Hazard ratio

A hazard ratio (HR) compares how fast events happen in two groups over time. The "hazard" is the instantaneous rate of an event, such as a heart attack, among people who have not yet had one. The ratio divides the hazard in the treatment group by the hazard in the control group. It usually comes from a Cox proportional hazards model, which assumes the ratio stays roughly constant through the trial.

• HR 1.0: no difference.
• HR below 1.0: events happened more slowly in the treatment group.
• HR above 1.0: events happened faster in the treatment group.

The example everyone quotes: SELECT

SELECT randomised 17,604 adults with established cardiovascular disease and overweight or obesity, without diabetes, to semaglutide 2.4 mg or placebo, with a mean follow-up of 39.8 months. The primary outcome, cardiovascular death, non-fatal heart attack or non-fatal stroke, occurred in 569 people (6.5 percent) on semaglutide and 701 (8.0 percent) on placebo. The hazard ratio was 0.80, with a 95 percent confidence interval of 0.72 to 0.90.

HR 0.80 means the event rate was about 20 percent lower on semaglutide. That is a relative reduction. The absolute difference was 1.5 percentage points over about three years: 8.0 minus 6.5. Both are true, and both matter. Relative numbers travel across populations; absolute numbers tell you how many people benefit.

Why timing matters

A hazard ratio uses when each event happened, not only whether it did. People join a trial at different times and are followed for different lengths, so a simple percentage at the end would treat a heart attack in month two the same as one in year three. Trial papers pair the hazard ratio with Kaplan-Meier curves, which plot cumulative events over time; curves that separate early and keep separating fit a steady benefit.

Common misreadings

• HR is not the same as relative risk. It compares rates over time, not just final proportions. With uncommon events the two are often close, but not always.
• A hazard ratio is not a chance for you. It is an average across a trial population with a particular risk level.
• Look at the interval. SURPASS-CVOT's upper limit of 1.01 crosses 1, so tirzepatide was shown not to be worse than dulaglutide, its pre-set test, but not shown to be better.
• Observational hazard ratios are associations. Insurance-record studies also report HRs, but the groups were not randomised, so the number can reflect who chose the drug as much as what the drug did.

The SELECT trial page and clinical trials page report each outcome trial with both the hazard ratio and the event rates.

Outcome trials measure events over years. Day to day, what you can track is the set of risk markers they move: weight, blood pressure, lipids and HbA1c. The bloodwork tracking page shows how lab results sit on the same timeline as your doses.

Related terms: confidence interval, surrogate endpoint, placebo-controlled trial, estimand. Back to the glossary.