D4.2 Validation methods for AI-based decision support systems

Deliverable 4.2 (D4.2 Validation methods for AI-based decision support systems) is part of Task 4.2 (WP4 DHT Assessment Framework, Toolkit and Manual), led by Fundación Progreso y Salud (FPS) with the involvement of TUD, UOXF, and NICE. The task addresses a critical gap between regulatory approval and local clinical adoption of AI-based Clinical Decision Support Systems (AI-CDSS). While many AI tools obtain CE marking or FDA approval, their performance and safety in local populations often remain untested, creating risks for patient safety and barriers to adoption.

To close this gap, D4.2 describes VALID-AI (VAlidation of Local Indicators in the pre-Deployment phase of AI-CDSS), a guideline for the external validation of AI-CDSS using local Real-World Data (RWD) in the pre-deployment, post-regulatory phase. VALID-AI provides methodological and practical guidance to ensure algorithms demonstrate accuracy, robustness, reproducibility, generalizability, fairness, and clinical relevance before being integrated into local workflows.