Rethinking cancer attribution in oncology VBC models: E&M vs treatment-based approach.

Abstract

1580 Background: Patients with multiple cancers are uncommon. In value-based care models such as EOM (Enhancing Oncology Model),episode attribution and financial accountability rely on assigning a cancer type based on the plurality of predefined E&M visits. However, this administrative approach may not accurately represent the cancer for which a patient is receiving active treatment, especially in the presence of concurrent malignancies. Despite this, little is known about how often the E&M based cancer type differs from the treatment-based cancer type. Methods: We analyzed EOM performance period data (July 2023–June 2025) from practices in The US Oncology Network. Injected based initiating episodes were identified using EOM Part B claims that occurred at practices’ sites, then matched to the Electronic Health Record by patient, drug, and initiating date. For oral agents, additional to patient, drug and prescribing provider, order dates within 60 days of the fill date in Part D Claims was used as a proxy in matching process. Cancer type associated with the initiating treatment was extracted from the EHR and then compared with EOM-attributed cancer types. Results: Episodes were predominantly triggered by injection-based chemotherapy with 13,803 initiated in physician office setting, and 5,441 were initiated by oral chemotherapy. Matching rates to an initiating event in EHR differed substantially by route:~ 90% for injection-initiated episodes vs. about 40% for oral-initiated episodes. Among injection-based episodes, 5% of episodes did not have a treatment-based cancer diagnosis in EHR and 4.5% had discordant cancer diagnoses, while oral-based episodes showed markedly higher rates of missing cancer diagnosis at 32%, with 24% mismatched diagnoses. Chronic leukemia and prostate had a higher proportion of oral initiations and correspondingly lower matching rates. Oral-initiated episodes demonstrated particularly low EHR matching in breast, lung, prostate, and small intestine/colorectal cancers. Conclusions: Cancer attribution in VBC models remains challenging, particularly for episodes initiated with oral therapies. As the use of oral agents continues to expand and patients increasingly present with concurrent cancers, attribution difficulties are likely to grow, resulting in greater discordance over time. Incorporating treatment-based cancer type into attribution could improve accuracy and better align assigned cancer types with actual care in value-based oncology models. EHR matching by route of chemotherapy initiation. Cancer Type Injection Episodes (n) Injection Concordance (%) Oral Episodes (n) Oral Concordance (%) Breast Cancer 3836 91.8% 727 23.9% Chronic Leukemia 171 83.6% 1370 50.1% Lung Cancer 3539 93.1% 323 31.3% Lymphoma 1724 84.6% 512 47.9% Multiple Myeloma 1976 93.2% 1211 73.6% Prostate Cancer 558 83.2% 1210 21.4% Small Intestine/Colorectal Cancer 1999 89.1% 88 21.6%