Caution in Interpreting Number Needed to Treat and Number Needed to Harm in Clinical Trials

We read with great interest the recent paper by Campese (2025) [1] and we would like to highlight an important caution regarding the use of the Number Needed to Treat (NNT) and Number Needed to Harm (NNH) in clinical decision-making. While these metrics translate relative risk reductions or increases into absolute measures, they can be misleading when compared across studies with populations of different baseline risk.

For example, consider two trials evaluating the same blood-pressure-lowering drug, both achieving a 25% relative risk reduction over 2 years for cardiovascular events.

High-risk population:

Control arm: 20 patients out of 100 experience an event (risk = 20%);

Treatment arm: 15 patients out of 100 experience an event (risk = 15%);

Absolute risk reduction (ARR) = 20 − 15 = 5%;

Relative risk (RR) = 15/20 = 0.75, i.e., 25% risk reduction;

NNT = 100/(20 − 15) = 20.

Low-risk population:

Control arm: 4 patients out of 100 experience an event (risk = 4%);

Treatment arm: 3 patients out of 100 experience an event (risk = 3%).

Absolute risk reduction (ARR) = 4 − 3 = 1%;

Relative risk (RR) = 3/4 = 0.75, i.e., 25%, risk reduction;

NNT = 100/(4 − 3) = 100.

Although the relative risk reduction is identical (25% in both trials), the NNT differs substantially due to baseline risk differences (see absolute risks in the control arms of the two populations, 20% versus 4%). The same principle applies to NNH: the apparent harm of a treatment may appear larger or smaller depending on the baseline risk of adverse events in the studied population. Comparing NNT or NNH across studies without accounting for baseline risk may therefore be misleading.

We encourage reporting both absolute and relative risk measures alongside baseline risk to ensure proper interpretation of treatment benefits and harms.