What is the number needed to treat (NNT)?

The NNT is the number of patients you would need to treat with an intervention, instead of the control, to prevent one additional bad outcome (such as a death or heart attack). It is the reciprocal of the absolute risk reduction: NNT = 1 / ARR. A smaller NNT means a more effective treatment.

How do you calculate NNT from a clinical trial?

First find the control event rate (CER = control events ÷ control total) and the experimental event rate (EER = treatment events ÷ treatment total). The absolute risk reduction is ARR = CER − EER. The NNT is 1 ÷ ARR, rounded up to the next whole patient. For example, CER 0.20 and EER 0.10 give ARR 0.10 and NNT 10.

What is a good NNT value?

It depends on the outcome and the cost or risk of the treatment. A low NNT (single digits) for preventing death is excellent; a higher NNT can still be worthwhile for a cheap, safe drug preventing a serious event. Always read the NNT alongside its confidence interval, the baseline risk, and the harms. There is no single universal threshold.

What is the difference between NNT and NNH?

NNT (number needed to treat) counts patients treated to prevent one bad outcome when the treatment helps. NNH (number needed to harm) counts patients treated to cause one extra bad outcome when the treatment increases risk. This tool switches automatically to NNH when the experimental event rate (EER) is higher than the control rate (CER).

Can the NNT be negative or infinite?

When CER equals EER the ARR is zero and the NNT is infinite, no measurable effect. When the treatment causes harm the raw 1/ARR is negative; by convention that is reported as a positive NNH instead. This calculator never shows a raw negative NNT: it labels the result NNT for benefit and NNH for harm.

How is the 95% confidence interval calculated?

Using Altman's 1998 method: SE(ARR) = √[CER(1−CER)/n_control + EER(1−EER)/n_treatment], then the ARR interval is ARR ± z·SE (z = 1.96 for 95%). The NNT bounds are the reciprocals of the ARR bounds. If the ARR interval crosses zero the result is not statistically significant and is written as 'NNTB … to ∞ to NNTH …'.

What does 'spans infinity' mean in an NNT confidence interval?

When a trial's absolute risk reduction is not significant, its confidence interval includes zero, a treatment effect of zero corresponds to an infinite NNT. Following Altman, the interval is reported across infinity, from a number needed to treat to benefit (NNTB) through ∞ to a number needed to treat to harm (NNTH), signalling the evidence cannot rule out either benefit or harm.

Does this calculator do meta-analysis or survival NNT?

No. This version handles a single 2×2 trial using cumulative event rates. Pooled meta-analysis NNT (which needs heterogeneity handling) and time-to-event NNT derived from Kaplan–Meier curves are deliberately out of scope so the result stays deterministic and faithful to the cited sources.

Evidence-based medicine

Number Needed to Treat (NNT) Calculator

Turn a trial's control and treatment event rates into NNT, NNH, absolute and relative risk reduction, and a 95% confidence interval, with the full working shown. Built for medical, nursing and pharmacy students appraising a randomised trial. No signup, sources cited.

By Induwara Ashinsana· Co-Founder, Ryzera TechnologiesUpdated Jul 12, 2026

NNT & NNH calculatorAltman 95% CI

BMJ / CEBM method

Control group

The comparison / placebo arm.

Events

Total

Treatment group

The intervention arm.

Events

Total

“Events” are the bad outcome you are counting (death, MI, relapse…), not the number who improved.

Examples

Confidence

NNT = 10

95% CI: 7 to 18

On average, treat 10 patients to prevent 1 additional bad outcome.

Control rate (CER)

20%

Treatment rate (EER)

10%

Absolute risk reduction

10%

0.1 as a decimal

Relative risk (RR)

0.5

Relative risk reduction

50%

1 / ARR (unrounded)

Rounded up for the headline

Working shown

CER = 0.2 EER = 0.1
ARR = CER − EER = 0.2 − 0.1 = 0.1
NNT = 1 / 0.1 = 10 → 10 (rounded up)
RR = EER / CER = 0.1 / 0.2 = 0.5
Cross-check via RR: 1 / (CER × (1 − RR)) = 10 → 10 ✓
SE(ARR) = √[CER(1−CER)/n_c + EER(1−EER)/n_t] = 0.02
95% CI(ARR) = 0.1 ± 1.96·0.02 = (0.06, 0.14)
NNT bounds = 1 / each ARR bound → 7 to 18 (both finite → significant)

Formulas: ARR/RR/RRR and NNT = 1/ARR (Cook & Sackett, BMJ 1995); confidence interval and the “spans infinity” convention (Altman, BMJ 1998); definitions cross-checked against the Oxford CEBM. Full links in the Sources section below. This tool handles a single 2×2 trial, not pooled meta-analysis or time-to-event NNT.

