Documentation & Transparency Across the AI Lifecycle

Why documentation matters

WHO describes documentation and transparency as essential for scientific and regulatory assessment and for trust between developers, manufacturers, and end-users. Without transparent records, it is difficult to judge whether evidence from a regulatory submission will generalize to your hospital ward, clinic, or patient mix — where performance often drops.

Effective documentation helps guard against bias and data dredging. Regulators expect to trace development steps and decision points, including:

• the clinical problem being addressed;
• the context in which the AI system will function;
• selection, curation, and processing of training datasets; and
• justifications when deviating from pre-specified plans.

Quality continuum across the lifecycle

WHO encourages a quality continuum from ideation through development, validation, deployment, and post-deployment learning (Figure 2 in the source). Key elements include:

• Pre-specification: development, validation, and deployment plans — including risk mitigation and plans for post-deployment updates;
• During development: processes, validation choices, datasets, and tracked deviations;
• Post-deployment: performance data, unexpected findings, and rationale for modifications.

Documentation should be proportional to risk. Early dialogue with regulators can clarify documentation needs before costly rework.

Transparency beyond regulation

For Hong Kong professionals, transparency also supports peer learning: publishing in journals, sharing datasets where appropriate, and reporting AI studies using CONSORT-AI and SPIRIT-AI standards. Communication should be adapted to end-users, patients, and communities — not only regulators.

Practical questions for clinicians

Before using an AI tool, ask the vendor or informatics team:

1. Is there a pre-specified intended medical purpose and development plan?
2. Can training and validation datasets be described (size, setting, demographics)?
3. Are deviations from the original plan documented with rationale?
4. Is post-deployment monitoring and update planning documented?

Source: WHO — Regulatory considerations on artificial intelligence for health (2023)