Number Needed to Treat
NNT — the number of people who must take a drug, undergo surgery, or receive a medical procedure before one person is helped.
Daniel Levitin argues this is the statistic patients should ask for under calm conditions (via pre-mortem), because stress makes on-the-spot reasoning unreliable.
Why It's Counterintuitive
Most people assume NNT should be 1 — my doctor wouldn't prescribe something that doesn't help me. Medical practice doesn't work that way. Scientists haven't mapped underlying mechanisms well enough; GlaxoSmithKline estimates 90% of drugs work in only 30–50% of people.
Levitin's Examples
Widely prescribed statin: NNT = 300 — 300 people take it for a year before one heart attack, stroke, or adverse cardiovascular event is prevented. Side effects occur in 5% → 15 people per 300. You are 15 times more likely to be harmed than helped. Levitin doesn't say whether to take it — he says have the conversation.
Prostate removal for cancer (men 50+): NNT = 49 — 49 surgeries per one person helped. Side effects in 50%: impotence, erectile dysfunction, urinary incontinence, rectal tearing, fecal incontinence — often lasting one to two years even for the "lucky" half.
How to Use It
Pre-mortem while calm:
- Ask for NNT
- Ask for side-effect rates
- Compare harm probability to benefit probability
- Discuss quality of life — shorter pain-free life vs. longer life with end-stage suffering
Medical ethics requires informed consent — access to this information is a right, not a courtesy.
Connections
- premortem — prepare the question chain before the doctor's office
- stress-impairs-judgment — why you can't manufacture this reasoning under cortisol
- base-rate-neglect — NNT forces population-level thinking over individual stories
- decision-quality-vs-outcome — sound process includes base rates before outcome