AI depends on medical records that don't always tell the full truth about women

An opinion piece published November 6 in STAT News argues that artificial intelligence tools designed to aid diagnosis and treatment are limited by a persistent flaw in the medical records on which they are trained: those records do not always capture the complete clinical picture for women patients. The authors contend that gaps and biases in documentation, accumulated over decades of clinical practice, carry forward into the datasets used to build AI systems, with particular consequences for how women’s pain and mental health symptoms are recorded and interpreted.

According to the piece, the problem is not merely technical but rooted in longstanding patterns of how clinicians have historically documented women’s complaints, patterns that predate any AI application. The authors write that for AI to reach what they describe as its full potential in women’s health, medical records need to tell the complete story, a standard they argue current documentation practices do not meet.

The opinion piece calls for changes to how clinical data is collected and recorded, framing more complete and accurate documentation as a prerequisite for AI systems that serve women patients as well as they serve men. The authors present this as an urgent matter given the growing reliance on AI across diagnosis, treatment planning, and risk assessment in clinical settings, where flawed inputs risk being amplified rather than corrected by automated systems trained on them.