You’re cruising through a biostats Q-bank and see NNT/NNH in the stem. Your brain says “easy,” but then the answer choices start mixing ARR, RRR, OR, HR, p-values, and “statistically significant.” That’s where people miss points: NNT and NNH are not just calculations—they’re interpretations of risk in a specific study design. Every distractor is usually tempting for a reason.
Tag: Biostatistics > Study Design & Probability
The Clinical Vignette (Q-Bank Style)
A randomized, double-blind clinical trial evaluates Drug X vs placebo to prevent stroke over 2 years in adults with hypertension.
- Drug X group: 40 strokes out of 1000 patients
- Placebo group: 60 strokes out of 1000 patients
Drug X also increases major GI bleeding:
- Drug X group: 30 major bleeds out of 1000
- Placebo group: 10 major bleeds out of 1000
Question: Which of the following best describes the number needed to treat (NNT) to prevent one stroke and the number needed to harm (NNH) for major bleeding?
Answer choices
A. NNT = 50; NNH = 50
B. NNT = 100; NNH = 100
C. NNT = 200; NNH = 50
D. NNT = 50; NNH = 100
E. NNT = 20; NNH = 20
Step 1: Translate the Stem Into Risks
Always convert counts to event risks first.
Stroke (benefit)
- Experimental event rate (EER) =
- Control event rate (CER) =
Bleeding (harm)
- Bleed risk Drug X =
- Bleed risk placebo =
Step 2: Compute ARR and NNT (Benefit)
Absolute risk reduction (ARR)
Number needed to treat (NNT)
Interpretation: Treat 50 patients for 2 years to prevent 1 stroke.
Step 3: Compute ARI and NNH (Harm)
Absolute risk increase (ARI)
Number needed to harm (NNH)
Interpretation: Treat 50 patients for 2 years and you’ll cause 1 additional major bleed (compared with placebo).
Correct Answer: A. NNT = 50; NNH = 50
This is the clean “mirror” situation: the drug provides an absolute stroke benefit of 2% and an absolute bleeding harm of 2%.
Why Every Distractor Matters (Systematic Breakdown)
A. NNT = 50; NNH = 50 ✅
- Matches and .
- Uses absolute risk differences, which is what NNT/NNH are based on.
B. NNT = 100; NNH = 100 ❌
This would correspond to and .
Why it’s tempting:
Students often misread the event rates or accidentally use:
- Stroke difference as and then divide by 2000 (wrong denominator), or
- Convert “per 1000” into “percent” incorrectly.
High-yield fix:
NNT is extremely sensitive to small arithmetic mistakes because it uses a reciprocal.
C. NNT = 200; NNH = 50 ❌
NNH is correct, but NNT = 200 implies (0.5%), not 2%.
Why it’s tempting:
This is what you get if you mistakenly use relative risk reduction (RRR) or otherwise shrink the absolute effect.
For stroke:
- RR =
- RRR = (33.3%)
If someone incorrectly treats RRR (33%) as though it directly gives NNT, they’ll go off the rails. NNT requires ARR, not RRR.
D. NNT = 50; NNH = 100 ❌
NNT is correct. NNH = 100 implies (1%), but the actual bleeding ARI is 2%.
Why it’s tempting:
A common error is subtracting bleeding risks in the wrong direction or mixing groups (e.g., by inventing an intermediate number).
High-yield fix:
For harm:
- ARI = risk in treated − risk in control (when treated increases adverse events)
E. NNT = 20; NNH = 20 ❌
This corresponds to and .
Why it’s tempting:
Students may confuse absolute difference in counts per 1000 (20/1000) with 0.05 instead of 0.02:
- Stroke: (not 0.05)
High-Yield NNT/NNH Facts USMLE Loves
1) NNT/NNH are built on absolute risk differences
Board mindset: if an answer choice only gives RR, OR, HR, or RRR, you probably still need ARR/ARI to get NNT/NNH.
2) Timeframe matters
NNT/NNH are always tied to:
- a specific duration (e.g., “over 2 years”)
- a specific population (baseline risk changes NNT)
If baseline risk is low, ARR shrinks → NNT increases.
3) What’s “good” depends on context
- Lower NNT = more benefit per patient treated
- Higher NNH = safer (harm is rarer)
A quick clinical interpretation tool:
- If NNT ≈ NNH, you need to weigh severity and reversibility of outcomes (stroke vs bleed), patient values, and alternatives.
4) NNT/NNH are most natural in RCTs (incidence known)
You need actual risks (incidence), which are easiest to get from:
- Randomized trials (prospective incidence)
- Cohort studies (incidence)
They’re not as straightforward from case-control studies because case-control designs begin with outcome status and typically yield odds ratios, not incidence.
5) Rounding on exams
Because uses a reciprocal, small differences matter.
Rule of thumb:
- If they want an integer, round up in clinical convention (more conservative), but many USMLE-style questions avoid ambiguity by using clean numbers.
Quick Table: NNT/NNH vs Other Common Measures
| Measure | Formula | Uses absolute risk? | Commonly from | Typical trap |
|---|---|---|---|---|
| ARR | Yes | RCT, cohort | Confusing with RRR | |
| RRR | No | RCT, cohort | Mistakenly used to compute NNT | |
| RR | No | RCT, cohort | Interpreted as absolute difference | |
| OR | No | Case-control (also others) | Approximates RR only when disease is rare | |
| NNT | Yes | RCT, cohort | Using RRR or OR instead of ARR | |
| NNH | Yes | RCT, cohort | Subtracting in wrong direction |
Takeaway Pattern for Q-Banks
- Convert to risks (per 100, per 1000 → decimals).
- Compute ARR (benefit) and ARI (harm).
- Take reciprocals for NNT/NNH.
- Use distractors to check what you might have mixed up (RRR vs ARR, OR vs RR, wrong direction, denominator errors).
When you treat the distractors as a checklist of common mistakes, these questions stop being “math problems” and start being pattern recognition—exactly what USMLE rewards.