Performance of a Logistic Regression Model Using Paired miRNAs to Stratify Abnormal Mammograms for Benign Breast Lesions

Abstract

Background

Mammography is effective in reducing breast cancer mortality, but it has false positive results that cause subsequent interventions such as biopsy or interval repeat mammography. Thus, there is a clinical unmet need for accurate molecular classifiers that can reduce unnecessary additional imaging and/or invasive diagnostic procedures for low-risk women.

Method

We performed miRNA profiling on a prospectively collected serum specimen obtained from each of the 432 subjects who received an abnormal mammogram or imaging result and then selected 265 subjects for further analysis. The miRNA classifier, named EarlyGuard, was generated based on a novel logistic regression model using “paired miRNAs” where the two miRNAs of interest exhibit the same properties.

Results

The classifier developed using the training set of 174 subjects enrolled at seven investigative sites resulted in a negative predictive value (NPV) and a sensitivity of 96.4% and 91.2%, respectively. The classifier was validated using the test set consisting of 91 subjects enrolled at three investigative sites, two of which were not included in the training set. The resulting NPV and sensitivity were estimated similarly to be 96.9% and 95.8%, respectively.

Conclusions

Our miRNA classifier has produced promising results that could be used in conjunction with mammography or other imaging procedures to reduce unnecessary invasive diagnostic procedures for women who are unlikely to have a suspicious or worse result on a subsequent diagnostic biopsy. Additional studies will be conducted in larger cohorts to determine if the sensitivity of the classifier will be improved.