Association between Different Indexations of Extravascular Lung Water (EVLW) and PaO2/FiO2: A Two-Center Study in 231 Patients

Background Variability of body weight (BW) and height calls for indexation of volumetric hemodynamic parameters. Extravascular lung water (EVLW) has formerly been indexed to actual BW (BWact) termed EVLW-index (EVLWI). In overweight patients indexation to BWact might inappropriately lower indexed EVLWIact. Several studies suggest indexation of EVLWI to predicted BW (EVLWIpred). However, data regarding association of EVLWIact and EVLWpred to mortality and PaO2/FiO2 are inconsistent. Two recent studies based on biometric database-analyses suggest indexation of EVLWI to height (EVLWIheight). Therefore, our study compared the association of un-indexed EVLW, EVLWIheight, EVLWpred and EVLWIact to PaO2/FiO2 and Oxygenation index (OI = mean airway pressure*FiO2*/PaO2). Methods A total of 2119 triplicate transpulmonary thermodilutions (TPTDs; PiCCO; Pulsion Medical-Systems, Germany) were performed in 50 patients from the evaluation, and 181 patients from the validation groups. Correlations of EVLW and EVLWI to PaO2/FiO2, OI and ROC-AUC-analyses regarding PaO2/FiO2<200 mmHg (primary endpoint) and OI>10 were performed. Results In the evaluation group, un-indexed EVLW (AUC 0.758; 95%-CI: 0.637-0.880) and EVLWIheight (AUC 0.746; 95%-CI: 0.622-0.869) provided the largest ROC-AUCs regarding PaO2/FiO2<200 mmHg. The AUC for EVLWIpred was smaller (0.713). EVLWIact provided the smallest AUC (0.685). This was confirmed in the validation group: EVLWIheight provided the largest AUC (0.735), EVLWIact (0.710) the smallest. In the merged data-pool, AUC was significantly greater for EVLWIheight (0.729; 95%-CI: 0.674–0.784) compared to all other indexations including EVLWIact (ROC-AUC 0.683, p = 0.007) and EVLWIpred (ROC-AUC 0.707, p = 0.015). The association of EVLW(I) was even stronger to OI compared to PaO2/FiO2. In the merged data-pool, EVLWIheight provided the largest AUC regarding “OI>10” (0.778; 95%-CI: 0.713–0.842) compared to 0.739 (95%-CI: 0.669–0.810) for EVLWIact and 0.756 (95%-CI: 0.688–0.824) for EVLWIpred. Conclusions Indexation of EVLW to height (EVLWIheight) improves the association of EVLW(I) to PaO2/FiO2 and OI compared to all other indexations including EVLWIpred and EVLWIact. Also considering two recent biometric database analyses, EVLWI should be indexed to height.


Introduction
Extravascular lung water (EVLW) is a measure of the interstitial, alveolar and lymphatic fluid content of the lungs. EVLW and its indexation to body weight (EVLWI) became routinely available after the introduction of single-indicator transpulmonary thermodilution (TPTD) [1][2][3][4][5]. A number of animal and clinical studies demonstrated an association of EVLW(I) to mortality and to parameters of pulmonary function such as PaO 2 / FiO 2 [5][6][7][8][9][10][11][12][13][14][15][16][17][18][19]. Variability of body weight (BW) and height strongly calls for biometric adjustment and indexation. Ideally, appropriate indexation should be based on a limited number of routinely available biometric data, and it should result in consistent normal values for patients with different height, weight, age, gender and race [20][21]. Originally actual body weight (BW act ) was used for indexation of EVLW. However, triggered by a rapidly increasing number of obese patients [22,23], the question arose as to which weight to choose for EVLW indexation since indexation to BW act might inappropriately decrease EVLWI act in obese patients. Based on a better correlation to mortality, a number of studies have suggested indexation of EVLW to predicted BW (EVLWI pred ) (see Table 1) [12: 14]. However, data are inconsistent regarding other endpoints: e.g. if available data provide worse [12], similar [14,16] or slightly better correlation of EVLWI pred to PaO 2 /FiO 2 than EVLWI act [13]. In one of the most recent studies [15], ' 'Chew et al. found that EVLW indexed to absolute body weight resulted in a stronger association with outcome'' including mortality compared to EVLWI pred [24]. Despite the overall strong predictive capacity of EVLWI in this and other recent trials [12][13][14][15][16][17][18][19], these inconsistencies demonstrate the need to optimize indexation of EVLW. Therefore, we recently analyzed a prospectively maintained database regarding the association of EVLW to biometric data [18]. This study demonstrated that height was the only biometric parameter independently associated to EVLW. These data were recently confirmed by Wolf et al., using a similar approach in a surgical group [19]. Despite these conclusive data, both studies did not investigate, if indexation of EVLW to height (EVLWI height ) provides better association to pulmonary function and outcome. Furthermore, all these trials were mono-centric.
