Calculator
Choose counts or rates. Counts enable confidence intervals automatically.
Example data table
| Scenario | Control events | Control total | Treatment events | Treatment total | ARR | NNT/NNH |
|---|---|---|---|---|---|---|
| Risk reduction example | 30 | 200 | 18 | 200 | 0.0600 | NNT 16.67 |
| Risk increase example | 10 | 150 | 18 | 150 | -0.0533 | NNH 18.75 |
| No difference example | 20 | 250 | 20 | 250 | 0.0000 | Not defined |
Examples are illustrative. Use your study numbers for decisions.
Formula used
- CER = control event rate = control events / control total
- EER = treatment event rate = treatment events / treatment total
- ARR = CER − EER
- NNT = 1 / ARR (when ARR > 0)
- NNH = 1 / |ARR| (when ARR < 0)
- RR = EER / CER (if CER > 0)
- Relative change = ARR / CER (if CER > 0)
- ARR CI (Wald) = ARR ± z × SE
- SE = √[ CER(1−CER)/nC + EER(1−EER)/nT ]
- NNT CI = reciprocal of ARR CI (stable only if ARR CI excludes 0)
Confidence bounds are approximate. For small samples or rare events, consider exact or bootstrap methods.
How to use this calculator
- Select Counts for study events and totals.
- Enter control and treatment outcomes carefully.
- Pick a confidence level, then submit.
- Read ARR first to interpret direction.
- Use NNT for benefit, NNH for harm.
- Download CSV or PDF for reporting.
FAQs
1) What does NNT mean?
NNT estimates how many people must receive a treatment to prevent one additional event compared with control, over the same follow-up period.
2) When do I use NNH instead?
If the treatment increases risk, ARR becomes negative. In that case, the calculator reports NNH, meaning how many treated people lead to one extra harmful event.
3) Why is NNT sometimes “not defined”?
When ARR equals zero, treatment and control risks match. Dividing by zero is undefined, so NNT/NNH cannot be computed meaningfully.
4) Should I round NNT?
For communication, many reports round up to the next whole patient. For technical work, keep decimals and show ARR and confidence bounds alongside NNT.
5) How are confidence intervals computed?
ARR uses a Wald interval based on the standard error of the risk difference. NNT bounds are the reciprocals of ARR bounds when the ARR interval does not cross zero.
6) Can I use rates without totals?
Yes, you can compute ARR and NNT/NNH using rates only. However, confidence bounds need sample sizes, so add totals to enable intervals.
7) What outcome should I enter?
Use a binary outcome, such as event or no event. Make sure it matches between groups and reflects the same follow-up duration and definition.