse cohorts, so two publications are not automatically independent tests. Differences across populations can be scientifically informative; failure to replicate is a reason to investigate, not automatically proof the original study was fraudulent.

11. Consider gene–environment relationships carefully

Back to contents

Genes operate within environments that include exposures, behaviour, healthcare access and social conditions. Gene–environment interaction refers to a change in an association according to a defined environmental factor under a specified statistical model. Such interactions can be difficult to measure and reproduce.

Define environmental variables, their timing, confounding factors and the interaction scale before testing. A striking subgroup contrast can arise by chance if many combinations were explored. Avoid treating social disadvantage as a genetic attribute or using genetic associations to minimise preventable environmental causes.

12. Separate heritability from inevitability

Back to contents

Heritability is a population-level statistical concept describing variation under specified conditions and methods. It is not the percentage of one person’s illness caused by their genes and does not mean a trait cannot respond to environmental change. Estimated values can depend on population and setting.

Explain the model and the population before quoting a heritability estimate. A highly heritable trait can still be affected by interventions. The useful question is how observed variation and causal mechanisms relate, not whether a trait can be labelled simply ‘genetic’ or ‘environmental’.

13. Evaluate polygenic risk scores with humility

Back to contents

Polygenic risk scores combine information from many variants into a statistical score associated with a particular trait or condition. NHGRI’s October 2026 definition emphasises that a PRS uses genomic information alone, not the full set of environmental and clinical factors influencing health.

A PRS needs validation for its intended population, outcome, time horizon and clinical use. Calibration, discrimination and subgroup performance matter. A higher score is not a diagnosis; it does not indicate that a person must develop disease or that treatment based on the score necessarily improves outcomes.

14. Assess portability and representativeness

Back to contents

Predictive associations estimated mainly in one set of populations may perform differently in others because allele frequencies, linkage patterns, environmental factors and data collection differ. Underrepresentation can lead to unequal performance and downstream health disparities if scores are implemented without appropriate evaluation.

Report who contributed to discovery and validation samples and how population descriptors were assigned. A broad regional label may hide substantial heterogeneity. External validation and calibration should be undertaken where a score would be used; promising retrospective performance in one setting is insufficient.

15. Use summary statistics with their metadata

Back to contents

GWAS summary statistics describe variant-level associations across an analysis, usually including effects, uncertainties and p-values, alongside metadata. The GWAS Catalog provides standardised formats and information about reference assemblies, harmonisation and sample ancestry. These features support reuse but do not remove the need to assess source studies.

When meta-analysing results, check effect alleles, genome builds, sample overlap, phenotype comparability and covariates. An allele reversal can produce an apparent conflict, while reused individuals can inflate certainty. Preserve data licensing, provenance and the analytical choices that reconcile files.

16. Distinguish Mendelian randomisation from automatic causality

Back to contents

Mendelian randomisation uses genetic associations as instruments under assumptions to study potential causal effects of exposures. The method can help with causal inference but can be undermined by horizontal pleiotropy, invalid instruments, population structure and other biases. It should not be interpreted as a mechanical conversion from association to proof.

State the exposure, outcome, selected instruments and assumptions. Test sensitivity to alternative instruments and model specifications where appropriate. A biological narrative remains provisional when evidence cannot distinguish the desired pathway from other genetic influences.

17. Protect participants and genomic data

Back to contents

Individual genomic information and linked health records are sensitive; identity risks may remain even when obvious personal fields are removed. Study permissions, data access, consent, retention and sharing rules need careful governance. Family relationships can also create interests that extend beyond the individual participant.

Minimise data exposure and follow applicable ethics and privacy duties. Share summary statistics responsibly under permitted terms, while guarding against unsupported interpretations about populations. Findings should never be used as a simplistic ranking of people’s worth, potential or inevitable health.

18. Make results transparent, reproducible and useful

Back to contents

The STREGA statement extends observational-study reporting guidance to genetic association research, drawing attention to methodological issues such as genotyping errors, population stratification and reporting. It is a reporting guide, not a stamp guaranteeing a study’s validity or clinical utility.

Preserve phenotype definitions, cohorts, variants, effect alleles, analysis code, multiple-testing strategy, uncertainty and replication status. Explain what would overturn the preferred interpretation. The gold standard is a credible public-health and biological investigation whose results remain open to revision rather than a deterministic claim about an individual.

