
- Published
- Updated
Treatment discontinuation, rescue medication and a missed assessment are different kinds of information. An event can change interpretation while an observation is still collected. Recording missingness separately from intercurrent events helps preserve the intended scientific question.
Define the question and assumptions
Define the treatment-effect question first. Maintain separate fields for treatment status, follow-up status and endpoint availability. Plan the intercurrent-event strategy alongside analytical handling of missing observations; the occurrence of an event does not automatically make every subsequent observation irrelevant or impossible to collect.
Hypothetical study example
In a hypothetical trial, a participant stops study treatment but attends the next endpoint visit. The team does not classify that measurement as missing merely because treatment ended. Its analytical role follows the prespecified question, while continued consent and safety follow-up needs are assessed separately.
Design review checks
- Classify treatment cessation, withdrawal from follow-up and unavailable measurements separately.
- Link intercurrent-event dates to the dates of endpoint assessments collected.
- Record known reasons and uncertainty for each missing observation where possible.
Record the decision and its limits
An accepted data model represents different combinations: treatment stopped with data available, or treatment continuing with an assessment missing. The analysis plan explains their scientific meaning. A single convenient withdrawn status can unnecessarily weaken interpretation and the ability to perform relevant sensitivity assessments using the information actually collected.
Frequently asked questions
- Does stopping treatment mean leaving follow-up?
- No. Treatment participation and data collection status need separate assessment.
- Does an intercurrent event always create missing data?
- No. A measurement may exist; its contribution to the scientific question needs separate definition.