The Potentiality of AI In Deepening Degenerative Healthcare

  • 10/09/2024
  • Category: Commentaries
No Comments

Author: Emira Anjani
Editor: Hafiz Noer

The convergence of AI in medical science is predisposed to revolutionize healthcare, particularly in its utilization as a tool to enhance the potency of ‘Precision Medicine’. Precision medicine is a biomedical paradigm that underscores interpersonal differences designed to facilitate personalized medical interventions tailored to the precise needs of patients.1 Fundamentally, irrespective of viral or bacterial infection, biological (sex) and non-biological (gender) factors constitute as relevant determinants of symptomatic and asymptomatic conditions in various illnesses. These include prevalence, risk factors, age of onset, necessary treatments, and manifestation of symptomologies, prognosis, and biomarkers.2 In their journal, Cirillo, D. et al (2020) presented various medical findings that proved a correlation between sex and gender with chronic diseases, such as autoimmunity, cancer, cardiovascular diseases, mental health disorder, diabetes, neurological diseases, and more. Under these circumstances, as a ‘one-size-fits-all’ approach is becoming obsolete against a myriad of variables, precision medicine has gained a strong reputation as a promising medical approach, especially when enhanced by AI, which increases accuracy in medical practices. However, it is not without flaws. Recent research highlights that while precision medicine offers improved precision, it still faces significant challenges, particularly in addressing structural issues related to gender disparity.3

Prior to proceeding, something must be clarified. In the interest of simplicity and coherence, this article departs from the understanding that sex and gender are interrelated.4 While the two factors are of its own distinct conception, they interact dynamically in shaping social roles, norms, identity, behavior, and social perception. By this virtue, from this point forward, sex and gender will be merged into a singular term of ‘gender,’ unless clarified otherwise. 

Data is The Problem

At its core, one should not make the reductionist mistake of analogizing technological accuracy with efficiency. It is true that if designed and configured properly, AI could be a powerful tool to the development of biomedicine in the absence of mitigating discriminatory and confounding factors. However, arbitrary reliance on AI as an augmentor of precision medicine could lead one into repudiation that technology acts as a double-edged sword that could amplify and perpetuate existing inequalities.5 The problem is, gender inequalities currently manifested in AI stemmed from a data pool that does not sufficiently represent women. At the current level of global social awareness on gender inequality under representation may pose a cliché problem. Yet, precisely because it is a cliché it signifies that structural patriarchy remains insufficiently mitigated hitherto. 

As a male-dominated field, a greater portion of biomedical research is predominantly concentrated on male clinical subjects and experimental models.6 In general, only about 5-10% of biomedical research concerns gender differences as a significant factor of epidemiology.7 As such, out of the 11 million studies published between 1960-2018 only a rough 30% represent male and female equally–and even less on just women.8 Several reasons are frequently cited to justify this disparity, including bioethical and socio-cultural factors such as social perceptions, stereotypes, and stigma.9 Others, which are expressed in preference to experiment with male rodents than females, include biotechnical concerns pertaining that hormonal cycles and reproductive systems are seen as confounding variables to research outcomes.10 

Meanwhile, the performance and reliability of AI is dependent upon the dataset it learned from. Hence the equation is straight forward: an AI tool that learns from a dataset consisting of an overrepresentation of one party will generate results commodious to said party. Commodious or effectivity, in this context, is not to be paralleled with accuracy. Parties of lower data proportion ought to receive accurate results (be it diagnosis, prognosis, or treatment recommendation) as per the tool’s available data. However, said result may not be catered effectively to the parties’ needs based on the variation configured by their biological and non-biological factors. For instance, the University College London (UCL) recently published that AI-powered liver screening devices exhibit risk of inaccuracy at 44% in women and 23% in men.11 Another study revealed that although coronary heart disease is the leading death cause of women, 67% of participants in clinical studies for treatment and assistive devices are conducted on males.12 Although sex contributes in determining pharmacokinetics and pharmacodynamics of cardiovascular medications, such as Statins, generated recommendations do not typically consider sex as a factor.13 Instead, therapeutic treatments and medications are often generalized.

De-biasing Bias

The first and most obvious solution to the presented issue is to debias AI programs by adding data inclusive to women–and its intersectionality, though it may not be as straightforward as it seems. The shortage of data, which this article emphasizes as the cause of AI bias, is part of a broader issue that stems from gender inequality. To develop effective mitigation strategies, it is essential to understand the extent of shortages or biases in the currently available data. This is not an impossible task, per se, as the Director of Research at the Distributed AI Research Institute, Alex Hanna, states that bias in datasets can be investigated using computational methods.14 However, gender bias in AI is more complex than it seems because it does not end exclusively at sex; rather it simultaneously transcends across the intersectionalities of gender, race, ability, and ethnicities. To de-bias data exclusively at the gendered dimension, one bears the likelihood of missing these intersections, which then arguably incite an ableist problem in turn. Hence there are no certain means of debiasing a dataset. 

Adjacent to engrossing efforts into the dataset, another important measure necessary to de-bias AI is to enforce changes in real life. Part of the problem in AI bias lies as well on the who question. That is, who contributes to the studies and formulates the dataset? The role of women is crucial not only in the dimension of practitioners and researchers but also in the greater context of STEM, particularly computational biology and bioinformatics, to challenge gender bias by ensuring that AI data formulation has accommodated various gender groups and its intersectionalities.15 

Ultimately, while the integration of AI into precision medicine holds great potential for enhancing personalized healthcare, it is not without challenges. The persistent gender biases within both biomedical research and AI datasets underscore the need for more inclusive approaches. Addressing these biases requires not only expanding and refining the data that AI systems are trained on but also promoting greater diversity among researchers and practitioners in the field. Only through such comprehensive efforts can we ensure that AI truly serves the diverse needs of all patients and contributes to a more equitable healthcare system.

