Our nation has been embroiled in social unrest for several years, reaching a boiling point in May 2020 with the death of George Floyd. Racial inequities permeate all facets of our lives, from housing and education to law enforcement and the workplace.
Health care is not immune to such inequalities. The United States is facing a crisis in health disparity — the unequal burden of illness, injury or mortality experienced among population groups.
Research by the W.K. Kellogg Foundation and Altarum, a nonprofit organization dedicated to advancing health among at-risk and disenfranchised populations, estimates that disparities lead to approximately $93 billion in excess medical care costs and $42 billion in lost productivity per year. Close to home, Cleveland rightfully boasts its role as a leading medical destination. However, the city has a poor record on health disparity. Understanding inequities and addressing them through the combination of appropriate medical therapies and technology such as artificial intelligence (AI) is paramount to population health management in Northeast Ohio, the nation and around the world.
There is mounting evidence that many of the disease risk calculators and companion diagnostic tests used today aren’t reliable for all populations. For instance, a recent study revealed that the Oncotype DX Breast Recurrence test — the most commonly ordered genomic test to ascertain risk of distant recurrence for early-stage breast cancer and gauge whether a person will benefit from chemotherapy — is less accurate for Black women than white women. Information used to develop the test derived from studies in which only 5% to 6% of the women participating were Black, which raises questions about the prognostic accuracy of the Oncotype DX Breast Recurrence Scores in women from minority groups.
AI tools can provide invaluable assistance to counteract potential inaccuracies, utilizing big data to more reliably predict disease risk, outcomes and therapy response. However, it’s important to note that AI tools alone aren’t the panacea to health disparities, particularly if their development is prone to the same bias as diagnostic tool construction.
News stories have fairly pointed out problems with AI when implicit bias infiltrates algorithms. For instance, multiple studies cited in an article in The Atlantic warn about the potential for racial disparities to crop up when using machine learning for skin cancer screenings because the data derives primarily from light-skinned populations. This is especially troubling since the mortality rate for Black Americans with skin cancer is notably higher than white Americans, according to the American Academy of Dermatology. Researchers and tech companies involved in the creation of algorithms need to remain vigilant against bias creep.
However, by responsibly and thoughtfully harnessing the power of big data and AI, researchers can create population-specific models that more accurately predict risk and recurrence of disease. In addition, equitable representation of all populations within models can help discern underlying biological differences in cancers and other diseases that might exist between different racial groups. For instance, an AI analysis of digitized images of cancer tissue conducted by the Center for Computational Imaging and Personalized Diagnostics (CCIPD) has shown critical variations between Black and white prostate cancer patients. Currently employed risk calculators, nomograms and companion diagnostic tests for prostate cancer don’t account for these variations.
Similarly, a study of more than 400 Black and white women presented at the 2021 American Society for Clinical Oncology’s annual meeting in June points to significant differences in the appearance of endometrial cancer between the two populations. By taking these differences into consideration, more nuanced, population-specific models could more effectively address biological differences and better predict recurrence of prostate cancer, endometrial cancer and other diseases in minority populations. This, in turn, could lead to better patient care and ultimately lives saved.
Partnerships are being created to foster work that counteracts health disparities. In April, the CCIPD signed a memorandum of understanding with Hampton University, a historically Black research university in Virginia, to develop AI tools for addressing cancer disparities. The technologies will help answer important questions about which Black men undergoing proton therapy are likely to respond to the treatment.
“I am thrilled that Hampton University will be working with leaders in AI in medicine,” said Bill Thomas, vice president for government relations at the university. “Together, Hampton University and Case Western Reserve University will be able to push the boundaries on addressing cancer disparities and make a positive impact on the health of underserved communities.”
There is a lot of long-overdue attention being paid today to social inequity and racial injustice in America. While society examines the treatment of minority populations, it can’t overlook one key segment — health care. Using responsibly developed AI tools that employ population-specific risk prediction models, we can recognize important biological variations, then develop and optimize tools that benefit all patients. It’s a small, but critical, step toward rectifying health disparity.
Madabhushi is the Donnell Institute Professor of Biomedical Engineering at Case Western Reserve University and director of the Center for Computational Imaging and Personalized Diagnostics. He also is a research health scientist at the Louis Stokes Cleveland Veterans Administration Medical Center.