Hazard ratio

Some studies follow people for months or years and record when an event happens as well as whether it happened. For this kind of time-to-event data, the standard summary is a hazard ratio from a survival analysis [2]. Hazard is close to the idea of risk, but it means the instantaneous risk at a moment, and it can change during follow-up [2].

Because the hazard ratio says how many times more or less likely the event is in one group than the other, a value of 1 means the same likelihood in both groups, a value above 1 means the event was more likely in the group being compared (for example, the supplement group), and a value below 1 means it was less likely [2]. Ratio measures such as the hazard ratio are usually analyzed on a log scale, and the Cox model, a common regression model for survival data, estimates the log hazard ratio directly [2].

Hazard ratios appear in randomized trials and in observational studies that examine how a treatment or exposure relates to an outcome [1]. A trial might report, for example, the hazard of a first event in a supplement group compared with a placebo group over several years.

Because a hazard ratio says how many times more or less likely the event is in one group than in another [2], it is a relative measure: it compares the groups, and on its own it does not give the number of people who had the event.

Many analyses also assume the hazard ratio stays constant over the whole follow-up, known as the proportional hazards assumption, even though the hazards themselves may vary over time [2]. Methods reviewers note that the hazard ratio is often misinterpreted, in ways that can undermine the critical appraisal of both randomized trials and observational studies [1].

NootropicsWork reports a hazard ratio with its confidence interval, the population, the comparison and the length of follow-up, and keeps findings from observational studies labeled as associations.

Confidence interval, randomized controlled trial, cohort study, statistical significance.