FormalPara Take-home messages

The lung-protective benefits of ECCO2R increase with higher alveolar dead space fraction, lower respiratory system compliance, and higher device performance. Alveolar dead space fraction and respiratory system compliance, rather than severity of hypoxemia, should be the primary factors determining whether to enroll patients in clinical trials of ECCO2R. Restricting enrollment based on the predicted treatment effect may enhance statistical power in a future trial of ECCO2R.

Introduction

In patients with acute respiratory distress syndrome (ARDS), lowering tidal volume (Vt) and driving pressure (i.e. end-inspiratory plateau pressure minus positive end-expiratory pressure: ΔP) to reduce pulmonary stress and strain can decrease ventilator-induced lung injury and improve survival [1,2,3]. However, lowering Vt is often associated with respiratory acidosis, even if respiratory rate (RR) is increased, and there is evidence that hypercapnia along with the use of high RR may worsen clinical outcomes [1, 2].

Removing carbon dioxide via an external membrane lung (extracorporeal CO2 removal, ECCO2R) can attenuate respiratory acidosis, permitting greater reductions in Vt, ΔP, and RR [3, 4], and thereby may improve outcomes in ARDS. ECCO2R, however, is invasive, costly, and carries significant risks including hemorrhage, hemolysis, and thrombosis [5]. Applying ECCO2R to reduce Vt to very low levels may also worsen lung mechanics by increasing atelectasis. To optimize the balance of benefit and risk, ECCO2R should ideally be applied specifically to patients who stand to accrue the greatest clinical benefit [6].

Based on a theoretical analysis of physiological equations defining alveolar ventilation [6], we hypothesize that for a given PaCO2 the reduction in Vt, ΔP, and mechanical power (PowerRS) enabled by ECCO2R depends on specific patient physiological characteristics [i.e., alveolar dead space fraction (ADF) and respiratory system compliance (Crs)], and on the CO2 clearance rate achieved by the ECCO2R device. In the current study, we aimed to test this hypothesis using data from the SUPERNOVA trial [7] to inform the design of a future trial of ECCO2R.

Methods

The SUPERNOVA trial

SUPERNOVA was a pilot trial evaluating the efficacy and safety of ECCO2R to achieve ultra-protective ventilation (Vt of 4 ml/kg predicted body weight [PBW]) in 95 patients with moderate ARDS (100 < PaO2/FiO2 ≤ 200 mmHg) from 23 centers [7]. At different centers ECCO2R was applied using either the Hemolung Respiratory Assist System (ALung Technologies, Pittsburgh, USA), the iLA Activve (Xenios, Heilbronn, Germany), or the Cardiohelp® HLS 5.0 (GETINGE Cardiopulmonary Care, Rastatt, Germany). The first device (lower CO2 extraction device) employs a membrane lung with a cross-sectional area of 0.59 m2 and is run at an extracorporeal blood flow between 300 and 500 ml/min. The other two devices (higher CO2 extraction devices) employ membrane lungs of 1.30 m2; in the trial they were operated with blood flows of 800–1000 ml/min. The primary endpoint in the trial was the number of patients who successfully achieved a VT of 4 ml/kg PBW with arterial pH > 7.30 and PaCO2 not increasing more than 20% relative to baseline condition (where VT was set at 6 ml/kg PBW, positive end-expiratory pressure (PEEP) was adjusted to obtain an end-inspiratory plateau airway pressure (PPLAT) between 28 and 30 cmH2O, and sweep gas flow was set to 0 l/min).

Physiological computations

Crs, anatomical dead space volume, ADF [6], ventilatory ratio (VR) [8], and PowerRS [9] were calculated as previously described using relevant physiological variables collected at baseline. Computations are detailed in the Online Supplement.

