- Published
- Updated
Carrying the last observation forward is computationally simple, but produces a value assuming the participant remains at the earlier state. Convenience does not establish suitability for the scientific question or appropriate representation of uncertainty in the resulting analysis.
Define the question and assumptions
Consider the natural course of the measurement and reasons for missingness. An early observation may be a strong assumption for a later period, particularly where change is expected. Select an appropriate approach with the statistician and preserve traceability distinguishing carried-forward values from measurements actually observed at that time.
Hypothetical study example
In a hypothetical trial, a participant leaves follow-up after early improvement. Carrying the value forward assumes that improvement persists, whereas withdrawal after deterioration can have different implications. The team evaluates both histories rather than assuming one convenient filling rule represents all clinically relevant missing-data scenarios.
Design review checks
- Retain the original date on which a carried-forward value was observed.
- Assess the trajectory before missingness together with known reasons for absence.
- Examine how conclusions change under alternative scientifically plausible assumptions.
Record the decision and its limits
An accepted plan does not present imputed observations as measured outcomes. Explain the rationale and implications for variance and uncertainty. Previous use in another study is not sufficient justification. Compare methods alongside whether they target the same question, rather than selecting whichever approach gives the most attractive result.
Frequently asked questions
- Is the last value the true later outcome?
- No. It represents an explicit or implicit assumption about an unobserved period.
- Can ease of calculation determine the method?
- Practicality matters, but it cannot replace scientific suitability.