Summary
Provides a risk-based approach to monitoring the accuracy and performance of predictive health AI after deployment, including model drift and transparency about performance and updates. It does not cover generative AI, AI scribes, or initial model training, and does not specify metric thresholds that trigger retraining.
Healthcare Implications
Developers and deploying healthcare organizations can use the standard to structure ongoing performance monitoring and review model changes. It complements the existing CTA-2135 row, which addresses verification and validation before deployment. Adoption is voluntary.