Add time, not just people
Imagine three fictional participants contribute two years, one year, and half a year of eligible follow-up. Together they contribute 3.5 person-years. They do not contribute three years simply because there are three people.
If one first event occurs during that time, the incidence rate is 1 divided by 3.5, or about 0.286 events per person-year. It can also be expressed as about 28.6 per 100 person-years. The second wording changes the scale, not the result.
Why follow-up stops matter
For a first-event analysis, someone generally stops contributing event-free time when the event occurs. Study end, loss to follow-up, and other rules can also stop observation. The protocol determines what time counts.
CDC's explanation of incidence rates places time directly in the denominator. That helps compare observation across participants, but it does not remove bias if those lost to follow-up differ in relevant ways.
Equal totals can conceal different histories
One person observed for ten years and ten people observed for one year each both supply ten person-years. They need not have the same risk pattern. Age, calendar period, and duration of exposure may differ.
An overall rate summarizes those histories. Analyses may need to account for changing rates rather than assuming one constant process.
Translate carefully for a public audience
“Five events per 100 person-years” is not automatically “five in every 100 people will have an event next year.” Estimating a cumulative probability requires assumptions or an analysis that handles the timing and relevant competing events.
The denominator tells you what was counted. Keeping it in the sentence prevents an exposure-time summary from becoming an unsupported personal forecast.
