How note bloat and copy-forward distort adverse event data

Note bloat is the build-up of text in clinical notes that was copied forward, pasted or pulled in by templates rather than written for the visit, so what is new becomes hard to find. In a clinical trial those notes are source records, so a carried-forward symptom can move an adverse event's onset date, end date, grade and attribution, or a copied negative can hide a new event. Across 104 million notes at the University of Pennsylvania Health System, 50.1% of the text was duplicated from earlier text about the same patient (Steinkamp et al., JAMA Network Open, 2022).

The remedy is not a ban on copy-forward. It is knowing, for every graded event, which visit the supporting text came from.

In short:

• What it is: copied and imported text that outnumbers what was written at the visit.
• What it moves: onset date, grade, attribution and end date, and sometimes whether an event is recorded at all.
• The check: find the visit where the sentence supporting the grade first appeared.

What is note bloat?

Note bloat is the accumulation of copied, carried-forward and imported text in clinical notes, until the findings that changed at a visit are hard to find among the ones that did not. Three habits add the text: copy and paste, copy-forward (starting today's note from the last one) and templates that import results, medication lists and histories. Each is reasonable on its own, since an oncology note builds on the last one. For the clinician, carried text is context; for a trial database, each note is observations made on its date, and the two readings meet in the hands of whoever is moving events from the EHR into the EDC.

A systematic review by the Partnership for Health IT Patient Safety found that between two-thirds and nine in ten clinicians routinely use copy and paste, and that the practice can promote note bloat, internal inconsistencies, error propagation and documentation in the wrong patient's chart (Tsou et al., Applied Clinical Informatics, 2017).

How common is copy-forward documentation?

Duplicated or redundant text made up about 30% to 78% of note text in four studies, measured in different ways across note types and health systems:

Study	What was measured	Finding
Wrenn et al., JAMIA, 2010	Notes from 100 randomly selected patient admissions	On average, 78% of signout note text and 54% of progress note text was duplicated from earlier documents
Cohen et al., BMC Bioinformatics, 2013	22,564 notes for 1,604 patients with chronic kidney disease at New York-Presbyterian Hospital	About 30% redundancy, which distorted text-mining results until redundant notes were removed
Rule et al., JAMA Network Open, 2021	2.7 million outpatient progress notes for 200,000 patients at Oregon Health & Science University, 2009 to 2018	Median length up 60.1%, from 401 to 642 words; median redundancy up from 47.9% to 58.8%
Steinkamp et al., JAMA Network Open, 2022	104 million inpatient and outpatient notes, University of Pennsylvania Health System, 2015 to 2020	50.1% of all text duplicated, rising from 33.0% in 2015 to 54.2% in 2020; 45.9% of duplicated text came from a different author

In Rule's data, notes written in 2018 averaged 29.4% directly typed text; the rest was templated or copied.

For trial data, two findings stand out. Records with more notes approached 60% duplicate text in Steinkamp's data, and a multi-cycle regimen produces many notes. Nearly half of duplicated text came from a different author, so a sentence in today's note may not be its signer's observation.

How does copy and paste in clinical notes affect adverse event data?

Copy and paste in clinical notes can move an adverse event's onset date, grade trajectory, attribution and end date, because a carried-forward sentence reads as a new observation at every visit that repeats it.

Progress notes are part of the trial record. Under 21 CFR 312.62(b), an investigator must keep adequate and accurate case histories, and the regulation names the physician's progress notes as part of them.

ICH E6(R3), adopted on 6 January 2025, says source records should be attributable, legible, contemporaneous, original, accurate and complete. A copied sentence is contemporaneous with the visit where it was first written, not with every visit that repeats it.

FDA's July 2018 guidance on using EHR data in clinical investigations asks sponsors to consider the reliability and quality of unstructured EHR data before using it as critical source data.

Four properties depend on which visit the supporting text belongs to:

1. Onset date. The start date decides whether an event is treatment-emergent, which cycle it is counted against, and the time relationship that causality assessment weighs. An event whose supporting sentence first appeared four weeks earlier has an onset the current note does not establish.
2. Grade trajectory. CTCAE grades are ordinal, so a change in grade between visits is itself information. If one sentence supports the grade at every visit, the log shows a plateau whether or not one exists.
3. Attribution. The WHO-UMC causality system separates "certain" and "probable" from "unlikely" partly on the time between drug intake and onset, so a moved onset can move the category. Attribution is already unstable: in the placebo arms of two phase III trials, when the same adverse event was reported for the same patient on multiple visits, its category changed at least once 36% and 31% of the time (Hillman et al., Journal of Clinical Oncology, 2010).
4. End date and duration. A symptom carried forward past its resolution keeps the event open. The end date arrives late, the duration grows, and the event can read as ongoing at a data cut.

Copying also hides events. A review of systems carried forward with "denies nausea" can sit in the note at the visit where nausea began, so an abstractor finds no event to record. By the ICH E6(R3) standard, that copied negative is neither accurate nor complete for its visit.

