Written and maintained by CASRAI Editorial Board
Last updated
STROBE-MR is the reporting guideline for studies that use Mendelian randomisation (MR) — the method that uses genetic variants as instrumental variables to estimate whether an exposure causally affects an outcome. It is a formal extension of the STROBE checklist, adding items that address what STROBE’s base 22 items were never written to cover: how genetic instruments were selected and validated, which of MR’s three core assumptions the study relies on, and what sensitivity analyses were run to check whether those assumptions hold. Like RECORD for routinely collected health data, STROBE-MR does not replace STROBE — it sits alongside it, with each MR-specific item cross-referenced to the base item it extends or supplements.
STROBE-MR at a glance
| Question | Answer |
|---|---|
| What it applies to | Studies that use Mendelian randomisation — genetic variants as instrumental variables — to estimate a causal effect of an exposure on an outcome, whether one-sample (individual-level data) or two-sample (separate GWAS summary statistics for exposure and outcome) design |
| Relationship to STROBE | A formal extension: 20 items recommended for MR studies, several new and several elaborating on an existing STROBE item |
| Citation (statement) | Skrivankova VW, Richmond RC, Woolf BAR, Yarmolinsky J, Davies NM, Swanson SA, VanderWeele TJ, Higgins JPT, Timpson NJ, Dimou N, Langenberg C, Golub RM, Loder EW, Gallo V, Tybjaerg-Hansen A, Davey Smith G, Egger M, Richards JB. “Strengthening the Reporting of Observational Studies in Epidemiology Using Mendelian Randomization: The STROBE-MR Statement.” JAMA 2021;326(16):1614–1621. |
| Companion paper | An accompanying Explanation and Elaboration paper, published the same year in the BMJ, walks through the rationale and worked examples for each item — use it alongside the checklist, not instead of it, when an item’s intent isn’t obvious from the wording alone |
| Maintained by | The STROBE-MR working group; catalogued by the EQUATOR Network |
Why Mendelian randomisation needed its own extension
STROBE was written for observational studies broadly — cohort, case-control, cross-sectional — on the assumption that the main threat to validity is unmeasured confounding between a measured exposure and outcome. Mendelian randomisation is a different kind of observational study: instead of using the exposure directly, it uses genetic variants that are robustly associated with the exposure as instrumental variables, exploiting the fact that alleles are randomly allocated at conception (analogous, proponents argue, to randomisation in a trial) to reduce confounding and reverse causation. That design swaps STROBE’s usual concerns for a different set: whether the genetic variants are valid instruments in the first place, whether the exposure and outcome samples overlap or come from different populations, and whether an observed association could be explained by the variants affecting the outcome through some pathway other than the exposure under study. STROBE-MR’s items exist to make a reader able to judge those questions from the paper itself, not take them on faith.
The three core assumptions STROBE-MR asks authors to address
Every MR study rests on three assumptions about the genetic instrument, and STROBE-MR requires each to be explicitly discussed, not merely satisfied silently in the analysis:
- Relevance — the genetic variants are robustly associated with the exposure (typically demonstrated with instrument strength statistics such as an F-statistic).
- Independence (exchangeability) — the variants are not associated with confounders of the exposure-outcome relationship.
- Exclusion restriction (no horizontal pleiotropy) — the variants affect the outcome only through the exposure, not through some other biological pathway.
The first is checkable directly from the data; the second and third are not fully testable and instead require triangulation across sensitivity analyses — which is why STROBE-MR devotes several items specifically to reporting which sensitivity analyses were run and what they showed, rather than treating the primary estimate as sufficient on its own.
What the 20 items cover beyond the base STROBE checklist
STROBE-MR follows the same title/abstract, introduction, methods, results, discussion, and other-information structure as STROBE, but several items are specific to instrument-based causal inference:
- Genetic instrument selection — how variants were chosen (a single variant, a polygenic score, or multiple independent variants), the source study or consortium the associations came from, and whether variants were clumped for linkage disequilibrium.
- Data sources and sample overlap — for two-sample MR specifically, whether the exposure and outcome genome-wide association study (GWAS) samples overlap, since sample overlap can bias estimates toward the confounded (observational) association.
- Population structure — the ancestry of the study samples, since allele frequencies and effect sizes vary by ancestry and mismatched populations between exposure and outcome data can introduce bias.
- Assumptions — an explicit statement of the three core IV assumptions above and how plausible each is for the specific instruments used.
- Sensitivity analyses — which pleiotropy-robust methods were run alongside the primary estimate: MR-Egger (which allows for and estimates directional pleiotropy), weighted median and weighted mode estimators (robust if a minority of variants are invalid), and outlier-detection methods such as MR-PRESSO, plus leave-one-out analysis to check whether any single variant drives the result.
- Interpretation caveats — whether the exposure effect is interpreted as lifelong/lifetime exposure to a genetically-predicted level, rather than the effect of a short-term clinical intervention, and what that distinction means for how the result should be read.
One-sample vs. two-sample MR reporting
STROBE-MR applies to both designs, but several items resolve differently depending on which is used. One-sample MR estimates the genetic-instrument, exposure, and outcome associations in the same individual-level dataset, which avoids sample-overlap bias but is often underpowered unless the dataset is very large. Two-sample MR, now the more common design in practice, draws the exposure-instrument and outcome-instrument associations from two separate GWAS — usually published summary statistics rather than individual-level data — which allows enormous effective sample sizes but reintroduces sample-overlap and population-mismatch questions that STROBE-MR’s data-source and population-structure items exist specifically to surface.
How STROBE-MR relates to other reporting guidelines in this space
STROBE-MR sits in the same family as other STROBE extensions — the RECORD and RECORD-PE extension for routinely collected health data is the closest structural parallel: both add a targeted set of items on top of the unmodified base STROBE 22-item checklist rather than replacing it. It is a distinct guideline from PRISMA (for systematic reviews) and from reporting guidance for randomised trials such as CONSORT: an MR study is an observational analysis of existing genetic and phenotypic data, not a trial and not a synthesis of prior studies, even though its causal-inference ambitions are sometimes compared to a trial’s.
Frequently asked questions
Do I need to complete the base STROBE checklist as well as STROBE-MR?
Yes. STROBE-MR is additive, not a replacement — most journals and the EQUATOR Network’s own listing expect both the relevant base STROBE items and the STROBE-MR-specific items to be addressed, since an MR study is still fundamentally an observational study with the added instrumental-variable design layered on top.
Is STROBE-MR required for journal submission?
Requirements vary by journal. Many epidemiology, genetics, and general medical journals that have adopted EQUATOR Network reporting guidelines as a submission requirement either name STROBE-MR explicitly for MR manuscripts or expect authors to follow the most specific applicable EQUATOR guideline, which for an MR study is STROBE-MR rather than plain STROBE. Check the specific journal’s author guidelines rather than assuming.
Does STROBE-MR apply to both one-sample and two-sample MR studies?
Yes, the checklist covers both designs, though some items (particularly around sample overlap and data source description) are most relevant to two-sample MR using published GWAS summary statistics.
What is horizontal pleiotropy, and why does STROBE-MR ask about it specifically?
Horizontal pleiotropy is when a genetic variant affects the outcome through a biological pathway other than the exposure being studied, which violates the exclusion-restriction assumption and can bias the causal estimate. Because it cannot be ruled out with certainty from a single analysis, STROBE-MR asks authors to report the sensitivity analyses (MR-Egger, weighted median, outlier detection) used to assess its likely impact, rather than asserting the assumption holds without evidence.
