USMLE Step 1Biostatistics and epidemiologyEP 04 of 08

Number needed to treat: ARR, RRR and NNH from trial data

A headline can say a drug halves the risk while 100 patients must be treated for five years to prevent one event. Absolute and relative figures answer different questions.

  • 5:00 video
  • 3 min read
  • 6 sources
  • Published 16 Nov 2026
Chapters (12)
By the end of this lesson

Calculate absolute risk reduction, relative risk reduction, number needed to treat and number needed to harm from trial data.

The short version

  1. Absolute risk reduction (ARR) is the control event rate minus the experimental event rate.
  2. Relative risk reduction (RRR) is the ARR divided by the control event rate, which equals 1 − relative risk.
  3. Number needed to treat (NNT) is 1 / ARR, rounded up, and belongs to a stated time window.
  4. Number needed to harm (NNH) is 1 / absolute risk increase (ARI), where ARI is the experimental rate minus the control rate.
  5. One relative reduction can hide very different absolute benefits, because baseline risk sets the NNT.

The headline puzzle: halved risk, but which number?

A headline reads: new drug halves the risk of a heart attack. The trial behind it ran for five years. Two in every hundred untreated patients had a heart attack, and one in every hundred on the drug.

Halved, yes. But whether the drug is worth taking depends on which number you read. Number needed to treat (NNT) is the figure that puts the benefit in patients, and it needs two earlier quantities first.[2][4]

Risk, control event rate and experimental event rate

Risk is the fraction of a group that has the outcome. A trial has two risks. The control event rate (CER) is the risk in the placebo group. The experimental event rate (EER) is the risk in the treated group.[1]

Absolute risk reduction
ARR = CER − EER

can be shown as a decimal, in percentage points, or as events prevented per 1,000 treated; it is the risk difference with its sign ignored when the effect is beneficial

In the headline trial, 2% − 1% is one percentage point, or 0.01.[1][3]

Relative risk reduction: the halving in the headline

Relative risk reduction
RRR = ARR / CER = (CER − EER) / CER = 1 − RR

RR = relative risk; the RRR does not incorporate baseline risk

Headline trial
  1. ARR = 0.02 − 0.01 = 0.01
  2. RRR = 0.01 / 0.02 = 0.50, or 50%
  3. RR = 0.5, so 1 − RR = 0.5 as well

That 50% is the halving in the headline. It is a proportion of the control risk, so it says nothing about how large the control risk was.[1][2]

Number needed to treat: formula, rounding and time window

Number needed to treat
NNT = 1 / ARR

by convention a non-integer result is rounded up to the next whole number

In the headline trial, 1 / 0.01 = 100. Treat 100 patients for five years to prevent one heart attack.[1][3]

If the result is not a whole number, round up: 6.2 becomes 7.[1]

The trap: same relative reduction, ten times the NNT

Take a sicker population, where, over the same five years, 20% have a heart attack untreated and 10% on the same drug. The relative risk reduction is still 50%. But the ARR is ten percentage points, and the NNT is 10.

Same RRR, different baseline risk
PopulationCEREERRRRARRNNT
Headline trial2%1%50%1 point100
Sicker population20%10%50%10 points10

The relative figure ignores baseline risk; the absolute figure depends on it. Baseline risk sets the NNT.[2][4]

Harm: absolute risk increase and number needed to harm

Treatments can also harm. When the outcome is bad and treatment makes it more common, the difference is the absolute risk increase (ARI): experimental rate minus control rate.[2]

Number needed to harm
ARI = EER − CER; NNH = 1 / ARI

NNH is read together with NNT to weigh benefit against harm

Bleeding example
  1. Over the same five years, major bleeding is 1% on placebo and 1.5% on the drug
  2. ARI = 0.015 − 0.010 = 0.005, half a percentage point
  3. NNH = 1 / 0.005 = 200

So, over five years, treating 100 prevents one heart attack, and treating 200 causes one extra bleed. A low NNT and a high NNH is the favourable direction, once you also weigh how serious each outcome is.[2][5]

Practice question

Original · USMLE style
Your turnQB-MS-S1-BIOS-004-01

A randomised trial enrolls 1,000 adults at high risk of ischemic stroke. Participants are assigned to a new antiplatelet drug (500 patients) or placebo (500 patients). During 3 years of follow-up, a stroke occurs in 60 patients in the placebo group and in 40 patients in the drug group. All patients are analysed in the group to which they were assigned.

