Collaboration across disciplines needed to guarantee fairness of AI in healthcare

While AI has shown potential in providing healthcare insights, concerns about bias remain. First author Ms Liu Mingxuan, a PhD candidate in the Quantitative Biology and Medicine Program and Center for Quantitative Medicine (CQM) at Duke-NUS, explains that a fair model should perform equally well across subgroups such as age, gender, and race. However, differences in performance may have clinical reasons and may not necessarily indicate unfairness.

Co-first author Dr. Ning Yilin, Research Fellow with CQM, suggests that focusing on equity, rather than complete equality, is a more reasonable approach for clinical AI. Adjusting the AI algorithm or its application can ensure that more vulnerable groups receive the care they need. Dr. Ning also highlights the importance of considering patient preferences and prognosis, as equal treatment does not always mean fair treatment.

The paper emphasizes the misalignments between AI fairness research and clinical needs. Associate Professor Liu Nan, a senior and corresponding author of the paper, comments on the difficulty of choosing suitable metrics to measure model fairness in healthcare, as they can often conflict with each other. He also stresses the importance of distinguishing between meaningful differences and true biases requiring correction in the medical context.

The authors stress the need for active engagement with clinicians in the development of fair AI models. They highlight the importance of evaluating which attributes are considered sensitive for each application. Clinicians can provide context, determine if differences are justified, and guide models towards equitable decisions.

In conclusion, the authors argue that achieving fair AI for healthcare requires collaboration between experts in AI, medicine, ethics, and beyond.

Co-author Associate Professor Daniel Ting, Director of SingHealth’s AI Office and Associate Professor at SingHealth Duke-NUS Ophthalmology & Visual Sciences Academic Clinical Programme, acknowledges the complexity of achieving fairness in the use of AI in healthcare. He stresses the need for collaboration between clinicians, AI experts, and industry experts to address fairness in AI and advance AI practices for patient care.

Co-author Clinical Associate Professor Lionel Cheng Tim-Ee, Chief Data & Digital Officer, Clinical Director (AI) Future Health System Department, and Senior Consultant at Singapore General Hospital (SGH), highlights the complexities of translating AI fairness techniques into ethical clinical applications. He emphasizes the collective commitment to developing trustworthy AI that enhances clinicians’ ability to provide quality and equitable care.

Professor Marcus Ong, Director of the Health Services & Systems Research (HSSR) Programme at Duke-NUS and Senior Consultant at SGH’s Department of Emergency Medicine, emphasizes the need for clinicians to be actively engaged in communication with AI developers to ensure models align with medical ethics and context. He emphasizes the importance of diverse experts in providing oversight and considering social and ethical nuances in the pursuit of equitable and unbiased AI for healthcare.

The perspective, published in npj Digital Medicine, is the result of collaboration between researchers from various institutions across Singapore, Belgium, and the United States. The authors from the SingHealth Duke-NUS Academic Medical Centre, along with experts from other institutions, worked together to provide valuable perspectives on fair AI in healthcare. The collaboration exemplifies the cross-disciplinary dialogues necessary to advance fair AI techniques.

Professor Patrick Tan, Senior Vice-Dean for Research at Duke-NUS, hopes that this collaborative effort will inspire further multinational partnerships in the pursuit of equitable and unbiased AI for healthcare.