Talk Clinical Data Competencies & Cross-Functional Interactions · Clinical Trial Operations · AI & Cognitive Tech
From LLMs to SLMs: Will Clinical Data Science redefine Clinical Trial Execution?
Sun, Sep 28, 2025 · 4:15–5:30 PM EDT Room 339+340, Baltimore Convention Center
About this session
Session chaired by Abby Abraham
Description:
The year 2024 has seen a transformative leap in artificial intelligence, driven by breakthroughs in Large Language Models (LLMs) like ChatGPT, Gemini, and Claude 3, alongside AI-augmented tools such as Co-Pilot and Perplexity. While industries like biopharma have harnessed generative AI for accelerated drug discovery and molecular modeling, clinical research lags behind, with adoption remaining fragmented and benefits largely theoretical. This session addresses the critical disconnect between AI’s potential and its practical implementation in clinical research, focusing on specific challenges and solutions.
Key Challenges
1. The Scalability Paradox of LLMs
While LLMs excel at processing unstructured data (e.g., text, images), clinical research relies heavily on structured, sparse datasets (e.g., trial outcomes, lab values). Training or fine-tuning general-purpose LLMs for these use cases often incurs prohibitive costs with diminishing returns.
2. Implementation Roadblocks
Business leaders demand AI to “transform” clinical operations overnight, but practical barriers persist:
- Data Silos and Legacy systems ( create interoperability issues.
- Value Realization Gap: Pilot projects often fail to scale due to mismatched expectations
3. Cost Barriers & Legacy Debt
- Financial Realities: Smaller sponsors/CROs face disproportionate costs due to:
- Legacy Technology Debt: Aging IT infrastructure (e.g., on-premise data warehouses) limits cloud-native AI integration.
- Regulatory Overhead: Validating AI/ML models for compliance (e.g., FDA’s 2023 AI/ML guidelines) strains budgets.
- Equity Implications: Risk of a “two-tier” ecosystem where only large players afford cutting-edge AI, exacerbating disparities in trial access and innovation.
Emerging Solutions of interest that require exploration and discussion:
- Small Language Models (SLMs) tailored to domain-specific, structured data. SLMs reduce computational overhead and improve interpretability.
- Agentic AI as a Bridge: Task-specific AI agents (e.g., for adverse event monitoring or protocol optimization) can deliver rapid, measurable ROI while sidestepping “boil-the-ocean” approaches.
- Insights into Cost Barriers: Understanding of financial challenges and legacy technology debt in organizations, helping to understand the solution and nuances of adoption strategies for different types and sizes of sponsors and service providers.
This session aims to develop frameworks for prioritizing use cases with the highest ROI and strategies to democratize AI access, ensuring equitable progress.
Speakers
-
Vice President - Clinical Science and Centralized Monitoring · Precision for Medicine
Speaker
