JAMA + AI One-Year Anniversary December 6, 2025

Dr Grosenick’s selection: “Predicting Diagnostic Progression to Schizophrenia or Bipolar Disorder via Machine Learning,” published in JAMA Psychiatry on February 19, 2025:

"This study used electronic health records from more than 24,000 psychiatric patients to train machine learning models that predicted who would go on to develop schizophrenia or bipolar disorder. The models showed that free-text clinical notes carried powerful signals for forecasting illness progression, performing better than coded data alone.

Instead of classifying existing illness, this important work moves toward prospective risk prediction using data already available in real-world clinical practice. It demonstrates the feasibility of AI as a tool for early identification and intervention in serious mental illness.

By setting a precedent for large-scale, validated prediction studies, this paper will inspire researchers to design longitudinal, clinically actionable AI models. It highlights the need to incorporate narrative clinical text into predictive pipelines and to rigorously test algorithms across health systems."

Dr Carroll’s selection: “FDA Draft Guidelines for AI and the Need for Ethical Frameworks,” published in JAMA Pediatrics July 7, 2025:

"This article reviews new FDA guidance for AI in drug development and calls for stronger ethical frameworks, especially for pediatric digital twins (virtual models of a person or organ used to simulate care decisions), to ensure safe and equitable use. This shift is important because it shifts the focus from just making AI models accurate to making them ethically responsible, particularly for children.

Once adopted, these guidelines will push regulators and developers to create guardrails for consent, data rights, and long-term protections in pediatric AI. Going forward, we need pilot projects that include families and ethicists, plus research on how children can control their data as they grow up."