An example: one fatigue sentence across four visits

An illustration, not patient data: at cycle 2, a participant's note says fatigue is not relieved by rest and limits instrumental activities of daily living: grade 2 Fatigue under CTCAE v6.0. The sentence is carried into the next three notes.

Visit	Fatigue text in the note	Other evidence at that visit	What an abstractor reading each note alone records
Cycle 2, day 1	Written new: not relieved by rest, limiting instrumental ADL	Participant describes it at the visit	Grade 2
Cycle 3, day 1	Same sentence, carried forward	Participant's check-in says fatigue is relieved by rest	Grade 2
Cycle 4, day 1	Same sentence, carried forward	Regimen changed after cycle 3; no new fatigue assessment	Grade 2
Cycle 5, day 1	Same sentence, carried forward	Participant's check-in says fatigue is relieved by rest	Grade 2

The log shows grade 2 fatigue that never moves. The record supports grade 2 at cycle 2, grade 1 (relieved by rest) at cycles 3 and 5 if a clinician confirms the participant's account, and no assessment at cycle 4. Which reading is right is the investigator's question, and it only gets asked if someone notices the sentence never changed.

How to check a note for copied-forward findings

For a coordinator abstracting events or a monitor doing source data verification:

1. Find the first appearance. Search earlier notes for the exact sentence that supports the grade. The first note containing it is the earliest source to check for the onset date.
2. Compare the wording across visits. Identical text is a prompt, not a verdict. Some events genuinely persist: a chronic neuropathy, graded as Peripheral sensory neuropathy, is often written the same way at each visit because it has not changed.
3. Look for new data at this visit. A new exam finding, laboratory value, vital sign or patient statement supports a current grade; a repeated sentence alone does not.
4. Check what changed around the event. A grade that holds while the regimen changed, or across a long interval, deserves a second look, as does a copied negative beside a newly reported symptom.
5. Read the patient's own report. A patient-reported symptom that moves while the clinician's grade stays put is independent evidence the note was not reassessed.
6. Use the EHR's provenance. Keeping copied text's provenance available is one of the Partnership's four safe-practice recommendations; where the EHR shows it, use it. FDA's EHR guidance says FDA may request EHR audit trail information during an inspection.
7. Raise a query when the source looks wrong. The investigator decides whether copied text was accurate for that visit, and a correction goes in as a dated, attributable addendum: ICH E6(R3) says changes to source records should be traceable and should not obscure the original entry.
8. Record the decision. Note which source line supports the grade, which visit it came from and who decided, so the reasoning is still there at the monitoring visit.

How Burna AI checks notes for copy-forward

Burna AI's copy-forward detection, part of its Documentation Quality Assessment, reviews each CTCAE term across a patient's graded visits and flags a term only when two or more of five signals are present:

• Grade stagnation: the same grade for three or more consecutive visits.
• Evidence repetition: near-identical supporting text from visit to visit.
• Temporal implausibility: the grade holds while treatment changed or significant time passed.
• Missing reassessment: no new clinical data supports the current grade.
• Patient-reported divergence: the patient's self-report shows the symptom shifting while the clinician's grade stands still.

Self-reports come from symptom check-ins in Healthnotes, the participant app a sponsor deploys through its sites; with none, the divergence signal is recorded as not available rather than read as agreement. By default, flags scoring below 60 out of 100 for confidence are held back and detection starts once a patient has three graded visits; both thresholds are configurable. Once a term is flagged, the patient's next grading run lowers the confidence on that term's grade and keeps the unadjusted confidence on the event for audit.

Every grade traces to the line in the medical record that produced it, with CTCAE v5.0 or v6.0 grading and the criterion attached. A flag lowers confidence and leaves the grade to the clinician. AI suggests, clinicians decide. Human-in-the-loop, always. Copy-forward detection shipped in March 2026 with four signals and added patient reports in May 2026.

Common questions about note bloat and copy-forward

Is copy-forward allowed in clinical documentation?

Generally, yes: published guidance governs copy-forward rather than banning it, assigning responsibilities to authors, organizations and EHR developers (Tsou et al., 2017), and each health system sets its own policy. For trial records, the question is whether each sentence is accurate for the visit it is filed against.

How do you record an adverse event onset date when the note was copied forward?

Find the earliest source that documents the event, and record the start date that source reports (for example, "since last Tuesday") rather than the visit date when it gives one. Follow the protocol and the case report form completion guidelines on whether a grade change opens a new entry, and when the source is unclear, ask the investigator.

Does a repeated sentence mean the grade is wrong?

No. A chronic adverse event can hold one grade for months, documented the same way each time. Repetition becomes a problem when nothing else at the visit supports the grade, treatment changed around it, or the patient reports something different.

If your coordinators spend monitoring visits working out which visit a finding belongs to, book a 15-minute conversation.