Absolute Effects: Why the Same Relative Risk Can Mean 5 or 100 Fewer Events

The Method in One Sentence

Apply a credible relative effect to a credible comparator risk, then report the resulting risk difference—with its uncertainty and time horizon—before converting it to a number needed to treat.

One Relative Effect, Three Clinical Stories

Assume a risk ratio of 0.75 and, only for illustration, treat it as transportable across three settings. The relative reduction is 25% everywhere. The absolute benefit is not.

Illustrative arithmetic, not a clinical recommendation. Each row assumes the same follow-up period and a constant risk ratio.

The Arithmetic Reviewers Should Reconstruct

For a risk ratio, the treatment risk is the comparator risk multiplied by the risk ratio. The risk difference is treatment risk minus comparator risk. When treatment lowers risk, the absolute risk reduction is the positive magnitude of that difference.

For the middle-risk row: 10% × 0.75 = 7.5%. The absolute reduction is 2.5 percentage points, or 25 fewer events per 1,000. The reciprocal is 40, so the NNT is 40 over the same follow-up period.

Baseline Risk Is an Input, Not Background Decoration

The comparator risk may come from the trial control group, a pooled control risk, a representative registry, or a locally relevant cohort. Each choice targets a different population and may produce a different absolute effect. Reporting an NNT without naming that source hides a major assumption.

A trial control-group risk is internally coherent but may not represent routine practice. A local risk estimate may be more decision-relevant but can import differences in case mix, calendar time, eligibility, adherence, or outcome definition. The source should be named, justified, and aligned to the target population.

Time Is Part of Every Absolute Effect

“Ten fewer events per 1,000” is incomplete without “over six months,” “by two years,” or another defined horizon. Risks accumulate with follow-up, so the risk difference and NNT generally change when the time point changes.

Do not casually turn a trial's two-year NNT into an annual figure by division, or project it to five years by multiplication. That assumes a stable effect and event process that the data may not support. Report the horizon actually estimated.

An NNT Is Not a Treatment Property

NNT inherits every assumption used to create the risk difference: population, comparator, outcome, follow-up, effect model, and baseline risk. It also becomes unstable when the risk difference is near zero. A point estimate of NNT without an interval can look reassuringly precise while the compatible effects include substantial benefit, little effect, or harm.

Translate the confidence limits of the relative effect into absolute risks using the same comparator risk. If the interval crosses no effect, a single benefit-labeled NNT is especially misleading; present the underlying risks and interval instead of forcing uncertainty into one reciprocal.

Do Not Treat Odds Ratios as Risk Ratios

Odds and risks are different quantities. When outcomes are uncommon, an odds ratio and risk ratio may be numerically similar. As the comparator risk rises, directly multiplying risk by an odds ratio can materially misstate the treatment risk.

Use the effect measure's correct conversion formula, preserve the analysis scale, and show the assumed comparator risk. A polished NNT built from the wrong conversion is still wrong.

The 90-Second Absolute-Effect Audit

What Reviewers Should Demand

For each important binary benefit and harm, ask for the event risk in both groups, the relative effect, the absolute difference, confidence intervals, and follow-up period. If the result will be applied across populations with different baseline risks, ask for a small set of justified scenarios rather than one universal NNT.

Then separate arithmetic from transportability. A correct conversion does not prove that the trial's relative effect, baseline risk, adherence, or outcome ascertainment applies to the target setting. The transparent table is the beginning of appraisal, not the end.

Sources and Evidence Maturity

• Cochrane Handbook, Chapter 14: Completing “Summary of findings” tables — current official methods guidance; recommends presenting both relative and absolute effects for dichotomous outcomes and naming the baseline-risk source.
• Cochrane Handbook, Chapter 15: Interpreting results and drawing conclusions — current official methods guidance; defines risk differences and NNT and explains why absolute effects should be shown across credible comparator risks.
• FDA, Benefit-Risk Assessment for New Drug and Biological Products (October 2023) — final guidance; requests treatment effects on both absolute-difference and relative scales for binary outcomes, with uncertainty and clinical context.
• FDA, Communicating Risks and Benefits: An Evidence-Based User's Guide (2011) — official communication guide; demonstrates how relative reduction, absolute reduction, and NNT can describe the same evidence differently.

Evidence note: the three-scenario table is hypothetical. It teaches the calculation and does not assume that a real treatment's relative effect is constant across risk groups.

Where Aqrab Fits

Aqrab can help reviewers find the comparator event rate, effect measure, confidence interval, follow-up period, subgroup definitions, and benefit-harm outcomes across a paper. Use that structure to catch missing denominators and hidden baseline-risk assumptions—not to automate a clinical decision. Try Aqrab on a trial report, or explore plans for repeatable evidence-review workflows.