Abstract
Breast cancer is considered a major women’s health issue in Taiwan. Particle swarm optimization (PSO) is a novel algorithm used to identify the optimize solution within a multidimensional search space. This study used complementary-logical PSO (CLPSO)–selection Cox proportional hazard (CoxPH) regression model to identify high-risk clinicopathologic characteristics in breast cancer overall mortality. This CLPSO-based Cox regression analysis were conducted using a prospective breast cancer registry database maintained by a Medical University Hospital. In total, 1896 patients with breast cancer treated between 2005 and 2017 were analyzed with an institute approval. CLPSO was used to identify the high impact characteristics on overall breast cancer mortality, and CoxPH was used to demonstrate the association between overall mortality and the selected characteristics. The C-statistic was used to estimate the performance of CLPSO– selection and general models. The CLPSO-selected high mortality risk characteristics included tumor size of >2.4 cm (HR = 2.37, p = 0.006), lymph node metastases (HR = 1.30, p = 0.038), dermal invasion (HR = 1.548, p = 0.028), and no hormone therapy (HR = 2.178, p = 0.003), which had achieved relatively higher accuracy than general model (76%). The study results demonstrated CLPSO–selection model could provide more accurate estimation for overall mortality in breast cancer with fewer characteristics and better prediction ability. Hence, CLPSO model may contribute for the clinical assessment in breast cancer patient mortality.