Development of a web-based multi-compatibility analysis software and an application example in healthcare

Objective: This study aimed to examine the multidimensional relationship structure among sociodemographic and clinical variables of breast cancer patients — including age, menopausal status, tumor size, number of involved lymph nodes, lymph node capsule status, tumor grade, breast quadrant, and breast side — using simple correspondence analysis and multiple correspondence analysis techniques. Materials and Methods: The study utilized an open-access dataset obtained from the Institute of Oncology, University Medical Centre Ljubljana, Yugoslavia. The dataset comprises 286 observations and 9 variables, including 68 breast cancer patients (23.8%) and 218 healthy individuals (76.2%). An original interactive web-based interface was developed using the R-Shiny framework as part of this thesis; all simple correspondence analysis and multiple correspondence analysis procedures were performed through this interface. Results: Multiple correspondence analysis revealed that the eigenvalues of the first two dimensions were 0.3097 and 0.2759, accounting for 13.5% of the total variance. The highest contributions to the first dimension were observed for the number of involved lymph nodes (0.166), lymph node capsule status (0.111), and tumor grade (0.056). On the perceptual map, postmenopausal status, high tumor grade, and large tumor size were positioned in proximity to cancer presence, whereas premenopausal status and low grade were located in the opposite direction. Conclusion: Lymph node involvement (contribution: 0.166) and capsule invasion (contribution: 0.111) were identified as the strongest explanatory variables. Tumor grade (0.056), tumor size, and menopausal status were determined as other major predictors of breast cancer progression. The first two dimensions explained 13.5% of the total variance, while the first ten dimensions accounted for 52%. Postmenopausal status, high grade, and large tumor size were positioned in the same direction as cancer presence, whereas premenopausal status and low grade occupied the opposite pole. Although the substantial overlap of confidence ellipses indicated limited discrimination, category-level patterns were found to be statistically significant. The findings confirm that the breast cancer risk profile is shaped by distinct clinical indicators and contribute to the existing literature. Keywords: Simple correspondence analysis, Multiple correspondence analysis, Breast cancer, Exploratory multivariate analysis, Categorical data visualization.