Rational prescribing questions often hand you a trial result and ask whether the drug is worth giving. Relative risk alone cannot answer that; you need the absolute figures. The trial below is fictional and written for teaching only.
The data. 4,000 patients with atrial fibrillation are randomised equally to a new anticoagulant or to standard care and followed for two years.
| Outcome | New drug (n = 2,000) | Standard care (n = 2,000) |
|---|
| Stroke | 120 (6.0%) | 160 (8.0%) |
| Major bleed | 50 (2.5%) | 30 (1.5%) |
Benefit.
- Relative risk = 6.0 ÷ 8.0 = 0.75
- Relative risk reduction = 1 − 0.75 = 25%
- Absolute risk reduction = 8.0% − 6.0% = 2.0 percentage points
- Number needed to treat (NNT) = 1 ÷ 0.02 = 50
Harm.
- Absolute risk increase = 2.5% − 1.5% = 1.0 percentage point
- Number needed to harm (NNH) = 1 ÷ 0.01 = 100
Put together. Treat 1,000 such patients for two years and you prevent 20 strokes and cause 10 extra major bleeds, so 2 strokes are avoided for each bleed caused. Whether that trade is acceptable depends on how severe each event is and on what the patient values, which is why a good answer ends with a sentence of judgement and not with a number.
Why baseline risk changes everything. Suppose a younger patient has a stroke risk of only 2.0% over the same period. If the relative reduction of 25% carries over, the absolute reduction is 2.0 × 0.25 = 0.5 percentage points and the NNT becomes 1 ÷ 0.005 = 200. The bleeding risk does not shrink in step, so the NNH stays near 100. Now 1,000 treated patients gain 5 strokes prevented at the price of 10 bleeds. One drug and one trial give a reasonable choice for the first patient and a poor one for the second.
Odds ratio, for comparison. Odds of stroke on the drug = 120 ÷ 1,880 = 0.0638; on standard care = 160 ÷ 1,840 = 0.0870; odds ratio = 0.73. It sits close to the relative risk of 0.75 because the event is uncommon. With frequent events the two drift apart, and quoting an odds ratio as if it were a risk ratio overstates the effect.
Where students lose marks on NNT questions.
- Calculating NNT from the relative risk reduction (1 ÷ 0.25 = 4) instead of the absolute one.
- Leaving out the time frame. An NNT of 50 over two years is a different claim from 50 over six months.
- Rounding down. Convention rounds an NNT up to the next whole patient, so 33.3 is reported as 34.
- Ignoring the confidence interval. If the interval for the absolute reduction crosses zero, the NNT has no clear meaning.
- Presenting benefit as a relative figure and harm as an absolute one, which flatters the drug.
This worked example is educational practice in appraising evidence. It is not prescribing advice, and real decisions follow current guidelines and the full trial report.