Person-years

Person-years

The sum of the observation time contributed by individuals under a study’s rules. One person observed for 1 year contributes 1 person-year; the total is often the denominator of an incidence rate.

How to read it

Check the event being counted, when the observation clock starts and stops, and the events per total person-time. Person-years are used in follow-up studies, trial safety analyses and studies using routine healthcare records. They are especially useful with unequal follow-up but can also be used with equal follow-up.

Why count time as well as people

In a teaching example, one person contributes 1 year and another contributes 2 years: 3 person-years in total. Counting only 2 people leaves out the difference in observation time.

An incidence rate divides the event count by eligible observation time. For a first event, follow-up usually stops at that event, loss to follow-up, death or the specified study end. Analyses of recurrent events may use different rules.

A mean a risk and a rate answer different questions

Mean weight loss describes the average change. Risk over a specified period describes how many people experience an event. A person-time rate describes event frequency per accumulated observation time. A study can report all of them.

Mean follow-up can be calculated, but is not itself an event rate. The exact arithmetic mean of the relevant observation times multiplied by the corresponding number of people recovers total time; median follow-up generally cannot be used this way.

The GLP-1 example

In its November 9, 2026 communication, FDA reported a pooled analysis of 126 randomized trials: 10 probable NAION events over 156947 person-years in the GLP-1 group and 5 over 131932 person-years in the combined comparator group. The rates were approximately 6.4 and 3.8 per 100000 person-years. The denominator includes accumulated observation time.

These figures came from randomized trials. FDA separately discussed Sentinel observational comparisons. Records describe a data source; randomization describes assignment; person-years describe how time enters a rate. These are different features of a study.

What the denominator cannot solve

Equal person-time does not make different populations or treatment periods interchangeable. If risk changes over time, period-specific analysis may be needed. Do not simply multiply an overall rate by any number of years and call it a personal cumulative risk.

Counting time before someone leaves a study does not remove bias if leaving is related to their outcome. Check population comparability, event definitions, missing follow-up and uncertainty.

Examples in other articles

Related readings will be added as the collection grows.