OBJECTIVES: Answering complex questions with real-world evidence requires identifying appropriate data. Insurance claims contain resource utilization and costs, but can lack clinical details like line of therapy (LOT). Medical records may contain clinical information but are less likely to have costs. The objective was to validate an algorithm for LOT in ovarian cancer (OC) claims, and compare the occurrence of specific clinical events between claims and medical records.

METHODS: Utilizing the Optum Research Database, a claims-based algorithm to identify LOTs and type of therapy (active vs maintenance) among OC patients (diagnosed and received chemotherapy 12/1/14-9/15/17) was developed. A retrospective claims analysis linked with abstracted medical record data from patients’ oncologist was used for validation. Percentages were used to compare events and procedures between the claims and record data, while Cohen's kappa statistic was used to measure level of agreement for LOTs.

RESULTS: In total, 294 patients were included. The algorithm demonstrated substantial agreement between claims and records for total number of lines of active and maintenance therapy (weighted Kappa 0.65 and 0.62 p<0.0001). Records identified 189/294 (64.29%) patients as advanced OC (stage III/IV). Compared to records, claims identified higher 24-month mortality (19.22% vs 3.58%) and surgery during 33 months of follow-up (47.03% vs 29.33%). For most clinical events including fatigue, nausea, and hematologic events, records identified more patients with events during follow-up.

CONCLUSIONS: Claims-based algorithms can reliably identify LOT in claims databases. Claims, which contain information from all patient interactions with the healthcare system, and not just a single oncologist, provided strong data on procedures and mortality compared to medical records. Medical records more accurately identified specific clinical events that may not appear in claims due to a lack of billable services. This research provides insights on the strengths and opportunities to consider when selecting specific datasets for oncology research.