Number Needed to Harm (NNH)

The number needed to harm (NNH) is the number of patients who need to be exposed to a risk factor, treatment, or intervention over a specific period before one additional patient experiences an adverse event that would not have occurred without the exposure. It is the inverse of the absolute risk increase (ARI) and provides an intuitive measure of the risk associated with an intervention or exposure.

NNH translates abstract risk differences into a clinically meaningful number that patients, clinicians, and policymakers can understand. Saying that "one additional patient in every 50 treated will experience a serious bleeding event" (NNH = 50) is more intuitive than stating that "the absolute risk increase is 2%." NNH allows direct comparison of the safety profiles of different interventions and facilitates shared decision making by helping patients weigh the benefits of treatment (captured by the number needed to treat, NNT) against its risks. Guideline developers, HTA bodies and regulators can use NNH alongside NNT when assessing the benefit risk profile of new treatments.

NNH is calculated as 1 divided by the absolute risk increase (ARI). The ARI is the difference in event rates between the exposed (or treated) group and the unexposed (or control) group: ARI = risk in exposed group minus risk in unexposed group. For example, if 8% of patients taking a new drug experience gastrointestinal bleeding compared to 5% of patients on placebo, the ARI is 3% (or 0.03), and the NNH is 1/0.03 = 33.3, rounded up to 34 (the convention for NNT; some authors round NNH down instead, so that harm is not understated, so state the rule used). This means that for every 34 patients treated with the new drug, one additional patient will experience gastrointestinal bleeding.

The confidence interval for the NNH is derived from the confidence interval for the ARI and may need special handling when the confidence interval for the ARI spans zero (which occurs when the risk difference is not statistically significant). In such cases, the NNH confidence interval includes infinity and extends to negative values (which correspond to NNT values), making interpretation complex. This situation is often handled by reporting the NNH as not statistically significant; Altman (1998) recommended instead giving the whole interval, labeling its two parts NNTH (number needed to treat to harm) and NNTB (number needed to treat to benefit), joined through infinity.

NNH depends on the baseline risk, the duration of follow up, and the specific adverse event being measured. A single treatment may have different NNH values for different adverse events (for example, NNH = 20 for nausea and NNH = 200 for serious hepatotoxicity). Clinicians compare the NNH for harms against the NNT for benefits to make treatment decisions: an intervention is most clearly justified when the NNT is much smaller than the NNH, meaning that many patients benefit for each one harmed.

Report NNH whenever a study demonstrates a statistically significant increase in adverse events associated with a treatment or exposure. It should be presented alongside the NNT to give a complete picture of the benefit risk tradeoff. NNH is most useful for communicating risk to patients in clinical settings, for comparing the safety of alternative treatments, and for formulary and coverage decisions. It is less useful when the adverse event rate is very low (producing a very large NNH that may understate clinical concern) or when the outcome is continuous (NNH requires a dichotomous outcome).

A common error is comparing NNH values from studies with different follow up periods or baseline risks without adjusting for these differences. An NNH of 100 over 5 years represents a different level of risk than an NNH of 100 over 1 year. Another mistake is reporting NNH without specifying the adverse event, the comparison group, and the time frame, all of which are needed for meaningful interpretation. Some researchers calculate NNH from relative risk measures without considering the baseline risk, which can produce misleading results. Finally, interpreting a large NNH as "safe" without considering the severity of the adverse event is inappropriate; an NNH of 500 for a fatal event may be more concerning than an NNH of 10 for mild nausea.

A meta analysis of anticoagulation therapy for atrial fibrillation finds that the annual rate of major bleeding is 3.5% with the new anticoagulant and 2.0% with aspirin alone. The ARI is 1.5%, and the NNH for major bleeding is 1/0.015 = 67, meaning that for every 67 patients treated with the new anticoagulant instead of aspirin for one year, one additional patient will experience a major bleeding event. The NNT for stroke prevention with the same anticoagulant is 25 per year, meaning that one stroke is prevented for every 25 patients treated. Setting the two side by side, about 2.7 strokes are prevented for each additional major bleed (NNH ÷ NNT = 67 ÷ 25); whether that favors treatment also depends on how serious a stroke is compared with a major bleed, and on individual risk factors.