Seoul St. Mary's Hospital develops AI to screen multiple myeloma patients for clinical trials

Researchers at Seoul St. Mary's Hospital Blood Cancer Center have developed AI-powered software that automatically screens multiple myeloma patients for clinical trial eligibility — cutting a process that clinical research coordinators once completed by hand over roughly two weeks down to a matter of seconds.

The hospital announced Wednesday that the software, developed under the leadership of hematology Professor Park Sung-soo, has been deployed at the Multiple Myeloma Center.

The team built it without external funding or support from other departments, starting development in April and completing a fully operational version in about two months. The team has also filed a patent application and registered a program copyright.

Multiple myeloma is a refractory blood cancer in which plasma cells in the bone marrow proliferate malignantly. Because the disease is prone to relapse and treatment resistance, drug development and clinical trials are active in the field. Previously, however, once a physician announced an open trial, coordinators had to manually review each patient's electronic medical records and check them against enrollment and exclusion criteria — a process that typically took two weeks.

Modern clinical trials demand not only detailed treatment histories — including the number of prior therapies, refractoriness to specific drugs, stem cell transplant experience, and time elapsed since the last treatment — but also real-time lab conditions such as kidney and liver function, blood counts, and infection status. Criteria can also vary by patient subgroup within a single trial, leaving room for human error and omissions.

The software automates the screening process in three stages. First, it compiles electronic medical records into a patient-by-patient database, organizing treatment counts, drug refractoriness, and stem cell transplant history. A large language model then reads the trial's inclusion and exclusion criteria and converts them into a machine-comparable format, also classifying which patient subgroup each individual belongs to. Finally, the system cross-references each patient's latest test results against the trial criteria.

The software automatically calculates double and triple refractoriness based on a patient's response to multiple drug classes, including proteasome inhibitors, immunomodulatory agents, and anti-CD38 antibodies. Patients with missing test results are flagged as "pending" rather than immediately marked ineligible, allowing the system to reassess them once new results become available. Patients who have already died are automatically excluded from the eligible pool.

The software's defining feature is its ability to cross-reference treatment history, the latest lab results, and trial-specific eligibility criteria all at once, fully automating the screening process.

Results are sorted into three categories — eligible, pending, and ineligible — with each patient's record showing which criteria were met and which were not, along with supporting evidence. Outputs are also generated as summary reports and Excel files. In practice, the system scanned a database of several thousand patients in tens of seconds and assessed a single patient in detail in just a few seconds.

The team ran more than 800 validation tests on the core logic and is conducting a separate study to verify clinical accuracy. The Multiple Myeloma Center has so far registered the criteria for 18 clinical trials and is running the software across all of them.

"Multiple myeloma is the blood cancer where new drugs are introduced most rapidly, but connecting patients to those opportunities has long depended on human hands and time," said hematology Professor Min Chang-ki. "This software was made possible by the asset we have built over many years — our multiple myeloma patient registry."

Meanwhile, the research team plans to finalize the clinical accuracy validation study while expanding the software's scope to other blood cancers and exploring integration with the hospital's in-house clinical research management system.