1 Introduction

Extravascular lung water (EVLW) is interstitial, intracellular, alveolar, and lymphatic fluid in the lungs, and thus reflects the amount of water in the lungs outside the pulmonary vasculature not including pleural effusions [1, 2]. In clinical practice, EVLW is indexed to biometric parameters (usually predicted body weight) to be able to use “normal ranges” of extravascular lung water index (EVLWI) despite inter-individual differences in biometric data [3, 4]. Accumulation of EVLW is a hallmark of both acute respiratory distress syndrome (ARDS) and hydrostatic pulmonary edema [1, 5]. EVLW(I) has been demonstrated to have a high predictive value regarding patient outcome in surgical patients [6, 7] and critically ill patients including patients with sepsis and ARDS [8,9,10,11].

At the bedside, EVLW can be estimated using single-indicator transpulmonary thermodilution (TPTD) [1]. TPTD is recommended for patients with circulatory shock not responding to initial therapy and shock complicated by ARDS [12]. In addition to EVLW, TPTD can be used to assess the pulmonary vascular permeability index (PVPI) that allows differentiating the pathophysiologic reason for increased EVLW [13, 14].

Computed tomography (CT) with quantitative analysis of the density of lung tissue has also been proposed to quantify pulmonary edema [15]. Quantitative CT allows the assessment of organ volumes and computation of the volume of gas, the volume of tissue, and the gas/tissue ratio [15]. CT of the chest has been suggested to quantify pulmonary edema both in experimental and clinical settings [16,17,18].

Diagnostic CT scans of the thorax are performed in many critically ill patients during intensive care unit (ICU) admission or treatment. To the best of our knowledge, there are no data in unselected ICU patients on the diagnostic value of quantitative CT analysis for the assessment of pulmonary fluid status using CT scans obtained in clinical routine. Therefore, we aimed to compare variables of pulmonary fluid status assessed using quantitative CT and TPTD in a clinical study in unselected critically ill patients.

2 Methods

2.1 Study design and setting

This observational clinical study was performed in patients treated in the medical ICU of a German university hospital (Klinikum rechts der Isar der Technischen Universität München, Munich, Germany). The study was approved by the ethics committee (Ethikkommission der Fakultät für Medizin der Technischen Universität München) and written informed consent was obtained from all patients or their legal representatives. Adult patients were eligible for study inclusion if they (a) were scheduled for CT scanning of the chest and (b) were monitored with TPTD (both for clinical reasons unrelated to the study). According to the study protocol, TPTD measurements (details see below) were performed directly before and after the CT examination.

2.2 Transpulmonary thermodilution measurements

We a priori defined TPTD as the reference method in our study. For TPTD we used the PiCCO system (Pulsion Medical Systems SE, Feldkirchen, Germany) as described previously [19, 20] with a 5-French thermistor-tipped catheter (Pulsiocath PV2015L20; Pulsion Medical Systems SE) placed in the abdominal aorta via the femoral artery. TPTD variables were calculated based on the analysis of the thermodilution curve after injection of 15 mL of iced 0.9% saline in the central venous circulation via a central venous catheter. The injection of the thermal indicator was performed in triplicate and each TPTD value represents the mean of the three consecutive measurements. EVLW was indexed to the predicted body weight resulting in EVLWI. As described previously [13] PVPI was calculated as the ratio of EVLW and pulmonary blood volume.

2.3 Computed tomography and derived calculations

CT was the test method in our study. CT scans were performed using a 64-slice multi-detector CT (Somatom AS, Siemens Healthcare GmbH, Forchheim, Germany). Chest CT was acquired with 120 kVp and automated tube current modulation. Images were reconstructed in transverse plane with a slice thickness of 3 or 5 mm. For this study, the CT scans were analyzed post hoc by two radiologists blinded to the clinical patient data and TPTD-derived parameters.

