The Number Needed to Treat, NNT for short, is an absolute measure of effect. It indicates how many persons would need to be treated over a clearly defined period to achieve a favorable event or prevent an unfavorable event in one additional person compared to a comparator treatment. The NNT can only be meaningfully interpreted together with the endpoint, comparison, observation period and underlying baseline risk.
Calculation from absolute risks
For a binary endpoint observed within a fixed period, the absolute risk difference between the control and treatment groups is first determined. If an unfavorable event is more frequent in the control arm than in the investigational arm, this difference corresponds to the absolute risk reduction. The NNT is the reciprocal of this difference. If the absolute risk reduction is 0.05, this results in an NNT of 20.
The calculation is only transparent if the direction of the event is stated. An NNT of 20 does not mean that exactly the twentieth treated person benefits. It describes an average comparison at the group level. For a favorable event, the formula can be expressed differently, but the principle remains the same: the NNT translates an absolute group difference into an more easily communicable number.
Baseline risk, period and precision
The NNT cannot be transferred without context. Given the same relative efficacy, it can be smaller in a population with a high baseline risk and larger in a population with a low baseline risk. Likewise, a longer observation period alters the cumulative risks and thus the NNT. Publications should therefore state the reference period, the endpoint and the event risks of both groups alongside the NNT.
Uncertainty must also be visible. Confidence intervals of the risk difference can include values ranging from benefit to no difference to harm. For the NNT derived from this, the presentation must then be explained particularly carefully because the reciprocal is not continuous at a difference of zero. A precise point estimate of the NNT without risks and confidence interval can make an effect appear overly clear.
Relation to other effect measures
The NNT is based on an absolute risk difference and complements relative effect measures such as risk ratio or hazard ratio. Relative measures can appear similar across populations, while clinical relevance varies significantly due to different baseline risks. For a balanced assessment, absolute risks, risk difference and, if applicable, a relative measure should be reported together.
For time-to-event data, an NNT without a time point is ambiguous. A hazard ratio therefore cannot be directly converted into an NNT. Estimated absolute risks at a clinically meaningful time point and a clear methodology are required. Accordingly, the Number Needed to Harm describes how many persons would have to be exposed to an intervention for an additional harm to occur; it is not a disproof of benefit, but part of the benefit-risk assessment.
The NNT is usually a derived metric and does not replace the pre-specified primary analysis. Study objective, estimand, endpoint and analysis model must be defined first. According to ICH E9(R1), estimation and sensitivity analyses should match the clinical question. If NNT values are presented additionally, the risk estimators and assumptions used for this must be documented comprehensibly.
Especially with missing data, a seemingly favorable absolute risk difference can be biased. The EMA recommends a pre-justified handling of missing data and sensitivity analyses for confirmatory trials. An NNT should therefore not be derived from an analysis whose risk of bias remains unclear. Its strength lies in the understandable classification of a robustly estimated absolute effect.
Relevance for clinical trials
The NNT supports the clinical interpretation of results when an event, a comparison and a period are clear. It can help in benefit-risk considerations, but must be read together with event risks, uncertainty and potential harms. For submission documents and study reports, it is crucial that it appears as a derived, transparently deduced metric and not as isolated proof of efficacy.
Full-service CROs such as Mediconomics support the selection of clinically understandable effect measures, the specification of supplementary analyses and statistical programming. They check the consistency of the used endpoints and periods, create tables with absolute and relative effects as well as confidence intervals and explain the derivation in the clinical study report.
Frequently Asked Questions (FAQ)
Is a smaller NNT always better?
For the same favorable endpoint and the same period, a smaller NNT indicates a greater absolute benefit. Comparisons across different endpoints or periods, however, are not permissible.
Can the NNT be negative?
In case of harm, the term Number Needed to Harm is usually used. The decisive factors are the direction of the risk difference and clear linguistic labeling.
Why does the NNT need a period?
Absolute risks change with observation duration. Without a time reference, it remains unclear which clinical benefit the number relates to.
Regulatory references
- ICH E9, Statistical Principles for Clinical Trials — requires the estimation and clinical evaluation of treatment effects.
- ICH E9(R1), Addendum on Estimands and Sensitivity Analysis in Clinical Trials — requires an analysis in line with the clinical question.
- ICH E6(R3), Guideline for Good Clinical Practice — regulates the pre-specified statistical methodology and documentation.
- Cochrane Handbook, Chapter 6 — methodically explains absolute and relative effect measures.