Patterns and predictors of left ventricular remodeling: A machine-learning based approach

The project aims 1) to identify different patterns of left ventricular (LV) remodeling and 2) to identify relevant clinical predictors for these remodeling patterns on a population-based level. To achieve these aims, data from two cohort studies, SHIP (Greifswald) and KORA (Augsburg), will be combined and analyzed using state-of-the-art machine-learning methods. To assess LV remodeling, detailed magnetic resonance imaging (MRI) data are available in both studies, which include measurements of wall thickness in 16 segments, LV volumes and mass and diastolic filling rates. These data will be evaluated by unbiased, unsupervised clustering methods to derive different LV remodeling phenotypes, which are characterized by specific wall thickness and end-diastolic patterns. To assess potential clinical predictors for these LV remodeling phenotypes, extensive measurements of laboratory biomarkers and further covariables are available in both studies. In a hypothesis-free manner, automated variable selection methods will be applied to evaluate the association and relevance of clinical predictor variables with the LV remodeling phenotypes. The final result of this project consists in a manuscript draft where the novel LV remodeling phenotypes and associated clinical predictors are presented. This cooperation provides an opportunity for combining Greifswald’s substantial expertise in complex statistics with the ample database given by the MRI measurements from both SHIP and KORA.