Manual segmentation was performed using Medical Imaging Interaction Toolkit (MITK) workbench (German Cancer Research Center, Heidelberg, Germany). Regions of interest (ROIs) were prescribed using the software’s region growing tool carefully excluding large vessels and extrapulmonary tissue. ROIs were prescribed individually for each axial slice and were initialized in well-ventilated lung tissue. Density of the lungs was determined in Hounsfield Units (HU). A value equal to 0 HU characterizes a voxel with a density equal to that of water and a value of − 1000 HU characterizes a voxel with a density equal to that of air. Based on the density data of the segmented images total lung tissue volume as well as percentages of hyperinflated (< − 900 HU), well-aerated (− 900 to − 500 HU), poorly aerated (− 499 to − 100 HU), and non-aerated (− 99 to + 100 HU) regions were calculated.

As described in detail previously [16] we calculated the tissue volume (TV) as:

$${\text{TV\,=\,}}\left( {{\text{volume}}\;{\text{of}}\;{\text{well-aerated}}\;{\text{lung}}\;{\text{tissue}} \times 0.3} \right)+\left( {{\text{volume}}\;{\text{of}}\;{\text{poorly}}\;{\text{aerated}}\;{\text{lung}}\;{\text{tissue}} \times 0.7} \right)+\left( {{\text{volume}}\;{\text{of}}\;{\text{non-aerated}}\;{\text{lung}}\;{\text{tissue}} \times 1.0} \right).$$

The tissue volume index (TVI) was obtained by indexing TV to predicted body weight.

In addition, we calculated the mean weighted index of voxel aqueous density (VMWaq) as a mathematical assumption of the relative contribution of water in the HU frame:

$${\text{VMWaq}}={{\left( {\left( {{\text{number}}\, {\left({\text{n}} \right)}\,{\text{of well-aerated lung tissue voxels}} \times 0.3} \right){\text{+}}\left( {{\text{n of poorly aerated lung tissue voxels}} \times 0.7} \right)+\left( {{\text{n of non-aerated lung tissue voxels}} \times 1.0} \right)} \right)} \mathord{\left/ {\vphantom {{\left( {\left( {{\text{number of }}\left( {\text{n}} \right){\text{well-aerated lung tissue voxels}} \times 0.3} \right){\text{+}}\left( {{\text{n of poorly aerated lung tissue voxels}} \times 0.7} \right)+\left( {{\text{n of non-aerated lung tissue voxels}} \times 1.0} \right)} \right)} {{\text{total number of voxels}}}}} \right. \kern-0pt} {{\text{total number of voxels}}}}.$$

2.4 Statistical analysis

Statistical tests were conducted in an exploratory manner on a two-sided 5% significance level. For statistical analyses we used IBM SPSS Statistics for Windows, Version 23 (IBM Corp., Armonk, NY, USA).

Descriptive data are presented as absolute and relative frequencies (categorical data) or as median and 25th and 75th percentile (continuous data).

We used the Spearman correlation coefficient to investigate bivariate correlations of quantitative measurements.

As primary endpoint we investigated the predictive value of the TVI for the prediction of EVLWI values of ≥ 14 mL/kg using receiver operating characteristics (ROC) curve analysis. We chose to use an EVLWI threshold of 14 mL/kg because this cut-off value has repeatedly been shown to be associated with increased mortality [21]. Considering the allocation ratio of 0.62 (mean EVLWI ≥ 14 and < 14 mL/kg), defining “good prediction” as an area under the ROC curve (ROC-AUC) of ≥ 0.80, and applying a 5% significance level, this primary endpoint of our analysis including 21 patients had a post-hoc power of 80%. In addition, we used ROC analysis to assess the predictive value of TV and VMWaq for the prediction of EVLWI values of ≥ 14 mL/kg.

3 Results

3.1 Patients

We included 22 critically ill patients in this study. One patient was excluded from the analysis because of technical problems with the CT analyzing software. Thus, we included 21 patients in the final analysis. The patients’ characteristics are presented in Table 1.

Table 1 Patients’ characteristics
Full size table

3.2 Pulmonary fluid status assessed using quantitative computed tomography

Data from quantitative CT analysis including information on the distribution of hyperinflated, well-aerated, poorly aerated, and non-aerated lung tissue are shown in Table 2 and Fig. 1.

