What is the number needed to treat?

The number needed to treat (NNT) is the number of people who must be treated for one extra person to benefit over a defined time.^[1][2][3] The NNT is related to absolute risk and is calculated as 1 ÷ absolute risk reduction.^[1][2][3] The NNT is useful for comparing the efficacy of treatments. A lower NNT is better because it means fewer people need to receive a treatment for one extra person to benefit.^[1][2][3]

For example, in the SECURE trial, adults who had had a heart attack within 6 months of the study were randomized to receive a daily “polypill” (aspirin, a statin, and ramipril) or standard treatment.^[4] Over three years, the primary outcome — a composite of cardiovascular death, nonfatal heart attack, nonfatal ischemic stroke, or urgent revascularization — occurred in 9.5% (118 of 1,237) of people receiving the polypill and 12.7% (156 of 1,229) of people receiving standard treatment. That’s an absolute risk reduction of 12.7% minus 9.5%, which is 3.2% (or 0.032). Therefore, the number of people needed to receive the polypill for one extra person to benefit (the NNT) is about 31 (1 ÷ 0.032) over three years.

The relatively low NNT is a good sign because only 31 people need treatment over three years to expect one extra person to benefit. Should everyone receive the polypill after a heart attack? Possibly! Based on the NNT, the polypill may be a promising intervention, but the final decision must be informed by additional information, such as adherence rates, access to the drug, and bleeding risk, which the trial did not evaluate.

What are some limitations of the number needed to treat?

Although the NNT can be a useful decision-making metric, it has some important limitations, which are often related to how NNT is calculated and reported:

1. NNT is highly sensitive to baseline differences in risk.^[5][6] NNT is calculated using the absolute risk reduction: the absolute risk in the exposed (or intervention) group minus the absolute risk in the unexposed (or control) group. This means it can produce different values if the risk in the unexposed group (the “baseline risk” or “base rate”) varies between trials and across different populations. For example, the NNT for an intervention to prevent heart attacks in 30-year-old women would be very high because heart attacks are rare in this population. Meanwhile, the NNT for the same intervention may be very low for 70-year-old men because their baseline risk is much greater.

2. NNT depends on the time frame.^[1][7] NNT is calculated for a specifically defined time period, so its value can vary depending on the duration of treatment. Studies that report NNT should always state the timeframe over which it is relevant.

3. NNT is often reported without a confidence interval.^[2] Confidence intervals are a common way to report uncertainty. Unfortunately, NNTs are often reported without their confidence intervals, giving a false sense of precision. Be aware of this when reading papers reporting an NNT. If the NNT has no confidence interval, look back at the confidence interval for the absolute risk reduction to get some indication of the precision of the NNT value.