Quantifying the effect of ECCO2R

There were two discrete interventions in SUPERNOVA: first, reductions in Vt permitted by ECCO2R, and second, reductions in Vt enabled by more permissive ventilation targets for PaCO2 (allowed to increase by up to 20% to achieve the Vt 4 ml/kg). To isolate the effect of ECCO2R on Vt independent of the change in PaCO2 target and to enable comparisons of treatment effect between patients, Vt before ECCO2R (at the moment before sweep gas flow commenced) and after ECCO2R (once the 4 ml/kg step was reached) were normalized to the Vt that would have been required to obtain a PaCO2 of 45 mmHg at RR of 30 breaths/min. PowerRS was standardized for a PaCO2 of 45 mmHg. Computations are described in detail in the Online Supplement.

The effect of ECCO2R was also quantified in terms of the probability of reaching the SUPERNOVA trial primary end-point at 8 h after initiating ECCO2R.

Determinants of the effect of ECCO2R

Baseline characteristics were compared between patients with smaller (< median) or larger (> median) changes in standardized ΔP after applying ECCO2R. Device performance was classified according to expected CO2 extraction capability (lower—Hemolung; higher—Cardiohelp and iLA activve) based on previously reported CO2 removal rates [10]. The independent effects of patient physiological characteristics (PaO2/FiO2, ADF, VR, Crs) and device performance (lower vs. higher CO2 extraction) on the change in standardized values of Vt, ΔP, and PowerRS obtained by applying ECCO2R were evaluated using pre-specified bivariate and multiple linear regression models (see Table E1 for description of multivariable models). The effects of these characteristics on the probability of achieving the primary endpoint was evaluated using multivariable logistic regression (see Table E1 for description of model).

Predicting the effect of ECCO2R on driving pressure

We previously showed that the reduction in ΔP that would be obtained by applying ECCO2R can theoretically be predicted according to the following relation, derived from the physiological equations defining alveolar ventilation [6]:

$$\Delta P_{{{\text{aw}},2}} - \Delta P_{{{\text{aw}},1}} = \frac{ - k}{{C_{\text{RS}} \cdot \left( {1 - {\raise0.7ex\hbox{${V_{\text{d,alv}} }$} \!\mathord{\left/ {\vphantom {{V_{\text{d,alv}} } {V_{\text{t}} }}}\right.\kern-0pt} \!\lower0.7ex\hbox{${V_{\text{t}} }$}}} \right) \cdot {\text{RR}}\, \times \,P_{{{\text{aCO}}_{2} }} }} \cdot \dot{V}_{{{\text{CO}}_{ 2} , {\text{ECML}}}} ,$$
(1)

where Crs is the compliance of the respiratory system, Vd,alv/Vt = ADF; RR = respiratory rate (breaths/min); \(\dot{V}_{{{\text{CO}}_{ 2} , {\text{ECML}}}}\) = CO2 removal rate by the ECCO2R device (ml/min); and k = 0.863. The predicted change in ΔP was computed for each patient using Eq. 1. Because the actual CO2 clearance rate achieved in each patient was not measured in SUPERNOVA, we assessed the impact of three different possible values for average CO2 removal (80 ml/min, 120 ml/min, and 150 ml/min).

The predicted and observed changes in ΔP obtained after applying ECCO2R were compared using linear regression, Bland–Altman analysis, and receiver operating characteristic curve (ROC) analysis. For the purpose of this analysis, we estimated that a sample size of 85 patients would yield 90% power to detect a correlation between predicted and observed changes in ΔP with r ≥ 0.4 at a Type 1 error risk of 1%. This sample size was also sufficient to estimate the Bland–Altman limits of agreement with confidence intervals of ± 0.1 cmH2O assuming the standard deviation of the difference between predicted and observed changes in ΔP was ≤ 2 cmH2O.

Because lowering Vt may alter Crs by relieving hyperinflation (increasing Crs) or exacerbating atelectasis (worsening Crs), the resulting change in ΔP may reflect both reductions in Vt and changes in Crs. Worsening compliance after lowering Vt would attenuate the effect of lowering Vt on ΔP and such changes in Crs might be prevented by increasing PEEP. To estimate the accuracy of predicting changes in ΔP with such a PEEP titration strategy, we compared the observed and predicted changes in ΔP using the baseline Crs value to compute ΔP before and after the application of ECCO2R.

