Set Expectations: AI Collaboration in Healthcare Research

According to the July 22, 2026 PLOS One study (Henkin et al.), early-stage interdisciplinary AI projects in UK healthcare confront communication, computational, and data-access frictions that slow progress and require role-based translation.

• Henkin et al. interviewed 13 participants between January 19, 2023 and March 3, 2023, reporting that stakeholders often experienced AI as a novel domain with differing expectations.
• Henkin et al. found that each consortium employed roughly 20–30 staff as of their study description, and that meetings (20–30 people) were the main site for cross-discipline knowledge exchange.
• Henkin et al. reported 9 face-to-face and 4 remote interviews in the study, and highlighted infrastructure limits such as constrained GPUs in Trusted Research Environments in their July 22, 2026 publication.
• Henkin et al. documented active translation work by PPIE coordinators and clinicians, concluding on July 22, 2026 that early expectation-setting and resourcing for translation are critical.

This section explains who was interviewed, when the interviews took place, and what the authors coded from those conversations.

According to the July 22, 2026 PLOS One article by Henkin et al., the team performed an inductive thematic analysis of 13 semi-structured interviews conducted between January 19, 2023 and March 3, 2023 across three NIHR-funded consortia.

According to Henkin et al., the study recorded 9 face-to-face interviews and 4 remote interviews and transcribed recordings using Otter.ai before manual correction.

According to Henkin et al., the researchers built a shared codebook using Taguette and Braun and Clarke's reflexive thematic analysis guidance, then identified three high-level themes: AI novelty, AI knowledge sharing, and multiple real-world healthcare perspectives.

Researchers should set explicit, early expectations about AI capabilities and limits based on the PLOS One study.

• According to Henkin et al. (PLOS One, July 22, 2026), differing expectations about AI capabilities created decision delays; teams should run an early "capabilities brief" and document model limits by role.
• According to Henkin et al., meetings with 20–30 participants were the primary collaboration setting and therefore teams should design short, role-specific briefings (e.g., 5-slide clinician brief) to reduce cognitive load.
• According to Henkin et al., computational constraints in Trusted Research Environments (limited GPUs and restricted languages) slowed iteration; project leads should budget compute and environment approvals at project start.

According to Henkin et al. (PLOS One, July 22, 2026), early-stage AI collaboration in healthcare is shaped by novelty, tooling gaps, and translation work that must be resourced explicitly.

Evidano helps teams operationalize the study's recommendations by automating interview synthesis, producing role-specific summaries, and creating auditable translations between technical and patient-facing outputs.

If you are running an interdisciplinary AI‑in‑health project and want faster, clearer cross‑discipline synthesis, Try Evidano for free.