Definition of NNT Posted on October 19, 2007 Helena Kraemer (bio) defines the concept of Number Needed to Treat.

Okay, I’m going to define [Number Needed to Treat]. I have a clinical record for each person, which describes their status at the end of the treatment and it includes things like cessation of symptoms or decrease of symptoms, remission or nonremission. It describes all possible side effects. It may describe quality of life measures and anything you can think of that has to do with the success of the treatment. And then when I, this is for the moment a Gedank experiment, I’m going to ask the patients or the experts to look at pairs of patients and simply compare them and say which of this pair is clinically preferable to you. Now unknown to the patient, one of them is a treatment and one of them is a control subject. They are blinded to that. And what I’m going to do is I’m going to compute the percentage of these paired judgments, in favor of the treatment. If it’s 50/50, the treatment is not so good. If it’s 100% in favor of treatment, wow. That doesn’t happen very often.

Okay, from that I can get number needed to treat because I can figure out how many times do I have to take a treated subject to get one more success than if I had taken this same number of control subjects and used the same criterion of success. Now that’s never happened. I’m just saying that’s a demonstration of how generally defined this measure is. That it could be used for any kind of an outcome measure. Moreover, if you tell you me your outcome measure is BMI, I can get an NNT. If somebody else comes and rescales the BMI instead of doing it in kilograms per meter, they do it in pounds per inch or whatever, the NNT comes out the same. It’s not necessarily true that the d comes out the same.

« Back to Article