Study finds women’s accounts of misdiagnosis often missing from medical records

Women’s accounts of misdiagnosis, broken trust and disrespect often failed to appear in their medical records, researchers Annemarie G. Hirsch and Jennifer Silva write in a November 6 STAT opinion essay. Their study paired interviews with 87 women with electronic records, exposing gaps they warn could shape what AI learns about patients—and whose experiences it carries forward.

The record shapes what AI can see

The authors describe two places where AI enters this information chain. AI scribes turn recorded conversations between doctors and patients into clinical notes. Other tools search and summarize existing records to help with diagnosis and treatment planning. Both depend on the information captured during care, rather than a complete, independent account of a patient’s life.

Hirsch and Silva argue that a medical record reflects clinical decisions as well as biology. The questions a clinician asks, the assumptions brought to an appointment and the details selected for a note all shape its contents. Patients also decide what feels safe to disclose. Fear of judgment or dismissal can leave important experiences unspoken, and therefore undocumented.

Their AI warning follows from that filtering: a system may reproduce how a woman is framed in her record rather than recover what was left out. The interview comparison examines documentation gaps; it is not a test showing an AI system caused or repeated them.

A diagnosis after years of explanations

The interviews extended beyond symptoms and appointments. Hirsch, an epidemiologist, and Silva, a sociologist, asked women about work, family, poverty and violence alongside pain, illness and encounters with healthcare. That approach gave them a second account to compare with the narrative preserved in each woman’s electronic record.

One participant, given the pseudonym Kate, lived with debilitating pelvic pain for more than a decade. According to the authors, doctors repeatedly linked her symptoms to obesity, and her notes recorded judgments about her eating habits. After bariatric surgery and substantial weight loss, the notes became more optimistic. Her symptoms persisted, however, and the record shifted toward mental-health explanations.

The authors identify an appointment attended by Kate’s new husband as the turning point. The clinician documented his concerns about painful intercourse and praised his questions. Doctors then conducted the examinations that led to an endometriosis diagnosis. In the authors’ account, the record preserved years of discussion about weight and mental health while missing the frustration and lost trust associated with remaining undiagnosed.

The proposed repair starts during care

For wider context, the authors cite the September Gallup-Pivotal Women’s Health Survey. It found that 34% of U.S. women said a provider had minimized or dismissed their symptoms in the previous five years. Another finding concerned diagnosis: 33% of women reported being misdiagnosed or unable to obtain a diagnosis, compared with 21% of men.

Hirsch and Silva’s proposed changes focus on how information enters the record, not just how software processes it. Their recommendations include:

• Make healthcare settings safe enough for women to describe their experiences openly.
• Listen when patients’ experiences do not fit a clinician’s expectations.
• Document delayed or incorrect diagnoses, without clinicians fearing a penalty for recording them.

With fuller records, the authors argue, AI could help identify patterns of missed diagnoses and reduce disparities in care. That is their proposed opportunity, not an outcome demonstrated by this comparison: better documentation could give the tools different evidence to work with.