The Bold Claims and Ground Realities
In the world of healthcare AI, bold claims are as abundant as they are ambitious. Companies like Alphabet’s Isomorphic and Lila promise breakthroughs that seem to leap straight from a sci-fi script. They talk of faster discoveries and life-changing medicines, investing heavily in the process. Anthropic’s recent $400 million acquisition of Coefficient Bio is a testament to this belief in AI’s potential. But the real test of AI in healthcare isn’t in the marketing brochures; it’s in the tangible outcomes. Did it lead to a medicine that saved lives? For most, the answer remains elusive.
The reality check is stark. Bringing a treatment to market requires navigating a labyrinthine process, often taking a decade and costing billions. Clinical trials, especially Phase 3, are a long haul, and diagnostics demand rigorous validation. The journey from AI model to medicine is fraught with challenges, and few have crossed that finish line. Companies like Isomorphic and Lila have yet to deliver a market-ready treatment, underscoring the gap between aspiration and realization.
Bridging the Gap with Real-World Trials
To translate AI’s potential into patient care, the industry must bridge the gap between model training and clinical application. This is where companies like Insilico Medicine and Recursion are making strides, advancing AI-discovered assets through clinical trials. At Owkin, we’ve taken our oncology drug, OKN4395, into the Phase 1a clinical INVOKE trial. This journey involves years of training AI on real patient data, leading to significant milestones like the CE mark for our MSIntuit CRC in European pathology practice.
The process is demanding but essential. Bringing AI to patients not only tests the technology but improves it. Our experience with diagnostic AI revealed the need for adaptability across diverse populations and technology setups. Developing robust methods to address these challenges has been critical, highlighting the iterative nature of integrating AI into healthcare.
Enhancing AI Through Real-Time Feedback
At Owkin, we believe in the power of real-time feedback from clinical trials to refine our AI. Unlike traditional trials that focus solely on success indicators, our INVOKE trial incorporates ongoing data from patients to enhance AI performance. When our AI’s predictions fall short, we retrain it with real data, creating a positive feedback loop. This iterative process not only improves AI accuracy but also expands its applicability, benefiting more patients over time.
This approach is gaining traction across the field, with variations in methodology. Some companies test AI results on in vitro systems before human trials, but ultimately, AI must prove its efficacy in human applications. The real-world feedback loop is crucial for validating AI’s role in drug discovery, diagnostics, and clinical applications.
Diverse Data: The Key to AI’s Success
Training AI on diverse and rich patient data is essential for bridging the gap between model predictions and clinical reality. The more comprehensive the data, the more accurate the AI’s insights. When existing patient data falls short, in vitro methods like patient-derived organoids can simulate human biological complexity, offering a wealth of clinical insights.
Testing AI predictions in real-world settings with human patients provides invaluable insights into the models’ strengths and limitations. At Owkin, this holistic approach helps us understand the true clinical challenges and refine our AI accordingly. It’s a demanding process, but one that is necessary to overcome the barriers to effective treatment delivery.
Facts Worth Knowing
- •💡 Bringing a new drug to market can take up to 10 years and $2 billion – source
- •💡 Few AI companies have brought treatments to market despite bold claims.
- •💡 Real-world testing and feedback loops are crucial for AI success in healthcare.
