As the use of artificial intelligence and predictive analytics in medicine becomes more commonplace in hospitals worldwide, the regulations around these approaches must keep pace - to ensure that the technology will improve the quality and efficiency of patient care. In a Policy Forum, Ravi Parikh and colleagues discuss the growing concern of a general lack of regulatory and clinical standards for predictive analytics in light of recent FDA approvals for its clinical use. Recent advances in artificial intelligence and computational power have opened new doors in the use of predictive analytics to increase prognosis efficiency and improve patient outcomes. However, the regulations and standards regarding the evaluation of the safety and impact of modern AI-based algorithms are very new - not as well-understood by clinicians and not held to traditional clinical trial standards, according to the authors. Despite this, several commercial devices based on advanced analytics have been recently approved by the Federal Food and Drug Administration (FDA), which has raised some concerns about the rigors of current regulatory processes. According to Parikh et al., existing FDA standards do not account for the unique dynamic characteristics of modern algorithms; for example, their mode of changing predictive performance in light of increasing amounts of data. To address these growing concerns, the authors propose five standards that could help guide the evaluation and regulation of clinical predictive analytics. Among these, Parikh et al. stress that AI-based systems need to be evaluated using established clinical standards and account for the agency of a clinician's actions within the data they are trained upon.
###
Journal
Science