Threshold Indices

NNT (Number Needed to Treat)


1. Description

NNT (Number Needed to Treat)[1][2] is derived from the probabilities of a favourable effect (benefits) for the treatment and comparator. NNT is sometimes known as NNTB (Number Needed to Treat to Benefit).[3] The difference between the two benefit probabilities, , gives the increase in certainty, . NNT is then calculated as the reciprocal of this difference, ; and can be interpreted as the number of patients that need to be treated (on average) for one favourable event (NNT) to be observed as a result of treatment.

2. Evaluation


2.1 Principle
  • The calculation of NNT is transparent due to its apparent simplicity.
  • The choice of input values used in the calculations need to be made more transparent e.g. in explicitly stating and justifying the source of data used.
  • Input values must be rates/probabilities of the events of interest.
  • Negative NNT is interpreted in the same way as an NNH.
  • Negative NNT is interpreted in the same way as an NNH.

2.2 Features
  • NNT can only handle one event at a time; consequently benefit and risk are described separately.
  • The final metric is easily understandable since it represents counts or the number of people.

2.3 Visualisation

A potential visualisation to represent NNT is here


2.4 Assessability and accessibility
  • Although the required data are straightforward, the source of the rates/probabilities can be questionable due to the quality of data sources and may be biased.
  • NNT are undefined when there is no difference between treatment and comparator group.
  • The confidence intervals have also been criticised when the rates difference includes zero, leading to the confidence intervals of NNT to include infinity.
This method was tested in the Natalizumab and Rimonabant case studies.

3. References

[1] Holden WL, Juhaeri J, Dai W. Benefit-risk analysis: examples using quantitative methods. Pharmacoepidemiol Drug Saf 2003 Dec;12(8):693-7.
[2] Laupacis A, Sackett DL, Roberts RS. An assessment of clinically useful measures of the consequences of treatment. N Engl J Med 1988 Jun 30;318(26):1728-33.
[3] Grieve R, Hutton J, Green C. Selecting methods for the prediction of future events in cost-effectiveness models: a decision-framework and example from the cardiovascular field. Health Policy 2003 Jun;64(3):311-24.