Health systems are increasingly experimenting with and implementing chatbots built on large language models to search and synthesize information from patients’ electronic health records (EHRs). The accessible portion of the STAT News report describes growing interest in these tools as a way to help clinicians navigate voluminous and fragmented records.
The article describes a specific clinical vignette at Stanford involving a tool named ChatEHR. Pathologists reviewing a lymph node biopsy had been unable to identify the patient’s cancer even after extensive staining. A physician used ChatEHR to query the patient’s record about prior skin lesions. After iterative queries, the chatbot surfaced a prior diagnosis from a different health system: sarcomatoid squamous cell carcinoma. The clinician reported that this finding explained the lymph node pathology and praised the tool’s value.
A key driver for adoption is the difficulty clinicians face in locating relevant information within modern EHRs. Clinical records have expanded in length and complexity, making it time-consuming to find historical diagnoses, procedures, or other critical details. According to the article, providers hope that generative AI–powered chatbots can act as tools to surface “needles in haystacks” buried in long records, improving diagnostic workflows and saving clinician time.
The report notes that health systems are pursuing both internally developed (“homegrown”) chatbots and vendor-built products to integrate AI-supported querying and summarization into clinical workflows. The accessible text references ChatEHR as one named example; beyond that, the paywalled portion of the piece likely discusses additional tools and vendors, but those specifics were not available in the provided source excerpt.
Reported benefits include faster retrieval of relevant clinical details and the potential to reveal important prior diagnoses or findings that clinicians might otherwise miss. The anecdote from Stanford illustrates how a chatbot can synthesize disparate documentation from multiple health systems and present an actionable finding that altered the clinical interpretation of tissue pathology.
While the accessible portion highlights promising use cases, it also emphasizes that these systems require ongoing oversight. The article’s subtitle states that the tools “need persistent monitoring,” signaling concerns about reliability, accuracy, and safe integration into care. Specific monitoring strategies, error rates, adjudication processes, or governance approaches were not included in the available excerpt.
The STAT News story is behind a STAT+ paywall; the provided source text is truncated. As a result, several important operational and clinical details were not reported in the accessible excerpt, including but not limited to:
Readers seeking those specifics will need to consult the full STAT+ article or additional reporting.
The accessible portion of the STAT News report documents early adoption of EHR chatbots powered by generative AI, illustrates a notable clinical anecdote with ChatEHR, and frames the technology as promising for saving clinician time and surfacing buried diagnostic information. It also underscores the need for persistent monitoring. Because the piece is paywalled, comprehensive data, broader examples, and operational details were not available in the provided source material.