NHS AI scribes misrecord drug names and diagnoses, Healthwatch warns
Healthwatch England, an NHS watchdog, has warned that AI scribes deployed across the NHS are misrecording drug names, diagnoses, and clinical information in patient consultations, with errors more often detected by patients than by healthcare professionals. The watchdog cited cases where incorrect diagnoses and omitted medication instructions appeared in records and called for clearer processes for patients to report and correct AI-generated errors.
What happened
Healthwatch England warned that AI scribes deployed by the NHS to transcribe patient consultations with doctors are making errors in recording drug names, diagnoses, and clinical information.
In one case, an AI scribe incorrectly recorded that a patient had demyelination, a serious nerve condition associated with multiple sclerosis, when the correct record should have stated 'null demyelination' following an MRI scan. In another case, an AI scribe confused the name of a prescribed drug with a different drug of similar name. A third case involved an AI-generated summary letter that omitted the consultant's instruction to seek a repeat prescription for migraine medication.
Healthwatch reported hearing multiple accounts from patients who identified errors in AI-transcribed records that were not caught by healthcare professionals.
Reported impact
Patients reported emotional distress from incorrect diagnoses appearing in their medical records. One NHS health professional described the experience of being given an incorrect diagnosis as 'very traumatising', particularly after being told the error was 'a typo'.
Healthwatch warned that unless such errors are detected by patients, they could persist in medical records and affect future patient care. The omission of prescription instructions could leave patients unable to obtain necessary medication.
Index analysis
The errors appear to stem from the AI system's difficulty in accurately transcribing or interpreting clinical terminology and medication names, particularly when similar terms or drug names are involved.
Multiple independent detection of errors by patients rather than healthcare professionals suggests the AI system may be producing transcriptions that are plausible enough to escape initial review by busy clinicians.
Governance controls
Healthwatch called for clarity on how patients can report and obtain corrections to errors made by AI scribing tools or the professionals using them.
The organisation noted that the Medicines and Healthcare products Regulatory Agency's guidance on AI scribes required investigation and oversight.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.