← All posts

When source data changes after a decision

Adjudication would be simpler if data stood still. They don't. A site enters an adverse event, the committee reviews and classifies it, and weeks later the site corrects the onset date, adds a lab result, or attaches a discharge summary that wasn't available before. Now the committee's decision rests on a record that no longer matches the source.

These updates are routine. How a study handles them says a lot about how defensible its adjudication record will be.

Why it matters

A final determination is only as good as the information it was based on. If the underlying data change and nobody revisits the case, the trial ends up with a classification that may not reflect the record. If the case is revisited by quietly overwriting the original decision, the trial loses the history of what was decided and why.

Neither is acceptable when the results may support a regulatory submission. What's needed is a process where:

  • changed cases are identified reliably, not by someone happening to notice;
  • reviewers can see exactly what changed;
  • the earlier decision stays on record;
  • any new decision is made and documented the same way as the first.

Step 1: detect the change

The first problem is knowing that something changed at all. In many studies, event data move from the clinical database to the adjudication team through periodic exports. Each new export should be compared with what was previously sent for review, and any modified or newly added records flagged automatically.

Relying on sites or data managers to send a note whenever they edit a record doesn't scale, and it is exactly the kind of manual step that gets missed late in a study.

Step 2: decide whether it needs re-review

Not every change affects a determination. A corrected typo in a site's address is not the same as a new hospitalization record. The charter or adjudication plan should say which changes trigger re-review. Two common approaches:

  • Re-review any change to the event record. Simple and conservative, but it can create a lot of extra work.
  • Re-review only changes to fields that matter for the decision, such as dates, diagnoses, outcomes, and source documents. More efficient, but those fields need to be defined in advance.

Whichever you choose, apply it consistently and write it down.

Step 3: show reviewers what changed

A reviewer looking at a re-opened case shouldn't have to compare two versions of a dossier line by line. The case should come back clearly marked as new or modified, with a field-by-field summary of what is different from the version they reviewed before.

It also helps for reviewers to see the prior final decision, clearly labeled as the earlier outcome and for reference only. That context lets them focus on whether the new information changes the conclusion, rather than starting from scratch.

Step 4: keep history intact

The original determination should never be edited or deleted. It was a valid decision based on the information available at the time, and the record should show it. The re-review produces a new determination, linked to the updated data, with its own reviewers, timestamps, and rationale.

The same principle applies at the end of the study. Once the final adjudication data have been exported for analysis, the record should be locked against new entries and deletions. If anything needs to be reviewed after that point, it should happen in a separate copy, so the locked record stays exactly as it was when it was analyzed.

Step 5: route it like any other case

A re-adjudicated case should go through the same consensus process as a new one: independent review, agreement or tiebreak, and committee discussion if needed. Shortcuts here, such as letting one reviewer quickly update the answer, undermine the consistency the process exists to protect.

Building it in

Handling changes by hand is possible on a small study. On a large one, with frequent exports and hundreds of events, it becomes a significant source of risk.

Reviewr™ is built for this: modified records return to the queue flagged as new or modified, with a field-by-field summary of what changed and the prior final decision shown for reference, and they are routed through the study's normal consensus rules. After the final export, records are locked, and any later review happens in a fresh copy. If late data changes are a concern on your study, let's talk.

Planning a CEC or DSMB?

Contact us