How it works

The calculator follows the standard evidence-based-medicine formulas published by the Oxford Centre for Evidence-Based Medicine (CEBM) and the two founding BMJ papers by Cook & Sackett (1995) and Altman (1998). Throughout, the counted “event” is the bad outcome, death, myocardial infarction, relapse and the like, not the number of patients who improved.

Event rates. The control event rate is CER = control events ÷ control total; the experimental event rate is EER = treatment events ÷ treatment total. In Rates mode you enter these two percentages directly.
Absolute risk reduction. ARR = CER − EER. This is the extra fraction of patients spared the bad outcome by the treatment.
Number needed to treat. NNT = 1 / ARR, always rounded up to the next whole patient. You cannot treat a fraction of a person.
Harm case. If EER exceeds CER the treatment increases the bad outcome. The tool then reports the number needed to harm, NNH = 1 / (EER − CER), and relabels the headline.
Relative measures.Relative risk RR = EER / CER and relative risk reduction RRR = ARR / CER = 1 − RR. When CER is zero these are undefined and shown as “n/a”.
Confidence interval (Altman 1998). The standard error of the ARR is SE = √[CER(1−CER)/n_c + EER(1−EER)/n_t]. The ARR interval is ARR ± z·SE, with z = 1.645, 1.96 or 2.576 for 90%, 95% or 99%. Inverting each ARR bound gives the NNT bounds.

When the ARR confidence interval crosses zero the treatment effect is not statistically significant: a true ARR of zero maps to an infinite NNT. Following Altman, the interval is then written across infinity, NNTB a to ∞ to NNTH b , because the data cannot rule out either benefit or harm. As an independent check, the tool also recomputes the NNT through the relative-risk route, 1 / (CER × (1 − RR)), and confirms it matches the 1/ARR result to the rounded patient.

Worked examples

1, Statistically significant benefit

Control 100/500 → CER = 0.20; Treatment 50/500 → EER = 0.10
ARR = 0.20 − 0.10 = 0.10 → NNT = 1 / 0.10 = 10
RR = 0.10 / 0.20 = 0.50; RRR = 50%
SE = √(0.20·0.80/500 + 0.10·0.90/500) = 0.022361
95% CI(ARR) = 0.10 ± 1.96·0.022361 = (0.0562, 0.1438)
NNT bounds = 1/0.1438 to 1/0.0562 = 6.95 to 17.80 → 7 to 18
Result: NNT = 10 (95% CI 7 to 18), significant

2, Not significant (spans infinity)

Control 15/100 → CER = 0.15; Treatment 10/100 → EER = 0.10
ARR = 0.05 → NNT = 20 (point estimate)
SE = √(0.15·0.85/100 + 0.10·0.90/100) = 0.046637
95% CI(ARR) = 0.05 ± 1.96·0.046637 = (−0.0414, 0.1414), crosses 0
NNTB side 1/0.1414 = 7.07 → 8; NNTH side 1/0.0414 = 24.15 → 25
Result: NNT = 20 (95% CI: NNTB 8 to ∞ to NNTH 25), not significant

3, Net harm (NNH)

Control 20/400 → CER = 0.05; Treatment 32/400 → EER = 0.08
EER > CER → harm; ARI = 0.08 − 0.05 = 0.03
NNH = 1 / 0.03 = 33.33 → 34 (rounded up)
RR = 0.08 / 0.05 = 1.60; relative risk increase = 60%
Result: NNH = 34, treat 34 patients to cause 1 extra bad outcome

Frequently asked questions

Sources & references

The formulas on this page were last cross-checked against these sources on 2026-07-12. NNT is defined for binary (event / no-event) outcomes from a single randomised trial; continuous outcomes and pooled meta-analysis are out of scope.

Related tools

Health

Odds Ratio & Relative Risk

Compute the odds ratio, relative risk, ARR, RRR and number needed to treat or harm from a 2×2 study table, each with a 95% confidence interval (Woolf & Katz methods). Free, no signup, sources cited.

Open tool

Health

QTc Calculator

Correct a measured QT interval for heart rate using all four standard formulas, Bazett, Fridericia, Framingham and Hodges, side by side, each flagged normal, borderline or prolonged for the patient's sex. Enter heart rate or RR interval; runs in your browser, sources cited.

Open tool

Health

STOP-BANG Score

Score the validated 8-item STOP-BANG questionnaire and get your obstructive sleep apnea risk band. BMI, age, neck size and sex are scored automatically from your measurements. You answer only the four symptom questions. Runs entirely in your browser.

Open tool

Comments & feedback

Spotted a bug or want an improvement? Tell us, our team reviews every comment, and good ideas get built. Comments are public and anonymous.

Found a bug, edge case, or want to suggest an improvement?

Email me at [email protected] , most fixes ship within 24 hours.