What is the NNT for this trial?
Enter the events in each arm and this returns absolute and relative risk reduction, the number needed to treat, and why the two look so different.
Why relative risk reduction flatters every treatment
Run the default numbers. Events fall from 6% to 3%. The relative risk reduction is 50% — the figure that reaches the press release. The absolute risk reduction is 3 percentage points, which means about 33 patients must be treated for one to avoid the event. Both numbers describe the same trial honestly. Only one of them tells a patient what to expect.
The gap widens as the event becomes rarer. A drop from 0.2% to 0.1% is also a 50% relative reduction, but the NNT is 1,000. This is why relative figures are the ones that get quoted, and why any appraisal checklist worth using asks for the absolute difference.
The number needed to treat is just the reciprocal of the absolute risk difference, rounded up, because you cannot treat a fraction of a person. It carries the same uncertainty as that difference, so it needs a confidence interval — and when the interval for the absolute difference crosses zero, the NNT interval becomes unbounded rather than merely wide. That is a real result, not a computational failure: the trial is compatible with the treatment helping, doing nothing, or harming.
Odds ratios appear because logistic regression produces them, not because they are easier to understand. When events are rare the odds ratio is close to the relative risk. When events are common it exaggerates, sometimes badly. If a paper reports only an odds ratio for a common outcome, be careful before you translate it into an NNT.
Frequently asked questions
Should I use the odds ratio or the relative risk?
Report the relative risk if you have the raw counts, which you do here. The odds ratio is worth having when you are comparing with a paper that reported one, or when the study was case-control, where relative risk cannot be calculated directly.
Why does my NNT have no upper limit?
Because the confidence interval for the absolute risk difference includes zero. As the difference approaches zero, the number you would need to treat approaches infinity. The honest reading is that this trial has not established an effect.
What is a good NNT?
There is no threshold. It depends entirely on the outcome and the cost of treating. An NNT of 100 is excellent for preventing a stroke with a cheap, safe tablet, and poor for preventing a mild rash with an expensive infusion.
Can I use this on a meta-analysis?
Only if you have the pooled event counts. If you have a pooled relative risk and a baseline risk instead, the absolute difference has to be derived from those, and the baseline risk you choose should reflect your own population, not the trials' average.
The book behind this tool
The Evidence-Based Practice Project Workbook
Take a clinical question from PICO to an implemented, evaluated change in practice.