Table 2 Transpulmonary thermodilution measurements and data from computed tomography
Full size table
Fig. 1
figure 1

Distribution of hyperinflated, well-aerated, poorly aerated, and non-aerated lung tissue. Distribution of hyperinflated (white), well-aerated (light gray), poorly aerated (dark gray), and non-aerated (black) lung tissue determined by quantitative computed tomography analysis (some columns do not sum up to 100% because of a small proportion of tissue with > 100 Hounsfield Units)

Full size image

Median TV was 733 (529–804) mL, median TVI was 10.1 (7.0–11.5) mL/kg, and median VMWaq was 0.32 (0.27–0.35).

3.3 Pulmonary fluid status assessed using transpulmonary thermodilution

Data on TPTD-derived EVLW, EVLWI, and PVPI are shown individually for each patient in Table 2.

Median EVLW before and after CT was 627 (526–959) and 751 (601–1100) mL, respectively. Median EVLWI was 10 (8–17) mL/kg both before and after CT.

3.4 Receiver operating characteristics curve analysis

According to ROC-AUC analysis, neither TV nor TVI (primary endpoint) significantly predicted EVLWI values of ≥ 14 mL/kg (Table 3). For VMWaq, ROC analysis yielded ROC-AUCs of 0.76 (p = 0.062) and 0.79 (p = 0.030) regarding an EVLWI value of ≥ 14 mL/kg before and after CT, respectively (for mean EVLWI ROC-AUC = 0.79, p = 0.030) (Table 3). Nevertheless, despite statistical significance, the ROC-AUCs were below the predefined threshold of 0.80.

Table 3 Prediction of extravascular lung water index ≥ 14 mL/kg using receiver operating characteristics curve analysis
Full size table

3.5 Correlation analysis

There was no significant correlation between TV and EVLW before CT (r = 0.13, p = 0.574), EVLW after CT (r = 0.05, p = 0.823), or mean EVLW (r = 0.05, p = 0.832).

There was no significant correlation between TVI and EVLWI before CT (r = 0.11, p = 0.626), EVLWI after CT (r = 0.06, p = 0.809), or mean EVLWI (r = 0.05, p = 0.840).

There was a statistically significant moderate positive correlation between VMWaq and mean EVLWI (EVLWI before and after CT) (r = 0.45, p = 0.042) and EVLWI after CT (r = 0.49, p = 0.025) but not EVLWI before CT (r = 0.38, p = 0.086).

4 Discussion

In this clinical study in unselected critically ill patients, we compared variables of pulmonary fluid status assessed using quantitative CT and TPTD.

CT-derived variables did not predict elevated TPTD-derived EVLWI values. In addition, there was no significant correlation of CT-derived TV and TVI with EVLW and EVLWI assessed using TPTD. Thus, in this study in unselected critically ill patients, variables of pulmonary fluid status assessed using quantitative CT could not be used to predict EVLWI.

Our findings in unselected critically ill patients are in contrast to previous studies that evaluated quantitative CT to estimate pulmonary fluid status under highly standardized conditions in experimental settings.

In an experimental study in 11 spontaneously breathing sheep, Kuzkov et al. [16] demonstrated that CT-derived TVI and EVLWI assessed using TPTD highly significantly correlated (r = 0.85, p < 0.001). CT scans in this study were performed during a 15-sec breath hold at functional residual capacity.

In a clinical study in 10 ARDS patients, Zhang et al. [18] also reported good correlation (r = 0.95, p < 0.0001) between TVI and EVLWI. In that study, all patients were mechanically ventilated and CT scans were performed during an end-expiratory pause [18].

In another previous study, Patroniti et al. [22] assessed pulmonary fluid status using double-indicator transpulmonary thermo-dye dilution (TPTDD; with indocyanine green dye in iced dextrose 5%) in 14 patients with ARDS and revealed that these measurements showed good correlation with those by quantitative CT. For our pragmatic clinical study, we deliberately chose to use the single-indicator TPTD method to estimate EVLWI because it is used in clinical practice to estimate EVLWI, has been validated against postmortem lung weight [23], and has been shown to reliably detect even small changes in lung water [24, 25]. One might, however, argue that single-indicator TPTD is not an established reference method to assess EVLWI. Indeed, it is important to understand that the TPTD method—in contrast to the TPTDD method—estimates EVLWI assuming a fixed relation between intrathoracic blood volume and global end-diastolic volume [26].