Therefore, this two-center study compared the association of PaO 2 /FiO 2 (and other outcome markers) to EVLW (un-indexed), EVLWI act , EVLWI pred , EVLWI height and EVLWI indexed to other biometric indices. To provide sufficient balance of biometric data, we investigated a group with a representative distribution of body mass index (BMI) [22], as well as an unselected second group from a second center.

Munich-evaluation-group
The institutional ethics committee approved the study (Ethikkommission Technische Universität München; Fakultät für Medizin; No. 3049/11). Patients on mechanical ventilation monitored using TPTD regardless of the study were included in the prospectively maintained database. The need for informed consent was waived due to the non-interventional design of the study. The patients included in this study completely distinct to the group previously analysed regarding the association of EVLW to biometric data [18].
To provide a representative distribution regarding bodyweight [22], we included 15 consecutive patients with BMI$30 kg/m 2 , 15 consecutive patients with 25#BMI,30 kg/m 2 and 20 consecutive patients with a normal BMI (,25 kg/m 2 ) irrespective of fulfilling the criteria of acute respiratory distress syndrome (ARDS). Conscious patients were asked for actual biometric data. In unconscious patients body weight and height were extracted from the patients records. In case of doubt height was verified using a flexible tape measure in the supine position.
A 5-F thermistor-tipped femoral arterial line (PV2025L20, Pulsiocath, Pulsion Medical Systems, Munich, Germany) connected to the PiCCO monitor device (PiCCO-Plus; Pulsion Medical Systems) was used for TPTD measurements. The mean EVLW was measured based on TPTD performed in triplicate with 15 ml cold saline 0.9%.

Antwerp-validation-group
Retrospective analysis of data from a prospectively developed independent cohort was performed for the first 7 days of ICU admission. Data of 181 critically ill patients requiring mechanical ventilation and TPTD-hemodynamic monitoring treated in two ICU's in ZNA Campus Stuivenberg, Antwerp, Belgium were collected prospectively. Ethics approval had been obtained and due to the retrospective analysis and non-intervention-nature of the study the need for informed consent was waived (project number EC 3765; Commisie voor Medische Ethiek, Ziekenhuisnetwerk Antwerpen 2020). Parts of the data not related to EVLWindexation have already been published in Annals of Intensive Care [25,26].
The measuring technique was identical to the Munichevaluation-group, with the only difference being that three 20 ml boluses of cooled saline were used for TPTD.
Mean length of the ICU-stay in the Munich group was 27.2621.4 days with a range of 3 to 120 days. In the Antwerp group the mean ICU stay was 25.9641.7 days with a range of 1to 429 days. Endpoints and Statistics 1) Weight and weight-correction formula based indexations. To compare the distribution of normal (EVLWI, 7 ml/kg), slightly elevated (7#EVLWI,10 ml/kg) and markedly elevated EVLWI (EVLWI$10 ml/kg) in dependency of the indexation, un-indexed EVLW was indexed according to actual (EVLWI act ), predicted (EVLWI pred ), ideal (EVLWI id ) and adjusted BW (EVLWI adj ) using the formulas mentioned in Table 1. The cut-offs of 10 ml/kg and 14 ml/kg have been demonstrated to be associated with ARDS [15] and mortality [5].
- ). For appropriate analysis of multiple serial measurements in 231 patients from the two groups several statistical analyses were performed: N 2b.) Inter-individual (''between-subject'') correlations: Furthermore, correlations of EVLW and differently indexed EVLWI to PaO 2 /FiO 2 and oxygenation index (OI = mean airway pressure * F i O 2 * PaO 2 21 ) were calculated. Since multiple serial measurements within 241 different patients were available, we analysed interand intra-individual correlations.
