FDA Clarifies Regulatory Expectations for AI Tools in Heart Disease Prevention

FDA Issues Draft Guidance on AI/ML Software for Cardiovascular Risk Assessment

The FDA issued draft guidance on April 3, 2023. This document outlines a framework for artificial intelligence and machine learning software as a medical device used in cardiovascular risk assessment. It focuses on data quality, algorithm transparency, bias mitigation, clinical validation in diverse populations, and predetermined change control plans. [1]

The guidance builds on good machine learning practice principles issued in 2021. Those principles came from a joint effort by the FDA, Health Canada, and the MHRA.

The FDA published 10 guiding principles for good machine learning practice on October 27, 2021. [2] The principles call for multi-disciplinary expertise, representative data sets, and independent review of models.

The 2023 draft applies these expectations to cardiovascular risk tools. It requires clinical validation across populations that reflect intended users by age, sex, race, and ethnicity so performance does not differ systematically among groups.

The draft sets expectations for transparency in how algorithms generate risk predictions. Developers must describe model inputs, outputs, and decision logic in submissions. The document also calls for bias detection and correction throughout the development process. [1]

These elements align with global standards. A June 2021 World Health Organization report described the need for governance structures that support fairness, transparency, and protection of patient autonomy in AI systems for health. [3]

AI models often incorporate new data or retrain after initial clearance. The draft instructs manufacturers to submit predetermined change control plans that define anticipated modifications and the protocols for verifying continued safety and effectiveness. [1]

Manufacturers must also establish plans for ongoing real-world performance monitoring once the software enters clinical use.

What this means

The draft translates the 2021 high-level principles into cardiovascular-specific expectations for premarket review. It places weight on the composition of training data and methods for bias assessment. The document indicates the evidence FDA will examine for these types of AI tools.

Key takeaways

• AI/ML tools for cardiovascular risk require rigorous, population-representative clinical validation. [1]
• Transparency, explainability, and bias mitigation count as key regulatory expectations for AI-enabled medical software. [1][2]
• Predetermined change control plans address management of post-approval algorithm updates. [1]
• Ongoing real-world performance monitoring supports safety and effectiveness of AI cardiovascular devices. [1]
• The framework builds directly on 2021 good machine learning practice principles issued with international partners. [2]

Limitations

This is draft guidance subject to revision after public comment. It focuses on software for risk assessment and does not cover all AI uses in cardiology such as imaging diagnostics. Long-term real-world evidence requirements remain evolving across jurisdictions.

Sources / References

1. Artificial Intelligence and Machine Learning Software as a Medical Device. U.S. Food and Drug Administration (FDA). https://www.fda.gov/medical-devices/software-medical-device-samd/artificial-intelligence-and-machine-learning-software-medical-device

2. Good Machine Learning Practice for Medical Device Development: Guiding Principles. U.S. Food and Drug Administration (FDA). https://www.fda.gov/medical-devices/software-medical-device-samd/good-machine-learning-practice-medical-device-development-guiding-principles

3. Ethics and governance of artificial intelligence for health. World Health Organization (WHO). https://www.who.int/publications/i/item/9789240029200