Simulating potential clinical trials designed based on predicted treatment effect

Sample size and screening size requirements were estimated for possible future ECCO2R trial designs with mortality as the primary endpoint. Trial designs employing different CO2 removal rates and varying degrees of predictive enrichment (restricting enrollment based on predicted ΔP response) were compared. For each trial design, the predicted change in ΔP that would be obtained by ECCO2R at the planned CO2 removal rate was calculated for all patients in the SUPERNOVA study population based on Eq. 1. To account for possible error in the predicted change in ΔP, random error (95% CI ± 4 cmH2O) was added to the predicted change in ΔP (4 cmH2O was chosen based on the limits of agreement for predicted and observed changes in ΔP). Each trial design was simulated 500 times, and the median, 5th percentile, and 95th percentile values for the predicted change in ΔP were used to predict the absolute risk reduction in mortality (ARR). To compute ARR, we assumed that a 7 cmH2O reduction in ΔP was associated with a hazard ratio for mortality (HR) of 0.68 as previously reported [11] and that ECCO2R carries a 1% increase in absolute risk of death due to treatment-related complications [5]. We further assumed a control group mortality rate based on the mortality rates observed for each subgroup of predicted responders in SUPERNOVA. Sample size estimates were computed from predicted ARR values assuming a 5% risk of Type I error and a 20% risk of Type II error. In a sensitivity analysis, ARR and sample size were re-estimated using an HR of 0.75 (based on the treatment effect observed in a previous trial of lung-protective ventilation) [6] instead of 0.68. The number of serious adverse events in patients randomized to ECCO2R was predicted based on the rate observed in SUPERNOVA (6/95, 6%).

All analyses were conducted using R version 3.5.1 (http://www.r-project.org).

Results

Patient physiological characteristics and responses to ECCO2R

Patient clinical and physiological characteristics are shown in Table 1. Due to missing data in plateau pressure measurements (n = 4), PaCO2 (n = 1), and computed alveolar dead space fraction (n = 2), and because sweep gas flow was not applied in 1 patient, the changes in standardized Vt and ΔP could be computed in 87 patients. The median time between baseline and post-ECCO2R measurements was 150 min (IQR 80–300 min). ECCO2R was applied using a higher CO2 extraction device in 62 patients (65%) and a lower CO2 extraction device in 33 patients (35%).

Table 1 Baseline clinical and physiological characteristics of the study population

The change in standardized Vt after applying ECCO2R varied widely among patients (Figure E1, median − 2.0 ml/kg, range − 1.1 to − 4.0 ml/kg). The change in standardized ΔP after applying ECCO2R also varied widely (Figure E1, median − 4.1 cmH2O, range − 11.1 to 1.7 cmH2O), partly because of large changes in Crs in some patients (Figure E1, Crs increased or decreased by more than 5 ml/cmH2O in 35% of patients). The change in standardized PowerRS also varied widely (Figure E1, median − 3.8 J/min, range − 8.7 to 0.3 J/min).

At 8 h from ECCO2R initiation, 74 patients (78%) achieved the primary end-point: 18 of 33 patients (55%) on the lower CO2 extraction device and 56 of 62 patients (90%) on the higher CO2 extraction devices (p < 0.001).

Demographics, cause of ARDS, baseline PaO2/FiO2, and severity of illness were similar among patients with smaller (> median of − 4 cmH2O) vs. larger (≤ median of − 4 cmH2O) changes in standardized ΔP after applying ECCO2R (Table 1). Baseline ΔP, plateau pressure, VR, and PaCO2 were significantly higher in patients with a larger reduction in standardized ΔP with ECCO2R, reflecting lower Crs and higher ADF (Table 1).

