Reading a "breakthrough" press release: five questions to ask first

Most science stories you see began as a press release. A university, a hospital or a company writes a short, upbeat account of a new result, sends it out under an embargo, and reporters with too little time and too many stories turn it into news. The release is not the study. It is an advertisement for the study, written by people whose job is to get attention for their institution.
That does not make releases dishonest. Many are careful. But the gap between what a paper shows and what a release implies can be wide, and the headline usually inherits the release's version. A 2014 study in The BMJ compared press releases from UK universities with the news stories based on them and found that when the release exaggerated, the news coverage was much more likely to exaggerate too. The fix starts upstream, with the reader.
Here are five questions that sort most releases quickly. After them comes a worked example on a release I wrote for this purpose.
The five questions
1. Mice or people?
The single most useful word to look for is the species. Results in cells, organoids or animals are real science and often important, but they are early steps. Treatments that look promising in mice fail in people far more often than they succeed, for reasons ranging from different biology to the simple fact that lab mice are genetically uniform and live in controlled cages. Our piece on why most drugs fail between phase 2 and approval covers the later attrition, but the drop from animals to the first human trials is steeper still.
If the release does not name the species in its first two paragraphs, assume it is not people until proven otherwise.
2. How many?
Count the subjects. Twelve mice, eight patients, a single cell line from one donor. Small studies can be valuable, but they produce noisy results, and a striking effect in a small group often shrinks when the study is repeated with more participants. Also check how many started versus how many were analyzed. If 40 enrolled and 28 appear in the results, the missing 12 are part of the story.
3. What exactly was measured?
Releases tend to name the hoped-for outcome rather than the thing that was measured. "Reverses heart failure" may mean a change in a blood marker. "Restores memory" may mean faster performance in a maze. These stand-in measures, called surrogate endpoints, can be useful, but they do not always move in step with the outcomes patients care about, such as living longer, feeling better or avoiding a hospital stay.
Look out for relative numbers too. "Cuts risk in half" sounds dramatic. If the risk fell from 2 in 1,000 to 1 in 1,000, the absolute change is one person per thousand, which may still matter but is a different story.
4. Compared with what?
Every result is a comparison, and the release should say against what. A randomized control group given a placebo or standard care is the strongest answer. A comparison with patients treated somewhere else in the past, known as a historical control, is weaker. No comparison at all, just "patients improved", tells you very little, because many conditions improve on their own and people in studies often do better simply from extra attention.
For a human trial, you can usually check the design yourself. Our guide to reading a ClinicalTrials.gov listing shows where to find the comparison group, the planned outcomes and the number of participants.
5. Who paid, and who benefits?
Look for the funding line, usually at the end of the release or in the paper's acknowledgments, and for any company the authors work for or hold shares in. Industry money does not make a result wrong. Much good research is company funded. But it is context, and a release from a small public company deserves an extra step: check what it has filed with the SEC around the same date. Our reference page on reading a biotech company's filings explains where trial results tend to appear in those documents, sometimes with more detail than the release.
A worked example (invented)
The release below is fictional. The institute, the scientists and the study do not exist. It is written to look like the real thing so the five questions have something to work on.
Headline: Halverton Institute scientists regrow hearing cells, offering hope to millions
In a landmark study, researchers at the Halverton Institute for Sensory Biology have regenerated the delicate hair cells of the inner ear, which are lost in age-related hearing loss and have long been thought impossible to replace. Using a novel two-drug combination, the team restored hearing function within weeks. "This could change how we think about deafness," said the study's senior author. Hearing loss affects millions of people worldwide. The findings appear in a leading journal of cell biology. The work was supported by Earwise Therapeutics, which has licensed the technology.
Now the questions.
Mice or people? The release never says. That silence is the first answer. In a real case, this is where you open the paper's abstract, which almost always names the species in its first lines. Suppose the abstract says the study used mice whose hair cells had been destroyed with a drug. That is a model of hearing loss, and a reasonable one, but it is not the slow, age-related damage the headline evokes.
How many? The release gives no number. Suppose the paper reports 14 treated mice and 12 controls, with hearing tests in 10 and 9 respectively by the end. A few animals dropped out, and the reasons should be stated in the methods.
What was measured? "Restored hearing function" turns out, in our invented paper, to mean a partial improvement in a brainstem electrical response to sound at certain frequencies, plus new cells that look like hair cells under the microscope. Those are meaningful lab results. They are not the same as a mouse hearing normally again, and the release's word "restored" claims more than either measure shows.
Compared with what? The paper compares treated mice with untreated mice that had the same damage. That is a proper control. Good. Note it, and give the study credit where it is due.
Who paid? A company that has licensed the technology funded the work. That is common and legitimate. It also means the company has a reason to want the result read generously, and that its next filing or investor presentation may describe the study in even brighter terms.
Put together, a fair one-line summary would read: "In mice with drug-induced hearing damage, a two-drug treatment produced new hair-like cells and a partial improvement in one measure of hearing, in a small company-funded study." Less exciting than the release. Much more useful.
Phrases worth a second look
| Phrase in the release | What it often turns out to mean |
|---|---|
| "Offers hope" | No patient has been treated |
| "Could lead to" | The authors are speculating about the future |
| "Landmark" or "breakthrough" | The press office's word, not the reviewers' |
| "Restores" or "reverses" | A partial change in one measure |
| "In a leading journal" | The journal is not named, so check it |
| "Novel" | New to this group, not necessarily to the field |
| "Millions affected" | The size of the disease, not of the study |
Where to check in under ten minutes
You do not need to read the whole paper. Find the title or the lead author's name in the release and search PubMed, which links to the abstract and often to a free full text. Read the abstract's first and last sentences for the species and the main result. Look at the methods for the number of subjects. For human trials, the registration number in the paper (it usually starts with NCT) opens the listing on ClinicalTrials.gov, where you can compare the outcomes the team planned to measure with the ones the release highlights. If the two do not match, that mismatch is often the most informative fact in the whole story.
This explainer is general information, not medical advice. Decisions about treatment, donation or trials belong with a clinician who knows your history.



