Statistics in Medicine and Health

Medical numbers arrive at the worst moments: a screening result, a prescription decision, a headline about the food you ate this morning. The same statistic can be framed to reassure or to frighten, and the framing usually serves someone. Reading these numbers calmly is a skill you can learn, and it changes real decisions about your body.

You do not need the biology behind a test or a drug to read its statistics well. You need three habits: ask what the base rate is, insist on absolute numbers alongside relative ones, and translate percentages into natural frequencies — counts out of a fixed group of people. Those habits protect you from the most common ways medical figures mislead.

Sensitivity, specificity, and the base rate

A test's sensitivity is how often it flags people who truly have the condition; its specificity is how often it clears people who truly do not. Both can be high and the test can still mislead, because what you actually want to know is the reverse question: given a positive result, how likely is it that I am sick? That answer depends heavily on the base rate — how common the condition is to begin with. When a disease is rare, even a good test produces many false alarms for every true case.

Absolute versus relative risk

Relative risk compares two groups: "50 percent higher" or "double the risk." It hides the starting point. Absolute risk tells you the actual chance: 1 in 10,000, or 2 in 10,000. A doubling sounds alarming, but doubling a tiny number leaves a tiny number. Whenever you meet a relative figure, ask "of what?" A percentage change is meaningless until you know the absolute baseline it moves.

Number needed to treat

Number needed to treat (NNT) is how many people must take a treatment for one of them to benefit. It is one divided by the absolute risk reduction. An NNT of 5 means a strong, worthwhile treatment; an NNT of 500 means most people who take it gain nothing, though it may still be worth it for a serious outcome. NNT converts abstract risk reductions into something you can weigh against cost and side effects.

Consider a screening test for a condition that affects 1 percent of people in a given group. The test has 90 percent sensitivity and a 9 percent false-positive rate. Someone tests positive; how worried should they be? Translate to natural frequencies. Imagine 1,000 such people. About 10 have the condition, and 9 of them test positive. Of the 990 who are healthy, about 9 percent — roughly 89 — also test positive. So about 98 people test positive, but only 9 truly have the condition. The chance that a positive result is a true positive is about 9 in 98, under 10 percent. The same numbers stated as percentages leave most people — including many clinicians — badly overestimating the risk.

Relative risk is not a trick, and absolute numbers are not always the small ones. Suppose a treatment lowers the risk of a bad outcome from 40 percent to 20 percent. That is a 50 percent relative reduction and also a 20-percentage-point absolute reduction, with an NNT of 5. Here the relative and absolute pictures both say the effect is large, because the baseline risk is high. The lesson is not "relative bad, absolute good." It is that relative figures only deceive when the baseline is tiny, so you must always check the baseline in both directions.

In October 1995 the UK Committee on Safety of Medicines warned that certain third-generation oral contraceptives roughly doubled the risk of venous thromboembolism (blood clots) compared with older pills. "Double the risk" made headlines and frightened many women into stopping the pill. But as the risk-communication researcher Gerd Gigerenzer has often noted, the absolute figures were small: the risk rose from about 1 in 7,000 women per year to about 2 in 7,000. The relative doubling was real; the absolute increase was roughly one additional case per 7,000 women. The panic had consequences. Analysts have linked the scare to a sharp rise in unintended pregnancies and abortions in England and Wales the following year — an outcome made worse because pregnancy itself carries a substantially higher clot risk than any of the pills. Presenting the same evidence as natural frequencies would very likely have prevented much of the harm.

• "A positive test means I have the disease." For rare conditions, most positives can be false. The base rate decides.
• "Double the risk is always a big deal." Only if the baseline is meaningful. Doubling 1 in 10,000 gives 2 in 10,000.
• "High sensitivity makes a test trustworthy." Sensitivity and specificity describe the test, not the odds that you are sick given your result.
• "If a treatment works, everyone who takes it benefits." Many effective treatments have a large NNT; most takers see no change.

Find one health headline that states a relative risk ("raises the risk of X by Y percent"). Search for the study's absolute numbers — the actual rates in each group. Rewrite the finding as natural frequencies: "out of 1,000 people, this many rather than that many." Notice how the story feels once the baseline is visible.

Think Like a Maester: When a risk is stated as a percentage change, ask "out of how many people?" before you feel anything.

1. Explain, in plain terms, the difference between a test's sensitivity and the chance that a positive result is correct.
2. Why does a rare condition produce many false positives even with a good test?
3. A drug is said to cut heart-attack risk by 30 percent. What single question must you ask before judging whether that matters?
4. Define number needed to treat and explain what an NNT of 200 tells you.
5. Restate this claim as natural frequencies: "The exposure doubles your risk, from 3 in 100,000 to 6 in 100,000."

Medical statistics mislead in predictable ways, and a few habits defend against all of them. A test's sensitivity and specificity are not the same as the chance you are sick given a positive result; that depends on the base rate, and for rare conditions most positives are false. Relative risk ("double," "50 percent higher") hides the baseline, so a frightening multiple can be a trivial absolute change — always ask "of what?" Number needed to treat turns a risk reduction into a human count you can weigh against harms and costs. Above all, translate percentages into natural frequencies: counts out of a fixed group of people. The 1995 UK pill scare shows the stakes — the same evidence, framed as absolute numbers, would have calmed a panic that caused real harm.

• Conditional Probability and Base Rates
• Reading Polls and Elections