The integration of machine learning-based & LLM based AI into clinical trial data management & clinical operations remains elusive, not just operational and technological challenges. The promise of AI transforming clinical trials is yet to unfold. Here are key gaps that still need to be addressed:

Domain-Specific Training Data: Machine learning models require huge amount of training data whereas LLMs, like GPT-4, are generalists. They’re trained on broad internet data, not the nuanced, regulation-heavy content that drives clinical trials. This lack of specialized data for training means these models often produce outputs that miss the mark, increasing the workload for teams already stretched thin. We need datasets that are steeped in clinical trial knowledge, not just internet trivia.

Poor Integration with Existing Tools: Clinical trial teams live and breathe Excel, Word, SAS and a suite of legacy systems. Whereas AI is usually supported via independent web-applications, disconnected from the day-to-day tools these teams rely on. The lack of seamless integration stifles productivity and leads to redundant work. To truly be useful, AI needs to embed itself into the daily grind, not exist as a shiny but isolated tool.

Lack of Interpretability and Trust: In a field where decisions have life-or-death consequences, a black-box model simply doesn’t cut it. Clinicians and trial managers need to understand how AI arrives at its conclusions. The industry needs AI models that offer transparency and trust, not just predictions wrapped in mystery.

Reliability and Validation: Companies aren’t going to use AI tools that haven’t been battle tested. Without validation, such tools will remain on the sidelines, limited to non-critical tasks. It’s time for standardized validation frameworks to bring AI into the mainstream.

Data Security Concerns: Clinical trial data is sensitive, and current AI solutions, especially cloud-based ones, pose too many risks. Security isn’t just a checkbox; it’s a deal-breaker. AI in clinical trials must prioritize secure solutions that protect patient data at all costs.

Cost & Scalability: Let’s face it—Setting up AI is resource-intensive and expensive. This limits their use to the big players, leaving smaller organizations out in the cold. Market need more efficient, cost-effective AI platforms that democratize access across the board.

Summary

The hype around AI in clinical trials is real, but so are the challenges. Until we address the lack of integration with everyday tools, and overcome the other technological and operational hurdles, AI will remain on the periphery of clinical trial operations.


While here are some of the interesting articles I noticed in the last week about use of AI in Clinical Trials. I hope you find these useful.

Not all AI health tools with regulatory authorization are clinically validated | Nature Medicine

By 2034, AI in Healthcare Market Size & Share Will Reach – GlobeNewswire

Clinical GenAI Solutions by Saama

Ushur Unveils AI-powered, HIPAA-secure Self-Service Solution to Modernize Clinical Trial …

Ushur launches self-service solution to modernize Clinical Trials engagement through AI …

Generative AI-Copiloted CSR Writing for Clinical Research – AlphaLive Sciences

The testing of AI in medicine is a mess. Here’s how it should be done – Nature

Enhancing diversity in clinical trials with AI and human expertise | ICON plc

Generative AI can not yet reliably read and extract information from clinical notes in medical …

Navigating the EU AI Act: Pharma’s Path to Responsible Innovation – PharmiWeb.com

Diagnostic Performance of AI-based Models versus Physicians Among Patients … – Frontiers

GSK & AstraZeneca Discuss Drug R&D AI Revolution | Healthcare Digital

How Harvard professors have scaled-up novel combination therapy discovery – Nature

Generative AI platforms drive drug discovery dealmaking – Nature