Time Series measurements of White Blood Cells (WBC) and plasma Cancer Antigen 15-3 (CA 15-3) in a patient with metastatic breast cancer, serving as a reference for assessing the clinical relevance of different clustering algorithms.

Clustering is highly valuable for analyzing cancer marker time series data. In a recent study, the significance of the K-means clustering algorithm was examined for the first time in evaluating time-series measurements of plasma Cancer Antigen 15-3 (CA 15-3) in a male patient with metastatic breast cancer. CA 15-3 is a glycoprotein commonly associated with breast cancer. The present study marks a significant advancement, as the time-series measurements of White Blood Cells (WBC) and plasma CA 15-3 from the same patient served as a reference for assessing the clinical relevance of various clustering algorithms. The time-series measurements of WBC are particularly noteworthy because the specific therapy administered to the patient has clearly defined all clustering parameters, including the number of clusters and their boundaries. This provides a unique opportunity for rigorous testing of nine different clustering algorithms. Remarkably, seven to nine of these algorithms demonstrate perfect alignment between the observed number of clusters and their boundaries, and the expected results based on the therapy administered to the patient. Furthermore, applying the nine clustering methods to the CA 15-3 time-series data revealed that K-means with Euclidean distance, K-means with Manhattan distance, and K-medoids are the most suitable algorithms for analyzing both the specific CA 15-3 and WBC time-series data.