(B-012) Comorbidity Assessment using Integrated EHR and Claims in a National Oncology Database

Background: Accurate assessment of comorbidities is essential in cancer (ca) research to better assess treatment selection and clinical outcomes. Existing comorbidity indices primarily derived from administrative claims data may be enhanced by integrating EHR data to improve completeness.

Objectives: To evaluate the utility of a composite measure of comorbidity burden that integrates information from an EHR source linked with administrative claims in a large population-based ca cohort.

Methods: A retrospective analysis of ConcertAI Patient360TM EHR-based dataset of adult patients (pts) with confirmed ca from academic and community sources across the US was conducted. The dataset is linked with administrative claims. Pts with bladder, breast, CRC, GEC, HCC, melanoma, NSCLC, SCLC, pancreatic, prostate, and renal ca diagnosis from 2014 to 2024 were included. A composite measure of presence of each Charlson comorbidity at baseline (12 months prior to date of initial ca diagnosis) was developed using following hierarchy: unstructured EHR via expert curation, structured EHR, and ICD-10 diagnosis and procedure codes from linked claims. Index ca as a comorbidity was excluded. Prevalence of comorbidities, total CCI score, and association between CCI and rwOS were summarized by ca type.

Results: The study included 164,913 pts across 11 ca types. Pts with HCC, SCLC, and NSCLC had the highest baseline comorbidity burden with mean CCI score of 2.6, 2.0, and 1.8 and with 15%, 9%, and 8%, respectively having a 5+ total CCI score. Pts with breast ca and melanoma had the lowest baseline comorbidity burden. Diabetes and chronic pulmonary disease were the top two prevalent conditions across all tumors, except HCC and pancreatic ca with liver disease. Claims identified incremental comorbidities vs. EHR alone increasing the number of pts with 3+ and 5+ total CCI score by at least 2-fold and 3-fold, respectively across most tumors. Peripheral vascular disease and mild/moderate renal disease were more frequently identified in claims whereas chronic pulmonary disease, diabetes, and other malignancies were more frequently identified in EHR. Increasing CCI score was associated with higher mortality risk, with the greatest risk in pts with 5+ (vs. 0) CCI score [lowest HR=1.41 in HCC to highest HR=5.02 in breast ca, p< 0.001].

Conclusions: Comorbidity burden is high for pts newly diagnosed with ca, varies by ca type, and is incompletely captured in EHR and claims separately. Provider documentation practices may contribute to EHR data gaps whereas underreporting or coding limitations may lead to gaps in claims data. An integrated comorbidity measure based on EHR linked to claims provides a more complete measure of competing risks in population-based ca datasets to understand impact on study findings.