OBJECTIVES: As electronic health records (EHR) do not explicitly capture lines of therapy (LoT) received by patients, algorithms aim to automatically extract this information. However, LoT algorithms depend on several arbitrary assumptions. We aim to validate a LoT algorithm in patients with a gynaecological cancer diagnosis using EHR.

METHODS: This validation study uses de-identified EHR for Ovarian and Endometrial cancer patients diagnosed between 2007 and 2023 from UK NHS partners collated as part of the Arcturis Data Platform. Upon random allocation of patients to either a baseline or refinement cohort, we will test variations of the algorithm’s parameters. Appraisal of various parameter combinations will be expressed via the percentage agreement of the number of lines and regimen definition between the output and a gold standard of blinded manual review. Anticipated areas of disagreement between manual review and the algorithm include the 120-day gap in treatment administrations that would advance a line; the 30-day window grouping agents that would indicate a combination therapy line; and differences in the regimen definition, especially if a drug is discontinued shortly after the commencement of a line.

RESULTS: From a total of 4,757 newly diagnosed ovarian cancer patients, and 3,884 newly diagnosed endometrial cancer patients, 1582 and 627, respectively, received chemotherapy. Preliminary analysis indicates that changes to the 120-day gap in therapy to between 90 and 180 days alters the percentage receiving only one therapy line by <2%; reducing the gap to 30 days minimally increases the percentage receiving 5+ lines by 4%. Likewise, algorithm output is robust to variations in the 30-day line regimen window with negligible difference in output between 30- and 90-days.

CONCLUSIONS: Initial results indicate minimal sensitivity of the algorithm to reasonable variations in the gap in therapy and line regimen window. Further validation as per the methodology is needed.