To correct for different numbers of measurements for each patient, the means of EVLW(I), PaO 2 /FiO 2 and OI were calculated for each individual patient (''one point per patient''). Subsequently the correlations between EVLW(I) and PaO 2 /FiO 2 and between EVLWI and OI were calculated. N 2c.) Intra-individual (''within subject'') correlations: The above-mentioned ''one point per patient'' analyses reflect the inter-individual association of EVLW(I) to PaO 2 /FiO 2 and OI. However, in cases of multiple serial measurements within different patients, between-subject heterogeneity may obscure correlations on an individual level (within subject correlation) which might be even more interesting than the inter-individual association. The effect of the confounder (between-subject heterogeneity) can be removed by calculating ''partial'' correlation between EVLW(I) and PaO 2 / FiO 2 (or OI) adjusting for heterogeneity of different patients (individual patient number/identifier as the adjustment factor).
3) Mortality analysis. Better prediction of mortality by EVLWI according to any indexation might be related to the direct association of the indexation to mortality. To overcome this problem, multiple binary logistic regression analysis regarding mortality included the first and last values of unindexed EVLW as well as BW act , height, gender and acute physiology and chronic health evaluation (version 2, APACHE-II).
All analyses were performed separately in both groups and in the merged data, with the only exceptions being intergroupcomparisons and mortality analyses which were restricted to the BMI-representative Munich-evaluation-group. Results of merged data were considered superior to those derived from sub-groups. No correction of p-values was applied to adjust for multiple testing. However, results of all statistical tests being conducted were thoroughly reported so that an informal adjustment of p-values can be performed while reviewing the data [27].
All statistical tests were conducted 2-sided and a p-value ,0.05 was considered to indicate statistical significance. The software used was IBM SPSS statistics, version 20.

Patients' characteristics
A total of 2119 TPTDs (each with triplicate TPTD) were performed in 231 patients from both groups. Table 2 summarizes the patients' characteristics of both groups.
1) Weight and weight-correction formula based indexations. Comparison of EVLWI act , EVLWI pred , EVLWI id and EVLWI adj: Table 2 demonstrates that in the Munichevaluation-group mean values of EVLWI pred , EVLWI adj and EVLWI id were significantly higher than EVLWI act in the subgroups of patients with BMI$30 kg/m 2 and with 25# BMI,30 kg/m 2 as well as for the totality of patients (intra-BMI-group-comparison). By contrast, EVLWI pred and EVL-WI adj were significantly lower than EVLWI act in the subgroup of patients with a normal BMI.
Similarly, in the Antwerp-validation-group mean values of EVLWI pred , EVLWI adj and EVLWI id were higher than mean the EVLWI act . Impact of indexation according to different weight correctionformulas for the classification of EVLWI: Distribution of EVLWIvalues classified as normal (EVLWI,7 ml/kg), moderately elevated (7 ml/kg#EVLWI,10 ml/kg) and markedly elevated EVLWI (EVLWI$10 ml/kg) significantly varied among the patients with a BMI$30 kg/m 2 as well as in the total patient groups depending on the weight used for indexation of EVLWI ( Fig. 1). For example in patients with a BMI$30 kg/m 2 , 51% (133/263) of EVLWI act measurements were within the normal range (EVLWI,7 ml/kg). By contrast only 16% (43/263; p, 0.001 vs. EVLWI act ), 14%, (38/263; p,0.001) and 30% of the measurements (79/263; p,0.001) were within the normal range if EVLWI was indexed according to predicted, ideal and adjusted BW, respectively. In addition to the different distributions of EVLWI classifications, different indexations obviously had an impact on the coefficient of variation (COV), in particular in patients with a BMI$30 kg/m 2 ( Table 2). COV amongst these    Regarding the prediction of the threshold ''OI.10'', in both collectives as well as in the merged data, the largest ROC-AUCs were obtained for EVLWI height and EVLW, with the lowest for EVLWI act (Table 4)     Antwerp-validation group and 0.762 (0.702-0.822), 0.756 (0.695-0.817) and 0.716 (0.649-0.782) for the merged data, respectively. For the merged data, the AUC was significantly higher for EVLWI height (0.762) compared to EVLWI act (0.716; p = 0.017), EVLWI adj (0.739; p = 0.05) and EVLWI BMI (0.719; p = 0.019).
2b.), 2.c) Inter-individual and intra-individual correlations of EVLW(I)to PaO 2 /FiO 2 and to oxygenation index: In general, the inter-individual association (represented by patients' means) of EVLW(I) to PaO 2 /FiO 2 and OI was not as strong as the intraindividual association represented by partial correlations. Means of EVLW(I) and PaO 2 /FiO 2 moderately correlated with r-values between 20.21 and 20.46. By contrast, intra-individual correlations represented by partial correlations provided r-values between 20.6 and 20.74 (Table 4).
