Quick answer
Relative risk compares event rates as a ratio, while absolute risk difference subtracts one group's event rate from the other's. A 50% relative reduction can describe a change from 40% to 20% or from 0.2% to 0.1%; the relative figure is the same, but the absolute differences are 20 percentage points and 0.1 percentage point. To interpret a peptide or GLP-1 trial claim, find both group event counts or percentages, the outcome definition, follow-up period, population, confidence interval, and missing-data rules. Number needed to treat or harm is derived from an absolute difference and must remain tied to that outcome and time horizon.
Key takeaways
- ✓Never interpret a relative percentage without the event rates in every comparison group.
- ✓Percentage points describe an absolute difference; percent change describes a relative difference.
- ✓Baseline risk, outcome definition, and follow-up time determine how an effect applies to the studied population.
- ✓NNT and NNH are reciprocals of an absolute risk difference and are not permanent properties of a drug.
- ✓Confidence intervals, harms, missing data, and clinical importance still matter after the arithmetic is correct.
01
Start with the two event rates
For a binary outcome, identify how many participants experienced the event in the intervention and comparison groups. Convert counts to risks only after checking the denominator and analysis population. If 10 of 100 people in one group and 20 of 100 in another have an event, the risks are 10% and 20%. The absolute difference is 10 percentage points, while the relative risk is 0.5 and the relative reduction is 50%.
Those descriptions are mathematically related but communicate different information. 'Half the risk' can sound dramatic while hiding whether the event was common or rare. 'Ten fewer events per 100 people over the study period' shows the observed absolute scale. A fair summary reports both and names the period and outcome.
02
Why baseline risk changes the practical meaning
The same relative effect can yield very different absolute differences when the comparison group's risk changes. A reduction from 40% to 20% is 20 percentage points; a reduction from 0.2% to 0.1% is 0.1 percentage point. Both are 50% relative reductions. Without the baseline rate, readers cannot see how much the event frequency changed in the studied population.
Baseline risk is not universal. Trial eligibility, disease severity, previous treatment, follow-up length, adherence, and background care can all affect it. Transferring an absolute result from a selected trial group to a lower- or higher-risk individual requires clinical judgment and additional evidence, not simple multiplication.
03
Percentage points are not percent change
When an event rate moves from 20% to 10%, it falls by 10 percentage points and by 50% relative to the original 20%. Calling that a '10% reduction' is ambiguous and can understate or misstate the result. Look for the original and final rates instead of trusting an unlabeled percentage in a headline, provider page, or social post.
Continuous outcomes need a different calculation. A mean change in weight, laboratory value, or symptom score is not a risk unless the study has defined a binary threshold such as achieving at least a specified change. Do not turn a mean percentage change into the percentage of people who benefited.
04
How NNT and NNH are derived
Number needed to treat is commonly calculated as the reciprocal of an absolute risk reduction expressed as a proportion. An absolute difference of 0.10 corresponds arithmetically to an NNT of 10. Number needed to harm applies the same idea to an increase in an unwanted outcome. Rounding, direction, and confidence intervals require careful handling.
An NNT is tied to one outcome, comparator, population, and period. It is not a score for overall drug value and should not be compared across studies with different endpoints or follow-up. A benefit NNT also does not cancel a harm NNH; readers need both sets of event rates and the seriousness of each outcome.
05
Read uncertainty and analysis choices with the effect
A point estimate is incomplete without uncertainty. Confidence intervals around relative and absolute effects show the range of estimates compatible with the data under the model. A wide interval may include meaningfully different conclusions. P-values do not supply the missing effect size or turn a small absolute difference into a clinically important one.
Check which participants were included, how missing outcomes were handled, whether the result was prespecified, and whether multiple endpoints or subgroups were tested. Relative and absolute calculations made from biased or selectively reported data remain biased. Correct arithmetic cannot repair a weak design.
06
A six-line translation for any trial headline
Write the claim in a fixed format: outcome, intervention group rate, comparison group rate, absolute difference, relative measure, and follow-up period. Then add the confidence interval and analysis population. If a report provides only a relative percentage, open the table, supplement, registry results, or regulatory review to find the underlying counts.
Finally, ask whether the exact studied product, route, dose, population, and comparator match the claim being made. Results for an FDA-approved product do not establish equivalence for a compounded formulation or a wellness product with a similar ingredient name. Treatment interpretation belongs with a licensed clinician; this framework is for checking how evidence is represented.
Red flags include a relative percentage with no denominator, combining several outcomes into one benefit claim, switching between risk and odds without explanation, omitting the comparator rate, or presenting a subgroup as though it were the whole trial. Also question an NNT that has no time horizon or confidence interval. Ask for the original table and calculate only from the prespecified result.
- →Defined outcome
- →Intervention event count and denominator
- →Comparator event count and denominator
- →Absolute difference in percentage points
- →Relative risk or relative change
- →Follow-up period
- →Confidence interval
- →Analysis population
Common questions
Frequently asked questions
What is the difference between absolute and relative risk?
Absolute risk is the event frequency in a group; an absolute difference subtracts group risks. Relative risk divides one group's risk by another's.
Is a 50% relative reduction always a large benefit?
No. It could represent a large or tiny absolute difference depending on the comparison-group risk, outcome, and time period.
What is the difference between percent and percentage points?
A move from 20% to 10% is a 10-percentage-point absolute decrease and a 50% relative decrease.
How is number needed to treat calculated?
It is commonly the reciprocal of the absolute risk reduction expressed as a proportion, but it must be interpreted with its outcome, comparator, population, time horizon, and uncertainty.
Can I calculate risk reduction from average weight loss?
Not directly. A continuous average and a binary event risk are different measures unless the study defines a threshold outcome and reports the proportion reaching it.
Does a small p-value mean a large absolute benefit?
No. Statistical significance does not state the effect size or clinical importance. Read the event rates, absolute difference, and confidence interval.
Primary sources
- Relative risk versus absolute risk: one cannot be interpreted without the otherPubMed, U.S. National Library of Medicine · checked September 2, 2026
- Clinically useful measures of effect in binary analyses of randomized trialsPubMed, U.S. National Library of Medicine · checked September 2, 2026
- Foundational Statistical Principles in Medical ResearchPubMed Central, U.S. National Library of Medicine · checked September 2, 2026
- How to read a published clinical trial: A practical guide for cliniciansPubMed Central, U.S. National Library of Medicine · checked September 2, 2026
Continue researching
Continue into provider research
Apply this guide’s verification questions to source-backed directory profiles and state coverage pages.