How many patients must be treated with the drug for 3 years to prevent one stroke?
Pick one option. Your answers stay on this device.

Common questions

How do you calculate the number needed to treat?

Divide 1 by the absolute risk reduction, where ARR is the control event rate minus the experimental event rate. Round a non-whole result up to the next whole number.

What is the difference between ARR and RRR?

ARR is the control event rate minus the experimental event rate. RRR is that difference divided by the control event rate, so it ignores baseline risk.

Why do we round the NNT up?

By convention a non-integer NNT is rounded up to the next whole number, so 6.2 becomes 7.

How do you calculate number needed to harm?

Find the absolute risk increase, the experimental rate minus the control rate, then divide 1 by it. For example, 1 / 0.005 = 200.

Why can a 50% relative risk reduction still be a small benefit?

Because baseline risk sets the absolute benefit. A 50% reduction from 2% to 1% gives an NNT of 100, while the same 50% from 20% to 10% gives an NNT of 10.

Transcript

Read the full narration, by chapter651 words

0:00The hook: a drug that "halves" heart-attack risk

A headline reads: new drug halves the risk of a heart attack. The trial behind it ran for 5 years. Two in every 100 untreated patients had a heart attack, and one in every 100 on the drug.

0:14Halved, yes: but which number do you read?

Halved, yes. But is the drug worth taking? That depends on which number you read. Today: calculate absolute risk reduction, relative risk reduction, number needed to treat and number needed to harm from trial data.

0:30Risk, control event rate (CER) and experimental event rate (EER)

Start with risk, from last lesson: the fraction of a group that has the outcome. A trial has two risks. The control event rate is the risk in the placebo group. The experimental event rate is the risk in the treated group.

0:46Absolute risk reduction (ARR)

Absolute risk reduction is the control event rate minus the experimental event rate. Our headline trial: 2% minus 1% is one percentage point, 0.01.

0:59Relative risk reduction (RRR)

Relative risk reduction is the absolute risk reduction divided by the control event rate. Zero point zero one over 0.02 is 50%. That is the halving in the headline. It also equals one minus the relative risk.

1:16Number needed to treat (NNT) and why we round up

Number needed to treat is one divided by the absolute risk reduction. One over 0.01 is 100. Treat 100 patients for 5 years to prevent one heart attack. If the result is not a whole number, round up: 6.2 becomes seven. And always state the time window, because a number needed to treat belongs to its follow-up period.

1:42The trap: same RRR, ten times the NNT

Now the trap. Take a sicker population, where 20% have a heart attack untreated, and 10% on the same drug. The relative risk reduction is still 50%. But the absolute risk reduction is 10 points, and the number needed to treat is 10. Same relative reduction, a tenfold different number needed to treat. The relative figure ignores baseline risk; the absolute figure depends on it.

2:11Harm: absolute risk increase (ARI) and number needed to harm (NNH)

Treatments can also harm. When the outcome is bad and treatment makes it more common, the difference is the absolute risk increase: experimental rate minus control rate. Say that over the same 5 years, major bleeding occurs in 1% on placebo and one and a half % on the drug. That is half a percentage point. Number needed to harm is one divided by the absolute risk increase: one over 0.005 is 200.

2:44Treat 100 to prevent one heart attack, 200 to cause one bleed

So, over 5 years, treating 100 prevents one heart attack, and treating 200 causes one extra bleed. A low number needed to treat and a high number needed to harm is the favourable direction, once you also weigh how serious each outcome

3:01Practice question

is. Your turn. A trial randomises 1000 adults at high risk of stroke to a new drug or placebo, 500 in each group. Over 3 years, 60 patients on placebo and 40 on the drug have a stroke. How many patients must be treated for three years to prevent one stroke? A, three. B, four. C, 13. D, 25. E, 33. Pause here and decide.

