Clinical Reasoning AI for Oncology Treatment Planning: A Multi-Specialty Case-Based Evaluation

A team of 36 researchers led by Philippe E. Spiess at Moffitt Cancer Center published a multi-specialty evaluation of OncoBrain, an AI clinical reasoning platform for oncology treatment planning. OncoBrain combines general-purpose large language models with a cancer-specific graph retrieval-augmented generation (RAG) layer, a gold-standard treatment-plan corpus as long-term memory, and a model-agnostic safety layer called CHECK for hallucination detection and suppression. The study addressed a critical gap: over 80% of U.S. cancer care is delivered in community settings, where survival outcomes remain worse than at academic centers due to the cognitive burden of integrating genomics, staging, radiology, pathology, and rapidly changing guidelines.

The evaluation involved 173 clinician-enriched case summaries across gynecologic, genitourinary, neuro-oncology, gastrointestinal/hepatobiliary, and hematologic malignancies. Three clinician groups—subspecialist oncologists (50 cases), physician reviewers (78 cases), and advanced practice providers (45 cases)—completed structured evaluations using a 16-item instrument. Results showed mean alignment with evidence and guidelines of 4.60, 4.56, and 4.70 on a 5-point scale across the three groups. Safety scores (absence of misinformation) averaged 4.80, 4.40, and 4.60. Workflow integration scored 4.50, 3.94, and 4.00, while perceived time savings ranged from 5.00 (subspecialists) to 3.60 (advanced practice providers). The authors conclude OncoBrain generates guideline-concordant, clinically acceptable treatment plans that are easy to supervise, justifying prospective real-world evaluation in community settings.