Marlex Lance Manalili, Reece Sergei Lim, Gerard Ompad
The Sessa Empirical Estimator is a data-driven method that constructs treatment episodes from observational data by incorporating the K-Means clustering algorithm. However, K-Means is limited by its assumption of spherical clusters, which often does not hold true for complex, real-world pharmacoepidemiological datasets exhibiting non-convex shapes and varying densities. This study extends the Sessa Empirical Estimator by integrating alternative clustering algorithms to accommodate such data characteristics. Various Cluster Validity Indices were also utilized to determine optimal cluster configurations. Statistical analysis, employing the Mann-Whitney U test, demonstrates a highly significant difference between the distributions of median duration estimates from the novel Sessa Empirical Estimator and the traditional approach. The proposed method showed that the novel Sessa Empirical Estimator consistently yields more compact distributions with fewer outliers, indicating superior consistency, precision, and reliability. This improved Sessa Empirical Estimator provides a more robust and accurate methodology for estimating medication exposure, enhancing the utility of observational data in pharmacoepidemiological research. © 2025 IEEE.
University of San Carlos, Department of Computer, Information Sciences and Mathematics, Cebu City, Philippines
