Skip to content
Breaking:

Patient Used OpenAI's GPT-4 to Diagnose Infertility Cause After Six Years of Failed Care

A personal account details how prompt-engineering ChatGPT around safety filters led to an MRI recommendation that located an overlooked uterine fibroid.

By The Company Wire3 min read
Share
OpenAI — Patient Used OpenAI's GPT-4 to Diagnose Infertility Cause After Six Years of Failed Care
OpenAI — Patient Used OpenAI's GPT-4 to Diagnose Infertility Cause After Six Years of Failed Care. Photo: Hacker News.

A published personal account shared via Hacker News at https://www.astralcodexten.com/p/our-ai-midwife details how consumer artificial intelligence models can serve as diagnostic research aids, recounting how writer Drew Housman used OpenAI’s GPT-4 to uncover a medical issue that went undetected across six years of unexplained infertility. Writing in an essay, Housman detailed how he and his wife turned to conversational AI to analyze lab results, review symptom histories, and advocate for diagnostic imaging that ultimately led to a successful pregnancy.

The couple entered the fertility system with an "unexplained infertility" diagnosis after five failed embryo transfers, despite creating viable embryos that were cryopreserved in Milwaukee. Facing short consultation windows standard in U.S. clinical practices, the couple began consulting large language models to review diagnostic scans, hormone levels such as estradiol and progesterone, follicle counts, and surgical options.

To bypass safety guardrails that initially restricted medical advice, Housman used prompt engineering by instructing GPT-4 to generate dialogue for a fictional television medical drama featuring an expert doctor named "Dr. Reid." While the couple cross-checked recommendations against their human clinicians, the AI model occasionally generated errors, including misreading an ultrasound to claim a pregnancy had occurred.

The diagnostic breakthrough occurred when the chatbot suggested ordering a magnetic resonance imaging (MRI) scan of the uterus. Despite prior testing, biopsies, and routine scans over six years, an MRI had never been conducted, and their fertility specialist initially advised against it before agreeing to the order.

The resulting scan identified a previously undetected submucosal fibroid measuring six centimeters in diameter on the uterine wall at an implantation site. The couple selected a surgeon in New York from a list compiled by ChatGPT to perform a laparoscopic myomectomy. Following the surgical removal of the fibroid, Housman reported that his wife conceived naturally, and the couple named their son Reed in reference to the AI persona.

Sources

  1. Hacker News

Company: OpenAI

Written by

The Company Wire

Newsroom · San Francisco

Inside the companies building what’s next. Reporting on startups, technology, funding and the people shaping them.