Key Facts

NNT translates the results of clinical trials into a clinically meaningful metric. It answers the question: 'How many patients do I need to treat to prevent one adverse outcome?' It is more intuitive than relative measures and essential for shared decision-making.

Calculation

• ARR = CER − EER (absolute difference in event rates)
• NNT = 1 / ARR
• NNH = 1 / ARI (where ARI = absolute risk increase for harm)

Example

• Control event rate = 20%, Treatment event rate = 15%
• ARR = 0.20 − 0.15 = 0.05 (5%)
• NNT = 1/0.05 = 20
• Interpretation: 20 patients need to be treated for 1 to benefit

RRR vs ARR

• RRR = ARR / CER = 0.05 / 0.20 = 25%
• RRR sounds more impressive but does not convey absolute benefit
• A 50% RRR could mean CER 2% → EER 1% (ARR 1%, NNT 100) or CER 40% → EER 20% (ARR 20%, NNT 5)
• Always report both relative and absolute measures

NNT for Common Interventions

• Statins for primary CVD prevention (5 years): NNT approximately 100-200
• Statins for secondary CVD prevention (5 years): NNT approximately 20-30
• Aspirin for secondary prevention: NNT approximately 67 (2 years)
• Thrombolysis for acute STEMI: NNT approximately 30-50
• Antibiotics for sore throat (symptom relief at day 3): NNT approximately 6

Using NNT in Clinical Practice

• Compare NNT with NNH to assess treatment benefit vs harm
• Net clinical benefit: if NNT < NNH for a serious outcome, treatment is likely beneficial
• Communicate to patients: 'If I treat 20 people like you, 1 will benefit; the other 19 would not have had the event anyway'
• NNT should be considered alongside patient preferences, comorbidities, and treatment burden

Limitations of NNT

• Time-dependent: NNT changes with duration of treatment
• Baseline risk-dependent: NNT differs for high-risk vs low-risk patients
• Assumes constant treatment effect across populations
• Cannot be directly added or subtracted between studies
• Confidence intervals for NNT can be counterintuitive (can pass through infinity)

Calculating NNT from Trial Data

1. Extract event rates from published data (2×2 table or Kaplan-Meier curves)
2. Calculate ARR = CER − EER
3. NNT = 1 / ARR (round up to next whole number)
4. Calculate 95% CI for NNT from CI for ARR

NNT from Odds Ratios

• When only OR is available (e.g. from meta-analysis), NNT can be estimated using: NNT = 1 / [CER × (1 − OR) / (1 − CER × (1 − OR))]
• Approximation when disease is rare: NNT ≈ 1 / [CER × (1 − OR)]

Confidence Intervals

• 95% CI for NNT calculated from 95% CI for ARR
• If CI for ARR includes zero (non-significant result), CI for NNT passes through infinity
• Expressed as: NNT = X (95% CI: Y to infinity to −Z)

Clinical Decision-Making with NNT

• Consider NNT in context of disease severity and patient baseline risk
• Higher baseline risk → lower NNT (treatment more beneficial)
• Balance NNT against NNH for specific side effects
• Use NNT for shared decision-making conversations with patients
• Integrate NNT with QRISK, FRAX, or other risk calculators for personalised treatment decisions

NNT in Guidelines

• NICE guidelines increasingly report NNT/NNH alongside recommendations
• Statin guidelines: QRISK ≥10% → offer statin (NNT varies with baseline risk)
• Anticoagulation in AF: CHA₂DS₂-VASc guided; NNT for stroke prevention varies with score
• Breast cancer screening: NNT to detect one cancer ≈ 500 per screening round; NNT to prevent one death ≈ 1,000-2,000 over 10 years

• NNT provides the most clinically useful measure of treatment effect for patient communication
• Treatments with low NNT for serious outcomes are strong candidates for widespread implementation
• NNT varies significantly with baseline risk — personalised medicine considers individual risk profiles
• Population-level NNT informs public health policy (e.g. vaccination programme NNT)
• The concept of NNT has improved transparency in clinical trial reporting since its introduction in 1988 (Laupacis et al.)

NNT Examples in Clinical Practice

RRR vs ARR Illustration

Same RRR (50%) but very different NNT (5 vs 100) due to different baseline risk