Number needed to treat

The number needed to treat is calculated as the reciprocal of the absolute risk reduction (ARR), where ARR is the difference in the rate of adverse outcomes between the control group (CER) and the treatment group (EER). For example, if a drug reduces the risk of stroke from 5% to 3%, the ARR is 2%, and the NNT is 50, meaning 50 patients must be treated to prevent one stroke. The NNT is a dimensionless integer, and it is typically rounded up to the next whole number. It is derived from randomized controlled trials and is most meaningful when the baseline risk is similar to that of the population to which the result is applied.

The NNT is a practical tool for clinicians and patients to weigh the benefits and harms of a treatment. A low NNT (e.g., 2 or 3) indicates a highly effective intervention, such as antibiotics for severe infections, while a high NNT (e.g., 100 or more) suggests a marginal benefit, common in preventive therapies like statins for low-risk patients. However, the NNT has limitations: it does not account for the severity of the outcome, the time horizon of the study, or the patient's baseline risk. It also assumes a constant treatment effect across populations, which may not hold. The number needed to harm (NNH) is a related measure for adverse effects, and the ratio of NNH to NNT (the likelihood of being helped or harmed, LHH) is sometimes used to balance risks and benefits.²

The NNT is used across medicine, from cardiology to psychiatry, to guide treatment decisions and health policy. For instance, in cardiovascular disease, the NNT for statins in secondary prevention is around 20 to 30, while in primary prevention it can exceed 100. In psychiatry, the NNT for antidepressants in moderate depression is about 7, but for mild depression it may be much higher, reflecting the importance of baseline risk. Variations include the NNT for a specific time period (e.g., NNT for one year) and the NNT for a specific outcome (e.g., NNT to prevent one death). The NNT can also be derived from survival data using the number needed to treat over time, which accounts for varying follow-up durations.

The NNT has a lesser-known history: it was first proposed in 1988 by Laupacis, Sackett, and Roberts, but a similar concept, the "number needed to be treated," appeared earlier in the 1970s in the context of hypertension trials. The NNT is also used in veterinary medicine and public health, such as in vaccination programs, where the NNT to prevent one case of influenza can be calculated. A notable edge case is when the treatment effect is not statistically significant; the NNT may be infinite or negative, which is often misinterpreted. Some researchers advocate for reporting the NNT with confidence intervals, as the point estimate alone can be misleading. Additionally, the NNT is sensitive to the choice of outcome; for example, using a composite outcome can lower the NNT compared to individual components.³

Absolute risk reduction

The difference in risk of an outcome between control and treatment groups.

Number needed to harm

The number of patients who need to be treated to cause one additional adverse outcome.

Likelihood of being helped or harmed

The ratio of NNH to NNT, used to balance benefits and risks.