Translating algorithmic fairness into health practice: formative evaluation of the fairness-to-action framework.

Introduction: Algorithmic fairness is increasingly acknowledged as a critical concern in digital health; however, existing knowledge on fairness remains challenging to operationalize and systematically embed into routine practice. Our study examines the knowledge-practice gap through a formative evaluation of the Fairness-to-Action framework and the development of a guidance artifact. Methods: The study was conducted within HealthEmove, a digital health initiative focused on electronic personal health records for mobile populations. Using an exploratory mixed-methods design, we interviewed 11 professional stakeholders using all 22 Fairness-to-Action prompts. We also collected prompt-level Clarity, Relevance, and Usability ratings and conducted stakeholder responsibility mapping. Results: Qualitative findings suggest that fairness knowledge was available in HealthEmove but unevenly distributed across roles, partially formalized, and not yet embedded in shared routines. Stakeholders' self-reported ratings were concentrated at the upper end of the Clarity, Relevance, and Usability scales and were interpreted as formative indicators of perceived fit within this case. Responsibility mapping made cross-role dependencies explicit and informed a proof-of-concept guidance artifact for the Junior Researcher role. The artifact underwent formative assessment by two of the 11 participants: the target role holder and one closely collaborating respondent. Their preliminary feedback suggests that clearer documentation expectations and onboarding may improve use. Discussion: This single-case study contributes a formative evaluation of the Fairness-to-Action framework and illustrates a procedure for turning its prompts into role-specific guidance within an early-stage digital health project. At the process level, it examines the fairness knowledge-practice gap by treating fairness knowledge as a resource that must be accessed, interpreted, and operationalized across project roles, rather than only specified as a technical or governance requirement.