Twenty Patients, Thirty Events: Read the Harms Denominator
October 28, 2026 · 5 min read · Aqrab editorial team
A trial says 20 people had a harmful event. Another table lists 30 events. That is not necessarily a contradiction: some people may have had the event more than once. Before comparing trial arms, identify what was counted, who was observed, and for how long.
One example, three denominators
Imagine a hypothetical randomized trial with 100 participants assigned to each arm. The counts below are invented to illustrate reporting, not evidence about any treatment.
| Arm | Participants observed | People with ≥1 event | Total events | Observed person-years |
|---|---|---|---|---|
| A | 100 | 20 | 30 | 80 |
| B | 100 | 15 | 15 | 100 |
For arm A, 20/100 = 20% of observed participants had at least one event; arm B has 15/100 = 15%. The recurrent-event rates are 30/80 × 100 = 37.5 and 15/100 × 100 = 15 events per 100 observed person-years. A rate counts events, not the probability that an individual experiences one. It can exceed one event per person, and it cannot be read as a percentage of people harmed.
The practical takeaway
When a harms headline says “incidence,” find the numerator, denominator, observation window, and rule for repeat events before repeating the comparison.
Person-time is useful, but not a repair for every gap
The unequal 80 and 100 person-years could reflect planned exposure, early withdrawal, death, or missing follow-up. The table does not tell us which. If people who develop a harm stop treatment early, observed exposure time is partly shaped by the harm itself. Dividing events by that time does not automatically remove selection or make the groups comparable. Conversely, a simple proportion over unequal observation time does not describe a common risk window. Request the planned and actual assessment periods, reasons for discontinuation, and a prespecified analysis aligned with the question.
These descriptive calculations also omit uncertainty intervals and the timing of events. They are not a treatment-effect estimate. For a decision, distinguish the risk of at least one event by a stated time from the burden of repeated events during observed follow-up.
Four questions for a harms table
- Who: How many participants were randomized, exposed, and included in each harms analysis?
- What: Are the figures people with at least one event, all events, or events of a specified severity?
- When: What are the risk window and actual follow-up or exposure time in each arm?
- After stopping: Were events still sought after treatment discontinuation, and how were repeated events handled?
Where Aqrab fits
Use Aqrab to organize a methods critique around the harms numerator, denominator, follow-up, and repeat-event rule. The critique can reveal an ambiguous comparison; it cannot recover unreported events or establish a causal harm effect from a summary table.
Source anchor
The CONSORT Harms 2022 statement and explanation asks trial reports to distinguish participants with at least one event from events per time at risk, state how recurrent events and differing exposure or follow-up were handled, and report the analysis denominators. It cautions that person-time can mislead when those who continue treatment are selected for experiencing fewer harms. This is reporting guidance, not validation of the invented counts.
Keep reading
Don't stop at one method.
Good methods judgment comes from contrast. Read the neighboring guides, see where the assumptions diverge, and avoid treating every observational problem like it needs the same hammer.
The Trial Ended. Who Made It Into the Open-Label Extension?
A denominator-first audit of long-term extension findings: trace who enrolled, who stayed, and what an uncontrolled extension can actually establish.
A 12-Month Benefit and a 30-Day Harm Are Not One Net-Benefit Number
A fictional trial timeline shows why benefit and harm counts measured over different windows cannot be silently collapsed into one net-benefit claim.
Clinical Importance: When a Significant Result Is Still Too Small to Matter
A practical guide to reading confidence intervals against a prespecified clinical-importance threshold. Separate evidence against the null from evidence that an effect is large enough to matter.