(C-112) Alignment Between Multiple Myeloma Line of Treatment Algorithms Developed for Structured Healthcare Data and an Oncology EHR

Background: Real-world studies on the safety and effectiveness of oncology treatments may require diverse databases; oncology electronic health records (EHR) can capture effectiveness with more validity, while structured claims or EHRs with healthcare data beyond the oncology clinic can capture safety events more comprehensively. When both datasets are required but linkage is infeasible, study approaches must be closely aligned in the different databases. In particular, the line of treatment (LOT) algorithm used to identify study treatment and index date in the structured data should reflect
the abstraction-informed LOT algorithm in the oncology EHR.

Objectives: To provide metrics on the alignment between a multiple myeloma (MM) LOT algorithm designed for structured data and an abstraction-informed MM LOT algorithm from an oncology EHR.

Methods: This study was conducted in US patients newly diagnosed with MM from 2011 through 2024 who had LOT information in approximately 280 US cancer clinics contributing to the deidentified Flatiron Health (FH) Core MM Registry database, which is from an oncology EHR. MM diagnosis served as the LOT algorithm index and was based on codes (ICD-9 203.0x, ICD-10 C90.0x, C90) and manual chart review. An LOT algorithm for structured healthcare data was developed using the OMOP Common Data Model (CDM) (OMOP algorithm) and was based on existing rules for the FH LOT algorithm (FH algorithm); different rules were required for identifying oral medications, corticosteroids, and procedures. Both LOT algorithms were applied to the FH data, which was transformed into the OMOP CDM. Metrics comparing the structured versus the oncology EHR LOT algorithms included: difference in line numbers, percent with aligning start time by line, difference in start dates by line, and difference in line length.

Results: There were 17,347 patients identified in the FH data, with 58,170 LOTs identified using the FH algorithm and 49,745 LOTs using the OMOP algorithm. The median number of LOTs per person in the FH algorithm was 2 versus 1 for the OMOP algorithm. The median length of time for the LOTs was 168 days in the FH algorithm vs. 132 days for the OMOP algorithm. Among patients with LOTs identified by the OMOP algorithm, 71% had a first line start within 30 days of the FH algorithm. The median difference between first line start time of the OMOP and FH algorithm was 0 days (IQR: 25 days) .

Conclusions: In this comparison of LOT algorithms, the structured OMOP algorithm identified fewer LOTs and found a shorter LOT length than the FH algorithm. Differences could arise due to chart-abstracted information informing the FH algorithm, different groupings of multi-phase treatments into a line, and different definitions of treatment continuity, for example, carrying a line forward until the next line starts. Future studies will assess validity versus clinician-assigned LOT.