The Statin Trial Had Two Primary Endpoints. Only One Is in the Headlines.

7 SEP 2026 — The STAREE trial randomised 9,971 Australians aged 70 and over, with no cardiovascular disease, diabetes or dementia, to atorvastatin 40mg or placebo, and followed them for a median of 5.9 years. It had two primary endpoints. The cardiovascular one was met and is being reported everywhere. The other one asked whether the drug kept people alive and independent, and the answer was no.

Both primary endpoints, in order

Major cardiovascular events — cardiovascular death, non-fatal heart attack, stroke or coronary revascularisation — occurred in 6.0 per cent of the atorvastatin group and 8.3 per cent of the placebo group. The hazard ratio was 0.70, confidence interval 0.61 to 0.82, p below 0.001. That is the 30 per cent reduction in the headlines, and it is credible. The trial was large and well conducted.

Disability-free survival — surviving without dementia or persistent physical disability — occurred in 12.8 per cent versus 13.6 per cent. Hazard ratio 0.94, confidence interval 0.84 to 1.05, p equals 0.25. That interval crosses one. The trial could not distinguish the drug from placebo on the outcome it was co-designed to measure.

Both were primary endpoints. Reporting one without the other tells half the story.

Thirty per cent of what

The gap between 8.3 and 6.0 is 2.3 percentage points. Divide one by the other and it is a 30 per cent relative reduction; subtract them and it is a 2.3 per cent absolute one. Both describe the same two numbers.

The number needed to treat puts it in proportion: about 43 people take atorvastatin daily for roughly six years for one of them to avoid a major cardiovascular event. The other 42 take it and get no cardiovascular benefit, though none of them knows in advance which group they are in.

An NNT of 43 over six years is not a poor result. Plenty of accepted preventive medicine sits in that range, and for a cheap generic with a long safety record it is a reasonable trade. A 30 per cent reduction and a one-in-43 chance land very differently on a 74-year-old deciding whether to start a daily tablet. The second is the quantity that describes their decision.

Mostly non-fatal, which changes what was prevented

The benefit was driven predominantly by non-fatal events. Fewer heart attacks and strokes that people survived, and fewer revascularisation procedures.

That is worth having. A non-fatal stroke is one of the more consequential things that can happen to someone in their late seventies, and preventing it is a good outcome by any measure.

It also explains the null result on the second endpoint. The drug prevented events people mostly survive, so it changed the character of the next six years without measurably changing whether those years were lived independently.

The harms are small, real, and were counted

Muscle, liver and diabetes-related adverse events all occurred more often on atorvastatin than on placebo. Serious adverse events were uncommon and identical between the groups, at 2.7 per cent each.

That is a mild harm profile and it is not nothing, particularly the diabetes signal in a cohort selected to be free of diabetes at entry. Statin-associated muscle symptoms are also the most common reason people stop taking these drugs, and a trial population is more persistent than a general practice population.

The real-world number needed to treat is therefore likely worse than 43, since adherence is always higher in a trial than in general practice.

What this does and does not settle

STAREE answers a question primary prevention trials left open. They have historically under-recruited people over 70, forcing guidelines to extrapolate into that age group from younger cohorts. STAREE tested it directly in almost ten thousand people and found a cardiovascular benefit that is unlikely to be chance.

It does not settle whether statins are worth starting at 70 for someone whose main concern is independence. On that question the trial gave a direct answer, and the answer was that it could not show a difference. The confidence interval, 0.84 to 1.05, is compatible with a small benefit and with a small harm.

The population is also narrower than the coverage implies. These were community-dwelling Australians recruited through general practice, free of cardiovascular disease, diabetes and dementia, and healthy enough to enter a six-year trial. That is not the average 75-year-old in a clinic.

This is a familiar pattern: a trial measures an intermediate event well and the outcome people care about poorly, and the intermediate one becomes the headline — the same shape as the obesity methylation work we covered last week. Here the outcome was actually measured, which is better, and the measurement was null.

What to watch

Watch the guideline response. The investigators have explicitly asked for updated guidance, and the two endpoints point in different directions. Any recommendation that cites the 30 per cent without the disability-free result will have chosen one half of a two-part trial.

And the subgroup data, when it is published in full. A cohort of 9,971 with a mean age of 74.7 contains people in very different positions, and whether the benefit concentrates in the higher-risk end of a low-risk population is the question a clinician actually has to answer.