The same trial result can sound dramatic or modest depending on how it is expressed. “Risk was cut in half” is a relative statement. “One fewer event occurred per 100 participants” is an absolute statement. Both can be mathematically correct, yet neither is interpretable without the outcome, comparator, denominator, time frame, and uncertainty.
Cochrane distinguishes ratio measures, such as a risk ratio, from difference measures, such as a risk difference.[1] Current CONSORT guidance asks trial reports to identify the effect measure, present results for each group, and give the estimated between-group effect with precision.[3][4] A research summary should therefore show the underlying event risks as well as any headline percentage.
Begin with risk in each group
For this hypothetical example, assume the event is undesirable and outcome status is known for every participant in the stated analysis population at the stated time. Under those conditions, observed risk is the number with the event divided by the stated participant denominator. If follow-up is censored or a competing event can prevent the outcome, a simple event/participant proportion may not be the appropriate time-specific risk; the report should name the survival or cumulative-incidence estimator used.[1][3][4]
The hypothetical group risks below are the foundation for several effect measures:
| Measure | Hypothetical result | What it expresses |
|---|---|---|
| Comparator risk | 20/1,000, or 2% | Observed risk under the comparator |
| Intervention risk | 10/1,000, or 1% | Observed risk under the studied intervention |
| Risk ratio | 1% ÷ 2% = 0.50 | Intervention risk relative to comparator risk |
| Relative risk reduction | 1 − 0.50 = 50% | Proportional reduction relative to comparator risk |
| Risk difference | 1% − 2% = −1 percentage point | Absolute difference between group risks |
The relative reduction is 50%, while the absolute reduction is 1 percentage point, or 10 fewer events per 1,000 participants over the specified period. Reporting only “50% lower” hides the scale on which the result occurred.
Relative effects answer a ratio question
A risk ratio compares the probability of an outcome between groups. A risk ratio of 0.50 means the observed risk in one group was half the observed risk in the reference group. It does not mean that 50% of participants benefited, that an individual’s risk was halved, or that the outcome became unlikely.
Direction also depends on how the outcome and comparison are defined. A ratio below one may favor the intervention when the event is undesirable, but the same numerical direction may be unfavorable when the event is a desired response. The labels and group ordering must remain visible.
Ratio measures are often useful across studies because they can be more stable than risk differences in some settings. That does not make them self-sufficient. Cochrane recommends interpreting relative effects alongside absolute effects generated from a relevant baseline risk.[2]
Absolute effects answer a difference question
The risk difference subtracts one group’s risk from the other. It can be presented as a decimal, percentage-point difference, or natural frequency such as “10 fewer per 1,000.” The sign depends on the subtraction order, so the groups and outcome direction must be stated.
Absolute effect is especially sensitive to baseline risk. Suppose a risk ratio of 0.50 applied in two different populations:
| Baseline risk | Corresponding risk at RR 0.50 | Absolute difference |
|---|---|---|
| 2% | 1% | 1 percentage point, or 10 fewer per 1,000 |
| 20% | 10% | 10 percentage points, or 100 fewer per 1,000 |
The relative effect is identical in the illustration, but the absolute difference is ten times larger in the higher-risk setting. This is why a relative percentage cannot be translated into an absolute claim without a defensible baseline risk.
Baseline risk is not a universal constant
In a randomized trial, the comparator-group event proportion is an observed baseline for that study’s population, comparator, follow-up period, and outcome definition. It may differ across settings because of eligibility criteria, disease severity, background care, calendar time, measurement, adherence, loss to follow-up, or chance.
A review may apply a pooled relative effect to one or more external baseline risks to show possible absolute effects. That calculation is conditional, not a direct observation from every represented population. The source and relevance of each baseline risk should be stated, and uncertainty in both the relative effect and baseline risk should be acknowledged.[2]
“Baseline” can also be ambiguous. It may mean a participant characteristic measured before assignment, the comparator event risk over follow-up, or a background population rate. A careful summary specifies which meaning is intended.
Risk, odds, rates, and hazards are different
Effect labels cannot be swapped because they all use the word “relative.”[1]
- Risk ratio compares event probabilities over a defined period.
- Odds ratio compares event odds, calculated as events divided by non-events.
- Rate ratio compares event rates that incorporate person-time and may allow repeated events.
- Hazard ratio compares hazards, instantaneous event occurrence within the relevant risk set, under a stated time-to-event model and assumptions.
For the same unadjusted two-group event table, an odds ratio can lie farther from 1 than the corresponding risk ratio, especially when event risks are not low. A hazard ratio is neither a risk ratio nor a time-specific event probability. With competing events, distinguish a cause-specific hazard from a subdistribution hazard and use an appropriate cumulative-incidence estimate for absolute risk; these quantities answer different questions.[1][3]
Number needed to treat is a transformation, not a new result
The reciprocal of the absolute value of a binary risk difference, rounded up to the next whole number when necessary, may be reported as NNTB for one additional beneficial outcome or NNTH for one additional harmful outcome. Cochrane recommends these direction-specific labels rather than NNH. The exact outcome, comparator, analysis population, and time frame must remain attached.[2][4]
For the hypothetical 1-percentage-point reduction in an undesirable event, NNTB is 100. This is an expected comparative quantity, not an identifiable “every hundredth participant.” If the risk-difference interval includes zero, the reciprocal interval is discontinuous: it extends through infinitely large NNTB and NNTH values rather than forming one ordinary interval across zero. Show the underlying group risks and interval.[2]
Uncertainty belongs on every scale
Observed group risks and derived effects are estimates. Confidence intervals should accompany the selected effect measure.[1][3] If an absolute effect is calculated from an external baseline risk and a relative-effect interval, the result also depends on how that baseline was chosen.
