A headline may say that a treatment “cuts risk by 25 percent.” Before deciding what that means, it helps to ask: risk of what, compared with what, in which people, and over how long? Those questions turn an impressive number into information that can support a conversation about care.
You do not need to evaluate every statistical technique to read more critically. Start with the study's question, methods, results, and limitations. Keep the original article nearby when possible, since a news summary may leave out details that affect the conclusion.
Identify the kind of evidence
In a randomized controlled trial, researchers assign participants by chance to treatment groups. A well-designed trial can help establish whether the treatment caused an observed difference. In an observational study, researchers examine what happens without randomly assigning the exposure or treatment. Differences between the groups may help explain an association.
Observational research is valuable, including for questions that cannot ethically or practically be randomized. Its findings need interpretation that accounts for possible alternative explanations. Check whether a headline's causal language matches the design. NIH's guide to clinical studies explains these distinctions.
Look closely at participants and the comparison
Who was included, and who was excluded? Consider age, illness severity, other medical conditions, and other treatments. Ask whether the care setting resembles the situation you are interested in. Results from one selected group may not transfer directly to another.
Identify the comparison: placebo, no treatment, usual care, or an active alternative. These answer different questions. Also look for blinding, sometimes called masking. Knowing which treatment a participant receives can influence how symptoms are reported or assessed. NCCIH describes why steps to reduce bias matter.
Put relative risk beside absolute risk
Consider this hypothetical example, created only to explain the arithmetic. Over one year, 12 of every 100 people in a comparison group experience an outcome, compared with 9 of every 100 in a treatment group.
- The absolute difference is 3 percentage points: 12 percent minus 9 percent.
- That is 3 fewer people experiencing the outcome per 100 people over one year.
- The relative risk reduction is 25 percent: the difference of 3 divided by the original risk of 12.
These describe the same result. The relative figure alone does not tell you the starting risk or the number of people affected. Keep the outcome, denominator, and time period consistent when making comparisons. This example is not a finding about any real medicine. NIH explains absolute and relative risk.
Ask what was actually measured
A study might measure symptoms, hospitalizations, daily functioning, or survival. It might instead measure a laboratory value or another surrogate endpoint intended to predict a clinical benefit. Some surrogates have strong supporting evidence in a specific context; others remain less certain.
An improvement in a measurement should be described accurately, without automatically expanding it into a claim about longer life or better functioning. Ask how closely the measured endpoint answers the question that matters to patients. FDA explains clinical and surrogate endpoints.
Read uncertainty and harms alongside benefits
A result is an estimate. A confidence interval helps show the range of effect sizes compatible with the data under the analysis assumptions. A wide interval signals that the magnitude is less precisely estimated. NCI discusses confidence intervals and risk measures.
Statistical significance does not establish that an effect is large enough to matter to a patient. Look at the size of the change and its practical importance. NCCIH's guide to study results distinguishes statistical from clinical importance.
Read adverse-event results, treatment discontinuations, and follow-up duration as carefully as the benefit. A short study or small sample may leave important safety questions unanswered. FDA's clinical-research overview explains why larger, longer studies can reveal less common or longer-term side effects.
Finish with the unanswered questions
Look for missing data, people who left the study, and limitations acknowledged by the authors. Read funding and conflict-of-interest disclosures. Ask whether other studies support the finding. The discussion section includes the authors' interpretation, and a single study rarely settles a clinical question. NCCIH explains discussion and conclusions.
Bring a concise question to a clinician or pharmacist: “Does this study apply to my situation, and how does it change the balance of benefits, risks, and alternatives?” This article is general education. A research headline alone is not a reason to start, stop, or change treatment.
Sources & further reading
- NIH: Understanding Clinical Studies
- NIH NCCIH: Minimizing Bias
- NIH News in Health: Understanding Health Risks
- FDA: Surrogate Endpoint Resources for Drug and Biologic Development
- NIH NCI: Cancer Screening Overview, Health Professional Version
- NIH NCCIH: Results
- NIH NCCIH: Discussion and Conclusion
- FDA: Step 3 Clinical Research