As a corollary to the Number Needed To Treat, the number needed to harm, in this programme, looks at the number of people you would expect to treat to get a bad result in one person – a result that is above and beyond a control condition. This might not sound as bizarre as it does when one considers that some treatments can vastly increase your chance of improvement whilst also increasing your chances of increased harm. For example, one can imagine a treatment for short-sightedness that vastly increases one’s chances for better vision over a one month period, yet which also increases one’s chances to become blind over that same 1 month period. The control group may have predominantly "stayed the same" without any operative intervention and while treatment might display a healthy NNT, it might also display an unhealthy NNH!
This programme computes the NNH by subtracting the control condition event rate for getting worse (deteriorating) from the treatment condition event rate for getting worse (deteriorating) and dividing 1 by the resultant number.
Negative numbers show that the control condition is more likely to harm than the treatment condition and so you don’t want to see any positive numbers under this statistic.
So, for the heart attack example (where we are defining only two types of outcome – life (good) and death (bad), the NNH would make use of the number of deaths in both condition and lead to the following result:
As with the NNT, it’s probably a good thing to have some idea about the 95% confidence interval of this figure. In other words, given the number of subjects who took part in the study, in what range does the true NNH lay (with 95% confidence).
This is computed using the same formula as for NNT, but instead looking at ‘Deterioration’ ARR rather than ‘Improvement’ ARR.