What to know
- An HR below 1 means a lower event rate in the first group; it is a relative measure.
- The same HR can mean 4 or 74 fewer events per 1,000 people, depending on baseline risk.
- An HR does not say how much sooner an event happens and assumes a roughly constant ratio over time.
01
What a hazard ratio measures
A hazard ratio (HR) compares how quickly an event happens in two groups during follow-up. The hazard is the chance of the event in the next short moment for people who have not had it yet. The Cochrane Handbook describes the hazard ratio as showing how many times more, or less, likely a participant is to have the event at a particular point in time. An HR of 1 means no difference, below 1 means a lower rate in the first group, and above 1 means a higher rate.
Studies report hazard ratios when people are followed for different lengths of time and not everyone has the event. People who leave the study or reach its end without an event are counted as censored: their event-free time still contributes to the analysis. That is why survival analysis, usually a Cox model, is used instead of simply counting events. A hazard ratio is a relative measure, so it never tells you on its own how many people benefited.
02
Reading HR 0.80 step by step with a real trial
The SELECT trial randomised 17,604 adults with cardiovascular disease and overweight or obesity, but without diabetes, to weekly semaglutide or placebo. Mean follow-up was 39.8 months. The main outcome, cardiovascular death, non-fatal heart attack or non-fatal stroke, occurred in 569 of 8,803 people on semaglutide (6.5%) and 701 of 8,801 on placebo (8.0%). The reported hazard ratio was 0.80, with a 95% confidence interval of 0.72 to 0.90.
- Direction: 0.80 is below 1, so the event rate was lower in the semaglutide group.
- Size, in relative terms: among people still free of the outcome at any point in follow-up, the event rate was about 20% lower, if the ratio was roughly constant over time.
- Uncertainty: the interval of 0.72 to 0.90 is compatible with rate reductions from about 10% to 28%. It does not include 1.
- Absolute terms: the event proportions differed by 1.5 percentage points over the trial. From the crude counts, that is roughly one fewer event for every 67 people treated for about three years.
The last step uses simple proportions and ignores differences in follow-up, so treat it as an approximate illustration rather than a figure reported by the trial. The point is the contrast: the same result is "20% lower" in relative terms and "1.5 percentage points" in absolute terms. Both are true, and a decision needs both, together with harms. In SELECT, 16.6% of people on semaglutide and 8.2% on placebo stopped the trial product because of adverse events.
Read the SELECT trial explained
Learn to read confidence intervals
Sources for this section: [3]
03
The same hazard ratio can mean very different absolute effects
The absolute benefit behind a hazard ratio depends on how common the event is without treatment. The table uses invented baseline risks, not data from any study, and applies a constant HR of 0.80 over the same follow-up period. When hazards are proportional, the treated group's event-free proportion equals the comparison group's raised to the power of the hazard ratio.
| Risk without treatment | Risk with HR 0.80 | Fewer events per 1,000 people | Risk ratio |
|---|---|---|---|
| 2% | 1.6% | 4 | 0.80 |
| 10% | 8.1% | 19 | 0.81 |
| 20% | 16.4% | 37 | 0.82 |
| 50% | 42.6% | 74 | 0.85 |
Two lessons follow. First, an identical hazard ratio can mean 4 or 74 fewer events per 1,000 people, so always look for the baseline risk and the time frame. Second, the hazard ratio and the risk ratio are close when events are rare but drift apart as events become common. Reporting "a 20% lower risk" is only an approximation, and it becomes less accurate as the event becomes more frequent.
04
What a hazard ratio does not tell you
It does not tell you how much sooner or later the event happens. Spruance and colleagues showed that the same hazard ratio can correspond to small or large differences in median time, and warned against phrases such as "twice as fast". When timing matters, look for median times, event curves or the proportion with the event at stated time points.
A single hazard ratio also assumes that the ratio stays roughly constant during follow-up. The Cochrane Handbook calls this proportional-hazards assumption a simplifying assumption. If the curves cross, for example when a treatment has early harms and later benefits, one number can hide the pattern. In observational studies, a hazard ratio shows an association adjusted for the measured factors, not proof that the exposure caused the difference. With a composite outcome, check which components drove the result.
05
A checklist for hazard ratios in papers and news
Use these questions whenever a study or headline reports a hazard ratio. Most answers are in the abstract, the results table or the event curves of the paper. If a news story leaves the numbers out, check the original paper before drawing conclusions.
- What exactly was the event, and was it a single outcome or a composite?
- Who was compared with whom, and was the comparison randomised?
- What were the event rates in each group, and over what follow-up time?
- What is the confidence interval, and does it include 1?
- How large is the absolute difference, and how common was the event without treatment?
- Do the event curves stay apart in a steady way, or do they cross?
- What harms, costs or burdens came with the benefit?
Sources
- Chapter 6: Choosing effect measures and computing estimates of effect (Cochrane Handbook for Systematic Reviews of Interventions, version 6.5)Cochrane (Higgins, Li and Deeks, editors) · 2023Source accessed:
- Hazard ratio in clinical trialsAntimicrobial Agents and Chemotherapy (Spruance, Reid, Grace and Samore) · 2004Source accessed:
- Semaglutide and Cardiovascular Outcomes in Obesity without Diabetes (SELECT)New England Journal of Medicine (Lincoff and colleagues) · 2023Source accessed:
Revision history
- Initial English, Spanish and Japanese article prepared with automated assistance. The cited sources were checked for the complete article on 2026-10-23T03:14:20Z (UTC), and the worked-example arithmetic was recalculated from the trial's published counts. The baseline-risk table is hypothetical. Publication and preparation use the current date; no historical date was assigned. No independent statistical or clinical review is recorded.