What is the STROBE checklist?

The STROBE checklist is a 22-item reporting guideline for observational epidemiology, published in October 2007, that specifies what a cohort, case-control or cross-sectional study must state in its manuscript. Eighteen items apply to all three designs; four — items 6, 12, 14 and 15 — carry design-specific wording. STROBE governs reporting, not study quality.

STROBE stands for STrengthening the Reporting of OBservational studies in Epidemiology. The statement was written by von Elm, Altman, Egger, Pocock, Gøtzsche and Vandenbroucke and published simultaneously across several journals, including The Lancet, BMJ, Annals of Internal Medicine, PLoS Medicine and Epidemiology. This page covers what the 22 items require, which four differ by design, which items authors most often omit, how missing compliance shows up in a manuscript, what reviewers write when it has been mishandled, and the two things STROBE explicitly does not do.

What STROBE is, and what it is not

The STROBE statement is a reporting guideline: a specification of what must appear in the text of an observational study, not a standard for how the study should have been designed or analysed. The STROBE Initiative states this in two clauses worth quoting exactly, because both are widely ignored: “the checklist is not an instrument to evaluate the quality of observational research”, and the “recommendations are not prescriptions for designing or conducting studies”.

Those two sentences separate STROBE from three things it is regularly confused with. STROBE is not a risk-of-bias tool such as ROBINS-I or the Newcastle-Ottawa Scale, which score the study. STROBE is not a methodological standard: an item can be fully satisfied by a transparent description of a weak design, and satisfying every item does not make the estimate valid. STROBE is not a registration or data-sharing policy; item 22 asks for the funding source and the role of the funders, and no item requires a protocol, a preregistration or a data availability statement.

The practical consequence is that a manuscript can score 22 out of 22 on STROBE and still be rejected on methods. STROBE makes the study’s weaknesses legible; it does not remove them.

The 22 items, by manuscript section

The STROBE checklist follows the structure of a paper, which is what makes it usable as a drafting tool rather than only as a compliance form. The wording below is condensed from the official cohort checklist; asterisked items (8, 13, 14, 15) additionally require that information be given separately for exposed and unexposed groups, or for cases and controls in a case-control design.

Items 18 to 21 are four separate items, not one discussion section. Manuscripts routinely merge limitations, interpretation and generalisability into a single closing paragraph, which leaves at least two of the three unaddressed in any reviewable form.

The four items that change with study design

Four STROBE items — 6, 12, 14 and 15 — are written differently for cohort, case-control and cross-sectional studies, which is why the STROBE Initiative publishes a separate checklist for each design as well as a combined one.

Item 6 (participants). The cohort version asks for eligibility criteria, sources and methods of participant selection, and a description of the methods of follow-up. The case-control version asks instead for the methods of case ascertainment and control selection, plus “the rationale for the choice of cases and controls” — a requirement that has no cohort equivalent and that many case-control manuscripts skip entirely. Sub-item 6(b) asks for matching criteria in both designs, but for the number of exposed and unexposed in a cohort and the number of controls per case in a case-control study.

Item 12(d) (statistical methods). The cohort version asks how loss to follow-up was addressed. The case-control version asks how the matching of cases and controls was addressed in the analysis. The cross-sectional version asks about analytical methods that take account of the sampling strategy. Three different questions occupy the same numbered slot, which is the single most common source of confusion when authors download the wrong checklist.

Items 14 and 15 (descriptive and outcome data). The cohort checklist adds sub-item 14(c), summarise follow-up time — average and total amount — which does not exist in the case-control version. Item 15 asks a cohort to report numbers of outcome events over time, and asks a case-control study to report “numbers in each exposure category, or summary measures of exposure”, because the sampling in a case-control study runs in the opposite direction.

Downloading the combined checklist and answering the cohort line of items 6, 12(d), 14 and 15 for a case-control study is a real and frequent error — the combined checklist stacks all three designs’ wording inside one Recommendation cell. It produces a completed form on which the design-specific lines of those four items were answered against the wrong design, because the combined checklist stacks the cohort, case-control and cross-sectional wording one under the other.

The items authors most often omit

Item 10, study size, is one of the items reviewers most often find unaddressed, and it is frequently left blank by mistake rather than by choice. Item 10 does not require a power calculation; it requires an explanation of how the number was arrived at, and “all eligible records in the registry between 2015 and 2022” is a complete and honest answer to it. Authors omit it because they assume it demands a formal calculation they did not perform, and a reviewer then reads the silence as a hidden sample size problem.

