How do electronic health records and AI work together?
The AI Glossary
In our classes, we often pause to explain the ideas behind AI and the language used to describe it. These posts grew out of those conversations. They’re for anyone who would like to understand AI better, with no technical background assumed.
An electronic health record, or EHR, is a digital system used to document and manage information about patient care. It may contain notes, diagnoses, medicines, test results and other records. AI can use information from the EHR, help people work with it or supply outputs that become part of the clinical workflow.
These are separate roles. A model trained on historical records learns from earlier examples. A model reading a record during use receives current context. An application that writes a draft note changes the information available for later care. The risks and permissions differ at each stage.
EHR data is not a complete, neutral account of a person’s health. It reflects what was measured, what was documented, where care occurred and which systems were connected. A missing diagnosis may mean no disease, no assessment or no accessible record. Repeated text may reflect copying rather than a new observation.
Information exchange also requires more than connecting two computers. Systems must agree on identity, dates, units and the meaning of fields. Interoperability means being able to exchange information and use it meaningfully. Shared standards help, but they do not automatically resolve conflicting or incomplete records.
AI can assist with finding relevant material, preparing summaries and extracting specified details. Those outputs should remain distinguishable from the underlying observations. A generated statement can otherwise be copied into later notes and acquire apparent authority simply through repetition.
For a referral service, an assistant might locate the latest relevant report and prepare a concise summary for review. It should show the source and distinguish a confirmed result from a question or historical diagnosis. Before anything is written back, the organization needs clear permissions, ownership and a correction process. The real challenge is not merely giving a model access to the EHR. It is preserving the meaning and provenance of information while helping people complete the next part of care.
Sources: ASTP/ONC — Health IT Playbook: Electronic Health Records; Panch, Mattie and Celi — The “Inconvenient Truth” About AI in Healthcare (2019); ASTP/ONC — Decision Support Interventions Certification Companion Guide
To learn more, take a look at the AI in Health Care Certificate of Specialization from the Advanced Learning Academy at Harvard T.H. Chan School of Public Health. You can take an individual course or work toward the full certificate.
AI is a fast-moving field. This explanation reflects our class discussions and related material at the time of writing. Terminology, capabilities and tools may change.
For suggested additions or corrections, please email me at tpanch@hsph.harvard.edu.
Dr. Trishan Panch
Program Director, AI in Health Care
Harvard T.H. Chan School of Public Health — Advanced Learning Academy