#317 Steven Brown: Why Modern Medicine Needs AI-Assisted Decision Making

What It Covers

Steven Brown, former chief AI officer at Abundance360, develops an AI-powered cancer management platform after his own diagnosis with a rare blood cancer. He creates a multi-agent system using multiple LLMs to analyze medical records, simulate specialist consultations, and surface treatment options that standard care protocols might miss, demonstrating how AI can empower patient-directed precision medicine.

Key Questions Answered

• •Multi-Agent Medical Analysis: Brown built a system using multiple AI agents modeled as different medical specialists (oncologist, hematologist, cardiologist, gastroenterologist) that analyze patient medical records from diverse perspectives. When agents converge on recommendations despite different viewpoints, confidence increases dramatically. The probability of multiple agents sharing the same hallucination approaches zero, creating a more reliable diagnostic framework than single-model queries. This methodology surfaces tests and treatment pathways that individual doctors might miss due to time constraints or specialty blind spots.
• •Context Window Medical Records: Modern LLMs with million-token context windows can process 750,000 words or hundreds of pages of medical documentation in a single prompt. Brown organizes patient medical records by categorizing recent relevant data (labs, imaging, genomics) rather than dumping entire electronic health records. This structured approach mimics how specialists prepare patient summaries for referrals, focusing on disease-relevant information while excluding outdated or irrelevant data. The system uses retrieval augmented generation to pull organized medical data into prompts without fine-tuning foundation models.
• •Precision Medicine Gap: Cancer treatments with thirty percent response rates mean seventy percent of patients receive no benefit from standard care protocols. Genetic mutations in cancer cells create vulnerabilities to specific treatments not included in standard protocols. Brown discovered his cancer had genetic markers making it less responsive to standard therapy but highly sensitive to an off-label treatment. Patients must actively request genomic testing and research mutation-specific therapies because oncologists managing diverse cancer types cannot track every precision medicine development for rare genetic variants.
• •Shared Decision Making Outcomes: Patient outcomes improve measurably when patients actively participate in treatment decisions rather than passively following doctor directives. Brown rehearses upcoming medical appointments by conversing with specialist AI agents, ensuring he asks informed questions during limited consultation time. This preparation transforms ten-minute doctor visits into focused discussions about specific treatment options rather than basic education sessions. The approach requires patients to become experts in their specific disease, enabling them to advocate for tests, treatments, and monitoring protocols doctors might not proactively suggest.
• •Treatment Optimization Details: Brown identified that his prescribed medication required high-fat food for proper absorption, a detail buried on page ten of prescription documentation that pharmacists summarized as take with food. After his cancer marker reduction plateaued, he consulted AI agents about potential resistance or dosing issues. Switching from low-fat meals to high-fat dinners with medication immediately restored downward marker trends. These granular treatment variables significantly impact outcomes but often go unaddressed in standard patient education and monitoring protocols.
• •Insurance Authorization Strategy: Academic medical centers can prescribe off-label precision medicines by writing appeal letters to insurance companies when genomic evidence supports treatment efficacy. Brown's specialized hematologist at a center of excellence agreed his AI-researched treatment approach was optimal but required insurance authorization because it was approved for different conditions. The appeal succeeded based on genetic mutation evidence. Patients at community oncology practices may never access these options because general oncologists stick to standard care protocols and lack resources to navigate off-label authorization processes.