Statistical learning confirms the diagnostic significance of the anemia panel in breast cancer

Diagnostic value of available tumor markers, such as cancer antigen CA 15-3 and carcinoembryonic antigen (CEA) in breast cancer is limited. There is an ongoing search for additional, potentially better diagnostic blood markers with improved clinical utility. The aim of this study is to evaluate performance of the approach based on routine blood tests accompanied by a statistical learning tool to the diagnosis of breast cancer.

Blood was collected from total of 104 subjects which were divided into two groups: breast cancer patients and a control group that consisted of asymptomatic volunteers and patients who had benign breast lesions at the time of blood collection. Random forest statistical learning method and the external method validation have been applied to evaluate diagnostic performance of 31 routine blood tests.

The applied statistical learning approach assigned the highest diagnostic importance to the anemia panel among all analyzed blood tests that also included CA 15-3. External validation has shown utility of selected statistical approach - we were able to select tests that provide a diagnostic accuracy comparable to some diagnostic tools described in literature and based on more demanding laboratory techniques, such as gene expression microarrays.

Inclusion of tests for anemia significantly improves diagnostic accuracy for the breast cancer in comparison to the diagnostic accuracy of the CA 15-3 alone. Application of the random forests also enables the reduction of number of laboratory tests needed for the establishment of diagnosis. Differences in relevant test values between the cancer and control group are small but application of multiparametric statistical learning ensured diagnostic accuracy of 72.0% associated by a sensitivity of 64.7% and specificity of 84.9%.