How to Read a Clinical Trial: Absolute vs Relative Risk, Explained

"New drug cuts risk in half." It sounds decisive, but the sentence hides the one number patients need: half of what? The answer separates a headline from a decision.

Relative versus absolute

Suppose 4 of 100 patients on standard care have an event, against 2 of 100 on a new treatment (hypothetical numbers). The relative risk reduction is 50%. The absolute risk reduction is 2 percentage points, so about 50 patients must be treated to prevent one event (1 ÷ 0.02). The Cochrane Handbook divides effect measures into ratio measures, such as the risk ratio and odds ratio, and difference measures, such as the risk difference. Both views are useful, and a good report gives both.

Odds are not risks

The odds ratio for the example is (2/98) ÷ (4/96) ≈ 0.49, close to the risk ratio of 0.50, because the event is rare. When events are common the two drift apart: with 40% against 20%, the risk ratio is still 0.50 but the odds ratio is 0.375. Reading an odds ratio as if it were a risk ratio then exaggerates the effect. The odds ratio calculator takes a 2×2 table and returns the ratio with its interval.

Is the difference real, or chance?

With 100 patients per group, the 4% versus 2% result cannot be told apart from chance: the p-value is about 0.41 and the 95% interval for the 2-point difference runs from about −2.7 to +6.7 points. Keep the same percentages but use 1,000 patients per group (40 events versus 20) and the p-value falls to about 0.009, with an interval of roughly 0.5 to 3.5 points. The effect has not changed; the precision has. The confidence interval calculator for two proportions shows both cases.

A confidence interval shows what a p-value hides: how big the effect is, and how uncertain.

Measured outcomes

Not every outcome is yes or no. Blood pressure, pain scores and recovery time are measurements. Two independent groups can be compared with the independent samples t-test calculator; when scores are skewed or ordinal, the Mann-Whitney test compares ranks instead. Report the difference between groups and its interval, not only the p-value.

Adjusting for other factors

When groups differ in age, severity or other characteristics, as often happens in observational studies, a logistic regression can estimate an odds ratio for the treatment that accounts for those variables. Randomization in a trial aims to balance such factors from the start, but analyses may still adjust for baseline differences.

Questions to ask of any trial

Was allocation random, and was the outcome assessed without knowing who got which treatment? How many participants dropped out? Is the main outcome something patients care about, or a laboratory marker? Are absolute risks reported next to relative ones? Is the interval narrow enough to rule out a trivial effect? This article is educational and is not medical advice; discuss treatment decisions with a qualified health professional.