Given the promising results from these previous studies our aim was to further evaluate the applicability of CT for the assessment of the pulmonary fluid status in critically ill patients. In a previous clinical study in critically ill patients, we observed that the CT-based estimation of EVLWI without analyzing software (qualitative CT) is not accurate compared with TPTD [17].

As a next logical step, in the present study, we used quantitative CT analyses to assess pulmonary hydration under clinical routine conditions in unselected critically ill patients. In these patients, especially in mechanically ventilated patients, diagnostic CT scans are usually performed without an end-expiratory respiratory pause under clinical routine conditions. To evaluate the clinical applicability of quantitative CT under clinical routine conditions, we deliberately chose to perform CT scans without an end-expiratory pause and to include both mechanically ventilated and spontaneously breathing patients. To further complicate matters, some of the critically ill patients included in our study had large-volume pleural effusions. Pleural effusions, however, were not included in the CT calculation of lung volumes and pleural effusions do not markedly contribute to TPTD-indicator dilution [2]. All but 2 CT-scans were performed with intravenous contrast agent (some with contrast agent bolus in the arteries/pulmonary arteries).

Therefore, CT-scans in the present study were performed according to clinical routine and not using a standardized protocol for study purposes. In this “clinical reality setting”, our results in unselected ICU patients indicate that quantitative CT might not be valuable to assess pulmonary edema (as defined by TPTD-derived EVLWI measurements).

Of note, CT revealed very small lung volumes in some of the patients included in the study. This is in line with the concept of the “baby lung” in patients with ARDS, a concept based on CT images of ARDS patients showing that ARDS not homogeneously involves the entire lung parenchyma but rather the dependent lung regions [27]. It has been shown that in ARDS—besides these dependent lung regions in which gas exchange is markedly impaired—there are normally aerated lung regions [27]. Because these normally aerated lung regions have been demonstrated to have dimensions of a 5- to 6-year-old child (300–500 g aerated tissue) the term “baby lung” was coined [27].

Differences between TV/TVI assessed using CT and TPTD-derived EVLW/EVLWI might also be explained by different measurement principles of the technologies. CT does not differentiate between lung compartments with interstitial fluid, pulmonary tissue, intravascular blood, and edema [16, 18] but detects fluid in the pleural space that is not detected by TPTD [1, 2].

Another explanation could be that the current interpretation of quantitative CT attributes non-aeration mainly to an excess in pulmonary water content. In ARDS, however, increased density and non-aeration also result from atelectasis and aerated areas are not necessarily replaced by fluid. This would also explain that the association of TVI and EVLWI seems to be better in healthy or slightly impaired lungs than in lungs with major pathologies. Under these conditions, non-aeration may result from pulmonary edema as well as atelectasis, e.g., the association of TVI and EVLWI in the animal study by Kuzkov et al. [16] seems to be better before induction of ARDS by oleic acid.

Besides using single-indicator TPTD as the reference method, our study has further limitations. Although the study was performed under routine clinical conditions the limited number of patients and the fact that all patients were treated in an ICU of a single university hospital might limit the generalizability of our findings.

5 Conclusion

In this clinical study in unselected critically ill patients, we compared variables of pulmonary fluid status assessed using quantitative CT and TPTD.

CT-derived variables did not predict elevated TPTD-derived EVLWI values. In addition, there was no significant correlation of CT-derived TV and TVI with EVLW and EVLWI assessed using TPTD. Thus, in this study in unselected critically ill patients, variables of pulmonary fluid status assessed using quantitative CT could not be used to predict EVLWI. To make rigorous conclusions about the value of quantitative CT analysis for the assessment of pulmonary fluid status more clinical data are needed.