Item 9, bias, is the second. A sentence such as “we adjusted for potential confounders” does not satisfy item 9, which asks what was done about sources of bias — selection, measurement and information bias included — not only about confounding.

Items 12(c) and 14(b) both concern missing data, from opposite directions: 12(c) asks how missing data were handled in the analysis, and 14(b) asks for the number of participants with missing data for each variable of interest. Reporting one without the other is common and immediately visible, because a reader can then see a method with no denominator, or a denominator with no method.

Item 16(c), translating relative risk into absolute risk for a meaningful time period, is qualified with “if relevant” and is therefore very often skipped. It is the item that most changes how a clinical reader interprets a hazard ratio.

Published adherence audits of STROBE exist across many fields, and their headline percentages should be read with care rather than quoted against each other. The scoring rules are not standardised: some audits score each of the 22 items as present or absent, some score every lettered sub-item separately, some award partial credit for a partially addressed item, and some exclude items judged not applicable to the study at hand rather than counting them as failures. A denominator of 22 and a denominator that counts each sub-item are two different measurements of the same manuscript, and an audit that drops “not applicable” items reports a higher figure than one that does not. Two audits of the same literature can therefore disagree by a wide margin without either being wrong.

A pattern worth checking in your own field, rather than an established rate you can quote: the items that ask an author to justify a decision — study size, efforts to address bias, and the pair of missing-data items — are the ones most often scored as failures, while the descriptive front-matter items (background, objectives, study design) are satisfied far more consistently. Verify it against an audit of your own literature before relying on it, and if you cite an adherence figure in your discussion, give the specific audit, its field, and its scoring rule alongside the number, because the number alone does not travel.

Item 16(a) is where methodological review actually begins

STROBE item 16(a) asks for unadjusted estimates, confounder-adjusted estimates with their precision, and an explicit statement of “which confounders were adjusted for and why they were included”. That last clause is the highest-leverage sentence in the whole checklist, and the one most often satisfied with the phrase “adjusted for age, sex and BMI” and no justification.

Reporting both the unadjusted and adjusted estimate is what makes the movement between them inspectable. A coefficient that changes direction on adjustment is the signature circumstance for collider bias and for adjustment on a mediator; a reader who only sees the adjusted number cannot notice it. Stating why each covariate was included is what separates an adjustment set chosen from an assumed causal structure from one chosen by stepwise selection — a distinction that no amount of confounding language in the discussion can substitute for.

Item 16(a) also creates the conditions for the Table 2 fallacy: a model fitted to estimate one exposure effect, presented as a table of adjusted coefficients that readers interpret as a set of causal effects. STROBE does not prohibit this, and complying with item 16 while committing the fallacy is entirely possible. Reporting completeness and causal correctness are separate axes.

How incomplete STROBE reporting is detected in a real manuscript

Detecting STROBE non-compliance in a submitted manuscript is a matter of searching for specific strings and reconciling specific numbers, not of forming an impression.

Search the title and abstract for a design term. Item 1(a) asks for the study’s design named with a commonly used term. Manuscripts that say “we analysed data from 40,000 patients” without the word cohort, case-control or cross-sectional fail item 1 in the first sentence, and this is the fastest single check available.

Reconcile three participant counts. The number in the abstract, the column total in Table 1, and the N in the regression model frequently differ. The gap is almost always complete-case exclusion, which item 14(b) requires stated per variable. An unexplained gap between an abstract N and a model N is the most reliable indicator that items 12(c) and 14(b) are both unaddressed.

Look for the dates. Item 5 asks for periods of recruitment, exposure, follow-up and data collection. Missing dates are how immortal time bias stays invisible: without the interval between cohort entry and exposure assignment stated explicitly, a reader cannot tell whether unexposed person-time was misallocated.

Check whether unadjusted estimates appear anywhere. Many manuscripts report only the fully adjusted model. Item 16(a) requires both.

Check whether category boundaries are given. Where a continuous variable has been split into tertiles or quartiles, item 16(b) requires the boundaries. A table of quartile odds ratios with no cutpoints cannot be compared to any other study.

Read the funding statement for the funder’s role. Item 22 asks for the source and the role. “This work was supported by grant XYZ” gives the source only.

PerfectPaper runs these reconciliations against the manuscript text itself, reporting where a numbered STROBE item has no corresponding sentence and where two reported counts disagree; the epidemiology review lane is where that check lives.

What a peer reviewer says when STROBE has been mishandled

Reviewer comments about STROBE compliance are unusually formulaic, because the reviewer is reading against a numbered list. Phrasings that recur:

“The abstract does not state the study design; please indicate this using a standard term (STROBE item 1a).”

