JMIR Publications

ABSTRACT

Artificial intelligence (AI) is a broad discipline that aims to understand and design systems that display properties of intelligence1. Machine learning (ML) is a subset of AI which describes how algorithms and models can assist computer systems in progressively improving their performance2. Based on publicly available information, in late September 2021, U.S Food and Drug Administration (FDA) listed 343 AI/ML-enabled medical devices marketed in the U.S. In healthcare, an increasingly common application of AI and ML is software as a medical device (SaMD) which has the intention to diagnose, treat, cure, mitigate, or prevent disease3. Regulatory frameworks for SaMD need to be adaptive while prioritising patient safety and effectiveness4,5,6. Regulatory challenges of SaMD include processing submitted evidence to verify generalisability, interoperability, data integrity, and data security. Constructing a fit-for-purpose regulatory framework for SaMD with a continuous learning algorithm is an added complexity. As regulatory agencies aim to advance health care delivery through SaMD adoption, with efforts to avoid unintended consequences, this commentary summarises the current regulatory frameworks for SaMD. First we describe the challenge of continuous learning algorithms, then highlight the new evidence standards and frameworks under development, and discuss the need for stakeholder engagement, concluding with two key steps that regulators need to address in order to optimise and realise the many benefits of SaMD.