Advancing cancer care through AI and interoperability to improve enhancing oncology model data submission: Results from a Cancer Moonshot initiative.

Abstract

599 Background: The White House Cancer Moonshot initiative aims to reduce cancer death rates by half before 2047 and emphasizes the importance of access to clinical data to improve patient outcomes through enhanced data standardization and interoperability. Effective data capture remains a critical challenge with manual abstraction consuming significant clinician administrative time. The US Oncology Network (The Network) partnered with Ontada to leverage artificial intelligence (AI) and Fast Healthcare Interoperability Resources (FHIR) standards to advance clinical data submission for the Center for Medicare and Medicaid Innovation (CMMI) Enhancing Oncology Model (EOM). Methods: A pilot study was conducted with The Network (Compass Oncology) to implement Ontada's AI-driven technology to submit 58 clinical data elements for 232 patients required for EOM Performance Period 1. Structured clinical data elements were sourced from iKnowMed Generation 2 EHR using FHIR mCODE (Minimal Common Oncology Data Elements) HL7 industry standards. Natural Language Processing (NLP) and AI models were developed to enhance data capture from unstructured clinical documentation for diagnosis date, primary tumor characteristics, regional lymph nodes, distant metastases, and histology. A time and motion study was conducted via manual chart abstraction to measure administrative time reduction and data quality improvements. Data was submitted to CMMI via FHIR. Results: Across five critical data elements representing 909 possible data points analyzed for this study, 53.5% were sourced from structured data vs. 40.6% from unstructured clinical documentation using AI. The time and motion study confirmed a 38% reduction in provider administrative time. The AI-driven submission successfully achieved 98% data completeness for EOM patients, exceeding the 90% threshold required by CMMI. Conclusions: Results from the pilot study underscore the transformative potential of AI and data interoperability in oncology care. By leveraging NLP technology and FHIR-compliant data standards, substantial improvements in data capture were achieved while significantly reducing administrative burden. This supports the Cancer Moonshot vision for standardized, interoperable oncology data and provides a scalable model for enhancing practice efficiency. The success of this pilot establishes a foundation for expanded implementation across additional practices to meet future EOM data submission requirements. Source of clinical data elements. Data Element Structured Unstructured-AI Manual Chart Abstraction Not Found Total Diagnosis Date 106 108 16 2 232 Primary Tumor (T) 99 47 3 0 149 Regional Nodes (N) 98 42 8 0 148 Distant Mets (M) 96 43 9 0 148 Histology 87 129 13 3 232 Total 486 (53.5%) 369 (40.6%) 49 (5.4%) 5 (0.5%) 909 (100%)

Ähnliche Arbeiten