A comprehensive Memorial Sloan Kettering Cancer Center real-world data model: Core clinical data elements.

Abstract

e18755 Background: The 2016 21st Century Cures Act supports the use of Real-World Data (RWD) for regulatory decision/approval. Due to technological advances, a vast amount of health-related data are now available, but most are not standardized nor readily useable for research. Also, currently available standardized RWD models are not applicable across cancer types or oncology specialties (surgery, medical oncology, radiation oncology, pathology, radiology, etc.). To address these deficiencies Memorial Sloan Kettering Cancer Center (MSKCC) built a comprehensive, pan-cancer, pan-specialty RWD model. Methods: The Core Clinical Data Element (CCDE) data model incorporates aspects of existing academic and biopharma data models, including PRISSMM framework, ASCO’s mCODE, and NAACCR tumor registry model. The data model encompasses 11 domains that are critical to the understanding of the patient’s cancer journey, including: demographic, comorbidities, diagnosis, pathology, imaging, genomics, cancer surgeries, radiation oncology treatments, medical oncology treatments, cancer status/progression, and additional health information. To align with current standards, we are using ICD-10, ICDO3, CTACE V5.0, HL7, SNOMED and LOINC code sets. Further, this adaptable model allows for 5-10 disease specific elements to accommodate for disease heterogenicity and capture the differences among cancer types. Results: The CCDE database includes 1,126 of total data elements. MSKCC has 52,704 patients with MSK-IMPACT (Next-Generation sequencing platform with 505 genes panel) testing of which, we have identified 1,132 bladder cancer patients with at-least one year of cancer care follow-up for the initial curation cohort. Patients were identified as having an OncoTree bladder tumor type code that is assigned by a pathologist who attests the diagnosis by reviewing results from clinical tests on tumor specimens. To the date, 641 patients including 46,415 curated forms have been curated (Table). Conclusions: The comprehensive MSKCC’s CCDE data model standardizes the common and critical pan-cancer and pan-specialty elements for RWD. The dataset resulting from this curation efforts will provide robust structured and unified genomic and phenomic data across tumor types for future research enabling greater collaboration across various cancer types as well as oncology specialties.[Table: see text]

Artificial Intelligence–Assisted Cancer Status Detection in Radiology Reports

“…The IRB protocol number is 22-106. We have approximately 350,000 radiology reports which were curated by human labelers to extract fields in the PRISSMM data model ( 1 ). A nonsignificant number of curated radiology reports of these patients originate from external providers and are not considered for this study.…”

“…Memorial Sloan Kettering Cancer Center (MSKCC) currently has approximately 100,000 patients with genomic testing (IMPACT; ref. 1 ) spanning all cancer types and continues to accrue more every day. Clinicians use this genomic data for research but lack structured clinical data to analyze alongside the genomic data.…”

Computational methods and translational applications for targeted next‐generation sequencing platforms

“…Data sharing protocols will also become more robust through the use of machine learning and natural language processing approaches for automatic data extraction from the EHR, which may become a new standard that replaces manual curation in future large collaborative projects. 81,82 From a technical point of view, the continuous decrease in genomic sequencing costs suggests that NGS assays will likely continue to evolve toward increasingly higher breadths and depths of coverage. Widespread use of WGS technologies within the clinic seems no longer a distant dream, but implementation of these approaches will pose formidable challenges for computational processing, long-term storage and meaningful biological interpretation.…”

“…Community‐driven efforts like OncoTree, 36 which provides a dynamic classification platform for rare and common cancer types, or GenomeNexus, which integrates multiple commonly used resources for variant annotation and interpretation, will play an important role to ensure this much needed interoperability. Data sharing protocols will also become more robust through the use of machine learning and natural language processing approaches for automatic data extraction from the EHR, which may become a new standard that replaces manual curation in future large collaborative projects 81,82 …”