Objective:

To establish a framework for defining public health action end points specific to AI outputs in digital public health initiatives.

Approach:

• Public Health Action End Point Definition: A public health action end point is linked to a predefined opportunity for action based on AI outputs, with recorded statuses such as completed, deferred, missed, unresolved, or excluded.
• Specification Elements: Key elements include output provenance, accountable actor identification, action period, and traceable action status.
• Nonaction Justification: Nonaction should not be labeled as justified to conceal failure; permissible explanations must be documented.
• Handling Repeated Outputs: Establish rules for repeated outputs before data collection, associating them with a single action opportunity unless distinct actions are anticipated.

Key Findings:

• AI outputs must be linked to specific action opportunities to assess their effectiveness.
• The framework emphasizes accountability and traceability of actions taken following AI outputs.
• Nonaction must be transparently documented to avoid misinterpretation of AI effectiveness.

Interpretation:

The framework aims to ensure that AI outputs are not mistaken for direct health benefits without clear action linkage.

Limitations:

• The framework does not address causal relationships between AI outputs and health outcomes.
• Ambiguous cases require independent adjudication, which may introduce variability.

Conclusion:

The proposed framework provides a structured approach to evaluate AI's role in public health actions, emphasizing the need for clear accountability and documentation.

Sources:

• WHO Guidance on AI for Health
• Lancet Public Health Article
• BMJ FUTURE-AI Guidelines
• J Med Internet Res Study