“No justification is provided for the study size. Please explain how the sample size was arrived at, even if no formal calculation was performed.”

“The authors do not describe any efforts to address potential sources of bias.”

“Numbers do not reconcile: the abstract reports 12,480 participants, Table 1 reports 12,104, and the adjusted model appears to include 9,867. Please report the number with missing data for each variable and state how missing data were handled.”

“Please report unadjusted as well as adjusted estimates, and state the basis on which each covariate was included in the model.”

“The discussion combines limitations, interpretation and generalisability. Please address external validity separately.”

“Please complete the STROBE checklist for cross-sectional studies and indicate the page number for each item.”

The last of these is procedural rather than substantive, and it is where many authors first meet STROBE — as a form requested at revision. Completing it retrospectively is when the omissions become visible, which is an argument for using the checklist while drafting rather than after review. If the objections arriving are about design rather than reporting, they belong to a different class; see reviewer objections about methods.

Extensions for specific designs and data sources

STROBE has spawned a family of extensions, each adding items to the core 22 for a particular design, data source or field, and the EQUATOR Network catalogue is the place to check whether one exists for yours. Five are worth knowing by name.

RECORD covers studies using routinely collected health data — administrative claims, electronic health records, disease registries — and was published by Benchimol and colleagues in PLoS Medicine in 2015. RECORD adds items on database and code list transparency, on the population selected from the database, and on data cleaning, all of which the core STROBE items assume away because they assume a purpose-collected dataset. Studies drawing on registries carry reporting obligations specific to secondary data that item 8 alone does not cover.

STROBE-MR covers Mendelian randomisation and was published by Skrivankova and colleagues in JAMA in 2021, with 20 main items. STROBE-MR asks authors to justify why Mendelian randomisation is a helpful method for the question and to report the assessment of instrument validity — requirements with no analogue in core STROBE.

STREGA extends STROBE to genetic association studies, adding items on genotyping methods, population stratification and Hardy-Weinberg equilibrium.

STROBE-nut covers nutritional epidemiology, and STROBE-EQUITY addresses the reporting of health equity considerations, an area where the core checklist is silent about how population subgroups are defined and analysed; see health equity reporting for the manuscript-level view.

A separate STROBE checklist exists for conference abstracts, which is shorter and is the version relevant when a study first appears as a meeting abstract rather than a full paper.

The misuse the STROBE authors themselves documented

The most consequential misuse of the STROBE checklist is its use as a methodological quality score in systematic reviews and meta-analyses. This is not a matter of opinion: da Costa, Cevallos, Altman, Rutjes and Egger examined how STROBE was being used in the literature and reported in BMJ Open in 2011 that “the STROBE reporting recommendations are frequently used inappropriately in systematic reviews and meta-analyses as an instrument to assess the methodological quality of observational studies”. The study was conducted within the framework of the revision of the STROBE recommendations.

The reason this matters is that the two things are uncorrelated in a specific and damaging way. A well-designed study reported tersely scores badly on STROBE; a badly designed study reported at length scores well. Ranking studies by STROBE item count in a meta-analysis therefore weights the review by writing thoroughness. Risk-of-bias assessment for non-randomised studies has purpose-built instruments — ROBINS-I for interventions, and design-appropriate alternatives — and STROBE is not among them.

The second misuse is treating the checklist as a submission formality. Filling in page numbers against 22 items after the manuscript is written finds omissions at the worst possible moment, when adding item 14(b) means recomputing the missing-data counts and item 16(a) means refitting unadjusted models.

Honest limitations of STROBE

STROBE has limitations its own authors state and that the surrounding evidence base supports, and stating them is more useful than treating the checklist as settled.

Evidence that STROBE improves reporting is largely observational and confounded. Before-and-after comparisons of journals that endorsed STROBE cannot separate the checklist’s effect from concurrent changes in editorial practice and in the research being submitted. Randomised evidence is thin and indirect: one masked randomised trial found that adding a reporting-guideline-based review (STROBE or CONSORT) to conventional peer review improved manuscript quality scores (Cobo et al., BMJ 2011;343:d6783), but no randomised study measures the effect of journal endorsement of STROBE on item-level reporting completeness, and adherence audits use incompatible scoring rules, so their percentages should not be pooled or compared across fields.