Determinants of effect of ECCO2R on tidal volume, driving pressure, and mechanical power

In bivariate linear regression, the change in standardized Vt after applying ECCO2R was associated with ADF, VR, and device performance, but not with Crs or PaO2/FiO2 (Fig. 1); similar findings were obtained in multivariable regression incorporating all these variables (Table E1).

Fig. 1
figure 1

Determinants of the change in tidal volume (standardized to PaCO2 45 mmHg and respiratory rate of 30/min) after applying ECCO2R

The change in standardized ΔP after the application of ECCO2R was associated with ADF, VR, Crs, and PaO2/FiO2 (Fig. 2); in multivariable analysis incorporating all these variables, the effect of PaO2/FiO2 was no longer significant (Table E1).

Fig. 2
figure 2

Determinants of the change in driving pressure (standardized to PaCO2 45 mmHg and respiratory rate of 30/min) after applying ECCO2R. After adjusting for alveolar dead space fraction and compliance, the effect of PaO2/FiO2 was no longer significant

The change in standardized PowerRS after the application of ECCO2R was associated with ADF, VR, Crs, PaO2/FiO2, and device performance (Figure E2); in multivariable analysis incorporating all these variables, the effect of PaO2/FiO2 was no longer significant (Table E1).

The combined effects of Crs and ADF on the change in standardized ΔP and standardized PowerRS are represented in Figure E3.

The probability of reaching the SUPERNOVA trial primary end-point was higher when ECCO2R was applied using a higher CO2 extraction device (Figure E4), particularly in patients with higher ADF (Fig. 3, p = 0.03 for interaction). This interaction with ADF persisted in multivariable analysis (Table E1). All patients on lower CO2 extraction devices required maximum sweep gas flow to achieve the study endpoint, whereas in patients on higher CO2 extraction devices, the sweep gas flow required to achieve the study endpoint progressively increased with increasing ADF (Figure E5).

Fig. 3
figure 3

Alveolar dead space fraction determines the degree of CO2 removal required to achieve the primary endpoint in the SUPERNOVA trial (tidal volume ≤ 4 ml/kg PBW and pH ≥ 7.3 and PaCO2 within 20% of baseline) at 8 h after initiation of ECCO2R (p = 0.03 for interaction). At lower alveolar dead space fraction, reductions in Vt from 6 to 4 ml/kg can be achieved with relatively less extracorporeal CO2 removal whereas patients with higher alveolar dead space fraction require higher CO2 removal to reduce Vt. The number of patients in each group is shown within the bar

Predicting the effect of ECCO2R on driving pressure

The observed change in standardized ∆P was correlated with the predicted change in driving pressure (R2 = 0.32, p < 0.001, Figure E6). In Bland–Altman analysis, predicted and observed changes in ΔP differed by up to ± 3.9 cmH2O (Figure E6, Table E2); the bias varied with the assumed rate of CO2 removal used to predict the change in ΔP (Table E2). Discrimination between higher and lower responders was moderately acceptable (area under receiver operating characteristic curve 0.72, Figure E6).

After recomputing the predicted change in ΔP using the same value for Crs at both baseline and at 4 ml/kg to evaluate predictive performance under conditions of stable Crs (e.g. to simulate the effect of titrating PEEP to prevent decreases in Crs resulting from ECCO2R-facilitated reductions in Vt), the correlation between predicted and observed values was considerably higher (R2 = 0.74), limits of agreement were narrower (± 2.1 cmH2O), and predictive discrimination improved substantially (area under receiver operating characteristic curve 0.92) (Table E2).

Trial design simulations

Figure 4 provides predicted sample size requirements computed in ECCO2R trial simulations using different values of predicted change in ΔP to define eligibility for enrolment. Results at predicted change in ΔP = 0 represent the sample sizes required in the absence of predictive enrichment. Greater CO2 removal and restricting enrollment to patients with a higher predicted ΔP response reduced predicted sample size requirements (Fig. 4). The effect of restricting enrollment to patients with a greater predicted reduction in ΔP on sample size requirements (Fig. 4) and hence on serious adverse event rates (Figure E7) was greatest at lower CO2 removal rates.

