Words that carry bias in clinical records
Aggressive, agitated, non-compliant. The research on descriptors in health records is uncomfortable, and the fix produces better notes as a side effect.
Most clinicians accept in principle that language in records matters. Rather fewer have seen the size of the effect, and it is larger than intuition suggests.
This is not about politeness. It is about accuracy, and about what the next clinician does after reading your note.
What the research found
The landmark study is Sun, Oliwa, Peek and Tung, Negative Patient Descriptors: Documenting Racial Bias In The Electronic Health Record, published in Health Affairs in January 2022.
The team analysed 40,113 history and physical notes covering 18,459 adult patients at a large urban academic medical centre, written between January 2019 and October 2020. They searched for fifteen descriptors and compared how often each appeared by patient ethnicity.
Compared with White patients, Black patients had 2.54 times the odds of having at least one negative descriptor in their history and physical notes (95% confidence interval 1.99 to 3.24).
The fifteen flagged descriptors were: (non-)adherent, aggressive, agitated, angry, challenging, combative, (non-)compliant, confront, (non-)cooperative, defensive, exaggerate, hysterical, (un-)pleasant, refuse, and resist.
A separate study in JAMA Network Open found stigmatising terms such as "nonadherence", "belligerent" and "abuser" appearing more often in the records of Black patients. Psychiatric work has shown related patterns: "agitation" applied where "anxiety" was used for other patients showing similar behaviour, and a London NHS study found ethnic differences in the use of "acute behavioural disturbance".
None of these studies suggest clinicians were choosing words maliciously. They describe an effect operating below the level of deliberate choice, which is precisely why a checklist helps.
Why wording changes care
A record is not a passive archive. It is read by the next clinician before they meet the person, and it shapes what they expect.
Goddu and colleagues demonstrated this directly. They gave clinicians the same case with only the language altered — one version using stigmatising phrasing, one neutral — and found that those who read the stigmatising version held more negative attitudes towards the patient and managed their pain less aggressively. Same patient, same facts, different words, different treatment.
In psychiatry the stakes attach to different decisions: whether someone is approached or contained, how a disagreement is interpreted, whether declining medication is read as a decision or a symptom. A word that primes the reader towards threat can shape all of them before anyone has said hello.
Records are cumulative. A descriptor entered once is read, absorbed, and often echoed in the next entry. Within a few admissions a single loosely chosen word can become an established characteristic that nobody can trace back to a source.
The words that carry the most weight
Not banned words. Words worth a second's thought, because each compresses an observation into a judgement and loses information doing it.
| Word | What it loses | What to write instead |
|---|---|---|
| Aggressive | What actually happened, and towards whom | The behaviour: raised his voice, stood abruptly, moved towards the door |
| Agitated | Whether this was distress, restlessness, akathisia or fear | Pacing, unable to remain seated, wringing hands, tearful and pacing |
| Non-compliant | Whether it was a decision, an oversight, a side effect, or a supply problem | Declined the medication offered; has not collected the prescription; stopped it after experiencing sedation |
| Refused | That declining treatment is a right | Declined; did not agree to |
| Uncooperative | What was asked and what happened | Did not wish to answer questions about his family; left the room after ten minutes |
| Manipulative | Everything. It is an inference about motive | What the person asked for, and what they did |
| Attention-seeking | Same problem, with contempt attached | The behaviour, and what they said they wanted |
| Denies | Neutrality — it implies disbelief | Reports no…; said he had not… |
| Claims | Same | Says; describes; reports |
| Poor historian | Whether the difficulty was the person's or the interview's | Was unable to give a sequential account; the account was difficult to follow |
"Denies" deserves a note of its own. It is so embedded in clinical writing that most people no longer hear it, but outside medicine it carries a strong implication of disbelief. "Denies chest pain" is harmless. "Denies the incident" is not, and in a psychiatric record where the person's credibility is already at issue, the difference is worth caring about.
Rewriting without losing the finding
The objection to all this is that neutral language is vaguer. Done badly, it is. Done well, it is more specific, because it forces you to say what happened.
Patient was aggressive and uncooperative throughout the assessment. Non-compliant with medication.
He raised his voice twice, stood up abruptly when his family were mentioned, and sat back down when asked. He did not wish to answer questions about his housing. He stopped taking his medication two weeks ago, describing sedation that made work difficult.
The second is longer. It is also the only one from which a reader can work out what to do — that the trigger was family, that he responded to a request, and that the medication problem is a side effect with a plausible solution rather than a refusal.
She was manipulative and attention-seeking, presenting to A&E for the third time this month.
This is her third attendance this month. She asked to be admitted and said she does not feel safe at home. She became tearful when discharge was discussed.
When to quote instead
Where the exact words matter — a description of mood, an account of an unusual belief, an explanation of why medication was stopped — quoting removes the interpretive step entirely.
He said the police "have been targeting me because of my race". He was clear that he does not accept the diagnosis: "I know they think I am ill, but I have never felt better in my life."
A paraphrase of the first sentence — "expressed persecutory ideas about the police" — is accurate but loses what the person believes and why, and it converts an allegation that may be true into a symptom. Quoting keeps the reader's judgement open.
What this is not asking you to do
It is not asking you to soften findings, avoid recording violence, or write euphemistically. If someone assaulted a member of staff, that goes in the record plainly, with what happened.
It is not asking you to avoid clinical terminology. "Blunted affect", "flight of ideas" and "thought blocking" are technical descriptions with agreed meanings, and they belong.
It is not asking you to omit the person's own words because they are unflattering.
When you reach for a word that summarises a person rather than describing an event, spend a second replacing it with what happened. That is the whole intervention, and it usually produces a better note as a side effect.
A short checklist
- Does this word describe an event, or characterise a person?
- Could a reader work out what actually happened from what I have written?
- Have I recorded who reported it, where it was not my own observation?
- Am I inferring a motive? If so, can it come out?
- Would I be comfortable with the person reading this sentence? Under the Data Protection Act 2018 they are entitled to, and increasingly they do.
- Would a quotation be more accurate than my paraphrase here?
The last question is the most useful. Most sentences that read badly in a psychiatric record are paraphrases that should have been quotations.
Common questions
What did the Health Affairs study on negative patient descriptors find?
Sun, Oliwa, Peek and Tung analysed 40,113 history and physical notes covering 18,459 adult patients at a large urban academic medical centre. They found that Black patients had 2.54 times the odds of having at least one negative descriptor in their notes compared with White patients, with a 95% confidence interval of 1.99 to 3.24.
Is 'non-compliant' acceptable in clinical notes?
It is widely used but collapses several different situations into one judgement. A person may have decided against the medication, run out of it, stopped because of side effects, or never received it. Recording which of those happened is both more accurate and more actionable than the label.
Should I stop using the word 'denies'?
It is entrenched and often harmless in physical medicine. In psychiatric records, where the person's credibility may already be in question, it carries an unhelpful implication of disbelief. 'Reports no thoughts of self-harm' says the same thing without the implication.
Does neutral language mean writing less clearly about violence?
No. If an assault occurred it belongs in the record plainly, with what happened, who was involved and what followed. The guidance concerns words that characterise a person rather than describe an event. Describing the event is usually clearer than labelling the person.
Can patients read what I write about them?
Yes. Under the Data Protection Act 2018 and UK GDPR people have a right of access to their health records, and many services now offer direct access as standard. Writing as though the person will read it is both a legal reality and a useful discipline.