Demographically informed models for improving synthetic haematocrit and extracellular volume estimation in cardiac computed tomography

Cardiac computed tomography-derived extracellular volume (CCT-ECV) is a promising biomarker for non-invasive quantification of myocardial fibrosis. However, serum haematocrit (Hct) is required for accurate CCT-ECV calculation, posing a potential barrier to clinical implementation. This study aims to develop a method for predicting synthetic Hct to derive accurate ECV values without blood testing and investigate the impact of clinical factors on model performance.

A total of 108 patients [70% male, body mass index (BMI) 27.2 (7.4) kg/m², age 81.9 (8.6) years] undergoing CCT prior to clinically indicated transcatheter aortic valve implantation for severe aortic stenosis were recruited. A non-contrast baseline scan, electrocardiogram (ECG)-gated CT angiography, and a late iodine-enhanced scan were performed on the same day as blood tests for serum Hct and used to compute voxel-wise ECV in the left ventricle. A univariable linear regression model was developed to predict Hct from Hounsfield units at the centre of the blood pool, outperforming previous models in literature. Sex stratification improved accuracy, with a significant difference in models for men at a BMI threshold of 30.7 (P = 0.035). In females, restricting to BMI > 22.4 improved performance. Age, estimated glomerular filtration rate, and creatinine did not improve predictions. The final model with combined sex and BMI stratification demonstrated better performance (ECV Pearson R 0.89, P < 0.001) than univariable and literature models.

This study highlights the necessity for sex-specific models to estimate Hct and accurately estimate ECV from CCT. Sex-specific BMI stratification further improves predictions; however, more research is required for females with a low or very high BMI.