Reading a health-study headline: relative risk, absolute risk and other traps

The short answer

A headline alone rarely tells you whether a study matters for you. Before it changes anything you do, find four things:

1. The real numbers. "Cuts risk by 50%" can mean 2 in 100 people dropping to 1 in 100. Look for what happened in each group, and over what time.
2. Who or what was studied. Results in mice, in cells or in people unlike you may not apply to you.
3. The kind of study. A randomized trial can test cause and effect. A study that finds two things "linked" usually cannot.
4. Where it stands. One study, a preprint not yet checked by outside experts, or a press release is an early signal, not a settled answer.

Do not start, stop or change a medicine because of a headline; take it to your clinician or pharmacist.

Trap 1: a percentage with no starting point

There are two common ways to describe the same result:

• Absolute risk is the chance that something happens to a person over a set time, such as "1 in 100 over five years."
• Relative risk compares two groups, such as "50% lower in people who took the pill."

Headlines often give only the relative number, which sounds bigger. A Food and Drug Administration (FDA) guide on risk communication notes that relative figures make a treatment seem more effective, and it advises giving absolute risks, not just relative ones.

Picture 100 people taking the pill for five years. On average, about 98 would not have had a heart attack anyway, 1 has one despite the pill, and 1 is spared. All 100 can get its side effects.

Now the harms. The same trial found serious stomach bleeding in 3 of every 1,000 people on placebo and 6 of every 1,000 on the pill. One story could call that "double the risk"; another, "only 3 extra people in 1,000." On one scale, for every 1,000 people taking the pill for five years: about 10 fewer heart attacks and about 3 more serious bleeds. The FDA guide recommends this: the same denominator ("out of 1,000") and the same time frame for every comparison. It also warns that some research articles give benefits in relative terms and harms in absolute terms, a mix that can make a treatment look better than it is.

Do the conversion yourself

1. Find the risk in the comparison group (no treatment, placebo or usual care) and the time period.
2. Find the risk in the treated or exposed group over the same time.
3. Subtract one from the other. That is the absolute difference.
4. For the number needed to treat, divide 1 by that difference written as a decimal, then round up. A difference of 1 in 100 is 0.01, and 1 ÷ 0.01 = 100.

If a story gives only a percentage change, look for both groups' numbers in the study's summary (the abstract) or results tables. Without a starting risk, you cannot judge how big the effect is.

The same headline can mean very different things

The Cochrane Handbook, a manual for reviewing medical evidence, notes that relative effects tend to be steadier across groups than absolute ones. The absolute benefit depends on how likely the outcome was to begin with.

Your own starting risk depends on things like age, health conditions and family history. A clinician can help you estimate it.

The trap also runs the other way. Example (fictional): "Popular snack linked to double the risk of a rare condition." If the risk goes from 1 in 10,000 to 2 in 10,000 over ten years, that is one extra case for every 10,000 people, an increase of 0.01 percentage points. And notice the word "linked."

Trap 2: "linked to" does not mean "causes"

In an observational study, researchers watch and measure what happens without assigning a treatment. In a randomized trial, chance decides who gets which treatment, so the groups start out alike and the comparison is fairer. The National Center for Complementary and Integrative Health (NCCIH), part of NIH, says well-planned clinical trials give the clearest information about whether a treatment works and is safe.

The weak spot of observational studies is that people who choose something often differ from people who don't. A 2015 post on NCCIH's research blog looked at a study that found a 29% lower risk of death over seven years among older Mexican American adults who used chamomile. The post pointed out that chamomile users may differ in many ways from non-users, and that statistical adjustments only account for differences researchers actually measured.

Words such as "linked to," "associated with" and "tied to" usually signal an observational study. These studies still matter: they spot patterns and raise questions worth testing, and for some questions, such as long-term exposures no one could ethically assign, they may be the main evidence. But one such study, on its own, cannot settle cause and effect.

Trap 3: mice, cells and test tubes

Many headlines describe preclinical research: studies in animals, or in cells in a dish ("in vitro"), done before testing in people. NCCIH cautions that such findings may mean little for your health until careful studies in people follow.

A 2024 review of 367 treatments first studied in animals found that about half went on to any study in people, about 40% reached a randomized trial, and about 5% were eventually approved, a median of 10 years after the animal work. If a study summary mentions mice, rats or cells, the finding is early.

Trap 4: too small, too short, or not about people like you

• Size. Small studies are more likely to produce results by chance, NCCIH notes.
• Length. Some benefits take time to appear, and some side effects take months or years.
• Fit. Were the people studied similar to you in ways that may matter, such as age, sex, race or health conditions?
• What was measured. Some studies track a stand-in measure, such as a lab value or tumor size on a scan, instead of outcomes people notice, such as living longer or feeling better. The National Cancer Institute (NCI) notes that these "surrogate endpoints" are not always true signs of how well a treatment works.
• Statistically significant is not the same as important. It means a difference is unlikely to be due to chance alone. A difference can be statistically significant and still too small to matter to patients.

Trap 5: preprints, press releases and the first study

Preprints. A preprint is a study posted online before peer review, the process in which outside experts check a paper before a journal publishes it. The preprint server medRxiv says preprints should not be relied on to guide medical care or health behavior, and should not be reported in the news as established information. This does not mean most preprints are wrong; a study of early-pandemic preprints that were later published in journals found most kept their main conclusions. It means no outside expert has checked the work yet.

One study. Peer review is a check, not a guarantee, and a single study rarely settles anything. Systematic reviews search for all the studies on a question and weigh their quality. Every Cochrane review includes a plain language summary that also says how confident the researchers are.

