Three broad classifications of acute respiratory failure etiologies based on regional ventilation and perfusion by electrical impedance tomography: a hypothesis-generating study

EIT measurements were performed with PulmoVista 500 (Dräger Medical, Lübeck, Germany). A silicone EIT belt with 16 surface electrodes was placed around the patient’s thorax at the 4th intercostal space level. All patients received standard care and no other research-related interventions were performed. EIT measurements were continuously recorded at 20 Hz when the patients were at relative stable condition after medical treatment. A bolus of 10 ml 10% NaCl was injected during a respiratory pause (for at least 8 s) through the central venous catheter. The respiratory pause was conducted via an end-expiratory hold maneuver with the ventilator in the intubated patients. The conscious patients were asked to hold their breath for 8–12 s at the end of expiration. The EIT data were digitally filtered using a low-pass filter with a cutoff frequency of 0.67 Hz to eliminate periodic cardiac-related impedance changes (for evaluation of both ventilation and perfusion). Perfusion evaluated via hypertonic saline bolus injection corresponded to a non-periodic impedance drop that was not influenced by low-pass filtering. Furthermore, the data were analyzed offline using customized software programmed with MATLAB R2015 (the MathWorks Inc., Natick, MA).

Regional ventilation map was calculated by subtracting the end-expiration from the end-inspiration image, which represents the EIT signal variation during tidal breathing. The tidal images before the apnea period were averaged to increase the signal-to-noise ratio:where V_i is pixel i in the ventilation image; N is the number of breaths within the analyzed period; ΔZ_i,Ins and ΔZ_i,Exp are the pixel values in the raw EIT image at the end-inspiration and end-expiration, respectively.

Due to its high conductivity, 10% NaCl acts as an EIT contrast agent and passes through the pulmonary circulation thereby producing a dilution curve after bolus injection during the apnea period based on the first pass kinetics theory [25, 26]. Regional perfusion map was generated by the slope of regional impedance–time curves after saline bolus injection [27]. In brief, the regional impedance–time curves during the descending phase were fitted with linear regression:where t is the time starting from one cardiac cycle after the initial descent in the global impedance curve caused by saline injection and ending at the trough of the global curve during the apnea period. Then the perfusion value of pixel i P_i in the perfusion image was equal to − a_i. Furthermore, ventilated and perfused regions were defined as follows: region k is ventilated if:

Similarly, region g is perfused if

Subsequently, three regions (number of pixels) were identified: regions that were only ventilated (R_V), regions that were only perfused (R_P) and regions that were both ventilated and perfused (R_V+P). To correlate with clinical events, the following EIT-derived parameters were calculated:

For the quantitative analysis of the regional ventilation and perfusion distributions, the following terms were defined:

Typical ventilation, perfusion and VQ matching images of one patient from the control group are shown in Fig. 4D.

A descriptive analysis was performed. Normal distribution was assessed with the Kolmogorov–Smirnov normality test. Normally distributed results are presented as mean ± SD, whereas non-normally distributed results are presented as median (25th, 75th percentile). The Kruskal–Wallis test was used to compare groups on continuous variables, and Bonferroni correction was used to adjust the P value for multiple comparisons. Chi-square and Fisher’s exact tests were used to compare categorical variables where applicable. Comparisons of two continuous variables were performed using Spearman’s correlation and linear regression. Since there was no a priori information regarding the etiology clusters available prior to this study, the sample size estimation was only based on PE diagnosis. According to the standard PE diagnosis rate (clinical diagnosis and/or confirmed by computed tomography pulmonary angiography), we conservatively estimated that a sample size of 69 patients (including 62 patients without PE) would be required to detect a 1% error rate (β = 0.8, and α = 0.05) based on 80% specificity in a 10% prevalence of acute PE. The ability of EIT parameters to identify three preset etiologies (PED, DLD and FLD) was examined including both ARF and control patients. The areas under the receiver operating characteristic (AUC) curves were compared using a Hanley–McNeil test [28]. The cutoff values were selected to achieve higher specificities. If the specificity was > 90%, the cutoff value was chosen to reach the highest Youden’s index. Otherwise, the cutoff value was chosen to reach the highest specificity. All comparisons were two-tailed, and P < 0.05 was required to exclude the null hypothesis. The statistical analysis was performed using the software package SPSS 24.0 (SPSS Inc. Chicago, IL) and MedCalc 11.4.3.0 Software (Mariakerke, Belgium).

