Panel, Conference Presentation
AI and Machine Learning in Cancer Medicine
Milken InstituteMarc Hurlbert, Maurice Ferré, Colin Hill, Mike Nohaile, Susan Swetter, Amanda Hall, Mark Shieldsmann, Lisa Desjardins
- Susan Sweater anticipates that while a consumer-facing smartphone app for skin cancer diagnosis and triage could reach 6.3 billion devices by 2021, the technology is not yet ready for diagnostic impact or prospective clinical validation in real-world settings, particularly noting current limitations with inflammatory diseases and risks of algorithmic bias such as misinterpreting rulers as malignant lesions.
- Colin Hill projects that within the next 18 to 24 months to a few years, real-world evidence will complement randomized controlled trials to drive labeling, impact approvals, and extend drug indications, followed by a 5 to 10-year timeframe where AI-driven drugs, personalized care pathways, and a precision healthcare renaissance with better outcomes and lower costs become standard practice.
- Maurice Ferre forecasts a long-term horizon of over 20 years for the healthcare system to fully adapt to the transformation required for effective big data and AI utilization, while Mike Nohaley expects a hybrid model integrating electronic medical records with omics and wearable data to emerge within 2 to 5 years.
- Mike Nohaley cautions that a "centaur model" where humans aid computation is the realistic path, disputing the immediate elimination of radiologists within 10 years and warning that the "move fast, break it" tech mentality will fail due to regulatory structures, data safety complexities, and the critical need for empathy and compassion that AI cannot replicate.
- Significant risks identified include the FTC fining developers for false diagnostic claims, algorithms that may underdiagnose melanoma by 30 percent, potential negative impacts on the doctor-patient relationship if workflow improvements lead to excessive patient volumes, and the possibility that AI algorithms in real-world settings may not outperform human sensitivity and specificity.
- Strategic plans involve multicenter prospective studies to validate high sensitivity and specificity, the generation of real-world evidence packages alongside trials to support approvals, and a shift from hypothesis-driven to data-driven research, with investment focus required on solutions possessing sufficient data granularity and existing AI tool availability.
- Colin Hill predicts a 3 to 5-year window for significant AI impact and notes that computing power is no longer a limiting factor, though data fuel is not yet fully complete but sufficient to scale, leading to an industry "golden era" where personalized medicine becomes the standard of care.
- Economic and systemic shifts are expected over a 5 to 10-year period as reimbursement landscapes change, payers adopt these approaches, and the current inefficiency of 18% GDP expenditure on healthcare becomes unsustainable, potentially driven by future economic factors and the need to match health interventions to individual patients more effectively.
- Tech companies are expected to likely acquire rather than replicate healthcare-focused entities and will not enter "wet labs" soon, constrained by the fact that lives are on the line, which prevents the total disruption seen in other sectors like ride-sharing or lodging.