Sex-Specific Patterns of Left Ventricular Remodeling Using Regional Wall Thickness Data and Their Associations with Cardiovascular Disease Risk

Background: Left ventricular (LV) remodeling is associated with impaired cardiac function and future cardiovascular disease (CVD). Current remodeling definitions use broad categorizations based on hypertrophy and mean wall thickness. Cardiac magnetic resonance (CMR) imaging provides detailed characterization of regional myocardial wall properties, which may enhance sex-specific cardiovascular disease (CVD) risk stratification. Objectives: We aimed to identify detailed LV remodeling patterns, their associations with CVD risk, and their clinical predictors. Methods: LV wall thickness data were obtained by CMR in three independent population-based cohorts (SHIP-TREND-0, n=931; SHIP-START-2, n=490; KORA-FF4, n=368). Sex-specific remodeling patterns were identified by k-means clustering and associated with established CVD risk scores and incident morbidity and all-cause mortality. Bootstrapped multinomial regression with LASSO regularization was used to select relevant clinical predictors of remodeling patterns. Results: The sample comprised 991 men (mean age 52.9 years, prevalent CVD 7.8%) and 798 women (52.5 years, 2%). Four remodeling clusters were found for men and women, respectively. For a subset of these clusters, significant associations with an increased CVD risk were found, e.g. in women, the high-risk cluster was associated with a 10.6 (95% confidence interval: 8.9, 12.3) percentage point increase in the 10-year Framingham Risk Score. Associations were independent of blood pressure and myocardial mass. Only in women, associations were also independent of average wall thickness and LV concentricity. Variable selection identified distinct clinical predictors of remodeling patterns. Conclusion: Particularly in women, regional LV wall thickness patterns detect unfavorable cardiac remodeling and might improve CVD risk stratification beyond existing strategies. Automated implementation during image acquisition and integration with shape-based models may facilitate clinical application.