Mayo
For women aged 40 to 49, ambiguous breast screening guidelines can muddy the waters when clinicians attempt to evaluate cancer risk and determine when it's appropriate to screen. An image-based deep learning model shows promise for helping providers make more informed decisions based on each patient's real likelihood of developing cancer.
"Screening recommendations can be confusing given the inconsistency coming from various sources, particularly for women in their 40s," said Ray Cody Mayo III, MD, associate professor from MD Anderson Cancer Center in Houston, TX. "What we know is that with appropriate risk prediction we have the ability to identify patients who are more likely to develop breast cancer within the next five years."
Given that information, Dr. Mayo said, it's up to the medical community to make sure these patients get screened appropriately.
Similar Risk Revealed for Women in Their 40s and Older Women
Dr. Mayo's research team examined how an image-based deep learning model could help identify patients aged 40 to 49 whose five-year risk is similar to that of patients aged 50 to 69. They examined more than 30,000 digital screening mammograms collected from six imaging programs over a five-year period. They extracted the patients' demographics from electronic health records and obtained cancer outcomes from local tumor registries.
The algorithm was designed to make predictions based on calibrated five-year risk and to classify risk levels as average, intermediate and high, based on National Comprehensive Cancer Network thresholds. "We estimated the percentage of exams for each threshold by age group and compared observed cancer rates—cancer-positive exams to all exams—across risk and age groups," Dr. Mayo explained. "Chi-squared tests were used to measure statistical differences across independent risk and age groups."
Predictably, the team found that cancer rates increased with each age group. But deep learning risk score thresholds also distinguished exams in patients with lower and higher cancer rates. For patients in their 40s, cancer rates were 0.7% in the average risk subgroup, 3.3% for the intermediate risk subgroup and 5.1% in the high-risk subgroup.
The forty-somethings' five-year cancer rate of 3.3% was similar to the rate—3.2%— identified for patients aged 50 to 69. Similarly, the high-risk group's 5.1% cancer rate was comparable to a 5.0% rate in intermediate-risk patients in their 50s and 60s.
"These results are not only interesting for women in their 40s, but the cohort of women in their 50s and 60s also shows that perhaps those at intermediate or high risk could potentially benefit from knowledge of their risk status," Dr. Mayo said. "For example, 50- to 69-year-old patients with dense breast tissue who have a deep learning algorithm assignment of intermediate or high risk may be more strongly considered for screening ultrasound."
As pointed out by moderator Wendy Burton Demartini, MD, in order to perform risk-based assessment, researchers need a baseline mammogram. Dr. Mayo agreed, "I think it absolutely lends itself to getting a baseline screening at age 40," he said. "The secondary benefit is that you now have a subset of women you can call back specifically. This is a way we can take clinically relevant, translatable action from this model."
Access the presentation, "Screening Mammography: Can a Deep Learning Model Support Expanded Screening of Patients in Their 40s?," (W7-SSBR09-5) on demand Meeting.RSNA.org.