Worked fictional case A: an odds ratio is not the probability of illness

A hypothetical case-control sample has 400 people classified as cases and 600 controls. Among cases, 100 carry a specified variant and 300 do not. Among controls, 90 carry it and 510 do not. The sample odds ratio for variant carriage is (100 × 510) ÷ (300 × 90) = 51,000 ÷ 27,000 ≈ 1.89. That is a measure of association between the variant indicator and case status in the constructed sample.

It is not a statement that 1.89 people will become ill or that carriers have an 89% probability of disease. Case-control recruitment fixes the case and control counts, so sample case fractions do not estimate ordinary population incidence. This simple example does not adjust for confounders or examine statistical uncertainty, linkage disequilibrium or causal mechanisms. Every count is invented.

Worked fictional case B: one million tests and false-positive expectations

Imagine one million independent tests in which every null hypothesis really is true. At an uncorrected p-value threshold of 0.05, the expected number of false positive results is 1,000,000 × 0.05 = 50,000. With a threshold of 5 × 10⁻⁸, the comparable expected count under these simplified assumptions is 1,000,000 × 0.00000005 = 0.05. This explains the logic behind stringent genome-wide testing standards.

Real variants are not independent, study designs differ and a low p-value does not prove a mechanism. The GWAS Catalog’s p < 10⁻⁵ top-association curation threshold is a different rule from commonly used genome-wide significance. This exercise should not be used to judge the clinical credibility of any individual published locus.

Practical genetic-epidemiology research audit

  • Identify source population, case definition and recruitment.
  • Review genetic assay quality and variant harmonisation.
  • Preserve ancestry methods and sample descriptions without conflating social and genetic categories.
  • Report effect allele, statistical model, effect size and uncertainty.
  • Control multiple testing and inspect population structure.
  • Separate reused cohorts from independent replication samples.
  • Validate prediction and calibration before discussing individual use.
  • Protect privacy, permissions and participant interests.
  • Report limitations following appropriate research guidance.

Frequently asked questions about genetic epidemiology

What is genetic epidemiology?

It studies how genetic variation and other factors relate to patterns of health and disease in populations.

Is genetic epidemiology the same as genomics?

No. Genomics studies genomes broadly; genetic epidemiology focuses on population patterns, associations and causal questions.

What is GWAS?

A genome-wide association study tests genomic variants for statistical relationships with defined traits across many participants.

Does a GWAS association prove that a gene causes disease?

No. A signal may tag correlated variation and be influenced by study design and confounding.

Is genetic ancestry identical to race or nationality?

No. These describe different genetic, social and political concepts and should not be substituted for one another.

What is the usual genome-wide significance threshold?

A commonly used threshold is p below 5×10⁻⁸; the appropriate approach depends on study design and testing.

Why does the GWAS Catalog use p below 1×10⁻⁵?

It is a threshold for curating certain top associations, not its definition of genome-wide significance.

Does an odds ratio directly tell a patient’s chance of disease?

No. Absolute risk requires population context, design, baseline rates and other relevant factors.

What is linkage disequilibrium?

A nonrandom association of alleles at different genomic positions that can make a marker track another variant.

What is a polygenic risk score?

A statistical score combining multiple genomic variants, whose practical interpretation depends on validation and context.

Are polygenic scores equally reliable across all populations?

Not necessarily. Data representation, linkage patterns and environmental context can affect performance.

What makes a genetic association robust?

Clear phenotypes, reliable genotypes, confounding control, multiple-testing discipline, independent replication and transparent reporting.

Connected Gold Standard reading: genetics to population evidence

Genetics introduces inheritance, genomics provides whole-genome measurement, epidemiology studies population patterns and biostatistics tests uncertainty. Bioinformatics helps manage genomic data, while health informatics addresses trustworthy health records. Read them together rather than using a single variant association as an explanation of a whole condition.

Primary scientific references and editorial scope

Sources reviewed on 24 October 2026. NHGRI updated its GWAS and PRS glossary entries on 22 October 2026. GWAS Catalog curation guidance distinguishes reported top associations from full summary statistics, and the older STREGA statement remains a reporting resource rather than a validity certification. The sample sizes, variant counts, odds ratio and testing calculations here are wholly fictional and must not be interpreted as any person’s genetic or medical risk.

Return to chapter index · The Gold Standard hub · How X Works hub