鶹ýӰ

AI Model Can Predict Breast Cancer Risk Regardless of Patient's Race

— Algorithm proved accurate where traditional risk models have failed, but needs more testing

MedpageToday

CHICAGO -- A deep-learning model using mammograms alone showed similar accuracy for predicting a woman's risk of breast cancer, regardless of her race, a researcher reported here.

For calculating 5-year risk of any breast cancer, the artificial intelligence (AI) algorithm demonstrated a similar area under the receiver operating characteristic curve (AUC) for white (0.71), Asian (0.68), and Black (0.73) women, according to Leslie Lamb, MD, of Massachusetts General Hospital and Harvard Medical School in Boston.

The model's performance was similarly accurate when looking at ductal carcinoma in situ (DCIS) or invasive breast cancer specifically, and show that mammograms contain highly predictive biomarkers of future cancer risk, she concluded in her Radiological Society of North America presentation.

Risk-based mammography screening has been elusive, as traditional risk models such as do not perform well on the individual level, said Lamb. Despite similar risks, white women are 2 to 3.5 times more likely to be classified as increased-risk than people of color.

"Consequently, patients in racial and ethnic minority groups may be inadvertently precluded from life-saving clinical interventions and protocols that reduce breast cancer risk, thereby exacerbating breast cancer disparities," she said.

In performing the study, Lamb and colleagues gathered data from five facilities at one institution, which included 129,340 routine bilateral mammograms from 2009 to 2018, including 83% from white women, 5% from Asian women, and 5% from Black women, with the remaining involving patients of other races or unknown.

Commenting on the findings, Stamatia Destounis, MD, of Elizabeth Wende Breast Care in Rochester, New York, called out the predominantly white patient population studied.

"While the numbers of exams scrutinized by the deep-learning algorithm in this study are huge -- there were 6,154 mammograms from Black women," said Destounis, who was not involved in the research. "The low percentage [5%] makes me wonder if we would have seen the same results had there been a better representation of Black women."

With at least 5 years of follow-up, a total of 3,401 cancers were diagnosed among the 71,479 patients in the study (mean age 59 years), including 2,615 invasive breast cancers (89.5% white, 3.7% Asian, 4% Black) and 786 cases of DCIS (88.7% white, 3.5% Asian, 4.9% Black).

Overall, the deep-learning model showed an identical AUC of 0.71 for the 5-year risk of any cancer, DCIS, or invasive breast cancer. And the model performed similarly across races:

  • AUC for any cancer: 0.71 for white, 0.68 for Asian, and 0.73 for Black
  • AUC for DCIS: 0.71, 0.71, 0.74, respectively
  • AUC for invasive cancer: 0.71, 0.67, 0.73

Lamb acknowledged the smaller proportions of patients of color and the single-institution design as study limitations and said that future studies will be needed to validate the results "in larger cohorts of patients of diverse races and ethnicities and comparing them to traditional models."

Strengths of the study included the large number of mammograms examined, she said, and that the current analysis did not include any mammograms from the training and testing sets of the deep-learning model.

  • author['full_name']

    Ed Susman is a freelance medical writer based in Fort Pierce, Florida, USA.

Disclosures

Lamb reported relationships with GE HealthCare.

Destounis reported no relevant relationships with industry.

Primary Source

Radiological Society of North America

Lamb L, et al "Deep learning model translates imaging biomarkers to predict future DCIS vs invasive breast cancer risk across races" RSNA 2023.