American Society of Clinical Oncology, Alexandria, VA
Katrina Caridad Rios, Arpitha Thakkalapally, Jacob Koskimaki, Mark Riffon, Robert S. Miller, George Anthony Komatsoulis, Danielle Potter
Background: Accurate calculation of key quality measures is critical for informing high-quality, value-based cancer care that is consistent with clinical guidelines. The American Society of Clinical Oncology (ASCO)’s CancerLinQ enables oncology organizations around the US to view near-real time quality measure dashboards sourced from structured electronic medical record (EMR) data; however, use of structured data in key fields is highly variable. Unstructured content, such as progress notes, contains important clinical information on treatment and disease status, which can then undergo curation. This process involves trained data abstractors searching for key data elements through a combination of manual review and natural language processing (NLP) to extract structured data from unstructured content. We hypothesize inclusion of curated data substantially augments structured data alone by more accurately representing the patient journey, thus improving validity of quality measures across EMRs. Methods: A total of 96,399 records across 57,232 patients from 4 EMRs vendors were analyzed from 2018-2019 across structured EMR and curated data. Each record represents 1 of 7 key data elements used to calculate the Staging Documented within One Month of First Office Visit quality measure. Structured documentation of these data elements determines if a patient is concordant with the measure, meaning they were staged within 31 days of their first visit after diagnosis, or non-concordant, meaning they were not staged within the appropriate window. Results: More than a quarter of records from patients concordant or non-concordant with the measure (28.85%) had key data elements sourced from curation. In total, 33% of all records among concordant patients were sourced from curation. Relying on structured data alone would show only 67% concordance versus 97.5% concordance among curated records. This demonstrates that appropriate care may often be delivered but documentation may be missing in a significant fraction of structured EMR data, thus limiting accurate reporting capabilities. Conclusions: NLP-assisted curation can meaningfully supplement structured EMR data by providing a more accurate picture of care rendered, which can have substantial impacts on clinical care, quality reporting, and business operations.
EMR | Distinct Patients | # of Structured Data Elements | # of Curated Data Elements | % Total Data Elements from Curation |
---|---|---|---|---|
EMR 1 | 12,668 | 10,842 | 4,454 | 29% |
EMR 2 | 12,555 | 13,011 | 4,849 | 27% |
EMR 3 | 26,594 | 26,589 | 14,938 | 35% |
EMR 4 | 5,415 | 4,465 | 2,891 | 39% |
Total | 57,232 | 54,907 | 27,132 | 33% |
Disclaimer
This material on this page is ©2024 American Society of Clinical Oncology, all rights reserved. Licensing available upon request. For more information, please contact licensing@asco.org
Abstract Disclosures
2024 ASCO Annual Meeting
First Author: Divya Deepak
2023 ASCO Annual Meeting
First Author: Arash Maghsoudi
2023 ASCO Quality Care Symposium
First Author: Caitlin Drumheller
2023 ASCO Quality Care Symposium
First Author: Anthony DiDonato