Within the Munich-evaluation-group only the patients' means of EVLW, EVLWI height and EVLWI adj significantly correlated with PaO 2 /FiO 2 (r = 20.34; p = 0.017, r = 20.32, p = 0.026 and r = 20.29; p = 0.041, respectively). In contrast, the patients' mean EVLWI according to all other indexations including EVLWI pred and EVLWI act did not correlate to PaO 2 /FiO 2 . Overall, the lowest r-values were found for EVLWI act . The highest coefficients of correlations to PaO 2 /FiO 2 were found for un-indexed EVLW and EVLWI height (Table 4).
However, in the same group all partial correlations of EVLW(I) and PaO 2 /FiO 2 were highly significant (p,0.001) and provided coefficients of partial correlation between 20.66 and 20.74.
Similar data were obtained for the correlation between OI and EVLW and its indexations (Table 3). Intra-individual partial correlations with r-values between 0.57 and 0.80 were more pronounced than inter-individual correlations represented by rvalues between 0.31 and 0.48 for correlations of patients' means. All these correlations were significant. In both groups as well as in the merged data the highest coefficients of correlation to OI were found for EVLWI height and EVLW, the lowest for EVLWI act (r-values for merged data 0.77, 0.78 and 0.69, respectively) ( Table 4; partial correlation).
Subsequently, a multivariable logistic regression analysis regarding mortality was performed. A model including APACHE-II, the first and the last EVLW provided high predictive capabilities regarding mortality (Nagelkerkes R 2 = 0.697). First (p = 0.021) and last (p = 0.004) EVLW were independently associated to mortality. The APACHE-II score slightly failed to reach significance (p = 0.064), but markedly contributed to the R 2 -value of the total model (R 2 = 0.628 without APACHE-II).

Discussion
Data regarding EVLW-indexation are contradictory. Therefore, our study investigated the association between different indexations of EVLW(I) to PaO 2 /FiO 2 and OI in two groups with mechanical ventilation and representative distribution of BMI. Our study demonstrated that 1) indexation of EVLWI to BW act is inferior to no indexation at all, 2) indexation to BW pred might provide a certain improvement compared to indexation to BW act and 3) indexation according to height or no indexation at all (EVLW) are superior to indexation to BW act or BW pred .
These results are -at first glance -surprising, as several studies have suggested BW pred as the appropriate indexation factor [12][13][14].
Historically, different techniques have been established to quantify pulmonary edema termed as ''EVLW'' which was originally determined without indexation. Early studies frequently used animal models with post-mortem gravimetric determination of EVLW as the gold-standard [1,2,4,10]. Regarding investigations in different species, indexation to BW act provided ''basic'' indexation allowing interspecies comparisons between different animal and human data.
However, in obese patients indexation to BW act might inappropriately diminish indexed EVLWI act . Based on a better prediction of mortality rather than on a better correlation to PaO 2 /FiO 2 , superiority of EVLWI pred to EVLWI act has been suggested: In the study by Phillips et al. [12] including 19 patients, EVLWI act was not related to mortality. By contrast, mortality of the seven out of 19 patients was univariately associated to EVLWI pred . However, EVLWI act obviously provided a better correlation to PaO 2 /FiO 2 than EVLWI pred (coefficient of correlation: 20.525 for EVLWI pred and 20.773 for EVLWI act ). With only 19 patients included, this study did not approach a multivariate mortality-analysis. Another study that included 44 patients (225 measurements) demonstrated better discrimination of ARDS-and non-ARDS-patients by EVLWI pred compared to EVLWI act [13]. A third study (44 patients; 44 measurements) showed the improved association of mortality by EVLWI pred compared to EVLWI act in a multivariate model [14]. However, similarly to the data of Phillips et al. this study did not demonstrate a better correlation of EVLWI pred to PaO 2 /FiO 2 compared to EVLWI act (coefficients of correlation 20.57 vs. 20.55). This is in accordance with two more recent studies suggesting a comparable [16] or even stronger [15] association of EVLWI act compared to EVLWI pred with mortality.
