Relative Risk, Absolute Risk, and Number Needed to Treat: How to Read a Statin Trial

What does "statins cut heart attacks by a quarter" actually mean?#

A claim like "statins cut heart attacks by a quarter" is a relative risk reduction, and a relative figure on its own tells you nothing about how often the event happened to begin with. To size a benefit honestly you convert that percentage into an absolute risk reduction (the plain difference in event rates between the treated and untreated groups) and then into a number needed to treat (one divided by that difference). Run those two conversions and the very same result can read as commanding or trivial, depending only on which of the three numbers reaches the headline. None of them lies. Each answers a different question, and the space between the answers is where most arguments about statins actually happen.

Key points#

• Relative risk reduction, absolute risk reduction, and number needed to treat are three views of one result, not three separate results.
• A relative figure stays large even when the underlying event is rare, so it travels well in headlines and poorly in decisions.
• Absolute benefit depends on baseline risk: the same relative effect helps a high-risk person far more than a low-risk one.
• For statins, pooled trials put the myocardial-infarction relative reduction near 29 percent but the absolute reduction near 1.3 percent.

Turn one trial into three numbers#

Every trial hands you the same starting material: the event rate in each arm. Imagine a study in which 4 in 100 untreated people have a heart attack over the follow-up window, against 3 in 100 treated people. Those two counts are everything the three numbers are built from.

The relative risk reduction reads them as a ratio. Dropping from 4 to 3 removes a quarter of the starting risk, so the relative risk reduction is 25 percent. This is usually the figure that reaches a press release, because it is the largest-sounding member of the set and it holds that size no matter how uncommon the event is.

The absolute risk reduction reads them as a subtraction. Four percent minus 3 percent is 1 percentage point. This is the number that keeps the underlying risk in the frame, which is why it matters. A 25 percent cut applied to a 4 percent risk and a 25 percent cut applied to a 0.4 percent risk deliver very different real benefit, even though the relative label reads identically on both.

The number needed to treat converts that gap into people. It is one divided by the absolute risk reduction, so 1 divided by 0.01 is 100. You would treat 100 people for the length of the study to spare one heart attack, while the other 99 reach the same outcome either way. The American Academy of Family Physicians walks through this exact chain: in one worked case a 12 percent relative reduction in death corresponds to roughly a 0.6 percent absolute reduction and a number needed to treat near 167 over about four years. One event, three sentences, each answering a different question.

Baseline risk is the hidden variable#

Here is what a lone headline can never carry. The same relative risk reduction yields a larger absolute benefit in people who start at higher risk. Relative reduction travels fairly steadily across risk levels; absolute reduction does not. Split trials by setting and the effect is visible. In primary prevention (people without established heart disease) the absolute reduction in heart attacks runs near 0.7 percent. In secondary prevention (people who already have vascular disease and therefore much higher baseline risk) it runs near 2.2 percent. The relative reductions in those two settings sit closer together than the absolute ones, which is precisely why you cannot read your own likely benefit off a relative number.

The JUPITER trial shows the same rule from the opposite direction. It enrolled apparently healthy adults with normal LDL cholesterol but raised C-reactive protein, and rosuvastatin cut the primary composite endpoint by a wide relative margin (New England Journal of Medicine, 2008). A later analysis in Circulation: Cardiovascular Quality and Outcomes put the five-year number needed to treat near 20 for that broad endpoint, favorable next to older primary-prevention figures. The relative effect was striking; the number needed to treat is what tells you how many people had to share in it.

What the pooled statin trials show#

Statins are the standard teaching case because their relative reductions are real and their absolute reductions are modest, so the framing gap is unusually wide. A 2022 systematic review and meta-analysis in JAMA Internal Medicine pooled 21 randomized trials, each at least two years long with more than 1,000 participants, and reported both kinds of number side by side.

For myocardial infarction, statin therapy carried a relative risk reduction near 29 percent, which sounds decisive, alongside an absolute risk reduction near 1.3 percent, which implies roughly 77 people treated for about four and a half years to prevent one heart attack. For all-cause death, the relative reduction was near 9 percent and the absolute near 0.8 percent. For stroke, roughly 14 percent relative against about 0.4 percent absolute. In every row the relative figure is the arresting one and the absolute figure is the sobering one, and both come from the identical trials.

A four-question habit for any trial#

• What were the raw event rates in each arm, not only the ratio between them?
• Is the quoted figure relative or absolute? If only the relative one appears, treat that as a prompt to go find the other.
• What was the baseline risk of the people studied? That is what decides whether a genuine relative effect becomes a large or a tiny absolute one for someone like you.
• Over what time window? A number needed to treat of 77 over five years is a different proposition from the same number over one year.

None of this argues for or against any particular statin. It argues for reading the number that answers your question instead of the number chosen to impress you, and for remembering that the three figures are three views of one result, not three different results.