NNT Calculator

Written by Great Medico

Updated September 24, 2026 •2 min read

Share Print Save

Epidemiology turns counts and probabilities into comparable measures. NNT Calculator performs the exact relationship supported here, while separating educational scenario math from an individualized medical forecast.

What this calculator does

NNT Calculator uses Control event rate and Treatment event rate to produce the number needed to treat used by this calculator. It only uses the fields shown in this calculator; it does not silently add tests, diagnoses, symptoms, or clinical variables that are absent from the interface. Exact NNT/NNH arithmetic from control and treatment event rates.

How to use it

Enter or select Control event rate and Treatment event rate. Match the numerator, denominator, percentages, and time window to the scenario you are trying to describe. If an input is unknown, it is better to leave the calculation unresolved than to invent a value.

How the calculation works

Absolute risk difference equals control event rate minus treatment event rate. The calculator returns 1 ÷ absolute risk difference (rates converted from percentages); a beneficial difference is labeled number needed to treat, while harm is labeled number needed to harm.

Example

If the control event rate is 20% and the treatment event rate is 10%, the absolute risk reduction is 10 percentage points (0.10). Dividing 1 by 0.10 gives an NNT of 10 people for the specified outcome and time horizon.

How to interpret the result

NNT/NNH is easiest to understand with the time horizon and outcome definition from the source study. A lower NNT for a beneficial outcome means a larger absolute treatment effect, but it says nothing by itself about adverse effects, baseline risk in another population, or durability.

Limitations and notes

Rates and scenario models depend on the population, time window, case definition, denominator quality, behavior, immunity, and assumptions. They should not be converted into an individual diagnosis or forecast unless the underlying model was validated for that purpose.