Risk Cut in Half? How to Read the Numbers Behind a Health Headline

“Cuts the risk in half” sounds decisive. It may describe an important benefit. It may also describe a change so small that most people would judge it differently if they saw the underlying numbers.

The phrase itself is not necessarily false. The problem is that it usually gives you a relative change without telling you the starting risk, the ending risk, the time involved, or the outcome being measured. Those missing pieces can turn an accurate statistic into a misleading impression.

You do not need a statistics degree to read these claims more clearly. You need the right numbers placed next to each other—and the patience to ask what they mean for a real person.

One result, two very different impressions

Consider a hypothetical five-year study. In the control group, 20 out of every 1,000 people had a heart attack. In the treatment group, 10 out of every 1,000 did.

• The relative risk reduction is 50 percent.
• The absolute risk reduction is 1 percentage point.
• Expressed as natural frequencies, the treatment was associated with 10 fewer heart attacks per 1,000 people over five years.

All three statements describe the same result. None is automatically the “correct” way to present it. But they do not create the same impression. Research on risk communication consistently finds that relative-risk formats can make benefits appear larger and make people more inclined to accept a treatment than absolute-risk formats do.

The absolute result does not make the treatment unimportant. Preventing 10 heart attacks per 1,000 people could matter greatly, depending on the treatment’s cost, burdens, harms, and the seriousness of the outcome. The point is that “50 percent lower risk” does not tell you enough to judge that tradeoff.

The four numbers that belong together

When a study, advertisement, press release, or news story reports a dramatic percentage, look for four pieces of information.

1. Baseline risk

Baseline risk is what happened without the intervention—or what is expected to happen in a comparable group. A 50 percent reduction means something very different when the starting risk is 40 percent than when it is 0.04 percent.

Baseline risk also changes from person to person. Age, diagnosis, disease severity, family history, prior events, and other factors can alter the chance of an outcome. That is why the same relative benefit can translate into a much larger absolute benefit for a high-risk person than for a low-risk person.

2. Absolute risk

Absolute risk tells you how many people actually experienced the outcome in each group. Natural frequencies—such as 20 out of 1,000 versus 10 out of 1,000—are often easier to understand than a relative percentage alone because they show the starting point and the change at the same time.

3. Relative change

Relative risk compares the two groups proportionally. It can be useful, especially when researchers combine results across populations with different baseline risks. It becomes misleading when it is presented by itself, because the reader cannot tell whether the underlying event was common or rare.

4. Time horizon

A risk is incomplete without a period of time. “A 2 percent risk” could mean over one month, five years, or a lifetime. Benefits and harms may also appear on different schedules. A medication might reduce an outcome over ten years while producing side effects within weeks. Both timelines belong in the decision.

When number needed to treat helps—and when it confuses

The number needed to treat, or NNT, estimates how many people would need to receive an intervention for one additional person to benefit compared with the control over a stated period.

In the hypothetical example above, the absolute reduction is 1 percentage point, so the NNT is 100 over five years. On average, 100 people would need the treatment for five years to prevent one additional heart attack compared with the control group.

NNT can be useful because it translates a percentage into a practical comparison. It can also confuse people when it appears without the underlying event rates. It is an average from a particular population, not a prediction that identifies which individual will benefit. It also changes with baseline risk and time. An NNT of 100 over five years cannot be compared casually with an NNT of 100 over six months.

Harms deserve the same treatment. Ask how many people experienced the relevant adverse effect in each group, how serious it was, whether it resolved, and over what period. A modest benefit may still be worthwhile when harms are rare and mild. A larger benefit may be a poor trade when the burdens are substantial. The values involved belong to the patient, not just the statistic.

Other traps hiding in a “positive” study

Relative versus absolute risk is only one layer. A result can be statistically positive and still be less useful than the headline suggests.

• A surrogate outcome: The intervention improved a laboratory marker, scan, or score, but the study did not show that people lived longer, functioned better, felt better, or avoided the complication that matters.
• A composite outcome: Several events were grouped together. The result may be driven by the least serious or most frequent component rather than the outcome emphasized in the headline.
• A subgroup finding: The overall trial was neutral, but one smaller subgroup appeared to benefit. Some subgroup effects are real; others emerge by chance when researchers make many comparisons.
• A wide confidence interval: The best estimate may look promising while the plausible range includes a very small benefit—or no benefit at all.
• Statistical significance without clinical importance: A large study can detect a very small average difference. “Statistically significant” does not tell you whether the change is noticeable or worth the cost, effort, and risk.
• Odds reported as risk: Odds ratios and risk ratios are not identical. When an outcome is common, an odds ratio can look more dramatic than the corresponding risk ratio.

None of these features automatically invalidates a study. They tell you what question the study actually answered and how much weight the result can carry.

A practical way to read the next headline

When you see “reduces risk by 30 percent,” “doubles the risk,” or “three times more likely,” pause before deciding whether the finding is large, frightening, or practice-changing. Ask:

• What happened in the comparison group?
• What happened in the intervention or exposed group?
• How many more or fewer people out of 100 or 1,000 experienced the outcome?
• Over what period?
• Was the outcome something people directly care about, or a surrogate?
• How precise was the estimate?
• What harms, costs, and burdens came with the benefit?
• Does my baseline risk resemble the people in the study?

If the article does not provide those numbers, follow the link to the abstract, study table, regulatory review, or evidence summary. If the information still is not available, the honest conclusion is not necessarily that the claim is wrong. It is that you have not been given enough information to know what the claim means.

This is not an argument for distrusting medicine, research, or expertise. It is an argument for communicating benefits and harms in a form that respects the reader’s ability to decide. Good risk communication does not push a person toward treatment or away from it. It makes the tradeoff visible.

For a broader framework, read Five Questions That Separate Trustworthy Health Information From Noise. The next time a percentage sounds alarming or miraculous, those questions—and the actual event counts—are a better starting point than the headline.

Sources & Further Reading

• Agency for Healthcare Research and Quality. Communicating Numbers to Your Patients.
• Cochrane Handbook, Chapter 15: Interpreting Results and Drawing Conclusions.
• U.S. Food and Drug Administration. Communicating Risks and Benefits: An Evidence-Based User’s Guide.
• Zipkin DA, et al. Evidence-Based Risk Communication: A Systematic Review. Annals of Internal Medicine. 2014.
• Trevena LJ, et al. How Difference Tasks Are Affected by Probability Format: A Making Numbers Meaningful Systematic Review. 2025.