Transcript:
Hello, and welcome to the NP Psych Navigator’s Clinical Insights video series. I’m Dr Leslie Citrome, clinical professor of psychiatry and behavioral sciences at New York Medical College in Valhalla, New York, and a consultant in clinical trial design and interpretation.
Today, we'll explore the concepts of “number needed to treat” and “number needed to harm” and how they may help healthcare providers understand clinical trial results in terms of clinical relevance.
When healthcare providers are weighing treatment options for patients in their practice, they often review the efficacy and safety of therapeutic interventions reported in clinical trials. To assess a therapeutic intervention, healthcare providers typically rely on “P-values,” representing statistical significance for the differences observed between interventions,1 such as those between a treatment arm and a placebo arm.
P-values are limited to understanding statistical significance, quantifying our degree of confidence that the results observed are real and not due to chance.1,2 However, P-values do not convey information about the size of the treatment effect. P-values do not describe the clinical relevance of the results observed.2
Having a method to quantify the clinical significance of trial results could assist healthcare providers in better interpreting and indirectly comparing efficacy and safety data from clinical trials when head-to-head trial data is not available.
We can express clinical significance by calculating the effect size, which represents the magnitude of a clinical response observed and thus is a way to measure the clinical relevance of trial results. Measures of effect size, such as standardized mean differences (like Cohen’s d), are used for this purpose.1,2 Standardized mean differences use standard deviation units, which are not intuitive.
Expressing effect size in terms of “patient units” using number needed to treat, NNT, and number needed to harm, NNH, can help healthcare providers interpret and communicate efficacy or safety data more easily.
NNT and NNH are used to assess the clinical significance of findings from medical research for efficacy and tolerability, respectively.2
Let’s first discuss number needed to treat. NNT is used to describe the efficacy of a therapeutic intervention. It answers the question “How many patients need to be treated with 1 treatment instead of an alternative before you can expect 1 additional patient with that positive outcome.”2
NNT is calculated by subtracting the rates of the outcome of interest for 2 different therapeutic interventions, such as medication versus placebo from a randomized clinical trial.1,4
To explain this in more detail, and as shown on the screen, you would first calculate the attributable risk, or AR, by subtracting the rate of outcome for the placebo from the rate of outcome for the treatment. Then, you would obtain the NNT by dividing 1 by the AR calculated in the first step.
If the NNT calculated is not a whole number, it should be rounded up to the next higher whole number as to not inflate the effect size between the 2 interventions. Whole numbers are used because we do not treat fractions of a patient.1
Let’s walk through an example NNT calculation for a hypothetical treatment, Drug A.
For Drug A, the rate of response in the treatment group was 40%, while the rate of response in the placebo group was 15%.
In order to calculate the NNT of Drug A compared to placebo, we would first calculate the attributable risk by subtracting the rate of outcome for the placebo group, 0.15, from the rate of outcome for the treatment group, 0.40, giving us the result of 0.25. Next, we would calculate the reciprocal of the AR, which would give us 4. So, the NNT of Drug A versus placebo would be 4.
The NNT of 4 here means that for every 4 patients who are treated with Drug A instead of placebo, you would expect to see 1 more patient who would benefit from Drug A.2,5
Now, let’s consider number needed to harm.
NNH is a metric used to describe adverse events associated with therapeutic intervention, and it answers the question “How many patients would you need to treat with a treatment versus an alternative before you expect to see 1 additional patient experiencing the adverse event in question.”2
NNH is calculated the same way as NNT, by first obtaining the attributable risk and calculating the reciprocal of the attributable risk.4
Similar to NNT, we round up the NNH to the next highest whole number.6
For an example of a NNH calculation, let’s return to our hypothetical treatment, Drug A.
The rate of an adverse event in the treatment group for Drug A was 18%, while the rate of an adverse event in the placebo group was 7%.
First, we calculate the attributable risk by subtracting the rate of an adverse event in the placebo group, 0.07, from the rate of an adverse event in the treatment group, 0.18. Then, we obtain the reciprocal of the AR, which gives us 9.09 for the NNH. For NNHs, we round up to the next higher whole number, 10, in this case.
The NNH of 10 here means that for every 10 patients who are treated with Drug A instead of placebo, you would expect 1 more patient who would experience the adverse event from Drug A.2
We’ve discussed what NNT and NNH express and how to calculate them, but what could they mean in terms of benefits and harm to the patients? How should they be interpreted? In other words, what would be considered a “good” NNT and NNH?
The lower the NNT values, the bigger the effect size difference between the 2 interventions and, presumably, the better the efficacy of the treatment when compared to the alternative.4
When considering if a treatment would be beneficial, NNT values under 10 indicate a meaningful difference between the 2 interventions.1,2
Conversely, the higher the NNH values, the less often you would expect to encounter 1 additional patient experiencing the adverse event of the treatment.4
NNHs greater than 10 are generally considered acceptable for psychotropic medications when compared against placebo on their rate of most commonly occurring adverse events.
It is important to note that there may be a wide range of adverse events associated with a treatment. The NNH values for specific adverse events will vary, and the NNH values for adverse events that are more concerning to a patient or have more severe consequences should be considered when evaluating the tolerability of a treatment.4
While NNT and NNH can each express the benefits and risks associated with a treatment, they can be combined to express what is called the “likelihood to be helped or harmed,” or LHH. LHH is the ratio of NNH to NNT.
LHH describes how much more likely is it for a treatment to be associated with a benefit than a harm. It is used to express the potential trade-offs between a specific benefit and a specific harm of a treatment.4
Again, LHH is obtained by calculating the ratio of NNH to NNT.4
For an example calculation of LHH, let’s again consider Drug A. The NNT for response is 4, while the NNH for its adverse effect is 10. To calculate the LHH, you would divide the NNH, 10, by the NNT, 4. This would give you an LLH of 2.5.
The LHH of 2.5 can be interpreted to mean that Drug A was 2.5 times more likely to be beneficial, leading to a response, than harmful, leading to an adverse effect, for the patient.5
As you have seen today, NNT and NNH can be easily calculated from the rates of outcomes of interest to express the relevance of clinical trial results.4 They can be especially useful for healthcare providers to effectively understand the benefits and risks of a treatment.7
There are some limitations of NNT and NNH that healthcare providers should be aware of, however. First, NNT and NNH can be calculated only for binary or dichotomous comparisons.1
Second, as NNT and NNH report absolute measures of the effect size, they do not provide information on the relative size of the treatment effect. Therefore, in published papers, the underlying rates used to calculate NNT and NNH should also be reported.2
Third, to indicate the precision of NNT and NNH estimates, they should be presented in published papers with 95% confidence intervals.1,7
Lastly, NNT and NNH are most informative when the subjects and the conditions tested in the clinical trials, such as patient demographics and medication dose and duration, are similar to the patient profile and the treatment parameters commonly encountered in the clinical setting.1,2
Number needed to treat and number needed to harm can be valuable metrics for assessing the clinical relevance of a clinical trial's statistically significant results.4 They can be used by healthcare providers to understand and communicate the potential benefits and risks of a particular treatment, with the goal of providing efficacious treatment while minimizing harm.
I hope that you found today’s presentation informative for your clinical practice and patient care. Thank you for joining me today!