NNT, absolute vs relative risk, and what to report to patients
A pharmaceutical rep shows you data for a new statin. They announce excitedly:
"Our drug reduces heart attacks by 50%!"
Sounds impressive. Should you prescribe it to your at-risk patients?
Before you decide, you ask: "What were the actual event rates?"
Relative risk reduction
Placebo: 2% had heart attacks
Drug: 1% had heart attacks
Absolute reduction: 1%
The same data can tell very different stories depending on how you frame it.
Screening reduces cancer deaths from 4 per 10,000 to 3 per 10,000
| Metric | Value | Sounds like... |
|---|---|---|
| Relative Risk Reduction | 25% | "Huge benefit!" |
| Absolute Risk Reduction | 0.01% | "Quite small..." |
| Number Needed to Screen | 10,000 | "For 1 life saved" |
The most patient-friendly way to communicate treatment effects.
Translation: You need to treat 100 patients to prevent 1 event.
NNT = 10 (treat 10 to help 1)
Green = helped by treatment | Gray = treated but no benefit
| Treatment | NNT | Interpretation |
|---|---|---|
| Aspirin for MI (acute) | 42 | Very effective |
| Statins (primary prevention) | 100-200 | Modest benefit |
| Mammography (50-59y, 10yr) | 1,000+ | Small absolute benefit |
Benefits don't exist in isolation. Every treatment has a cost.
Consider a blood thinner for stroke prevention:
NNT to prevent 1 stroke
NNH for major bleed
For every 2 strokes prevented, there's 1 major bleed caused
Ideally: NNT << NNH (help many, harm few)
Danger zone: NNT ≈ NNH (helping as many as harming)
Never acceptable: NNT > NNH (harming more than helping)
Odds ratios are not risk ratios, but they're often treated as if they are.
A case-control study reports: OR = 2.5 for the association between a medication and falls in elderly patients.
Does this mean the medication doubles fall risk?
Probability of event in exposed / Probability in unexposed
Intuitive: "2x more likely"
Odds of event in exposed / Odds in unexposed
Less intuitive but necessary for case-control studies
When outcome is common (50% baseline):
| Group | Events | Odds |
|---|---|---|
| Control | 50% | 50/50 = 1.0 |
| Exposed | 75% | 75/25 = 3.0 |
OR = 3.0, but actual risk increase = 1.5x (not 3x!)
Same data, different framings - which helps patients make informed decisions?
The drug reduces cardiovascular events from 8% to 6% over 5 years.
"This drug reduces your risk by 25%!"
Relative risk - sounds bigger than it is
"The p-value was highly significant at 0.001."
Statistical jargon - meaningless to patients
"Without treatment: 8 out of 100 people like you will have a heart attack or stroke in 5 years."
"With treatment: 6 out of 100 will."
Natural frequencies - intuitive
"If I treat 50 people like you for 5 years, 1 person will avoid a heart attack or stroke. The other 49 would have been fine anyway or will have one despite treatment."
NNT framing - honest about limitations
Use natural frequencies (X out of 100) and NNT. Add context about side effects and NNH. Let patients weigh the trade-offs.
For each scenario, identify the best interpretation or catch the misleading statistic.
Always ask for absolute risk reduction. A 50% relative reduction could be 10% → 5% (big deal) or 0.002% → 0.001% (trivial).
Number Needed to Treat is the most patient-friendly metric. "I'd need to treat X people for 1 to benefit."
Every benefit has a cost. Compare NNT to NNH for a complete picture.
OR ≈ RR only when outcomes are rare. For common outcomes, OR overstates the effect.
Tell patients "X out of 100" rather than percentages or relative risks.