Fig. 4
figure 4

Predicted effect of ECCO2R on ΔP (top left panel) and mortality risk (top right panel) in trial designs using different enrollment criteria and different degrees of device performance (CO2 removal). The error bars represent the 5th and 95th percentile values for the median change in ΔP predicted after incorporating random prediction error of ± 4 cmH2O. The predicted effect on mortality was used to compute sample size requirement (bottom left panel). Because only a subset of patients meet the threshold for inclusion, the number of otherwise eligible ARDS patients who need to be identified by screening (bottom right panel) is higher than the sample size for randomization

Based on these calculations, a trial designed applying ECCO2R of at least 3 ml/min/mmHg (approximately 135 ml/min assuming an average PaCO2 of 45 mmHg) to patients like those in SUPERNOVA and using a predicted reduction in ΔP ≥ 5 cmH2O to determine eligibility for randomization would need to identify 1340 otherwise eligible patients with ARDS, of whom 548 (41%) would meet enrollment criteria and be randomized to refute a predicted 12% absolute risk reduction in mortality. In a sensitivity analysis employing a less optimistic mortality effect (HR 0.75), the same trial design would need to identify 2460 eligible patients, of whom 1006 would meet enrollment criteria and be randomized (see Supplemental Tables and Online Supplement for details).

An online calculator to predict treatment response in individual patients has been made available at https://bit.ly/2RRHevj.

Discussion

This study demonstrates that the effect of ECCO2R on Vt, ΔP, and PowerRS varies widely in patients with moderate ARDS as substantially determined by ADF, Crs, and the CO2 extraction capability of the ECCO2R device. These findings suggest that patients with higher ADF or lower Crs and patients treated with higher CO2 extraction are most likely to benefit from ECCO2R (Fig. 5 and Figure E8). We also found that the a priori model for predicting the change in ΔP obtained by ECCO2R was only moderately accurate, possibly because a number of key variables (e.g. actual CO2 removal rate) were not measured in the SUPERNOVA trial and because of changes in Crs after applying ECCO2R and reducing Vt. Finally, simulations of treatment effect in different trial designs contingent upon a range of assumptions (discussed below) suggest that a future trial of ECCO2R-facilitated ultra-protective ventilation powered to detect impact on clinical outcomes, incorporating both predictive enrichment and enhanced CO2 clearance, is feasible.

Fig. 5
figure 5

Determinants of the effect of ECCO2R on driving pressure. The expected reduction in driving pressure at a given level of CO2 removal according to compliance and alveolar dead space fraction computed from the effects observed in SUPERNOVA is represented as a heat map; isopleths of driving pressure reduction are represented by the dashed lines. Heats maps are shown for two different levels of CO2 removal: the approximate rate in the SUPERNOVA trial (~ 100 ml/min), and the distribution predicted for a 50% increase in extracorporeal CO2 clearance

Even in an age of ‘big data’, the randomized clinical trial remains the most powerful and reliable tool to test new therapies in clinical practice. However, the many ‘negative’ trials in critical care have spurred important innovations in trial design [12, 13]. Predictive enrichment is one approach to enhance the probability of demonstrating benefit (if any) by selecting patients most likely to respond. The importance of enrichment in the case of ECCO2R is highlighted by the widely varying effect of ECCO2R on ΔP observed in this study and the known risk of treatment-related complications.