Demographics, clinical characteristics of four groups (the control group and the three ARF subgroups: PED, DLD and FLD) are shown in Table 1. Differences in EIT-related parameters among the four groups are shown in Table 2. The PED group had a significantly higher DeadSpace_% and perfusion-defect score than the other groups (Figs. 2 and 3). The DLD group had significantly lower Defect_Q and Defect_V+Q score than the FLD and PED groups, and it had a similar VQMatch_%, Shunt_%, Defect_V and Defect_Q scores as the control group (Figs. 2 and 3). The FLD group had a significantly higher Defect_V score than the other groups (Fig. 2).

PaO₂/FiO₂ was significantly correlated with VQMatch_% (r = 0.324, P = 0.001), DeadSpace_% (r = − 0.301, P = 0.002) and Defect_Q score (r = − 0.212, P = 0.027). Regional ventilation, perfusion distributions and the corresponding V/Q (%) ratio were not significantly correlated with PaO₂/FiO₂.

The PED patients (14/93) had a significantly higher DeadSpace_%, Defect_Q score, and LR-V/Q (%) and lower VQMatch_% and Shunt_% than non-PED patients. EIT-related parameters have a significantly higher performance regarding the diagnosis of PE than d-dimer. DeadSpace_% resulted in the highest AUC. A cutoff value of DeadSpace_% > 30.37% was used for the diagnosis PED, resulting in a sensitivity of 78.6% and a specificity of 92.4% (Table 3).

The DLD group (21/93) had a significantly lower Defect_V, Defect_Q, and Defect_V+Q scores than the non-DLD groups. The AUC for using Defect_V+Q to diagnose DLD was 0.893. The cutoff of Defect_V+Q was < 0.5 for the diagnosis of DLD, resulting in a sensitivity of 66.7% and a specificity of 90.3% (Table 3).

The FLD group (58/93) had a significant higher Defect_V, Defect_V+Q scores, Shunt_% and lower DeadSpace_% than non-DLD patients. The AUC for using Defect_V to diagnose FLD was 0.844. The cutoff of Defect_V was > 2.5 for the diagnosis of FLD, resulting in a sensitivity of 39.7% and a specificity of 97.1% (Table 3).

These related characteristic EIT parameters that produced specificities > 90% of ventilation/perfusion phenotypes were retained for three broad ARF etiologies (Fig. 4). Moreover, typical ventilation, perfusion, VQ matching images and CT images of individual patient subgroups are shown in Fig. 4. Defects of regional perfusion with normal ventilation were observed in the PED patient (Fig. 4A). No cognizable defect of regional ventilation and perfusion was observed in the DLD patient (Fig. 4B). Defects of regional ventilation with relatively normal regional perfusion in left lung were found in FLD patients with acute left lung atelectasis (Fig. 4C).

Previous experimental studies have shown that regional lung perfusion after pulmonary embolism-like events could be effectively detected by EIT and bolus saline [12, 14, 34]. Recently, several clinical cases reported using EIT and bolus saline to detect PE in clinical practice [16, 18, 19, 36]. Our data supported the conclusion that the DeadSpace_% and Defect_Q score had a significantly better performance than d-dimer for PE diagnosis (Table 3). A high d-dimer concentration was considered an indication for anticoagulation treatment in ARF patients with suspected PE [37, 38]. Since many factors such as operation, tumor, and inflammation could also independently cause an increase in d-dimer, it is not surprising that d-dimer did not show a good ability to diagnose PE in the mixed patient population. Further study is required to examine whether EIT-based PE diagnosis could change clinical decisions and limit unnecessary anticoagulation in a critical condition, especially pertaining to postoperative patients.