Endorsement does not mean enforcement. A journal that lists STROBE in its instructions to authors frequently does not check it at review, and a completed checklist submitted with page numbers is rarely audited against the pages it cites. The failure mode is mundane rather than deceptive: an author fills in a page number for the item they believe they addressed, and no one re-reads that page against the item’s actual wording. Whether a specific journal enforces STROBE — and at what stage, submission or revision — is worth checking in its own author instructions before submission, because the answer determines whether the checklist is a drafting aid or a revision chore.

Several items are optional in practice. Item 13(c) says “consider use of a flow diagram” and item 16(c) is prefaced by “if relevant”. Optional items are omitted at high rates, and the checklist provides no mechanism to distinguish “not relevant here” from “not done”.

Core STROBE predates current expectations. The 2007 statement contains no item on data availability, code availability, preregistration of an observational protocol, or reporting of a prespecified analysis — expectations that many journals now impose independently. A STROBE-complete manuscript can still fail a journal’s data and code availability policy.

STROBE cannot detect causal overreach. Nothing in the 22 items prevents a paper from reporting an association impeccably and then interpreting it causally in the abstract. Item 20 asks for a “cautious overall interpretation”, which is a matter of degree, not a testable requirement; causal language discipline is the separate check that catches it.

How to use STROBE while drafting

Use the design-specific checklist, not the combined one, and download it before writing the methods rather than at revision. The four items that differ by design — 6, 12, 14, 15 — are the ones a combined checklist lets you answer for the wrong design without noticing.

Write item 5 and item 10 first. Relevant dates and the origin of the study size are the two things authors reconstruct least accurately months later, and both are frequently unrecoverable once the analysis has moved on.

Keep a running participant reconciliation from the first analysis, so that item 13’s stage-by-stage numbers and item 14(b)’s per-variable missingness fall out of the pipeline rather than being assembled by hand at the end. The flow diagram in item 13(c) is then a rendering of numbers already recorded.

Draft items 18 to 21 as four separate paragraphs with those four headings in the manuscript file, and merge them only if the journal’s format requires it. Sections written as one paragraph are almost never expanded back into four.

Review my manuscript

Frequently asked questions

What does STROBE stand for?

STROBE stands for STrengthening the Reporting of OBservational studies in Epidemiology. The STROBE statement is a 22-item checklist first published in October 2007 — simultaneously in The Lancet, BMJ, Annals of Internal Medicine, PLoS Medicine and Preventive Medicine, and subsequently in Epidemiology, the Bulletin of the World Health Organization and the Journal of Clinical Epidemiology, specifying what cohort, case-control and cross-sectional studies should report. Eighteen items are common to all three designs and four carry design-specific wording.

How many items are in the STROBE checklist?

The STROBE checklist contains 22 numbered items, several with lettered sub-items. Eighteen items are identical across cohort, case-control and cross-sectional designs; items 6, 12, 14 and 15 are worded differently for each, which is why the STROBE Initiative publishes three design-specific checklists alongside a combined version.

Is the STROBE checklist a quality assessment tool?

No. The STROBE Initiative states that “the checklist is not an instrument to evaluate the quality of observational research”. Using STROBE to score methodological quality in a systematic review was documented as a frequent misuse by da Costa and colleagues in BMJ Open in 2011. Purpose-built risk-of-bias instruments exist for that purpose.

Which STROBE checklist do I use for a case-control study?

Use the case-control checklist published by the STROBE Initiative, not the cohort or combined version. The case-control wording of item 6 asks for the methods of case ascertainment, control selection and the rationale for that choice; item 12(d) asks how matching was addressed in the analysis; item 15 asks for numbers in each exposure category.

What is the difference between STROBE and CONSORT?

STROBE is a 22-item reporting guideline for observational studies — cohort, case-control and cross-sectional. CONSORT is the corresponding guideline for randomised controlled trials. A study without randomised allocation is reported against STROBE; a randomised trial is reported against CONSORT. Both are reporting specifications rather than measures of study quality.

Do journals require the STROBE checklist?

Many journals list STROBE in their instructions to authors and request a completed checklist with page numbers at submission or revision, though endorsement and enforcement frequently differ, and a listed guideline is often not checked during review. Whether a given journal requires it should be verified in that journal’s own author instructions before submission.

Are there STROBE extensions for specific study types?

Yes, a number of them. RECORD covers routinely collected health data, published in PLoS Medicine in 2015. STROBE-MR covers Mendelian randomisation, published in JAMA in 2021 with 20 main items. STREGA covers genetic association studies, STROBE-nut covers nutritional epidemiology, and STROBE-EQUITY addresses health equity reporting.