So I'm a numbers guy and I go by what the numbers tell me. Most researchers and medical professionals are moving away from ARR and NNT numbers as a way to interpret the results of a study. The reason is simple; there are too many calculations for these numbers so compared to each other they mean nothing. For instance, the position of the NIH;
Absolute risk reductions and consequent NNT values associated with statin therapy among those with elevated high-sensitivity C-reactive protein and low low-density lipoprotein cholesterol are comparable if not superior to published NNT values for several widely accepted interventions for primary cardiovascular prevention, including the use of statin therapy among those with overt hyperlipidemia. Now let's consider NNT which many assume is that inverse of ARR or;
NNT = 1/ARR.
The problem is that is just one way to calculate NNT and is not commonly used by statisticians but is usually used to look for flaws in a trial or to minimize the results as it's an easy target. The real question is what NNT calculation was used? Was it:
Choen's d or Kreamer & Kupher?
http://www.plosone.org/article/info:doi/10.1371/journal.pone.0019070 Strauss Survival method; NNT 1/Arr-1St-Sc is it one of these with certain co=variables introduced;
NNT [risk difference> = 1 / RD ` NNT [relative risk of event> = 1 / (pc*RRR) ` NNT [relative risk of no event> = 1 / ((1-pc)*(RRne-1)) ` NNT [odds ratio> = (1-(pc*(1-OR)) / ((1-pc)*pc*(1-OR)) Or how about confidence factors, any added;
Confidence Interval Methods Confidence intervals for odds ratios were calculated using the methods described by Martin and Autsin (1991). Confidence intervals for relative risk are calculated using Koopman's likelihood-based approach advocated by Gart and Nam (Gart and Nam, 1988; Koopman, 1984). Confidence intervals for risk difference and number needed to treat are based on the iterative method of Miettinen and Nurminen (Newcombe, 1998a; Gart and Nam 1990; Miettinen and Nurminen, 1985) for constructing confidence intervals for differences between independent binomial proportions. Confidence intervals for individual risks are calculated by the Clopper-Pearson method for binomial proportions (Newcombe, 1998b). The notation of harm/benefit suggested by Altman (1988) is used here instead of quoting negative estimates and confidence limits. Again, the best number to look at is simple, relative risk reduction which is very basic and based on the data collected, people with low cholesterol are 42% less likely of dying of heart disease. It's not all about ARRs or NNTs, it really is this simple and not part of any hype by the drug companies. As a researcher who does this every day I can tell you this with total confidence.