  1. Cirillo, D. et al. (2020) ‘Sex and gender differences and biases in artificial intelligence for biomedicine and healthcare’, npj Digital Medicine, 3(1), p. 81. Available at: https://doi.org/10.1038/s41746-020-0288-5; Chen, D. (2024) ‘Ethical frameworks of informed consent in the age of pediatric precision medicine’, Cambridge Prisms: Precision Medicine, 2, p. e6. Available at: https://doi.org/10.1017/pcm.2024.3. ↩︎
  2. Cirillo, D. et al. (2020) ‘Sex and gender differences and biases in artificial intelligence for biomedicine and healthcare’, npj Digital Medicine, 3(1), p. 81. Available at: https://doi.org/10.1038/s41746-020-0288-5 ↩︎
  3. Ruth March (2023) The changing landscape of precision medicine, Astra Zeneca. Available at: https://www.astrazeneca.com/what-science-can-do/topics/technologies/precision-medicine-history.html (Accessed: 24 August 2024);  Cirillo, D. et al. (2020) ‘Sex and gender differences and biases in artificial intelligence for biomedicine and healthcare’, npj Digital Medicine, 3(1), p. 81. Available at: https://doi.org/10.1038/s41746-020-0288-5; Chen, D. (2024) ‘Ethical frameworks of informed consent in the age of pediatric precision medicine’, Cambridge Prisms: Precision Medicine, 2, p. e6. Available at: https://doi.org/10.1017/pcm.2024.3. ↩︎
  4. Martha Anker (2011) ‘Taking sex and gender into account in emerging infectious disease programmes: An analytical framework’. World Health Organization. Available at: https://iris.who.int/bitstream/handle/10665/207693/9789290615323_eng.pdf?sequence=1. ↩︎
  5. Jongsma, K.R., Sand, M. and Milota, M. (2024) ‘Why we should not mistake accuracy of medical AI for efficiency’, npj Digital Medicine, 7(1), p. 46. Available at: https://doi.org/10.1038/s41746-024-01047-2. ↩︎
  6. McGregor, A.J. et al. (2016) ‘How to study the impact of sex and gender in medical research: a review of resources’, Biology of Sex Differences, 7(S1), p. 46. Available at: https://doi.org/10.1186/s13293-016-0099-1. ↩︎
  7. Galea, L.A. and Parekh, R.S. (2023) ‘Ending the neglect of women’s health in research’, BMJ, p. p1303. Available at: https://doi.org/10.1136/bmj.p1303. ↩︎
  8. Hariharan, K. (no date) How Will AI Affect Gender Gaps in Health Care? Available at: https://www.marshmclennan.com/insights/publications/2020/apr/how-will-ai-affect-gender-gaps-in-health-care-.html (Accessed: 24 August 2024). ↩︎
  9. Hay, K. et al. (2019) ‘Disrupting gender norms in health systems: making the case for change’, The Lancet, 393(10190), pp. 2535–2549. Available at: https://doi.org/10.1016/S0140-6736(19)30648-8. ↩︎
  10. Balch, B. (2024) Why we know so little about women’s health, AAMC. Available at: https://www.aamc.org/news/why-we-know-so-little-about-women-s-health (Accessed: 24 August 2024). ↩︎
  11. Straw, I. and Wu, H. (2022) ‘Investigating for bias in healthcare algorithms: a sex-stratified analysis of supervised machine learning models in liver disease prediction’, BMJ Health & Care Informatics, 29(1), p. e100457. Available at: https://doi.org/10.1136/bmjhci-2021-100457. ↩︎
  12. Xhyheri, B. and Bugiardini, R. (2010) ‘Diagnosis and Treatment of Heart Disease: Are Women Different From Men?’, Progress in Cardiovascular Diseases, 53(3), pp. 227–236. Available at: https://doi.org/10.1016/j.pcad.2010.07.004; Dhruva, S.S., Bero, L.A. and Redberg, R.F. (2011) ‘Gender Bias in Studies for Food and Drug Administration Premarket Approval of Cardiovascular Devices’, Circulation: Cardiovascular Quality and Outcomes, 4(2), pp. 165–171. Available at: https://doi.org/10.1161/CIRCOUTCOMES.110.958215. ↩︎
  13. Stock, E.O. and Redberg, R. (2012) ‘Cardiovascular disease in women’, Current Problems in Cardiology, 37(11), pp. 450–526. Available at: https://doi.org/10.1016/j.cpcardiol.2012.07.001. ↩︎
  14. Hanna, A. (2023) ‘Understanding Gender and Racial Bias in AI’. Available at: https://www.sir.advancedleadership.harvard.edu/articles/understanding-gender-and-racial-bias-in-ai. ↩︎
  15. Holman, L., Stuart-Fox, D. and Hauser, C.E. (2018) ‘The gender gap in science: How long until women are equally represented?’, PLOS Biology. Edited by C. Sugimoto, 16(4), p. e2004956. Available at: https://doi.org/10.1371/journal.pbio.2004956; Buslón, N. et al. (2023) ‘Raising awareness of sex and gender bias in artificial intelligence and health’, Frontiers in Global Women’s Health, 4, p. 970312. Available at: https://doi.org/10.3389/fgwh.2023.970312. ↩︎