Moreover, a high DeadSpace_% could be observed in severe ARDS patients, which might result from lung microvascular embolism. Marui et al. used a similar EIT method to show an elevated ventilation–perfusion mismatch, with a larger prevalence of ventilated nonperfused lung units (dead space) in comparison to perfused nonventilated units (shunt) in COVID-19 patients with ARDS [17]. This condition was not taken into consideration in the present study. Here, we emphasized the potential advantage of using EIT to determine submassive/massive PE in critically ill patients with heavy devices and a high risk of transfer for CTPA examination.

There are a wide variety of causes of DLD in ICU patients, which are typically diagnosed with chest radiography, CT, and ultrasound [39]. In the present study, DLD of ARF was classified as diffuse pulmonary alveolar/interstitial/lobular disease without a significant focal lesion. Since the damage mainly affects the alveoli and interstitial tissue in a diffuse manner and spatial resolution of EIT is limited, an obvious defect of regional ventilation and perfusion is not expected to be found in the DLD patients, which is in harmony with and explains our results (Table 2). The DLD group had a low Defect_V and Defect_Q scores, which were similar to the control group. We presented a concept of low Defect_V+Q score that supports DLD diagnosis in the primary assessment of ARF patients. The AUC for using a low Defect_V+Q score (< 0.5) to diagnose DLD was relatively high (0.893). Moreover, estimation of extravascular lung water could be useful for further differential diagnosis in DLD patients. Quantitative evaluation of extravascular lung water by EIT has been reported in human and animal studies [21, 40]. Further studies with larger samples of DLD should be conducted to examine the efficacy of EIT-based parameters (including extravascular lung water assessment) on DLD subcategory classification.

EIT has been proven to effectively measure ventilation defects by impedance monitoring in different pathophysiologic conditions, such as pleural effusion, pneumothorax, pneumonia, consolidation, atelectasis etc. The present study also supported that EIT could detect regional ventilation defects caused by various causes. Both extrapulmonary (pneumothorax and pleural effusion) and intrapulmonary (atelectasis/pneumonia/consolidation) lesions were included in the FLD. As expected, the highest Defect_V scores were found in the FLD group compared to the other groups. The cutoff Defect_V score was > 2.5 for the diagnosis of FLD, resulting in a sensitivity of 39.7% and a specificity of 97.1% in the present study (Table 3). Moreover, the regional perfusion varied in the ventilation-defect quadrants. Studies have suggested that regional perfusion may increase in damaged lung areas in pneumonia and ARDS [29, 41] but could also fall in the injured regions due to obstruction or compression of pulmonary capillaries [42, 43]. A recent animal study found experimental atelectasis with minimal tidal recruitment/derecruitment, and mechanically inspiratory breaths redistributed blood volume away from well-ventilated areas [44].

The greatest contribution of our study is that the phenotype of ventilation/perfusion images and related parameters from EIT were determined for three broad ARF etiologies. Moreover, these results were consistent with the pathophysiologic mechanism. Someone might argue that these related parameters had high specificities > 90% but relatively low sensitivities (< 80%). Several potential reasons may be considered: functional EIT measurement of ventilation and perfusion does not completely reflect anatomical location of ventilation and perfusion changes. Small lung lesions might not cause a significant change in regional perfusion or ventilation. Functional changes in ventilation and perfusion are complicated and variable in different lung lesions. Correlation between PaO₂/FiO₂ and EIT parameters were weak, which coincided with a recent study from Spinelli et al. [45]. They found a weak negative correlation between the percentage of only perfused units and PaO₂/FiO₂ ratio (r = − 0.293, P = 0.039) in ARDS patients. We suspect that with limited spatial resolution, EIT might not be able to capture small V/Q mismatch. Besides, V/Q mismatch does not account for 100% the decrease of PaO₂/FiO₂. Ventilation and perfusion could be both significantly decreased and still matched. The PaO₂/FiO₂ would not be satisfactory in this case.

