Algorithm Change Protocol
The detailed procedural section of a PCCP that specifies how planned modifications to an AI/ML model will be developed, validated, and implemented.
Definition
An Algorithm Change Protocol (ACP) is one of three required components of a Predetermined Change Control Plan (PCCP) for AI/ML-enabled device software under FDA's 2024 final guidance. The ACP specifies the methodology for implementing the planned modifications described in the Description of Modifications: data management protocols, retraining triggers, performance evaluation methods, statistical analysis plans, acceptance criteria, update implementation procedures, transparency to users, and traceability. A well-constructed ACP is what lets manufacturers iterate on AI/ML models within an approved PCCP without filing supplements for each retrain.What this means in practice
The ACP is the section of a PCCP where most of the technical and operational rigor lives. FDA reviewers focus on whether the ACP makes future changes predictable, evaluable, and traceable. Common ACP weaknesses include vague triggers ('when performance drifts'), missing acceptance criteria, no plan for handling subgroup performance degradation, and no specification of how user communication will be handled. A strong ACP often runs 30+ pages and is the centerpiece of the PCCP review.- •Treating the ACP as a high-level outline, FDA expects specific protocols with quantitative criteria.
- •Omitting subgroup analysis methodology, disparate performance changes across patient subgroups are a focus area.
- •Failing to specify rollback procedures, what happens when an update doesn't meet acceptance criteria in production?
Related terms
Grouped by themeEditor's picks
· Hand-selected related conceptsMedical device that uses artificial intelligence or machine learning to perform its intended use.
Methods (SHAP, saliency maps, prototype explanations) that help users understand why a model made a given prediction.
Guiding principles for the development of AI/ML-enabled medical devices.
Distinction between models whose behavior is fixed at release vs. those that continue learning.
More in Software & AI
· Same categoryStructured documentation of a machine learning model's intended use, performance, and limitations.
Ongoing surveillance of input data and model outputs to detect performance degradation post-deployment.
FDA mechanism to pre-authorize specific modifications to AI/ML-enabled devices.
AAMI Technical Information Report providing guidance on applying Agile software development practices within an IEC 62304-compliant medical device software lifecycle.
Primary references
3 sources- 1
PCCP Guidance for ML-Enabled Devices (Final 2024)VerifiedFDAfda.gov
- 2
MDCG Software GuidanceVerifiedMDCGhealth.ec.europa.eu
- 3
FDA - Software as a Medical Device (SaMD)VerifiedFDAfda.gov
Inline markers like [1] jump to the matching reference above.