AI takes over medical records at university hospital, saving doctors up to 30 days a year

Lee Min-woo, a neurology professor at Hallym University Sacred Heart Hospital, demonstrates the use of an artificial intelligence system to write electronic medical records. Courtesy of Hallym University Medical Center

At Hallym University Sacred Heart Hospital in Anyang, Gyeonggi Province, writing patient medical records now takes just seconds.

When neurology professor Lee Min-woo clicks the “HAI” button at the top of the hospital’s electronic medical record (EMR) program, the day’s progress notes appear instantly, complete with sections on the patient’s reported symptoms, examination findings, clinical assessment and future care plans.

After a quick review and final approval, one day’s record for an inpatient is complete.

The change is driven by Hallym University Medical Center’s generative artificial intelligence (AI) platform, HAI, which automatically drafts EMRs.

While some hospitals have used AI for parts of the process, this is the first in Korea to apply it across the entire inpatient care cycle, from admission to discharge.

The move comes amid a shortage of residents, who traditionally handled most recordkeeping, following last year’s labor disputes between doctors and the government.

To build the system, the hospital linked 97 categories of patient data from its information system, including nursing notes and major test results from the past 24 hours, to generate a draft.

The AI was trained by comparing its output with records written by doctors, fine-tuning it to match the style and detail clinicians expect.

“We’re continually refining it so the notes come out exactly as we want without any changes,” said Lee, head of the HAI task force.

Currently used in neurology, obstetrics and gynecology, and otolaryngology for select conditions, the system will soon expand hospital-wide.

In neurology, where writing an inpatient EMR takes about 10 minutes, Lee estimates the AI saves him 90 minutes a day, equivalent to 20 to 30 days a year.

Hospital officials project it could save about 83,000 hours annually if used across all departments. “That time can go toward seeing more patients, research or teaching,” Lee said.

The hospital chose to run HAI on an “on-premises” system using a large language model from Korean AI firm Konan Technology, rather than cloud-based services like ChatGPT or Gemini, to protect sensitive patient data.

“Because nothing leaves the hospital network, there’s no risk of patient information being exposed outside,” Lee said.

This article from the Hankook Ilbo, the sister publication of The Korea Times, is translated by a generative AI and edited by The Korea Times.