Skip to content

Gender bias in medical care and AI

Did you know that women are more likely to die from a heart attack than men? The main reason for this is that our medical knowledge is based on clinical studies that have primarily included white men. As a result, it is more difficult for medical professionals to recognise a heart attack if presenting symptoms are different from what they have learned as being standard. This has been identified as one of the main reasons for the disparities in the diagnosis and treatment of heart disease in women. As several studies have shown, cardiac arrests are often not recognised in female patients because of different presenting symptoms1. Failing to diagnose heart disease can lead to women being less likely to receive life-saving treatment and being more likely to die as a result of a cardiac arrest2, in particular when being treated by a male physician3. Gender inequities in medical care and health outcomes have also been found across other medical conditions such as critical illness (including cardiac arrest)4 and hip or knee arthroplasty5.

With the rise of AI in the medical domain, there is growing concern that this bias will find its way into AI. For instance, if AI is trained on existing data and clinical practice patterns, it may be less likely to recommend testing for cardiac arrest in female patients compared to male patients. Additionally, healthcare professionals may be more likely to accept decisions made by AI that reinforce current clinical practice resulting in a self-reinforcing feedback loop that perpetuates healthcare disparities. To prevent bias from creeping into promising new technologies such as AI, there is a need to enforce responsible AI practices and to minimise bias in the datasets that are used to train AI. By doing that, AI has the potential to improve medical care and health outcomes. To give an example, frequently used chest X-ray databases for AI are 60% male, leading to lower diagnostic performance for women. Increasing the representativeness of databases to 50% female has been shown to improve diagnostic performance of AI for both female and male patients6

In order for AI training datasets to be diverse and representative, we need to ensure that clinical trials are diverse and representative and that gender is included in research designs and reported in research outcomes. Additionally, gender differences in presenting symptoms and responsiveness to treatments need to be explicitly integrated into medical education programs. A less obvious but effective solution is to increase the number of female medical professionals, in particular in areas that are still male dominated such as intensive care and surgery. As research has shown, mortality rates for female patients with a heart attack decrease if they are treated by female physicians or by male physicians who work with more female colleagues7. Together, these measures will hopefully decrease gender inequities in medical care and health outcomes and ensure that both women and men benefit from new technologies such as AI.

  1. Dey, S., Flather, M.D., Devlin, G., Brieger, D., Gurfinkel, E.P., Steg, P.G., Fitzgerald, G., Jackson, E.A., & Eagle, K.A. (2009). Sex-related differences in the presentation, treatment and outcomes among patients with acute coronary syndromes: the Global Registry of Acute Coronary Events. Heart, 95(1), 20-26; Goldberg, R.J., O’Donnell, C., Yarzbeski, J., Bigelow, C., Savageau, J., & Gore, J.M. (1998). Sex differences in symptom presentation associated with acute myocardial infarction: a population-based perspective. American Heart Journal, 136(2), 189-195; Shulman, K.A., Berlin, J.A., Harless, W., Kerner, J.F., Sistrunk, S., Gersh, J.,…Escarce, J.J. (1999). The effect of race and sex on physicians’ recommendations for cardiac catherization. The New England Journal of Medicine, 340, 618-626. ↩︎
  2. Kudenchuk, P.J., Maynard, C., Martin, J.S., Wirkus, M., & Weaver, W.D. (1996). Comparison of presentation, treatment, and outcome of acute myocardial infarction in men versus women (the Myocardial Infarction Triage and Intervention Registry). The American Journal of Cardiology, 78(1), 9-14. ↩︎
  3. Greenwood, B.N., Carnahan, S., & Huang, L. (2018). Patient-physician gender concordance and increased mortality among female heart attack patients. Proceedings of the National Academy of Sciences, 115(34), 8569-8574. ↩︎
  4. Fowler, R.A., Sabur, N., Li, P., Juurlink, D.N., Pinto, R., Hladunewich, M.A., Adhikari, N.K.J., Sibbald, W.J., & Martin, C.M. (2007). Sex-and age-based differences in the delivery and outcomes of critical care. Canadian Medical Association Journal, 177, 1513-9151. ↩︎
  5. Hawker, G.A., Wright, J.G., Cote, P.C., Williams, J.I., Harvey, B., Glazier, R., & Badley, E.M. (2000). Differences between men and women in the rate of use of hip and knee arthroplasty. The New England Journal of Medicine, 342, 1016-1022; Parsley, B.S., Bertolusso, R., Harrington, M., Brekke, A., & Noble, P.C. (2010). Influence of gender on age of treatment with TKA and functional outcome. Clinical Orthopaedics and Related Research, 468(7), 1759-1764.  ↩︎
  6. Larrazabal, A.J., Nieto, N., Peterson, V., Milone, D.H., & Ferrante, E. (2019). Gender imbalance in medical imaging datasets produces biased classifiers for computer-aided diagnosis. Proceedings of the National Academy of Sciences, 117(23), 12592-12594. ↩︎
  7. Greenwood, B.N., Carnahan, S., & Huang, L. (2018). Patient-physician gender concordance and increased mortality among female heart attack patients. Proceedings of the National Academy of Sciences, 115(34), 8569-8574. ↩︎