STREGA (STrengthening the REporting of Genetic Association Studies)

What it is

— STREGA (STrengthening the REporting of Genetic Association Studies) is a reporting-guideline extension of STROBE (the Strengthening the Reporting of Observational Studies in Epidemiology statement) tailored to genetic association studies. It does not replace STROBE's 22 items; it adds genetics-specific reporting expectations to a subset of them. STREGA was produced by an international, multidisciplinary workshop convened in Ottawa (2006) and published in 2009 simultaneously across several journals (Little, Higgins, Ioannidis, and colleagues; PLoS Medicine, European Journal of Epidemiology, and others). It is maintained as a STROBE extension within the EQUATOR Network library. Its purpose is narrow and specific: to make the genotyping technology and quality control, the handling of population stratification, the treatment of Hardy-Weinberg equilibrium, the multiple-testing/multiple-comparison strategy, the modeling of genetic contrasts/haplotypes, and the replication and synthesis of findings transparent and reproducible in reports of gene-disease and gene-environment association studies. It is a reporting standard, not an analysis method and not a study-quality score.

When to use

— Apply STREGA when you are reporting (or peer-reviewing, or registering the reporting expectations for) an observational genetic association study — a candidate-gene study, a genome-wide association study (GWAS), or a meta-analysis/synthesis of genetic associations — most commonly built on cohort, case-control, or cross-sectional designs. In the RWE/HEOR world the realistic trigger is pharmacogenomic and biomarker-association work: a claims- or EHR-linked biobank study testing whether a genotype predicts drug response, an adverse drug reaction, or a disease phenotype; a registry-plus-genomics study; or any submission/publication where a genotype is the exposure or effect-modifier of interest. Decision rule for the right family member: use STROBE for a non-genetic observational study; use STREGA when genotype is central and you need the genetics-specific items layered on top of STROBE; use RECORD / RECORD-PE when the study is built on routinely collected health data and the data-provenance items dominate (and combine RECORD with STREGA if a routinely-collected-data study also carries a genetic exposure); and use STROBE-MR — not STREGA — when the design is a Mendelian randomization study using genetic variants as instruments, because the reporting burden there is about instrument validity, not association reporting. STREGA is for the association report; STROBE-MR is for the instrumental-variable causal report.

What it requires

— STREGA keeps STROBE's backbone (title/abstract, background, objectives, eligibility, variables, data sources/measurement, bias, study size, statistical methods, participant flow, descriptive and outcome data, limitations, generalizability, funding) and adds genetics-specific reporting at the points where genetic studies go wrong: (1) Genotyping and laboratory methods — the platform/assay, call thresholds, blinding of genotyping to outcome, and the genotyping error rate / call rate / quality-control procedures, including how SNPs or samples failing QC were handled. (2) Hardy-Weinberg equilibrium (HWE) — whether HWE was tested (typically in controls), the method and threshold, and how departures were interpreted (a classic flag for genotyping error or population structure). (3) Population stratification — how confounding by ancestry was addressed (e.g., restriction, family-based design, genomic control, principal components / ancestry adjustment). (4) Multiple testing / multiple comparisons — the number of variants and models tested and the correction or significance threshold (e.g., genome-wide significance), to constrain false-positive reporting. (5) Modeling of the genetic contrast — the inheritance model assumed (additive, dominant, recessive, genotypic), how haplotypes were inferred, and treatment of gene-gene and gene-environment interaction. (6) Replication and synthesis — whether findings were replicated in an independent sample, and, for pooled work, how between-study heterogeneity and meta-analysis were handled. For pharmacogenomic RWE specifically, the genetics items must be reported alongside the ordinary RWE reporting burden the linked data create — data fitness, the validity of the claims/EHR-defined outcome or drug-response phenotype, time-zero alignment, and confounding control — because a genotype-outcome association inherits every weakness of the phenotype it is regressed on.

When NOT to use — limitations and common misapplications

— (1) It is a reporting checklist, not a risk-of-bias instrument and not a quality score. A fully STREGA-compliant paper can still report a badly confounded, underpowered, or unreplicated association transparently; completeness of reporting is necessary, not sufficient, for validity. Do not use STREGA to grade studies — that is the job of appraisal tools (e.g., Newcastle-Ottawa for the underlying observational design). (2) Wrong design. Using STREGA for a non-genetic observational study where plain STROBE (or RECORD/RECORD-PE for routinely-collected data) is the correct standard is over-reach; using STROBE alone for a genetic association study under-reports the genotyping-QC, HWE, stratification, and multiple-testing items that STREGA exists to force. (3) Wrong genetic extension. Applying STREGA to a Mendelian randomization study instead of STROBE-MR misses the instrument-validity reporting (relevance, independence, exclusion-restriction, pleiotropy) that is the whole point of an MR report. (4) Checklist-as-theater. Ticking items while leaving the genotyping error rate, the HWE result, the stratification adjustment, or the multiple-testing threshold vague defeats the purpose; the value is the specific numbers, not the page count. (5) Reporting compliance is not causal inference — a STREGA-complete pharmacogenomic claims study with an unvalidated response phenotype and unaddressed ancestry confounding is still not a credible causal claim.

How it maps to this catalog

— STREGA is a reporting layer; in this repo its substantive requirements are implemented (and should be pre-specified and appraised) by these concepts: - Genotyping QC and the validity of the genetic/phenotype measurement: `algorithm-validation` supplies the validation discipline (sensitivity/specificity/PPV) that the genotyping-error-rate and phenotype-definition items demand by analogy; `biomarker-defined-cohort-rwe` operationalizes constructing a cohort around a genotype/biomarker exposure. - Data fitness for a linked-genomics RWE study: `fit-for-purpose-data-assessment-rwe` covers whether the linked claims/EHR/biobank substrate is adequate for the genotype-outcome question. - The structured question and confounding/stratification frame: `picots-framework-rwe` declares population/exposure(genotype)/comparator/outcome/timing/setting; `baseline-characteristics-and-covariate-balance-rwe` supports the ancestry/covariate reporting that population-stratification control requires. - Missing data and synthesis: `multiple-imputation-longitudinal-rwe` for missing genotype/covariate data, and `meta-analysis-obs` for the replication-and-synthesis item when genetic associations are pooled across studies/biobanks. - Distinguish from the instrumental-variable cousin: `instrumental-variables-pharmacoepi-rwe` is the analytic basis of Mendelian randomization — when genetics are used as instruments rather than as the exposure of interest, the report belongs under STROBE-MR, not STREGA.

Applied note (pharmacogenomic claims/EHR/biobank RWE)

For a study linking a genotyped biobank to claims/EHR to test a genotype-drug-response or genotype-ADR association, STREGA forces you to report the genotyping platform, call rate, and error rate; the HWE test in an appropriate reference group; the ancestry-adjustment (principal components) used to defuse population stratification; the genome-wide or candidate-set multiple-testing threshold; the inheritance model; and whether the signal replicated. None of that displaces the ordinary RWE reporting burden of the linked data — the validity of the algorithm-defined drug-response or outcome phenotype, time-zero alignment, and confounding control — and a credible report must satisfy both layers.