Nevertheless, mortality is multifactorial, also depending on ''donot-resuscitate'' statements and might be directly associated to some of the components of indexation (weight, BMI) [28][29][30][31][32][33][34][35][36]. E.g. further analysis of the data by Phillips et al. demonstrates that mean EVLW and BMI were increased to a similar degree in the non-survivors compared to survivors (45% and 31%, respectively). Therefore, mortality is not an obvious endpoint to compare the appropriateness of different indexations of EVLWI, particularly when applied in small mono-centric collectives.
In our study-groups un-indexed EVLW -next to EVLWI heightprovided the highest predictive capability regarding PaO 2 /FiO 2 and OI. This indicates that particularly weight-based indexation might be a confounder rather than an improvement of the interindividual comparison of EVLW(I) in an adult population. In the analysis of patients' means, PaO 2 /FiO 2 was significantly associated to unindexed EVLW, EVLWI height and EVLWI adj , whereas the correlation was not significant for all other indexations. These findings suggest that the association of EVLW(I) to PaO 2 /FiO 2 might be obscured by inappropriate indexation. In general, the association of EVLW(I) was closer to OI compared to PaO 2 /FiO 2 . Including mean airway pressure in addition to PaO 2 /FiO 2 , OI also reflects the Positive End Expiratory Pressure (PEEP), peek and plateau pressure, ventilation mode (there is a usually higher mean airway pressure in controlled compared to assisted ventilation) as well as I:E ratio. Since the association of EVLW(I) to OI was not extensively investigated in the previous studies, the close association in our study might even strengthen the role of EVLW(I) as a parameter of pulmonary (patho)physiology.
With regard to indexation of other pulmonary parameters, the strong performance of height as an indexation for EVLWI is not surprising. As stated in recent consensus guidelines ''lung volumes are related to body size, with standing height being the most important factor'' [21].
Furthermore, a look at the ''weight correction-formulas'' demonstrates that BW pred and BW id (Table 1) do not contain any weight at all, but simply adjust height for gender and subtract a length-constant [20,21,28].
In addition to weight and height, the third major determinant of most indexation formulas is gender, which has impact on BW pred , BW id , BW adj and TLC. With regard to the above-mentioned formulas, indexation according to BW id increases EVLWI by 5.5% for women. EVLWI pred of women with a height between 150 and 190 cm is increased by 5-10% compared to men with the same height. However, this marked impact of gender on EVLW is not substantiated by our data: Multiple regression analysis regarding EVLW in our merged data including the variables age, height, weight, PaO 2 /FiO 2 and gender demonstrated that gender was not independently associated to EVLW.
Finally, the question remains, whether in adults ''no indexation at all'' is the answer. Regarding our data in two adult groups with a high variability in body weight and BMI, but lower variability in height, this might be a reasonable option. However, it must be kept in mind that the variability in height was low in these groups: e.g. mean height in the Munich-evaluation-group was 170610 cm (median 171 cm, range 150-190 cm).
On the other hand it is self-evident that indexation will improve the predictive capabilities in a group with a higher variability of parameters closely associated to EVLW such as height. There is elaborate data on the pulmonary function parameters in children and adolescents: Normal values in these groups with high variability in height and weight are mainly adjusted to height [20,21].

Limitations of the study
Despite the inclusion of two different groups and the large sample size compared to the previous data, our study has several limitations. Our Munich-evaluation-group was a preselected group of non-operative mechanically ventilated patients with a prolonged ICU-stay. Although this drawback might be -at least in part -outweighed by a re-evaluation in a large group of nonselected anesthesiology patients, the data of both groups might not apply to patients without pulmonary impairment. On the other hand, the significance of modest correlations with r-values as low as -0.29 require cautious interpretation, since large numbers of patients promote significance of modest associations. Furthermore, these data are mainly derived from Caucasians. Despite a ''considerable lack of data on lung-volumes in non-Caucasians'' [20,37] at least two studies give hints on differences regarding TLC between whites and blacks [38], Polynesians, Northern Indians and Pakistanis [39][40][41]. Finally, we cannot extrapolate our results to a pediatric population in which indexation to height may be much more appropriate than unindexed EVLW which was comparable to EVLW height in our adult groups.

Conclusions
EVLW is a marker significantly associated to pulmonary function and mortality. Regarding the prediction of PaO 2 /FiO 2 and OI, indexation of EVLWI act is inappropriate. EVLWI pred provides a slight improvement. The highest predictive capabilities in an adult population were found for EVLWI height and unindexed EVLW. Therefore, our data suggest that EVLW should be indexed to height (EVLWI height ) or remain unindexed in adults.