OpenAI revealed that ChatGPT Health has been integrated with Epic's electronic health record system, allowing doctors to import patient data while complying with regulations, and to use AI to query and organize medical records. Epic covers over 325 million patients, and this integration signifies that ChatGPT is further entering the mainstream American healthcare workflow.
Can read medical records but cannot write them back.
According to OpenAI, doctors can retrieve appointment records, test results, medication information, and specialist documents through ChatGPT, and quickly generate summaries. The system can also help to organize the patient's past medical history, identify any changes, and prepare for subsequent consultations.
In some deployments, doctors can directly use ChatGPT to complete pre-consultation reviews and organization of clinical timelines without leaving the patient's medical record page. OpenAI emphasizes that this access currently only provides read-only permissions, and AI does not write any content back into the medical record system.
Add a new medical public data plugin
In addition to medical record integration, OpenAI has also launched the Healthcare Public Data plugin, which can extract information from multiple public medical databases, including ClinicalTrials.gov, CMS Coverage, RxNorm, DailyMed, and PubMed.
The company stated that this plugin can be used to integrate information such as trial enrollment criteria, drug identification, versions of medical insurance coverage policies, and records of healthcare providers, helping medical staff to complete data retrieval and summarization more quickly.
Expand the scope of use for compliance workspaces
OpenAI also indicates that institutions that have been authorized to sign Business Associate Agreement are allowed to use ChatGPT Work, Codex, applications, and connectors within their work areas to support medical processes that comply with regulatory requirements.
The company also disclosed that it has collected over 4,300 feedbacks from doctors regarding 27 clinical use scenarios, including pre-visit reviews, clinical timelines, medication reviews, and handover summaries. Among these responses, 99.1% were assessed as safe.
However, OpenAI also mentions that even though the proportion of insecure responses is very low, they can still cause actual harm in medical scenarios. This means that while expanding access, security control will still be a key focus of deployment in the medical AI.
