The STROBE checklist is a 22-item reporting guideline that sets out what an observational study must disclose for a reader to judge it and reproduce its reasoning. Published by von Elm and colleagues in 2007, it covers the three analytical observational designs: cohort studies, case-control studies, and cross-sectional studies. STROBE governs how you report a study you have already done. It does not tell you how to design one, and it does not score quality, which is a separate job handled by appraisal instruments.
That distinction is the reason so many manuscripts come back with reporting queries even when the underlying research is sound. A well-designed cohort study can still be rejected because the reader cannot tell how many participants were lost to follow-up, which variables were adjusted for, or how missing data were handled. STROBE exists to close exactly those gaps, and completing it honestly before submission is the cheapest quality gain available to an observational study.
There is no single STROBE file, and taking the wrong one is the most common practical mistake. Four of the 22 items are worded differently depending on design, because the sampling logic differs:
- Cohort studies. Participants are grouped by exposure and followed for outcomes. The design-specific wording asks how follow-up was handled, and for exposed and unexposed groups to be described separately.
- Case-control studies. Participants are selected by outcome status and exposure is looked at retrospectively. The wording asks for the rationale behind case and control selection and for any matching criteria.
- Cross-sectional studies. Exposure and outcome are measured at one point in time. The wording asks how the sample was drawn and for the analytic methods accounting for the sampling strategy.
A combined checklist covering all three exists, and a separate shorter version covers conference abstracts. If your design is a variant, for example a nested case-control or a case-cohort study, take the checklist for the closest parent design and state the variation explicitly in the methods rather than silently reinterpreting items. Our guide to choosing and reporting a cross-sectional design covers the analytic decisions that sit behind these items.
The items follow the shape of a manuscript, so working through them in order doubles as a structural check on the paper. Described in our own words, the sections ask for the following. The authoritative wording lives with the guideline itself, and the checklist is copyright of the original authors, so download the official file rather than relying on any summary including this one.
Title and abstract. One item, asking that the design be named in the title or abstract and that the abstract give a balanced summary of what was done and found.
Introduction. Two items covering the scientific background for the investigation and a clear statement of specific objectives, including any prespecified hypotheses.
Methods. Nine items, the heaviest section and the one that decides whether the study is reproducible. It covers the study design, the setting and relevant dates including recruitment and follow-up periods, participant eligibility and selection, the definition of every outcome, exposure, predictor, confounder and effect modifier, the sources and measurement methods for each variable, the efforts made to address potential sources of bias, how the study size was arrived at, how quantitative variables were handled and grouped, and the statistical methods including subgroup analysis, handling of missing data, loss to follow-up, and any sensitivity analysis.
Results. Five items, covering participant flow at each stage with numbers and reasons, the characteristics of participants including numbers with missing data, outcome event counts or summary measures, the main results with unadjusted and confounder-adjusted estimates plus their confidence intervals, and any other analyses performed.
Discussion. Four items: key results against the objectives, the limitations including the direction and magnitude of any likely bias, a cautious overall interpretation, and the generalisability of the findings.
Other information. One item asking for the funding source and the role of funders.
Completed checklists are frequently accurate about the easy items and vague about the hard ones. Four failures recur:
- Participant flow without numbers. Saying participants were "followed up where possible" satisfies nobody. The item wants counts at each stage and the reasons for each loss, which is why a flow diagram is usually the efficient answer even though STROBE does not demand one.
- Confounders introduced in the results. If the first mention of an adjustment set is the regression table, a reader cannot tell whether those variables were chosen in advance or after seeing which combination moved the estimate.
- Missing data described as "excluded". Complete case analysis is a legitimate choice, but it is a choice, and the item asks how missing data were addressed rather than whether they were removed.
- Limitations that are not limitations. "A larger sample would be desirable" is a formality. STROBE asks for the direction and size of potential bias, which means naming the specific threat and saying which way it would push the estimate.
Reporting guidelines are design-specific, and using the wrong one reads as carelessness. STROBE is for observational studies. For randomised trials the equivalent is the CONSORT checklist, and for the trial protocol written beforehand it is SPIRIT. For a systematic review of any of these, the reporting guideline is PRISMA 2020, with PRISMA-P covering the review protocol. Qualitative work uses the COREQ checklist or SRQR, diagnostic accuracy studies use STARD, and single patient reports use CARE.
The confusion worth clearing up is between reporting and appraisal. STROBE asks whether a study is adequately described. Appraisal instruments ask whether it was well conducted. A study can be reported impeccably and still be at high risk of bias, which is why a systematic review that includes observational studies needs both: STROBE to judge the reporting, and a tool such as the Newcastle-Ottawa Scale or ROBINS-I for non-randomised intervention studies to judge the conduct. Presenting a STROBE score as a quality score is a methodological error, and reviewers do catch it.
STROBE is at its most useful read backwards. If you are extracting data from observational studies into a synthesis, the 22 items are a ready-made list of what you should be able to find in each paper, and the items you cannot answer are the ones to record as unclear rather than guessing. That turns reporting quality into a documented, reproducible judgement instead of an impression, and it feeds directly into the certainty assessment. Our walkthrough of the GRADE approach to certainty of evidence explains how reporting gaps propagate into a downgrade.
Complete the checklist last, against the final draft, because every entry has to point at text that actually exists. For each item, record the page and section rather than writing "yes", since a page number is checkable and a tick is not. Where an item genuinely does not apply, write not applicable and give the reason in a few words. Submit the completed file as a supplementary document, and keep it updated through revision, because reviewers who asked for new analyses will expect the checklist to reflect them. If a journal asks for the checklist as part of a resubmission, the response to reviewers is the place to state plainly which items changed and where.