AI and health equity in cancer: 3 areas of progress

Prof Jakob Kather’s research team at Else Kröner Fresenius Center for Digital Health at TUD Dresden University of Technology in Germany is using AI to revolutionize oncology treatment through biomarker extraction.

Dr Karin Dembrower, Senior Breast Radiologist at St Göran’s Hospital in Sweden, is overseeing a program using AI technology for breast cancer screening with promising results.

Dr Judy Gichoya, an Associate Professor at the Emory University School of Medicine and a leading figure in healthcare AI and translational informatics, is exploring how AI can be used to address global health challenges and optimize resource allocation for maximum impact.

The three experts at the intersection of medicine and computer science were featured on a recent webinar presented by The Lancet Group: AI and Health Equity in Cancer 새 탭/창에서 열기. Moderated by Dr Ben Abbott 새 탭/창에서 열기, Executive Editor of The Lancet 새 탭/창에서 열기, and Dr Rupa Sarkar 새 탭/창에서 열기, Editor-in-Chief of The Lancet Digital Health 새 탭/창에서 열기, the forum explored the opportunities and challenges of using AI in cancer care and oncology, with an emphasis on health equity. Here are some highlights.

AI in precision oncology

“In 2010 … we had very limited options; there were only a certain number of chemotherapy drugs available,” he said. “But in 2023, these guidelines were much, much more extensive. If you now look at the guidelines for how we treat a patient, for example, with lung cancer or any other cancer, then you can see that we find these huge decision trees.”

“So we have to make many, many decisions to choose between many different types of treatments. The question is: How do we make this decision? One thing that can help us with this is biomarkers. A biomarker is something that you measure in cancer tissue or in cancer patients that helps you make the right treatment decisions and prescribe the right tree.”

Delving into the practical application of generalist models like ChatGPT in oncology, Prof Kather emphasized the need for additional context to enhance responses: “We need to use it in the right way in order to explore what this technology can do for us.”

AI for breast cancer screening

Reflecting on the equity considerations in AI implementation, Dr Dembrower stressed the importance of diverse training datasets and robust validation processes to address disparities. She raised concerns about access to AI-integrated screening, noting disparities in awareness and attendance based on socioeconomic status and cultural factors.

“We want to have an AI algorithm that is trained on a diverse population where minorities are represented,” she said. “And it’s not always easy to get access to diverse datasets. And that is something we have to be aware (of) and take into account the validation process …”

Dr Dembrower also addressed the challenges around access. For example, in Sweden there are only a few hospitals working with AI-integrated screening. “Although women can decide where they want to be screened,” she said, “our experience is that women with more resources (are) more likely to find out where AI is integrated … compared to women with lower social economic status. And we have also noticed that there is a lower attendance in the screening program of women with lower socioeconomic status because of language, culture and an inability to access the screening facility.”

Where does AI bring the most ROI?

Dr Gichoya underscored the potential of AI in enhancing patient engagement and improving access to healthcare information: “Today, you don’t need to rely on English as the only language where you can get information. You can ask questions: Where can I get screening? How do I plan for my screening?

“But I want to stop and pause here because when you think about these large language models — this is one form of AI that can be used especially for patient engagement — we see that the answers that are represented represent the majority opinion.”

Because guidelines differ around the world, for example, “it’s difficult to say what guidelines should you use.”

She pointed out that because LLMs learn from preexisting patterns, they’re not always applicable universally. She mentioned “historical biases that have pervaded in our medical literature” and challenges involving health equity.

“It’s very difficult to translate AI across different settings,” she said. For example, “we still find, disparities where we see Black women tend to have still worse outcomes. And it is a cascade effect from who gets screening and to who gets treatment. And so just fixing one problem is not going to end up saving our lives.”

Dr Gichoya raised thought-provoking questions about the practical implications of AI implementation in diverse healthcare settings and the need for tailored solutions that consider local contexts and guidelines.

She highlighted the complexities of translating AI advancements across different healthcare settings and populations, emphasizing the importance of considering the broader implications of AI implementation beyond diagnosis. She challenged the audience to critically evaluate the impact and resource allocation associated with AI development and implementation, urging a thoughtful approach to maximize the benefits of AI in healthcare.

“There’s always an opportunity cost,” she said. “So what after you get an AI diagnosis? Can we treat? Do we have the necessary resources? And where are we diverting resources when we spend all this amount of money on AI?”

She urged a nuanced approach to AI development and regulation to ensure equitable and effective use of technology in improving global health outcomes.