Endometrial cancer — in which tumors develop in the inner lining of the uterus — is the most prevalent gynecological cancer in American women, affecting more than 66,000 women a year. Black women are particularly at risk, with an 80% higher mortality rate than other demographic groups and a greater chance of contracting more aggressive cancer subtypes. Regardless of lifestyle choices and health care equity, studies still show Black women have lower survival rates. A team of Emory researchers wondered: Could that poorer prognosis in Black women be caused by pathologic and genetic differences as well?
"Racism and equitable access to health care certainly play a big role in the increased mortality for populations of color," says Anant Madabhushi, executive director of the Emory Empathetic AI For Health Institute. "But with endometrial cancer, it may not completely explain the difference in mortality. One of our underlying hypotheses is that beyond social determinants of health, there are also potentially biological differences between different populations that need to be studied in a very precise way."
Searching for differences in cancer development
In their research, published recently in NPJ Precision Oncology, Madabhushi's team examined tissue slides of endometrial tumors from populations of both African American and European American women. They looked for differences in overall structural features and complex microscopic interaction in the tumor between the body's natural immune cells and different subcellular features such as connective tissue.
Anant Madabhushi
"We started with images of the tissue, which we had for both Black and white women with endometrial cancer," Madabhushi says. "Machine learning identified a set of features that were unique to Black women, a set unique to white women, and a set of features if you combine the two populations together."
Machine learning helped sort the results into different risk models for the two groups, based particularly on differences between the behavior of tumor infiltrating lymphocytes, the white blood cells that attack the tumor as part of the body's natural response to cancer. In the data used to train AI models, white women were more likely to have tumors associated with better survival rates, while Black women had a higher proportion of subtypes with higher mortality.
The AI-developed risk profiles found that in Black women, those lymphocytes tended to interact more with the cellular supporting tissue called stroma, but in white women, they interacted with epithelial tissue, the layer of protective cells that line internal and external body surfaces. The investigators found that a risk model that combined both groups didn't accurately predict risk for Black women. Only the risk profile developed specifically from the data from Black women was able to more precisely predict greater risks for that group.
"We absolutely could not have made these discoveries without AI," says Madabhushi, who's affiliated with Emory's Winship Cancer Institute. He doesn't hesitate to call it a "stunning" insight into ways cancer develops in different populations.
"There really are differences with regard to the immune architecture in Black women compared to white women," he says. "The epithelium is traditionally where all the action is and where pathologists tend to look. But it turns out that the immune architecture, particularly in the stroma, was stunningly different between Black women and white women. When we leveraged that pattern, we were able to create a predictive model that worked much more accurately in Black women in prognosticating outcome for endometrial cancer."
New discovery could aid cancer immune therapy
The discovery has implications for the developing science of immune therapy, which harnesses the body's natural defenses, including tumor infiltrating lymphocytes, to fight cancer.
"As we think about therapeutics for Black women with endometrial cancer, we may have to explicitly consider the findings in this study in the way we design immunotherapies for Black women as opposed to the way we've previously done it, which is sort of a one size fits all," says Madabhushi.
He adds that gynecologic cancers haven't been as publicly visible as breast cancer tends to be. One of the goals of the Empathetic AI for Health Institute is to look at projects that disproportionately affect underrepresented populations.
"If you want to pick an example of a disease that disproportionately affects women of color, It has to be endometrial cancer," he says. Though the new study doesn't explain everything about how tumors develop in different populations, Madabhushi says it clearly establishes that cancer progression isn't a single thing but is driven by genetic differences.
