Blog
Subscribe to my Substack
Can Machine Learning Give Us Faster and Cheaper Clinical Trials?
On Trial Design, Historical Controls, and Clinical Prediction Models
In drug development, where the median clinical trial costs about $48 million, biotech companies face the major challenge of balancing expenses with the need for robust, well-designed, and sufficiently large clinical trials. This reality has motivated software companies like Unlearn.AI to pioneer machine learning (ML) applications to boost clinical trial efficiency. Yet, implementing such cutting-edge methodology is not without significant practical challenges.
Today’s post delves into the merits and limitations of Unlearn’s ML-based Prognostic Covariate Adjustment (PROCOVA). I argue that the theoretical cost-savings of PROCOVA are unlikely to be realized in practice. Given the often high stakes of clinical trials, the risk of adopting innovative ML-based methodologies may not be worth the purported benefits.
The Hard Truth about Artificial Intelligence in Healthcare: Clinical Effectiveness is Everything, not Flashy Tech
Lessons from HeartFlow’s Incursion into Cardiovascular Imaging
AI in healthcare faces a critical issue: our obsession with cutting-edge technology often overshadows the actual impact on patients. Successfully bringing AI medical devices to market entails much more than excellent diagnostic performance; it requires rigorous clinical trials and comprehensive cost-effectiveness analyses. HeartFlow’s AI-powered cardiac imaging product FFRCT is a perfect example of that. In this blog post, I critically review FFRCT and discuss broad lessons for the future of AI medical devices