One Decision That Changes a Study: The Outcome Window

Two windows, two questions

Consider a fictional study of events after an index admission. “Any event within 30 days” and “any event within 90 days” are different outcomes, even if both are called a complication rate. The longer window creates more opportunity for events and may require more complete follow-up.

This is a conceptual example. No event counts or treatment effects are being claimed.

What changes in the workflow

The data collection form needs a clear index date, event dates, follow-up end, and rules for incomplete observation. The analysis needs an approach that matches the question and the information available. A person without 90 days of observation cannot automatically be classified as having no 90-day event.

A longer window may also change how competing events, repeat admissions, or exposure changes matter. These are design and analysis decisions, not formatting details.

Rehearse the awkward cases

Create fictional records with an event just before the boundary, just after it, and with observation ending early. Ask the team how each would be classified. If the answers differ, refine the definition before collection.

Draft a table shell with an explicit outcome window and denominator. Then verify that the planned dataset supplies the necessary information. This small exercise can reveal an expensive omission before it becomes a real-data problem.

If the window changes later

Record the reason, timing, and what had already been examined. Preserve the original plan and distinguish the original analysis from amendments or exploratory alternatives. A change made after observing results must not be represented as prospectively specified.

Use the Data Collection Builder, download the decision log, or scope a Research Rehearsal.

Reporting reference: STROBE and its explanatory resources. Reporting the decision transparently is necessary; it does not by itself establish that the decision was methodologically appropriate.