Bot or Not?: AI in Health Care | Future of Health Summit 2024

Panelists and Core Mission

• The panel included Dipanita Das (CEO/Co-founder, Sociero), Dr. John Halamka (President, Mayo Clinic Platform), and Vidya Rahman (CMO, Teladoc Health).
• All three panelists agreed that the convergence of multimodal data availability and rapid computing advances has created a threshold for a transformative phase in healthcare.
• Vidya Rahman emphasized a goal to "curate the world's data, create fair AI, and get it used," noting that 2024 shifted from AI reluctance to an imperative to move forward.
• Dr. Halamka identified that while storage, compute, and policy are no longer primary constraints, "psychology, culture, and resistance to change" remain the limiting factors.

Data Challenges and Standardization

• Data Quality: AI performance is directly tied to data quality; disparate, non-interoperable, and "dirtier" data leads to unreliable systems.
• Lack of Depth: Current datasets often possess high volume but lack "depth" (granularity), failing to capture the specific attributes needed for true personalization.
• Heterogeneity Gaps: Clinical trial data is predominantly urban and professional, lacking the heterogeneity required to ensure algorithms work across rural, diverse, and underrepresented populations.
• Real-World Evidence: There is a critical need to embed real-world evidence into care pathways rather than relying solely on traditional clinical trial data.
• Prevention Focus: A strategic shift is required to utilize lifestyle and wellness data (from wearables) for daily behavioral empowerment and prevention, rather than just reactive care.

AI Implementation and Human-in-the-Loop

• Co-Creation Model: Physicians and healthcare providers must act as "co-creators" and partners in AI solution design, rather than passive recipients.
• Personalization at Scale: AI enables the ability to personalize care for large populations, moving beyond "one-size-fits-all" protocols to nuanced treatment plans based on specific patient clusters.
• Trust and Validation: Dr. Halamka noted that an algorithm developed in one location (e.g., Minnesota) may not function correctly in another (e.g., Washington D.C.) without rigorous local validation and trust.
• Generative AI in Workflow: Teladoc is using generative AI to tailor patient communication strategies based on predictive analytics, moving from generic interventions to personalized engagement (e.g., specific messaging for diabetic patients).

Specific Use Cases and Success Stories

• Oncology Precision: Mayo Clinic's retrospective analysis of pancreatic cancer cases allowed them to use AI to diagnose the disease at Stage Zero, compared to the historical norm of Stage Four detection.
• Endoscopy Enhancement: AI-assisted endoscopy can detect polyps five times more effectively than human practitioners alone, addressing the 15% miss rate of standard colonoscopies.
• Radiology Collaboration: AI tools can flag anomalies (e.g., lymph nodes) by comparing current images with historical data, allowing radiologists to catch details they might miss when viewing scans in isolation.
• PMS and Hormonal Health: AI analysis identified that severe PMS symptoms in a specific patient were linked to a drastic estradiol crash, revealing a treatable hormonal imbalance that human doctors had overlooked.
• Diabetes Clustering: Unsupervised AI learning revealed five distinct clusters of obesity-related cardiometabolic risk, challenging the assumption that high BMI always correlates with high risk.

Analogy and Strategic Outlook

• "AI Water Bottles": Panelists urged against adopting AI for its own sake (the "AI + water bottle" trend), emphasizing that solutions must address specific business problems like margin pressure, recruitment shortages, and burnout.
• Fintech Analogy: Dipanita Das compared the potential trajectory of AI in healthcare to the evolution of fintech, where secure data handling enabled massive shifts in payments and services.
• Musical Harmony: Vidya Rahman used the analogy of learning Carnatic music with a guru; just as AI helps a musician identify pitch errors for improvement, AI can help decode historical healthcare mistakes to create better future protocols.
• Global Cooperation: Dr. Halamka called for a "global federated network of data" to reduce competitive angst, drawing parallels to the Davos Alzheimer's Collaborative to achieve country-scale cooperation while protecting privacy.

The "Moonshot" Questions

• Key Evaluation Question: Panelists agreed the primary litmus test for any AI health venture is: "What specific problem are you solving for, and how does the technology directly connect to that problem?"
• Trust Inquiry: A secondary critical question is, "Can you trust the development of this model?" requiring scrutiny of data sources (e.g., validating if an algorithm was trained on only 100 patients).
• Prevention Moonshot: Vidya Rahman's vision is "Precision Health" focused on prevention, empowering global consumers to use data to become stewards of their own health before illness occurs.
• Accessibility Moonshot: Dipanita Das envisions a future where AI bridges the gap between complex medical science and patient understanding, making information accessible and comprehensible for patients and caregivers.
• Scale Moonshot: Dr. Halamka's objective for the Mayo Clinic is to "touch four billion people by 2030" through curated, globally accessible, and safe data ecosystems.