Represent interval-censored event times correctly

Preserve what was actually observed

An event detected at a scheduled visit may have occurred any time since the previous assessment. That is interval censoring, not an exact event at the detection date. Start by recording the last time the event was known absent and the first time it was known present.

The survival package Surv documentation distinguishes exact, right-censored, left-censored and interval-censored observations. The data structure and estimator must support the observation type. Passing a detection date into an ordinary exact-event analysis changes the information being analysed.

Build lower and upper bounds

Create explicit lower and upper time bounds and a status definition. Confirm that both times use the same origin and units. For a right-censored observation, the event is known not to have occurred up to the last assessment; for a left-censored observation, it had already occurred by the first relevant assessment. An exact event has a known time rather than a nonzero-width interval.

Check impossible bounds, inconsistent assessments and changes in visit schedules. Keep the original visit data so a reviewer can trace every interval. Do not replace missing lower bounds with zero unless the study design establishes that interpretation.

Hypothetical example

A fictional participant tests negative at day 30 and positive at day 60. The event is known to lie between those assessments under the test and endpoint assumptions. Assigning day 45 as the event date creates precision that was not observed.

A midpoint analysis may be explored as a clearly labelled approximation in some settings, but it should not be presented as the measured event time. If visit frequency differs by group or health status, the observation process also needs consideration.

Select a compatible analysis

Confirm that the chosen survival procedure accepts interval-censored data and that its assumptions fit the scientific question. Ordinary Cox routines for exact or right-censored data are not automatically compatible. Record the software version, interval encoding and handling of boundary conventions.

Report the assessment schedule, event-detection rules, censoring types and analysis method. Explain how imperfect tests or informative assessment timing could affect interpretation. Use the follow-up reconciliation worksheet to identify missing visits and contradictory records. The useful final dataset retains uncertainty about timing rather than hiding it inside a fabricated date, and the report should make that uncertainty visible to readers.

Source verification: 6 November 2026. Examples are hypothetical.