Clinical evaluation of GenAI adaptive cancer therapy

The learning and inferencing capabilities of large language models (LLMs) that underlie generative artificial intelligence (GenAI) can be exploited to optimize cancer treatments and improve patient outcomes. The ability of these models to learn from large amounts of clinical, molecular, and radiomic data on cancer patients and their treatments is driving research interest in their application to treatment decision-making. The learning and predictive power of LLMs make them uniquely suitable for supporting adaptive cancer therapy. However, the clinical validation of GenAI support for clinical decisions in oncology needs to address the complexity and unique challenges of GenAI clinical interventions. The United States Food and Drug Administration (FDA) guidelines on the clinical evaluation of software as a medical device (SaMD) are explored as a basis for the clinical evaluation of GenAI-assisted adaptive cancer therapy. Metrics are proposed to address clinical associations and analytical validation along with an outlook on randomized clinical trials. This article provides a much-needed and timely perspective on the clinical evaluation of GenAI-assisted cancer treatments and provides insights into overcoming the inherent challenges of GenAI in its acceptance and adoption in real-world clinical settings. • LLMs enable data-driven optimization of adaptive cancer therapies. • GenAI supports clinical decision-making through multi-modal learning. • FDA SaMD guidelines inform GenAI clinical evaluation frameworks. • Proposed metrics address analytical and clinical validation needs. • Insights offered for real-world adoption of GenAI in oncology care.