3:35Answer and explanation

A is out: that is one over the relative risk reduction, the wrong quantity to invert. B is out: four percentage points is the absolute risk reduction itself, not its reciprocal. C is out: that is one over the drug group's event rate, not one over the difference. E is out: 33% is the relative risk reduction, a percentage, not a number of patients. So the answer is D. Sixty over 500 is 12%, 40 over 500 is 8%, the absolute risk reduction is 4%, and one over 0.04 is 25.

4:18Recap and next lesson: study designs

Three things to keep. One: absolute risk reduction is control minus treated; relative risk reduction is that difference over the control rate. Two: number needed to treat is one over the absolute risk reduction, rounded up, tied to a time window; number needed to harm is one over the absolute risk increase. Three: one relative reduction can hide very different absolute benefits, because baseline risk sets the number needed to treat. Next, study designs: from case report to randomised controlled trial.

Sources

Numbers in the text, e.g. [1], point here
  1. Cochrane Handbook for Systematic Reviews of Interventions, chapter 15 (Interpreting results and drawing conclusions), section 15.4: risk difference, relative risk reduction and number needed to treat (NNTs are rounded up). training.cochrane.org/handbook/current/chapter-15
  2. Invest Educ Enferm. 2026: absolute risk reduction, relative risk reduction, NNT and NNH (PMC13078847). pmc.ncbi.nlm.nih.gov/articles/PMC13078847/
  3. Chiropr Osteopat. 2010: risk difference and number needed to treat (PMC2893201). pmc.ncbi.nlm.nih.gov/articles/PMC2893201/
  4. eClinicalMedicine. 2026: similar relative effects give different absolute benefits at different baseline risks; an NNT is specific to its time period (PMC13681460). pmc.ncbi.nlm.nih.gov/articles/PMC13681460/
  5. Rambam Maimonides Med J. 2015: number needed to treat and number needed to harm (PMC4524406). pmc.ncbi.nlm.nih.gov/articles/PMC4524406/
  6. Absolute risk reduction; Relative risk reduction; Number needed to treat (Wikipedia, secondary). en.wikipedia.org/wiki/Absolute_risk_reduction · en.wikipedia.org/wiki/Relative_risk_reduction · en.wikipedia.org/wiki/Number_needed_to_treat

All numbers on screen are teaching values from an invented trial: 2% vs 1% and 20% vs 10% over five years, bleeding 1% vs 1.5%; the practice question uses its own values. No real trial is described.

The practice question is original and its patient is synthetic. Every fact is checked against the sources above in independent accuracy reviews. Always follow your local guidelines and your own laboratory’s ranges.

Image credits. Illustrations: Servier Medical Art (smart.servier.com), CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/), adapted (cropped, recoloured, outlined): myocardial infarction (anteroseptal infarct), tablet, hourglass, ischemic stroke, elderly man. Health Icons (healthicons.org), CC0 1.0: blood drop (recoloured). The charts, dot grids, formulas, ledger, result slip, question card, recap card and all handwriting are drawn in code (original). No AI-generated teaching images.

Corrections. None so far. Spotted an error? Email info@medservatory.com with the lesson code MS-S1-BIOS-004.

For exam study only, not medical advice. Read the medical disclaimer.

  1. EP 01Sensitivity vs specificity: the 2×2 table, SnNout and SpPin5:34
  2. EP 02PPV, NPV and prevalence explained5:27
  3. EP 03Relative risk vs odds ratio: the 2x2 table explained5:15
  4. EP 04Number needed to treat: ARR, RRR and NNH from trial data5:00
  5. EP 05Study designs explained: case-control, cohort, cross-sectional and RCT5:22
  6. EP 06Types of bias: selection, recall, lead-time and confounding5:30
  7. EP 07Type I and type II errors: alpha, beta, power and the p-value5:16
  8. EP 08Confidence intervals and significance: is the null value inside?5:12