Maintenance
Part of Science health news verification: a practical guide
Science and health story sign-off checklist for editors
A science and health story sign-off checklist for editors: test the draft wording, regulator terms, evidence base and visuals before a health claim runs.
What to take away
- Edit the claim wording first and the study second.
- Match every regulator word in the draft (approved, cleared, registered) to one agency record.
- Refuse sign-off until the file holds five itemsstudy version, design, outcomes, participant counts, central result.
- One new paper does not replace earlier evidence unless support is shown.
- Headline, visuals and correction plan carry the same limits as the body.
Test the draft wording against the record
A sign-off editor starts from the sentence, not the paper. Every strong word in a health draft makes a promise, and the reporting file must hold the document that keeps it. The science and health verification guide covers the reporter's side.
Word in the draft / What the file must show
- Approved, cleared
- The agency record for that product, category and intended use
- Safe
- Harm data and follow-up length, or a bounded description instead
- Causes, prevents
- A design that can support a causal verb
- Cuts risk by
- Baseline risk and absolute change, not the relative figure alone
- Proves, overturns
- The earlier evidence and any current synthesis
- Works, benefit
- A patient outcome, not a biomarker or surrogate outcome
Match the verb to the design before anything else. A laboratory, animal, observational or experimental finding supports different verbs, and a story that names the design in plain words usually reads more honestly. The evidence types compared show what each design can claim.
Five items the reporter's file must hold
An editor cannot re-audit the paper, but can refuse a file that lacks the basics. Ask for each item with a reviewer name, a date and the location of the evidence. A checked box should mean the underlying document was inspected, not that someone recognized the publisher.
For the full audit of the paper itself, use the walkthrough on auditing a clinical study story. Here the question is only whether the file proves the work was done.
1. Study version
Study identity and version
- Title, authors, institution, DOI, date
- Preprint, accepted, final versions distinguished
- Correction, concern, retraction status checked
- Paper, supplement, protocol, registration saved
- Version used by story identified
- Preprint, accepted manuscript and final paper distinguished
- Correction, expression of concern and retraction status checked
- Institutional release kept separate from the paper
2. Design and question
Research question and design
- Population, intervention, comparator, outcome, period
- Design namedlab, animal, observational, etc.
- Design can answer the story's question
- Recruitment and selection examined
- Confounding, selection, reverse causation considered
- Population, comparator, outcome and period written plainly
- Design can answer the question the story asks
- Confounding, selection and measurement error considered
3. Planned and reported outcomes
Protocol and outcome audit
- Primary and secondary outcomes from earliest plan
- Outcome definitions, instruments, time points match
- Eligibility, enrollment goal, stopping rule compared
- Amendments dated and explained
- Subgroup and exploratory analyses labeled
- Unreported planned outcomes queried
- Primary and secondary outcomes copied from the earliest plan
- Amendments dated and explained
- Subgroup and exploratory analyses labeled
4. Participant counts
Participant and sample flow
- Screened, eligible, assigned, treated, followed, analyzed
- Exclusions and missing data explained
- Attrition compared across groups
- Sample size and power tied to main question
- Study population matches people in story
- Cell or animal findings not called human outcomes
- Screened, assigned, followed and analyzed counts reconciled
- Study population matches the people named in the story
- Cell, animal or model findings not described as human outcomes
5. Central number
Statistical result
- Main calculation reproduced from reported inputs
- Events and denominators retained
- Absolute and relative effects both assessed
- Confidence interval or uncertainty carried into story
- Statistical significance separated from clinical importance
- Multiple comparisons and subgroup claims cautious
- Model assumptions and sensitivity analyses reviewed
- Main calculation reproduced from reported inputs
- Absolute and relative effects both assessed
- Statistical significance separated from clinical importance
Regulator words in plain English
Regulator verbs are not interchangeable. The FDA's page on product approval explains that premarket requirements differ by product category: new drugs and some devices require review, while cosmetics and dietary supplements generally do not receive the same prior approval.
Verify the category before attaching an FDA verb to a product. Then check three more points before the draft passes.
- Product identity, manufacturer, formulation, dose, route and jurisdiction matched
- Approved, authorized, cleared, registered, listed and investigational used precisely
- Trial registration not described as approval or endorsement
- Regulatory date is current for publication
Weigh one paper against the body of evidence
A new paper is one data point. FDA's framework for the scientific evaluation of health claims assesses study type, methodological quality, sample size, relevance, replication, consistency, and the totality of the evidence.
The guidance addresses food-label health claims, but those review dimensions are also useful checks against building a news conclusion from one favorable result.
- Earlier studies and current evidence synthesis found
- Contrary results described
- Mechanistic plausibility not substituted for demonstrated patient benefit
Headlines, visuals and corrections
Read the headline last, but hold it to the strictest standard. Headline, deck, body, pull quote, caption, metadata and social copy must agree, and the material limit has to sit near the finding rather than in the final paragraph.
- Causal verbs match the design
- Population and period remain visible
- Figure traced to the paper and original data when available
- Anomaly in an image not reported as misconduct without investigation
- Conflicts reported as risk information, not proof that a finding is false
- Correction plan covers all formats and distribution surfaces
Common questions
Must every item be checked for a short news brief?
The depth can vary, but study identity, design, central result, uncertainty, conflicts, current status, and claim wording cannot be skipped safely.
Does FDA registration mean FDA approval?
No. The meaning depends on the product category and program. Use the specific agency record and exact regulatory term.
Is an image discrepancy proof of fabrication?
No. Preserve the image, compare originals, ask the authors and journal, and distinguish a visible anomaly from a finding about cause or intent.
What if the study conflicts with a systematic review?
Explain the difference in population, intervention, outcome, design, and date. Do not select the newer result solely because it is new.
