Researchers warn gender gap in medical data to affect AI accuracy

For years, researchers have raised concerns that medical research does not pay enough attention to biological differences between men and women, including differences in treatment responses and medication effectiveness. Now, the issue is gaining attention in another area of healthcare: first-aid training, particularly the mannequins used to teach people how to perform CPR.

The issue could have wider implications as artificial intelligence becomes increasingly integrated into healthcare, as AI systems are trained on medical research and health data that may reflect existing gaps in medical knowledge.

A 2024 study of 20 CPR mannequin models sold worldwide found that three-quarters were described as male or had no sex specified. Only one of the 20 models offered a breast overlay, according to an article by The Conversation citing the findings.

They reflect a broader tendency in medical education and research to treat the male body as the standard.

The lack of female representation in CPR training can have consequences. A US study of 19,331 out-of-hospital cardiac arrests found that 39% of women who collapsed in public received bystander CPR, compared with 45% of men.

For people suffering cardiac arrest, even a short delay can affect survival. Another US study found that patients who received bystander CPR four to five minutes after a witnessed cardiac arrest had 27% lower odds of surviving to hospital discharge than those who received it within one minute.

In England, fewer than one in 12 patients for whom ambulance services attempted resuscitation survived for 30 days, according to figures reported in 2024.

Researchers and medical educators have suggested that if bystanders hesitate to perform CPR on women because they are uncomfortable with the presence of breasts, training with more representative mannequins could help reduce that barrier.

The historical lack of female representation in medical research also has implications for healthcare AI. Medical AI systems learn associations from health data, but medical records reflect clinical decisions as well as biological differences.

If women and men with similar symptoms have historically been assessed differently, an algorithm could learn those patterns without understanding why they exist.

An experimental study involving medical students and residents, for example, found that when symptoms of coronary heart disease were presented alongside psychological stress, women received fewer coronary heart disease diagnoses and cardiology referrals than men. Their symptoms were also more likely to be interpreted as psychogenic.

An AI system trained on medical records shaped by such decisions could potentially learn those patterns as if they reflected differences in disease rather than differences in clinical assessment.

Reviews of AI in medicine have warned that unbalanced datasets can result in uneven performance. Assessing that performance can also be difficult. A 2024 review of 692 AI-enabled medical devices approved by the US Food and Drug Administration found that demographic information and details of performance studies were often missing from public documents.

Reasons for gender discrepancy

Women have historically been underrepresented in medical studies. Hormonal changes during the menstrual cycle were often viewed as a source of unwanted variation.

Following the thalidomide tragedy of the late 1950s and early 1960s, the FDA recommended in 1977 that women who could become pregnant be excluded from early drug trials. The policy was reversed in 1993 amid concerns that it was limiting knowledge about how medicines affected women.

Women of childbearing potential were routinely excluded from industry-sponsored trials. Male animals have also been favoured in laboratory research, despite evidence that females are not more biologically variable, while many studies have failed to report the sex of cells used in experiments.

These gaps can affect treatment. Women clear the sleeping drug zolpidem more slowly than men, and regulators recommended in 2013 that the starting dose for women be cut in half.

A 2020 analysis found that women experienced adverse drug reactions nearly twice as often as men, with differences in how drugs move through the body accounting for many of the variations.

Heart-attack symptoms further illustrate the importance of accounting for biological differences. An analysis of more than one million people with acute coronary syndromes found that 74% of women and 79% of men experienced chest pain. However, women were more likely to report other symptoms, including neck or jaw pain, fatigue and shortness of breath.

By Nazrin Sadigova