Biostatistics What are number needed to treat (NNT) and number needed to harm (NNH)? What are number needed to treat (NNT) and number needed to harm (NNH)?
Number Needed to Treat (NNT)70,72
NNT offers a method for clinicians to use to interpret clinical significance of an intervention
NNT conveys an estimate of a treatment’s clinical effect81
Typically, NNT is applied to a primary endpoint in a study, or one that has clinical relevance
NNT answers the question:
How many patients would you need to treat with Intervention X before 1 patient would experience a benefit?
Number Needed to Harm (NNH)72
Conveys an estimate of the potential for a treatment to cause harm
NNH answers the question:
How many patients would you need to treat with Intervention X before 1 patient would experience harm? (e.g., an adverse event [AE])
Number Needed to Treat (NNT)70,72
NNT offers a method for clinicians to use to interpret clinical significance of an intervention
NNT conveys an estimate of a treatment’s clinical effect81
Typically, NNT is applied to a primary endpoint in a study, or one that has clinical relevance
NNT answers the question:
How many patients would you need to treat with Intervention X before 1 patient would experience a benefit?
Number Needed to Harm (NNH)72
Conveys an estimate of the potential for a treatment to cause harm
NNH answers the question:
How many patients would you need to treat with Intervention X before 1 patient would experience harm? (e.g., an adverse event [AE])
How are NNT and NNH interpreted?72
How are NNT and NNH interpreted?72
NNT
A single-digit NNT is usually “good enough” for randomized, placebo-controlled clinical studies that evaluate response/non-response to treatment
- For example, the treatment is at least 10% better than placebo, resulting in an NNT of less than 10, and the lower the number the better.
Important to know rates of outcomes used to calculate NNT, not just the difference
- For example, both 20% vs. 10% and 80% vs. 70% will produce the same NNT, but the clinical treatment scenarios are very different
NHH
An acceptable NNH for drug vs. placebo depends on the outcome in question
- Outcome must be clinically significant to the patient
- Some patients may be more at risk for AEs
- NNHs of 10–100 may be acceptable for AEs that lead to discontinuation but that are not associated with serious immediate health risks
- NNHs of ≥1,000 are usually required for AEs that pose a significant health risk
Generally want NNH > NNT so that the benefits occur more frequently than the harm
Lower NNHs may be acceptable in certain situations. For example, NNH may be lower than NNT when comparing a beneficial outcome with a mild-moderate, temporary AE that does not lead to discontinuation (e.g., a mild dry mouth vs. response to an antidepressant medication)
A single-digit NNT is usually “good enough” for randomized, placebo-controlled clinical studies that evaluate response/non-response to treatment
- For example, the treatment is at least 10% better than placebo, resulting in an NNT of less than 10, and the lower the number the better.
Important to know rates of outcomes used to calculate NNT, not just the difference
- For example, both 20% vs. 10% and 80% vs. 70% will produce the same NNT, but the clinical treatment scenarios are very different
An acceptable NNH for drug vs. placebo depends on the outcome in question
- Outcome must be clinically significant to the patient
- Some patients may be more at risk for AEs
- NNHs of 10–100 may be acceptable for AEs that lead to discontinuation but that are not associated with serious immediate health risks
- NNHs of ≥1,000 are usually required for AEs that pose a significant health risk
Generally want NNH > NNT so that the benefits occur more frequently than the harm
Lower NNHs may be acceptable in certain situations. For example, NNH may be lower than NNT when comparing a beneficial outcome with a mild-moderate, temporary AE that does not lead to discontinuation (e.g., a mild dry mouth vs. response to an antidepressant medication)
Calculating NNT72
Calculating NNT72
NNT
f X = Frequency of outcome for Intervention X
f Y = Frequency of outcome for Intervention Y
Attributable Risk (AR) = f X – f Y
NNT = 1/AR (Rounded up so that it is a whole number)
Example
In a clinical trial, remission rates were 50% with Drug A and 20% with Drug B.
To answer the question
How many patients would you need to treat with Drug A instead of Drug B
before you would expect to have one additional patient in remission? requires calculation of NNT.
f A = 0.50
f B = 0.20
AR = 0.30
NNT = 1/0.30 = 3.33; rounded is 4
This NNT suggests that a therapeutic advantage may be expected in every fourth patient (frequent) treated with Drug A compared with Drug B
f X = Frequency of outcome for Intervention X
f Y = Frequency of outcome for Intervention Y
Attributable Risk (AR) = f X – f Y
NNT = 1/AR (Rounded up so that it is a whole number)
Example
In a clinical trial, remission rates were 50% with Drug A and 20% with Drug B.
To answer the question
How many patients would you need to treat with Drug A instead of Drug B
before you would expect to have one additional patient in remission? requires calculation of NNT.
f A = 0.50
f B = 0.20
AR = 0.30
NNT = 1/0.30 = 3.33; rounded is 4
This NNT suggests that a therapeutic advantage may be expected in every fourth patient (frequent) treated with Drug A compared with Drug B
NNT
Successful interventions for “very treatable” acute conditions (e.g., acute agitation) are generally expected to have NNTs vs. placebo in the range of 2–3; those for “somewhat treatable” conditions (e.g., osteoarthritis pain) are generally expected to be in the 4–6 range
Higher NNTs may be acceptable in certain situations such as difficult-to-treat conditions where other interventions have failed and few options remain (e.g., treatment-resistant major depressive disorder)
NNT
Successful interventions for “very treatable” acute conditions (e.g., acute agitation) are generally expected to have NNTs vs. placebo in the range of 2–3; those for “somewhat treatable” conditions (e.g., osteoarthritis pain) are generally expected to be in the 4–6 range
Higher NNTs may be acceptable in certain situations such as difficult-to-treat conditions where other interventions have failed and few options remain (e.g., treatment-resistant major depressive disorder)
Likelihood to be helped or harmed (LHH)72
Likelihood to be helped or harmed (LHH)72
LHH is a measure of the ratio of NNT and NNH
LHH = NNT/NNH
LHH much >1:
- NNT much larger than NNH
- Norm when comparing a desired outcome with a very severe AE
LHH a little >1
- NNT a bit larger than NNH
- Common for acceptable interventions when comparing a desired outcome with an AE that is usually mild or moderate but that may still lead to discontinuation
LHH ≤1
- NNH larger than NNT
- Generally acceptable only when comparing a desired outcome with an AE that is usually mild or moderate but that is usually temporary and does not lead to discontinuation, or there is a particularly urgent need for benefit that mitigates an otherwise prohibitive risk of harm
NNT/NNH ratios are important with ADHD interventions, as some options may have larger NNHs for a particular AE while remaining a viable treatment option for specific patient populations
LHH is a measure of the ratio of NNT and NNH
LHH = NNT/NNH
LHH much >1:
NNT much larger than NNH
Norm when comparing a desired outcome with a very severe AE
LHH a little >1
NNT a bit larger than NNH
Common for acceptable interventions when comparing a desired outcome with an AE that is usually mild or moderate but that may still lead to discontinuation
LHH ≤1
NNH larger than NNT
Generally acceptable only when comparing a desired outcome with an AE that is usually mild or moderate but that is usually temporary and does not lead to discontinuation, or there is a particularly urgent need for benefit that mitigates an otherwise prohibitive risk of harm
NNT/NNH ratios are important with ADHD interventions, as some options may have larger NNHs for a particular AE while remaining a viable treatment option for specific patient populations