Systematic misclassification of left ventricular mass by echocardiography is driven by ventricular geometry: a CT-based reference analysis

Background Left ventricular (LV) mass is a cornerstone of cardiovascular risk stratification. Transthoracic echocardiography (TTE) estimates LV mass using simplified geometric assumptions, yet the extent to which these assumptions introduce systematic bias in contemporary populations remains poorly defined. Purpose To quantify the magnitude, direction, and structural determinants of discordance between TTE- and CT-derived LV mass, and to evaluate the clinical implications of this misclassification. Methods We analysed 2,121 individuals undergoing CT coronary angiography in the BioHEART Study, with automated deep-learning–derived LV mass, including 463 with paired TTE. LV mass determinants were assessed using multivariable regression. Inter-modality discordance was defined as the relative difference between TTE and CT measurements and modelled against clinical and structural parameters, including LV geometry. Model performance and variance explained were assessed using adjusted R². Results Clinical determinants, including sex, anthropometry, hypertension, heart failure, atrial fibrillation, and renal function, were independently associated with LV mass and explained a substantial proportion of its variance (adjusted R² 0.72–0.75). In contrast, comparisons between modalities revealed a systematic, directionally consistent bias. TTE-derived LV mass deviated by >25% from CT in 23% of cases, with a characteristic pattern of overestimation at lower LV mass and underestimation at higher LV mass, consistent with non-linear geometric distortion. Importantly, discordance was driven overwhelmingly by ventricular geometry rather than clinical phenotype. Increased wall thickness and chamber size were the dominant predictors of measurement error (all p<0.001), collectively explaining most of the variance in discordance (adjusted R² 0.93). After accounting for geometric parameters, traditional clinical variables contributed minimally, indicating that misclassification arises primarily from intrinsic limitations of geometric modelling in TTE. These findings identify a subgroup of patients (those with altered ventricular geometry) in whom TTE-derived LV mass is most unreliable, with potential implications for risk classification and downstream management. A clinically derived prediction model demonstrated robust estimation of LV mass using routinely available variables, providing a scalable alternative in settings without advanced imaging. Conclusion TTE-derived LV mass is subject to systematic, geometry-dependent misclassification, with predictable over- and underestimation across the LV mass spectrum. These findings challenge the assumption that echocardiographic LV mass is a reliable surrogate for true myocardial mass and highlight the need for geometry-aware approaches or alternative modelling strategies to improve risk stratification.Figure 1Figure 2