When an RR or OR interval is converted using an external baseline risk, state that the usual transformed interval propagates the relative-effect interval but does not incorporate uncertainty in the assumed comparator risk; discuss that source uncertainty separately.[2]
A point estimate can suggest a reduction while its interval remains compatible with little difference, no difference, or an increase. The appropriate description is uncertainty, not proof of benefit and not proof of no effect. FDA’s ICH E9 guidance connects estimation, uncertainty, prespecification, and bias in the interpretation of clinical trials.[5]
Precision does not address every limitation. Selective reporting, missing data, outcome changes, measurement error, and design bias can remain even when an interval is narrow.
Presentation can change perception
FDA’s December 2023 guidance is limited to covered DTC promotional communications; it recommends pairing relative-frequency presentations with corresponding absolute probabilities.[6] As an editorial synthesis of that communication principle with CONSORT and ICH E9, not an FDA-prescribed format for research summaries, a neutral evidence summary can show:[3][4][5]
- the outcome and assessment time;
- the number analyzed in each group;
- events and denominators in each group;
- one clearly labeled relative effect;
- one clearly labeled absolute effect; and
- confidence intervals and the prespecified analysis status.
The same format should be used for favorable and unfavorable outcomes. Absolute framing should not be reserved for small benefits while relative framing is used for large harms, or vice versa.
A practical reading sequence
When a study headline gives a percentage, reconstruct the full comparison:
- Identify whether the number is a risk, odds, rate, hazard, or change on a continuous scale.
- Find the event counts, denominators, and analyzed population for each group.[4]
- Confirm the comparator, outcome direction, and follow-up period.
- Translate ratios into absolute terms only with an explicit baseline risk.
- Read the confidence interval on the same scale as the estimate.
- Check whether the result was primary, secondary, exploratory, or post hoc.
- Review missing data, multiplicity, and other limitations before drawing a conclusion.
Important limitations
- Relative and absolute effects describe the same comparison on different scales; neither establishes causality when the design cannot support it.
- A baseline risk observed in one trial may not represent another population or setting.
- Odds ratios, risk ratios, rate ratios, and hazard ratios are not interchangeable.
- Number-needed-to-treat values depend on outcome definition, comparator, baseline risk, and time frame.
- A null or uncertain interval is not evidence that the true effect is exactly zero.
- Effect estimates do not establish the identity, composition, quality, or comparability of a separately sourced material.
- This page explains research reporting and does not provide medical advice or treatment guidance.
Frequently asked questions
Is a 50% relative reduction the same as a 50-percentage-point reduction?
No. A 50% relative reduction halves the comparator risk. The absolute percentage-point change depends on that comparator risk.
What is baseline risk?
In this context, it is the event risk under the comparator or reference condition over a defined period. Its source, population, and outcome definition matter.
Is absolute risk reduction always more useful than relative risk?
They answer complementary questions. Relative effects describe proportional contrast; absolute effects show the event difference at a stated baseline risk. Strong reporting presents both.
Can an odds ratio be read as a risk ratio?
Not generally. Odds and risks are calculated differently, and their ratios can diverge substantially when an outcome is common.
Does a hazard ratio of 0.70 mean 30% fewer participants had the event?
No. It summarizes a time-to-event contrast under a model. Event proportions at a particular time require separate estimates.
Why can the same risk ratio produce different absolute risk differences?
Because for a fixed risk ratio the risk difference still depends on comparator risk. In the illustrated RR 0.50 case, the absolute reduction equals comparator risk × (1 − 0.50). This is not a generic conversion rule for odds ratios or hazard ratios.
Does a favorable study effect apply to a catalog material with the same ingredient name?
No. A study concerns its defined intervention, formulation, sourcing, population, and procedures. It does not establish identity, quality, or equivalence for another material.
Primary and official sources
- Cochrane Handbook, Chapter 6: Choosing Effect Measures
- Cochrane Handbook, Chapter 15: Interpreting Results
- CONSORT 2025 Explanation and Elaboration: Outcome Comparison
- CONSORT 2025 Explanation and Elaboration: Numbers Analysed
- FDA: ICH E9 Statistical Principles for Clinical Trials
- FDA: Presenting Quantitative Efficacy and Risk Information
Review and corrections
Sources checked: September 1, 2026. Questions or corrections can be sent through the Serum Research contact page.
Editorial responsibility: Serum Research prepared and maintains this educational page. No independent scientific, statistical, regulatory, or clinical review is claimed. See how Serum Research reviews evidence or return to the Research Peptide and Compound Library.