How One Health System Is Piloting AI In Clinical Trials

By John Oncea, Chief Editor, Clinical Tech Leader

Most coverage of AI in clinical research describes either sweeping adoption or sweeping skepticism. What’s happening at Blessing Health System right now is neither; it’s a small, deliberately governed pilot that hasn’t yet earned the right to expand.

“It is being used. I do not know that it’s improving workflow,” said Kiley McGlauchlen, Director of Clinical Research at the Quincy, IL-based health system, describing an ambient AI tool currently in beta for physician dictation. “I think it’s being piloted among a very few key individuals. If it appears to be beneficial, it will open up to a wider range.”

Governance Before Scale

The pilot group is intentionally narrow – a handful of physicians, including the system’s Chief Medical Information Officer (CMIO), who also sees patients directly. Rather than rolling the tool out and measuring adoption after the fact, Blessing built oversight in from the start: a governance committee made up of hospital executive and clinical leadership reviews how the AI’s suggested dictation language compares against the organization’s existing clinical practice guidelines before any decision is made about wider deployment. The pilot participants are, in effect, testing the tool against their own judgment before the organization trusts it more broadly.

That governance-first posture is notably different from the more common industry pattern of deploying a tool broadly and then managing problems as they surface. For a health system operating inside a heavily regulated clinical trial environment, where any output eventually touches a patient record or a study file, McGlauchlen’s team appears to be treating that caution as a feature rather than a delay.

A Second, Quieter AI Pilot: Regulatory Document Management

A less visible but arguably more concrete use of AI is underway in Blessing’s clinical research document management. Site regulatory files require organizing dense volumes of documents – medical licenses, physician CVs, training certifications, delegation-of-authority logs, financial disclosure forms – each of which needs to be filed, named, and tracked precisely enough to survive a sponsor audit.

Blessing is piloting AI-generated suggestions for how those documents should be categorized, named, and filed within Veeva Vault, the electronic trial master file and site regulatory file platform used widely across the industry, using Veeva’s Vault AI capability and built around a standardized naming convention. A training certificate, for example, might be named to capture the sponsor, the training topic and version, the staff member’s name, and the completion or expiration date in one consistent string; the kind of naming discipline that makes a document findable months or years later without opening it first. Letting AI suggest that categorization and naming, rather than leaving it to individual judgment call by call, is a small but meaningful step toward consistency across a growing regulatory file.

The timing lines up with a broader shift in the rules governing this exact area. The recently revised ICH Good Clinical Practice guidance, E6(R3), added new language on data governance shared between investigators and sponsors, guidance that didn’t exist in the prior version, E6(R2). For sites like Blessing that are simultaneously modernizing their document systems and absorbing new regulatory expectations around data governance, the two are arriving at the same time by necessity rather than coincidence.

Taken together, the two pilots point to a consistent philosophy: let AI touch the lower-stakes, high-volume work first – document filing, dictation drafts subject to physician review – and hold it to the same clinical-guideline standard as everything else before letting it touch more. For a health system that spent five years building its research infrastructure from nothing, that patience looks less like caution for its own sake and more like the same discipline that got the department here in the first place.