Paired interviews reveal 85% of women's reported misdiagnoses absent from electronic records

The STAT analysis of 87 women shows systematic under-documentation: 53 trust-breaking encounters appeared in only seven records, while disrespect was never noted despite 30 reports. This gap arises because records capture clinician assumptions, such as attributing pelvic pain to obesity or mental health, rather than patient narratives or unasked questions about violence, poverty, and work. AI scribes now transcribing visits risk encoding these omissions at scale, particularly for endometriosis and chronic pain where diagnostic delays already average seven to ten years.

Gender bias patterns compound the issue. Prior observational data from large cohorts indicate women receive fewer indicated diagnostic tests for cardiac and autoimmune conditions even after symptom reporting, a finding replicated across multiple health systems. When EMRs omit patient-expressed mistrust, downstream algorithms trained on these data will underweight female symptom credibility, widening outcome disparities rather than correcting them.

Next steps require prospective validation studies that link interview-verified patient experience to AI output accuracy, with stratification by sex and condition. Regulators should mandate audit trails for AI-generated notes that flag discrepancies between transcribed dialogue and final documentation.