Table 2. Action research cycles in the... | Emerald Publishing

Table 2.

Action research cycles in the RAidD project

Cycle	Period	Main activities	Primary forms of resistance identified
1. Initial diagnosis and problem codefinition	December 2022–September 2023	Mapping of existing practices in nephrology, hepatology and diabetology; collective definition of database requirements; weekly recorded meetings; interviews with clinicians and IT researchers; analysis of clinical documentation and organizational protocols; collaborative design of the first data model; start of clinical data collection and harmonization; collective reflection sessions to consolidate a shared understanding of methodological and organizational challenges	Epistemic: misalignment in data definitions, clinical reasoning and comparability of variables across specialties; uncertainty over data ownership and responsibility
2. Development and early experimentation	Late 2023–2024	Planning and implementation of synthetic data for training and testing FL models; codesign workshops with clinicians and IT researchers; weekly monitoring meetings to evaluate reproducibility of results; individual interviews to capture perceptions and expectations; final reflection phase leading to redefinition of data collection strategies and preparation for real-world experimentation	Institutional: constraints related to data availability, privacy and regulation; skepticism toward model validity and generalizability; negotiation of acceptable performance metrics and validation criteria
3. Clinical validation	2025–2026 (ongoing)	Progressive integration of AI algorithms into clinical workflows; training sessions, interdisciplinary workshops and feedback activities; pilot tests in each department; collection of qualitative and quantitative data on impact and sustainability; joint reflection among clinicians, IT researchers and researchers to identify improvements and ensure coherence with ethical, institutional and professional requirements	Professional/ethical: concerns about interpretability, accountability, trust and the legitimacy of AI-supported clinical decisions at the point of use

Cycle	Period	Main activities	Primary forms of resistance identified
1. Initial diagnosis and problem codefinition	December 2022–September 2023	Mapping of existing practices in nephrology, hepatology and diabetology; collective definition of database requirements; weekly recorded meetings; interviews with clinicians and IT researchers; analysis of clinical documentation and organizational protocols; collaborative design of the first data model; start of clinical data collection and harmonization; collective reflection sessions to consolidate a shared understanding of methodological and organizational challenges	Epistemic: misalignment in data definitions, clinical reasoning and comparability of variables across specialties; uncertainty over data ownership and responsibility
2. Development and early experimentation	Late 2023–2024	Planning and implementation of synthetic data for training and testing FL models; codesign workshops with clinicians and IT researchers; weekly monitoring meetings to evaluate reproducibility of results; individual interviews to capture perceptions and expectations; final reflection phase leading to redefinition of data collection strategies and preparation for real-world experimentation	Institutional: constraints related to data availability, privacy and regulation; skepticism toward model validity and generalizability; negotiation of acceptable performance metrics and validation criteria
3. Clinical validation	2025–2026 (ongoing)	Progressive integration of AI algorithms into clinical workflows; training sessions, interdisciplinary workshops and feedback activities; pilot tests in each department; collection of qualitative and quantitative data on impact and sustainability; joint reflection among clinicians, IT researchers and researchers to identify improvements and ensure coherence with ethical, institutional and professional requirements	Professional/ethical: concerns about interpretability, accountability, trust and the legitimacy of AI-supported clinical decisions at the point of use

Source(s): Authors own work