How to read the evidence · 2 of 4
What a hazard ratio means
A hazard ratio (HR) compares how fast an event happens in two groups. Below 1, the event happens more slowly in the experimental group than in the control group; above 1, faster; at 1, no difference. It is reported with a 95% confidence interval: a range of values compatible with the data. If the interval includes 1, the data are also compatible with no difference.
Teaching example: a model with a 12-month control median. Not trial data.
- Control median
- 12 months
- Second arm median
- 20.0 months
- Event-free at 24 months
- 25.0% vs 43.5%
A hazard ratio of 0.60 means the event happens at a lower rate in the second arm than in the control arm, at any moment during follow-up, on average over the trial. It is a summary of the whole curve, not the ratio of anyone's survival times, and it assumes the ratio stays about constant over time (proportional hazards). Real trial curves can cross or plateau, so a single hazard ratio can hide that.
Data table (modelled values)
| Months | Control | Second arm (HR 0.60) |
|---|---|---|
| 0 | 100.0% | 100.0% |
| 6 | 70.7% | 81.2% |
| 12 | 50.0% | 66.0% |
| 24 | 25.0% | 43.5% |
| 36 | 12.5% | 28.7% |
| 48 | 6.3% | 18.9% |
Things a hazard ratio does not tell you
- How many months were gained. That depends on the shape of the curves; read the medians or landmark rates beside it.
- Whether the benefit lasts. If the curves cross, a single number averages different periods together.
- Whether the trial was well designed or the endpoint matters to patients. The ratio of a weak endpoint is still a weak endpoint.
Sources and further reading
These pages explain general methods in plain language and were written for this site; the sources below are the references to read next. Some classic papers have no abstract online, so the explanations stay at textbook level.