Press releases. A 2014 BMJ study of 462 press releases issued by 20 UK universities in 2011 found that about a third stated cause and effect more strongly than the research paper did, and about a third stretched animal findings to people. When a press release exaggerated, most of the related news stories did too.

Who paid. NCCIH suggests extra caution when the funder could profit from the result. Most journal articles list funding and financial ties.

One headline, start to finish

• The study: a randomized trial in 120 adults ages 18 to 40, lasting three months, paid for by the extract's maker and posted as a preprint.
• The numbers: 10 of 60 people on placebo (16.7%) and 5 of 60 on the extract (8.3%) had at least one cold. That is 50% fewer in relative terms, about 8 percentage points fewer in absolute terms, and a number needed to treat of 12 for three months.
• The size: only 15 people with colds in total. A gap this size in a study this small could easily be due to chance, and a standard statistical test finds it is not statistically significant.
• The fit: no one over 40 and no one with a long-term illness took part.

Verdict: interesting, not settled. Wait for larger, independent trials. If you want to try it anyway, ask a pharmacist or clinician first, especially if you take other medicines; NCCIH notes that supplements can change how medicines work.

A two-minute checklist

• Was the study done in people?
• Was it a randomized trial, or an observational "link"?
• What happened in each group, out of how many people, over what time?
• How many people took part, for how long, and were they like me?
• Did it measure something I would notice, or a lab number?
• Are harms given in the same kind of numbers as benefits?
• Is it a preprint, a journal article or a press release? Who paid for it?
• Is this the first study, or do reviews of many studies agree?

Where to find the fuller story

• The study itself. Search its title on PubMed, the National Library of Medicine's free index of medical research. It shows the abstract and often links to the full text.
• The whole body of evidence. Search Cochrane's plain language summaries by condition or treatment.
• Benefits and harms side by side. The Harding Center for Risk Literacy's "fact boxes" often show outcomes out of 1,000 people with and without a test or treatment. Many are written for Germany, so screening ages may differ from U.S. advice.
• Background skills. NCCIH's Know the Science pages explain study designs and health news.

For judging websites and social posts in general, see How to Evaluate Online Medical Information and How to Spot and Avoid Health Misinformation. If a chatbot summarized a study for you, see how to check an AI chatbot's health answer; for films and series, see how to check the claims in a health documentary.

Before you change anything

MedlinePlus, the National Library of Medicine's consumer health site, warns that stopping or changing a medicine on your own can be dangerous and can make your condition worse. Bring the headline or link to your clinician or pharmacist, and ask:

• "What is my own risk of this, over what time, with and without the treatment?"
• "Out of 100 people like me, how many would benefit, and how many would have side effects?"
• "Did this study include people like me?"
• "Does this change what you would recommend for me?"

Sources

1. Absolute risk — National Cancer Institute, NCI Dictionary of Cancer Terms, accessed October 26, 2026.
2. Relative risk — National Cancer Institute, NCI Dictionary of Cancer Terms, accessed October 26, 2026.
3. Observational study — National Cancer Institute, NCI Dictionary of Cancer Terms, accessed October 26, 2026.
4. Randomized clinical trial — National Cancer Institute, NCI Dictionary of Cancer Terms, accessed October 26, 2026.
5. Preclinical study — National Cancer Institute, NCI Dictionary of Cancer Terms, accessed October 26, 2026.
6. Surrogate endpoint — National Cancer Institute, NCI Dictionary of Cancer Terms, accessed October 26, 2026.
7. Understanding Health Risks — NIH News in Health, National Institutes of Health, October 2016.
8. Communicating Risks and Benefits: An Evidence-Based User's Guide — U.S. Food and Drug Administration, August 2011 (chapter 7, "Quantitative Information," and chapter 18, "News Coverage").
9. Chapter 15: Interpreting results and drawing conclusions — Cochrane Handbook for Systematic Reviews of Interventions, version 6.5, chapter last updated August 2023.
10. Know the Science: How To Make Sense of a Scientific Journal Article — National Center for Complementary and Integrative Health, accessed October 26, 2026.
11. Checklist for Understanding Health News Stories — National Center for Complementary and Integrative Health, accessed October 26, 2026.
12. Reduced Mortality Risks and Correlation vs. Causation — National Center for Complementary and Integrative Health research blog, July 22, 2015.
13. Analysis of animal-to-human translation shows that only 5% of animal-tested therapeutic interventions obtain regulatory approval for human applications — Ineichen BV and others, PLOS Biology, June 2024.
14. The association between exaggeration in health related science news and academic press releases: retrospective observational study — Sumner P and others, BMJ, December 2014.
15. About medRxiv — medRxiv, accessed October 26, 2026.
16. Tracking changes between preprint posting and journal publication during a pandemic — Brierley L and others, PLOS Biology, February 2022.
17. Our evidence: plain language summaries — Cochrane, accessed October 26, 2026.
18. Fact Boxes — Harding Center for Risk Literacy, accessed October 26, 2026.
19. About PubMed — National Center for Biotechnology Information, National Library of Medicine, accessed October 26, 2026.
20. How Medications and Supplements Can Interact — National Center for Complementary and Integrative Health, accessed October 26, 2026.
21. When you feel like changing your medicine — MedlinePlus Medical Encyclopedia, National Library of Medicine, reviewed September 4, 2024.
22. Poison Help — Health Resources and Services Administration, accessed October 22, 2026.
23. America's Poison Centers — America's Poison Centers (national association of U.S. poison centers that answer the Poison Help line), accessed October 26, 2026.