Multivariate adaptive bivariate regression splines (MABRS) binary response for modeling stroke and hypertension in RSKD Dadi City Makassar

Abstract

Classification of health conditions with more than one correlated response variable is an important challenge in medical data analysis. This study proposes a Multivariate Adaptive Bivariate Regression Splines (MABRS) approach to classify stroke type (ischemic vs. hemorrhagic) and hypertension status simultaneously. Utilizing clinical data from stroke patients, the MABRS model was built based on the optimal parameter combination for each response. The results showed that the stroke type classification achieved a fairly good performance (accuracy 76.82%), with the most influential variables being obesity, hypercholesterolemia, and diabetes mellitus. In contrast, the hypertension classification model performed poorly (51.21% accuracy), although diabetes mellitus, gender, and age were identified as the main predictors. The MABRS approach has the advantage of capturing nonlinear relationships and interactions between predictor variables, while considering the correlation between two binary response variables in a unified model framework. These findings confirm the potential of MABRS in uncovering complex relationships between clinical variables and supporting data-driven medical decision-making, particularly in the management of comorbidities in stroke patients.