Science reporting guides work best when evidence and advice stay separate

What to take away

• Name the evidence before describing the result.
• Separate a measured association from a demonstrated cause.
• Give absolute numbers, a time frame, and the population studied.
• State what researchers know, what remains uncertain, and what could change.
• Put practical health advice in the words of the responsible authority.

A science or health story often begins with a result that sounds simple. A drug reduced an outcome. A virus was found in wastewater. A food was associated with disease. The reporting job is to show what produced that result, how far it reaches, and what a reader should do with it.

The method matters because the same number can carry very different weight in a randomized trial, an observational study, a laboratory experiment, or a surveillance report. A sound background briefing identifies the evidence first and saves the strongest language for the strongest design.

Photo and credit

The photograph shows laboratory work. It does not illustrate a particular study, diagnosis, institution, or finding in this article.

Start with an evidence card

Before drafting, put the claim into a compact record.

Field	Question to answer
Claim	What is the narrowest accurate statement?
Evidence	Trial, cohort, survey, model, laboratory test, case report, or surveillance?
Population	Who or what was studied, and where?
Comparison	Compared with what group, exposure, or baseline?
Outcome	What was measured, and how was it defined?
Size	How many people, samples, places, or observations?
Time	Over what period was the outcome measured?
Uncertainty	What interval, error range, sensitivity test, or stated limitation applies?
Status	Preprint, conference abstract, peer-reviewed paper, agency report, or guidance?

If a field cannot be filled, say so. An unanswered question is more useful than a polished assumption. The evidence card works like the statistical claim checklist: a blank field is a reporting task.

Match the verb to the design

Words such as "caused," "prevented," and "proved" make causal claims. Use them only when the study design, analysis, and supporting record justify them. Observational work commonly supports language such as "was associated with" or "occurred more often among."

Laboratory findings can explain a mechanism without showing how often an effect occurs in people. An animal study can justify further research without predicting a clinical benefit. A model produces an estimate under stated assumptions. It does not observe the future.

Translate size without stripping context

Report the numerator, denominator, baseline, period, and comparison whenever the source gives them. A relative change alone can make a small difference sound large. A percentage without its base can hide whether it refers to people, tests, samples, visits, or survey responses. These are the same traps cataloged in statistics reporting problems.

Useful translations include:

• events per 1,000 people in each group;
• average change on a named scale;
• range of plausible estimates;
• number needed to treat or harm, when reported and suitable;
• difference between statistical significance and practical size.

Do not convert an estimate into a count unless the units and population support the calculation.

Treat uncertainty as information

Uncertainty is not a disclaimer attached at the end. It helps readers understand what the evidence can support today. Record uncertainty beside each major claim, then decide whether it belongs in the headline, lead, chart, or explanation.

The CDC's guidance on acknowledging uncertainty in health communication tells communicators to explain what is known, what is not known, and what is being done to learn more. It also notes that recommendations can change as new information arrives. That advice supports direct revision language rather than pretending that early guidance was final.

Check experts and institutions

Ask an expert to identify the strongest evidence, the largest limit, and the claim they would refuse to make. Record the person's field, institutional role, and relevant financial or professional interests. Expertise in one medical specialty does not automatically cover epidemiology, statistics, regulation, or patient care.

Institutional prestige cannot repair a weak method. Read the study and supporting material, not just a news release. If access is limited, state what was available and avoid conclusions that depend on missing methods or tables.

Separate evidence from advice

A study result and public guidance answer different questions. Research may estimate an effect. Guidance weighs evidence alongside feasibility, supply, adverse effects, equity, law, and current conditions.

WHO's handbook for guideline development documents that weighing process step by step, down to a chapter on issuing a strong recommendation when the evidence is of low quality. The chapter title alone tells a reporter that a recommendation encodes more than an effect estimate.

Attribute recommendations to the responsible health agency or clinical body. State the date and population covered. If agencies disagree, identify the difference instead of blending the advice into one sentence. Attribution mechanics follow the guide to developing news stories: name the source where the claim appears.

Build a correction path

Science coverage changes. Preserve the first publication time, label material updates, and explain what changed. Correct a wrong number in place, but keep a transparent note when the correction alters the story's meaning. The correction mechanics are demonstrated in the developing story desk case, down to the note's wording.

For a developing finding, schedule checks for peer review, revised datasets, regulator action, and later safety signals. The follow-up is part of the reporting plan.

Common questions

Can one study settle a health question?

Rarely. Its weight depends on design, size, execution, consistency with other evidence, and relevance to the population in question.

Should a preprint be covered?

It can be, if its unreviewed status is prominent and the story explains why the result matters before formal peer review. The claim should not outrun the available methods and data.

What belongs in a health headline?

Name the population or evidence type when omission would exaggerate the finding. Avoid advice or causal language that the evidence does not support.

How should changing guidance be reported?

Give the old and new guidance, dates, stated reason for the change, affected population, and remaining uncertainty.