Implementation of an AI-enabled clinical workflow augmentation platform in community oncology: Early effects on clinical operations, documentation quality, and administrative burden.

Abstract

739 Background: Community oncology practices face growing documentation burdens spanning revenue cycle management, quality reporting, and clinical operations, contributing to administrative burden and incomplete data. In one multi-practice study, the average fill rate for required oncology data elements (oMIPS) was just 23%. Such gaps create barriers to guideline-concordant care. Artificial intelligence (AI) has emerged as a strategy to reduce burden and improve data completeness, yet real-world evidence in community oncology remains limited. Tennessee Oncology evaluated Apollo Oncology, an AI-enabled workflow platform integrating with the electronic medical record at the point of care. Built for community oncology, Apollo targets highest-burden workflows: chart preparation, diagnosis and staging documentation, identification of in-house next-generation sequencing (NGS) care gaps, and ensuring appropriate diagnoses for specialty lab orders. Methods: A retrospective analysis was conducted following Apollo implementation across 18 physicians and 2 advanced practice providers at Tennessee Oncology, a large community oncology practice. Apollo automates chart preparation and data entry with the clinician in the loop, surfaces quality opportunities, and prevents avoidable medical necessity denials via prospective order validation. Primary outcomes included clinician adoption and workflow efficiency; secondary outcomes included in-house NGS testing activity (Feb 1–May 1, 2026). Results: Apollo was utilized on over 3,400 encounters, with utilization increasing steadily despite stable clinician volume via weekly feedback-driven updates. Apollo prepared charts for over half of encounters, identified over 2,000 previously undocumented diagnoses, and staged or updated staging for over 350 cancers. Clinicians saved an estimated 150+ clicks per week on diagnosis and staging alone. In-house NGS orders increased approximately 40% versus the prior year. Prospective order validation prevented approximately 90 lab order denials per month—49% of all preventable denials. Conclusions: Implementation of an AI-enabled workflow platform demonstrated sustained clinician adoption and measurable operational impact in a large community oncology practice. Early results suggest AI-assisted clinician-in-the-loop workflows reduce administrative burden, improve documentation completeness, and facilitate high-quality care. Continued evaluation will assess long-term effects on quality metrics, provider experience, and financial outcomes.