Health systems are experimenting with artificial intelligence tools that can probe and condense patients’ often-bloated electronic health records (EHRs), according to reporting by STAT. Several hospitals have deployed proprietary and vendor-built chatbots powered by large language models to answer clinicians’ questions and surface information that may otherwise be buried.
Clinical benefit: finding needles in haystacks
The promise is illustrated by a case at Stanford where a physician used one such tool, known as ChatEHR, to search a patient’s records. The chatbot identified a prior diagnosis — sarcomatoid squamous cell carcinoma — recorded in another health system, information that helped explain puzzling lymph node findings and influenced clinical reasoning, the STAT report says.
“If that doesn’t prove the value of ChatEHR, I don’t know what does!”
Proponents say the technology can save clinicians time, reduce the cognitive burden of sifting through lengthy charts and occasionally reveal critical historical details that change diagnoses or management plans.
Not a panacea: accuracy, hallucination and context
Despite promising anecdotes, the tools are not infallible. Large language models can produce confidently worded but incorrect answers, a phenomenon often called hallucination, and may misinterpret clinical nuance. The STAT excerpt notes the technology’s ability to synthesise records but stops short of providing systematic validation data in the public reporting available.
Clinicians and health system leaders face hard questions about when to trust model outputs, how to validate them, and who is responsible for errors that arise from AI-assisted decisions. Current evidence remains limited and largely anecdotal in the publicly available account; rigorous evaluation in diverse clinical settings is essential before broad roll-out.
Operational and governance challenges
Implementation touches on multiple operational issues:
- Data integration: Many EHRs are fragmented across institutions, limiting tools’ ability to access a full longitudinal record unless robust health information exchange exists.
- Privacy and security: Use of third-party AI systems raises concerns about patient data handling, storage and compliance with data protection laws.
- Clinical workflow: Tools must fit into clinicians’ routines without adding verification burdens or alert fatigue.
These questions are particularly pertinent in contexts with mixed digital maturity across public and private sectors. Any AI application that queries sensitive health data must meet local legal and ethical standards.
What clinicians and patients should watch for
Health systems exploring or piloting chatbot search tools should prioritise:
- Independent validation studies comparing AI outputs with human review and real-world outcomes.
- Transparent documentation of model capabilities, limitations and provenance of training data.
- Clear governance frameworks that delineate responsibility for errors and processes for clinician oversight.
Adoption without these safeguards risks delegating clinical reasoning to opaque systems and may amplify biases present in source records or model training sets.
Balancing innovation with patient safety
The STAT report captures a broader trend: health systems are attracted to generative AI for its potential to make sense of voluminous records and speed up information retrieval. At the same time, scant public evidence about performance across different patient populations and clinical contexts counsels caution.
For patients and clinicians, these systems should be seen as decision-support tools rather than replacements for clinical judgement. If you have concerns about how your records are used, or if an AI-assisted decision affects your care, discuss this with your clinician or the health facility. Always consult a healthcare professional rather than relying on algorithmic summaries for diagnosis or treatment.
The STAT piece provides a snapshot of early adoption and debate; robust, peer-reviewed research and transparent governance will be needed to convert promising anecdotes into safe, equitable practice.
