Jupe,

Good question. First off, I am no biostatistician. However, I have been following cardiovascular outcomes trials for the past few years. What I describe below is my understanding, though it may not be, and probably is not, 100% correct. 

Let's use 2 trial examples. EXAMINE and EMPA-REG OUTCOME, both published in the New England Journal of Medicine. You are correct that the trial outcome is technically time to first occurrence of 3-point MACE. These results are often expressed as Kaplan–Meier curves that plot cumulative incidence of events over time (see Figure 2 for EXAMINE and Figure 1 for EMPA-REG OUTCOME). Since not all patients in a large cardiovascular clinical trial will be enrolled at the same time, patients will be dosed for varying periods of time. For example, in BETonMACE first patient was randomized to placebo or apabetalone in November 2015 and enrollment ended in March 2018 (excluding the handful of Chinese patients that finished enrolling in June 2018). So when all is said and done, patients in BETonMACE will be dosed for between ~ 1 year and 3.5 years (dosing in all patients continues until trial reaches 250 events, so these are just estimates). But the point is enrollment occurred over a ~28 month period. 

So when reporting data in these CVOT trials, what most often is reported is the event rate at the median dosing period (dosing period for the patient in the middle of the range). Then, the event rates for the placebo and treatment groups at this median dosing time point are compared to calculate a relative risk reduction (RRR)and absolute risk reduction (ARR). One can also calculate the number needed to treat (NNT) from the ARR. NNT is the inverse of the ARR (1/ARR). 2 examples below for composite of 3-point MACE. 

Let's look at EXAMINE. "For this complete analysis, the primary end point occurred at similar rates in the alogliptin and placebo groups (in 11.3% and 11.8% of patients, respectively, after a median exposure of 18 months." What they did was look at the event rates at 18 months. To get an RRR, one would divide 11.3% by 11.8% and subtract this from 1, or perform (11.8%-11.3%)/11.8%. Either way, you get 0.04237, which you can convert to a percentage by multiplying by 100 to give 4.237% RRR. In the case of EXAMINE, the 4.237% RRR did not achieve statistical significance. The ARR is 11.8% - 11.3%, or 0.5% ARR. The NNT is 1 divided by 0.005 (aka 0.5%). NNT = 200. 200 folks would need to be treated for 18 months to prevent 1 patient from experiencing an event.  

Now, let's look at EMPA-REG OUTCOME. Same exercise. "The primary outcome occurred in a significantly lower percentage of patients in the empagliflozin group (490 of 4687 [10.5%]) than in the placebo group (282 of 2333 [12.1%])......The median duration of treatment was 2.6 years." To get the %RRR, 1 - (10.5/12.1), or alternative (12.1-10.5)/12.1, gives 0.1322, which converted to a %RRR is 13.22 RRR. They squeaked by with statistical significance with P=0.04 for superiority. The ARR is 12.1% - 10.5%, or 1.6% ARR. The NNT is 1 divided by 0.016 (aka 1.6%). NNT = 62.5. 62.5 folks would need to be treated for 2.6 years to prevent 1 patient from experiencing an event. 

BearDownAZ