The perceived risk of missing a breast cancer diagnosis with imaging studies often leads to unnecessary breast biopsies, according to a new report published in the American Journal of Roentgenology.
The authors showed how statistical methods can be used to downgrade the risk classification of breast masses to reduce the need for unnecessary biopsies. Clinicians from San Antonio, TX-based Seno Medical and medical center collaborators from the University of Texas (MD Anderson and U of Texas Health Sciences Center) co-authored the report.
The authors studied a statistical calculation known as the negative likelihood ratio (NLR), which can be calculated from a diagnostic test's sensitivity and specificity, and outlined how the breast imaging and reporting data system (BI-RADS) 4A subcategory has low enough and narrow enough range of pre-test probabilities to allow downgrading to a post-test probability of 2% or less after a negative diagnostic imaging test with an adequately low NLR.
Each BI-RADS category is associated with a specific range of risks of breast cancer. The approach includes the following steps:
- Classify lesions according to BI-RADS category 4 subcategories. Subcategory 4A represents the subcategory where the range of PPVs is both low enough and narrow enough to allow downgrading to BI-RADS category 3.
- Achieve the positive predictive value (PPV) is within the American College of Radiology (ACR) benchmark PPV range for BI-RADS subcategory 4A, which is greater than 2% but no greater than 10%.
- Ensure the NLR is adequate for a negative test finding to reduce the post-test probability to 2% or less, using Bayes' theorem.
Currently, anytime a breast imaging study determines that the risk of breast cancer is greater than 2%, many clinicians recommend a biopsy, said Thomas Stavros, MD, the chief medical officer at Seno Medical and a professor specialist at the University of Texas Health Services in San Antonio.
"Sometimes ancillary diagnostic breast imaging studies are performed to reduce risk to less than 2%, but it is difficult to know exactly how much risks are reduced even after a negative ancillary diagnostic imaging examination," Stavros said. "However, the use of the NLR [in Bayes' theorem] along with BI-RADS 4 subcategories can help to reduce the number of false-positives without experiencing excessive negative results that would lead to cancer going undiagnosed."
This approach simplifies a much more involved, five-step process that involves Thomas Bayes, an English statistician and minister known for formulating a specific case of the theorem that bears his name, the Bayes' theorem. Radiologists are taught to use Bayes theorem and follow multiple steps to eventually arrive at the probability that a suspicious lesion is cancerous.
"Mathematically it's not really difficult but it's hard to remember and it's easy to make mistakes and it's time-consuming so radiologists just haven't used it much," Stavros told MD+DI.
In fact, only about 33% of radiologists currently use BI-RADS 4 subcategories, he said.
The researchers are in the process of creating a webpage to help radiologists calculate the post-test probability based on a specified pre-test probability and the NLR. Or, the radiologist could enter a desired post-test probability, such as 2%, and an NLR, to calculate the maximum pre-test probability that can be reduced to 2% by that NLR. It could also be used to figure out the NLR necessary to achieve the post-test probability based on a specified pre-test probability and the desired post-test probability.
All three calculation options would simply require the input of two values to get a third value, without the headache of performing manual Bayes theorem calculations, Stavros said.
"But I think NLR, together with Bayes' theorem, and the use of BI-RADS 4 subcategories, especially when achieving the ACR benchmark range of 2% to 10% for 4a, is the key to actually reducing our false positive rates, without missing more cancers," Stavros said.