Objective:

To introduce SAFE_DTx, a safety-oriented optimization framework for digital therapeutics (DTx) that prioritizes patient safety through the use of intermediate feedback signals.

Approach:

• Framework Introduction: SAFE_DTx focuses on estimating short-term feedback responses for therapy options and uses a planner to select optimal actions while adhering to safety constraints.
• Intermediate Feedback Focus: The framework emphasizes optimizing for intermediate feedback rather than solely relying on final clinical outcomes.
• Separation of Prediction and Planning: SAFE_DTx structurally separates clinical surrogate prediction from predefined safety-constraint filtering.

Key Findings:

• Current DTx personalization strategies face challenges due to limited long-term outcome evidence and unpredictable AI behavior.
• SAFE_DTx allows for fine-grained treatment adjustments while maintaining safety limits.
• The framework integrates elements from operations research and safe reinforcement learning to enhance patient safety.

Interpretation:

SAFE_DTx operationalizes adaptive principles for clinical DTx by focusing on patient safety through a structured approach to feedback and planning.

Limitations:

• The framework does not claim algorithmic novelty in its learning mechanisms.
• Existing DTx personalization approaches may not incorporate the proposed feedback prediction and constraint-aware planning architecture.

Conclusion:

SAFE_DTx represents a significant step towards ensuring patient safety in AI-driven digital therapeutics by prioritizing intermediate feedback and safety constraints.

Sources:

• Digital therapeutics from bench to bedside
• Artificial intelligence, bias and clinical safety
• A primer on reinforcement learning in medicine for clinicians
• Unleashing the potential of reinforcement learning for personalizing behavioral transformations with digital therapeutics
• A comprehensive survey on safe reinforcement learning
• Offline safe reinforcement learning for sepsis treatment
• New approach to equitable intervention planning to improve engagement and outcomes in a digital health program