From Potential to Practical at DIA 2026: Structure Before AI

Part 2 of a Three-Part Series on Unlocking AI’s Role in Modern Clinical Trials

As artificial intelligence becomes more embedded in clinical development, the focus is shifting from capability to trust. At DIA 2026, discussions emphasized that scaling AI requires more than performance—it requires shared frameworks, clear definitions, and consistent approaches to validation.

Strengthening data integrity: structuring patient experience data 

Patient experience data is now a foundational component of clinical development, shaping regulatory confidence, market access and how treatment benefit is understood. Although collected electronically for decades, its use at scale is more recent. 

The DIA 2026 session From Insight to Impact: Leveraging Patient Experience Data Through Technology and AIunderscored a key distinction: Electronic data capture alone does not ensure data quality. Data must be structured and interpretable to support confident decision-making.

This challenge reflects the current state of the ecosystem. Technologies such as electronic clinical outcomes assessment are widely adopted but still evolving. While eCOA has been used for decades, its expansion at scale has introduced new complexity, as expectations for automation and performance have outpaced the underlying processes and scientific oversight required to support them. As a result, assumptions about automation and scalability do not fully reflect operational reality, where manual configuration, protocol-specific setup and workflow dependencies continue to shape outcomes.

In practice, even well-designed studies can encounter issues, not because data are absent, but because they are fragmented or disconnected from the broader study context. Patient behavior also has an impact. Disengagement, fatigue and inconsistent reporting can introduce variability that is difficult to detect when systems are viewed in isolation. 

High-frequency data collection, such as daily diaries, offers greater insight into patient experience but also introduces additional variability. These fluctuations may reflect changes in behavior or engagement rather than underlying clinical outcomes. In this context, the question is no longer simply whether data have been collected but whether they reflect genuine clinical change or patterns introduced by how they are generated.

The work behind reliable and scalable AI 

Historically, monitoring approaches have focused on individual systems like eCOA, providing a partial view of study performance. But risks are rarely isolated within a single dataset. A patient may appear compliant in one system (e.g., eCOA) while discrepancies in timing, missing assessments or protocol alignment only emerge when viewed across the full data ecosystem. By connecting these signals, trial sponsors can move from retrospective assessments to a more complete, real-time understanding of study health. Importantly, this approach establishes the conditions needed for structured, connected data environments that support reliable analysis and more advanced analytics.

Achieving this requires deliberate effort. Data must be organized, normalized and aligned across systems in a consistent and interpretable way. Otherwise, even advanced analytics can amplify noise rather than clarify it. The effectiveness of AI depends as much on the quality and organization of the data it uses as on the sophistication of the models themselves.

Building this capability requires domain expertise and intentional design to connect systems, define meaningful signals and ensure data accurately reflect patient experience. This thoughtful scaffolding transforms fragmented inputs into a usable, auditable and governable environment, enabling more transparent and interpretable AI-driven insights.

When this structure is in place, the impact becomes measurable. Cross-system monitoring approaches improve compliance, reduce site-level data issues to low single digits and enable earlier identification of risks that could otherwise affect endpoint quality. These improvements also establish the conditions required to scale more advanced analytics.

With clean, connected datasets, machine learning models can move beyond describing what has occurred to anticipating where risk will emerge, including identifying patients who may disengage and flagging sites showing early signals of data quality issues. For example, more than 30% of patients using GLP-1 therapies for obesity discontinue within the first month and nearly 60% stop within three months, underscoring the importance of identifying and prioritizing potential high-risk periods for proactive monitoring and targeted intervention, if needed. 

But AI does not replace human expertise. AI requires ongoing human oversight to ensure insights are interpreted appropriately and translated into meaningful action.

Ultimately, the shift is not simply toward collecting more data but toward structuring it in a way that makes it usable, creating a pathway from insight to reliable impact and from established tools to more predictive, scalable approaches to clinical development. This foundation is essential not only for improving trial execution, but for ensuring that AI can be applied consistently and at scale.

Lindsay is a behavioral scientist and Principal at IQVIA, where she provides scientific and technical leadership in the use of technology to advance patient-centered clinical research, with a focus on electronic clinical outcomes assessments (eCOAs). With nearly 20 years of experience spanning global health, epidemiology and clinical research, her work centers on applying behavioral science and data-driven innovation to improve patient outcomes, reduce burden and strengthen the quality and reliability of clinical data.