Because of the limitation of using functional ventilation/perfusion defects to reflect the ARF etiologies of anatomical location, we focused on the diagnostic specificity of EIT images. The described EIT method should be taken as the preliminary assessment for broad ARF etiologies but not for precise etiology identification. We stressed that these significant signs (high DeadSpace_%, Defect_V and Defect_Q scores) were related to the potential etiologies of ARF. These related parameters could work as feature indicators for the identification of broad etiologies of ARF and provide a probable diagnostic direction at bedside. Moreover, the relatively normal ventilation and perfusion distribution should be interpreted with caution in ARF patients. Both DLD and small lung lesions could have a “normal” distribution of ventilation and perfusion. Our proposed method has potential advantages in the differential diagnosis of life-threatening ARF, especially for unstable patients who require bedside monitoring tools. A scoring system for ventilation and perfusion distribution defects was defined in the present study. There were several potential advantages of the scoring system: (1) the scoring system is feasible and convenient at the bedside. (2) Regional ventilation and perfusion data from EIT are complex and often hard to interpret. For examples, Deadspace_% can categorize PE patients but no other etiologies. With the regional ventilation and perfusion images, it is hard to identify DLD directly. Hence, from the perspective of overall evaluation, the proposed scoring system has the potential to identify the predefined three ARF etiologies, which was supported by the present results.

This work has some limitations. First, our study was carried out in a single center, and a small number of patients were included. The primary aim of study was to investigate the relationship of oxygenation and regional ventilation and perfusion assessed by the indicator based EIT method. The hypothesis of three broad ARF etiologies was generated with some of the results and analysis done. Second, the three broad etiologic diagnoses of ARF were established in a combined assessment by the ICU team based on an objective examination and standards (Additional file 1: Table S1). Strict predefined criteria of CT scan for the ARF etiology was lacking. Unfortunately, a validation for all the study subjects using radiography (e.g., CTPA and CT) was ethically impossible based on the study design. PE might also have been present in patients with DLD or FLD. The potential confounders would need to be considered when interpreting the results. Nevertheless, the proposed EIT method was used to diagnose broad ARF etiologies and not for precise and comprehensive diagnoses. Although the investigators were not totally blinded to the patient’s clinical presentation, the EIT profiles were established based on objective data. Third, only one main ARF etiology was identified based on a comprehensive clinical judgment in the present study. Patients with multiple etiological diagnoses (e.g., diffuse ARDS—superimposed gradient and dorsal collapse, PE patients with DLD or FLD) might have a potential impact on the results. Fourth, the thresholds of quadrant ventilation and perfusion distribution and the corresponding scores might not be optimal. Data percentile distributions in the control group and clinical simplicity were taken into consideration when the scoring system was defined. With the aim of obtaining a high diagnostic specificity, the value of the lowest 5th percentile of quadrant distributions in the control group was selected as the cutoff value for the abnormal distribution defect. The quadrant distribution is relative, so a low distribution in one quadrant is accompanied by an elevated distribution in the other quadrants. For simplicity, we only scored the quadrants with a decreasing distribution. A certain limitation of the proposed scoring system is that the sample size of control patients was small. Further studies are, therefore, required to validate or optimize the scoring system. Fifth, most patients were sedated and intubated in the present study. We found that it could be difficult in some awake ARF patients to perform a sufficient breath hold. Moreover, these patients normally generate respiratory muscular contractions that could modify the venous return and cardiac output. Hence, the saline-contrast EIT method might be more suitable for sedated patients under mechanical ventilation. Sixth, in the present study, lung perfusion was measured only at the end-expiration. Different airway pressures at end-inspiration and end-expiration could cause variation in lung perfusion. It is debatable whether lung perfusion by EIT should be measured at end-inspiration or end-expiration [46] or whether it matches average perfusion during normal tidal breathing. Inspiratory hold could be more readily implemented for patients with mild sedation, but a higher airway pressure at end-inspiration may cause an impairment of global circulation. The following potential benefits of lung perfusion at end-expiration were identified: (1) little impact on venous return and circulation and (2) a significant change in chest impedance might be easier caused by saline bolus at lower global impedance of end-expiration. In some preliminary data, we did not find differences in perfusion when obtained during a short interruption of ventilation at end-expiration and end-inspiration (the difference in global inhomogeneity index for perfusion was < 10%). We suspected that small tidal volume did not have a pronounced influence on regional lung perfusion. Further studies are required to compare the clinical relevance of end-inspiration occlusion and end-expiration methods.

The extensive limitations might limit the clinical applicability of the study. Advanced mathematical analysis and machine learning maybe helpful to explore the topic provided more patient data.