Computed tomography to estimate cardiac preload and extravascular lung water. A retrospective analysis in critically ill patients

In critically ill patients intravascular volume status and pulmonary edema need to be quantified as soon as possible. Many critically ill patients undergo a computed tomography (CT)-scan of the thorax after admission to the intensive care unit (ICU). This study investigates whether CT-based estimation of cardiac preload and pulmonary hydration can accurately assess volume status and can contribute to an early estimation of hemodynamics.

Thirty medical ICU patients. Global end-diastolic volume index (GEDVI) and extravascular lung water index (EVLWI) were assessed using transpulmonary thermodilution (TPTD) serving as reference method (with established GEDVI/EVLWI normal values). Central venous pressure (CVP) was determined. CT-based estimation of GEDVI/EVLWI/CVP by two different radiologists (R1, R2) without analyzing software. Primary endpoint: predictive capabilities of CT-based estimation of GEDVI/EVLWI/CVP compared to TPTD and measured CVP. Secondary endpoint: interobserver correlation and agreement between R1 and R2.

Accuracy of CT-estimation of GEDVI (< 680, 680-800, > 800 mL/m²) was 33%(R1)/27%(R2). For R1 and R2 sensitivity for diagnosis of low GEDVI (< 680 mL/m²) was 0% (specificity 100%). Sensitivity for prediction of elevated GEDVI (> 800 mL/m²) was 86%(R1)/57%(R2) with a specificity of 57%(R1)/39%(R2) (positive predictive value 38%(R1)/22%(R2); negative predictive value 93%(R1)/75%(R2)). Estimated CT-GEDVI and TPTD-GEDVI were significantly different showing an overestimation of GEDVI by the radiologists (R1: mean difference ± standard error (SE): 191 ± 30 mL/m², p < 0.001; R2: mean difference ± SE: 215 ± 37 mL/m², p < 0.001). CT GEDVI and TPTD-GEDVI showed a very low Lin-concordance correlation coefficient (ccc) (R1: ccc = +0.20, 95% CI: +0.00 to +0.38, bias-correction factor (BCF) = 0.52; R2: ccc = -0.03, 95% CI: -0.19 to +0.12, BCF = 0.42). Accuracy of CT estimation in prediction of EVLWI (< 7, 7-10, > 10 mL/kg) was 30% for R1 and 40% for R2. CT-EVLWI and TPTD-EVLWI were significantly different (R1: mean difference ± SE: 3.3 ± 1.2 mL/kg, p = 0.013; R2: mean difference ± SE: 2.8 ± 1.1 mL/kg, p = 0.021). Again ccc was low with -0.02 (R1; 95% CI: -0.20 to +0.13, BCF = 0.44) and +0.14 (R2; 95% CI: -0.05 to +0.32, BCF = 0.53). GEDVI, EVLWI and CVP estimations of R1 and R2 showed a poor interobserver correlation (low ccc) and poor interobserver agreement (low kappa-values).

CT-based estimation of GEDVI/EVLWI is not accurate for predicting cardiac preload and extravascular lung water in critically ill patients when compared to invasive TPTD-assessment of these variables.