A theoretical analysis of the physiological equations defining alveolar ventilation suggested that the effect of ECCO2R on ΔP would be positively correlated to ADF and inversely related to Crs [6]. ADF determines the magnitude and frequency of tidal inflation required for pulmonary ventilation, while Crs determines the pressure required to achieve this tidal inflation. Accordingly, reducing the requirement for pulmonary ventilation by ECCO2R will reduce the requirement for tidal volume and pressure in proportion to ADF and Crs. This study confirms this hypothesis. Although we did not directly measure ADF, empirical estimates of ADF were strongly correlated with changes in Vt, ΔP, and PowerRS, and this finding was corroborated by observing similar effects with VR (a surrogate marker for physiological dead space) [8]. The effect of ECCO2R on ΔP and PowerRS was correlated with baseline respiratory compliance as expected, given the mathematical and physiological coupling between these variables. Importantly, although baseline PaO2/FiO2 ranged between 92 and 242 mmHg in SUPERNOVA, the effect of ECCO2R on ΔP was unrelated to the severity of hypoxemia (the usual measure of ARDS severity) after adjusting for ADF and Crs. This suggests that ADF and Crs, rather than severity of hypoxemia, should be the primary factors in determining whether to enroll patients in clinical trials of ECCO2R.

The success of predictive enrichment is contingent upon two conditions: (a) the availability of a predictive ‘biomarker’ that reliably reflects the causal mechanism conditioning outcome [14]; and (b) reliable and feasible detection or prediction of the biomarker response. The predictive validity (‘credentials’) of various biomarkers of ventilator-induced lung injury remains the subject of considerable debate: options include Vt, ΔP, and PowerRS. There is evidence that both the magnitude and frequency of tidal ventilation contribute to lung injury as reflected by PowerRS [15]—ECCO2R could permit both tidal volume and respiratory frequency to be lowered, leading to substantial reductions in PowerRS. Our trial design simulation analysis focused on driving pressure given the body of evidence in its favor and data linking ΔP with mortality [2, 11].

Even if a biomarker with acceptable credentials is available, the effect of treatment on that biomarker must be reliably determined before randomization. While a ‘test dose’ of some therapies could be used to assess biomarker response (e.g. the gas exchange response to higher PEEP in mechanical ventilation) [16], a ‘test dose’ of ECCO2R cannot be applied because it is invasive, costly, and associated with potentially serious adverse events. Predicting the response to ECCO2R is therefore crucial to incorporating predictive enrichment in the design of trials evaluating ECCO2R-facilitated ultra-protective ventilation.

Despite a sound theoretical basis and confirmatory associations between treatment effect and both ADF and Crs, a previously derived model based on these parameters was only moderately successful in predicting the ECCO2R response in this study. Limits of agreement between predicted and observed changes in ΔP were wide and the ability to discriminate between patients with or without a significant treatment response (arbitrarily defined based on the median change in ΔP) was limited. Several factors likely account for this. First, ADF was not measured, but was estimated empirically—a future trial could easily incorporate direct estimates of ADF based on end-tidal CO2 measurements. Second, Crs changed before and after ECCO2R application in many patients, possibly because lowering tidal volume relieved hyperdistention (compliance increased by 5 ml/cmH2O or more in 16% of patients) or caused alveolar derecruitment (compliance decreased by 5 ml/cmH2O or more in 19% of patients). Using a constant value for Crs substantially improved predictive discrimination. A future trial could incorporate a physiological strategy that aims to maintain [17] or even improve [18] Crs to maximize predictive accuracy and also to minimize ΔP as part of its interventional approach.

Third, accurate prediction of device performance (CO2 removal) is crucial for accurate predictions of patient response. CO2 removal rates were not directly measured in SUPERNOVA, but predicted device CO2 extraction strongly modified the ability to achieve ultra-protective ventilation targets, particularly in patients with higher ADF. In patients with lower ADF, considerable reductions in Vt can be achieved while applying little or no CO2 removal (as indicated by the relatively low sweep gas flow requirements in these patients, Figure E4), whereas patients with higher ADF can attain only small decreases in Vt apart from CO2 removal (illustrated in Figure E6). Given the importance of device performance, particularly in patients with high ADF, device performance (i.e. CO2 removal) must be accurately predicted when deciding whether to apply ECCO2R. CO2 elimination rates by membrane lungs are a complex function of several factors including sweep gas flow rate, membrane area, extracorporeal blood flow rate, venous CO2 tension, and membrane materials and design. The respective contributions of each of these variables are well understood from a biophysical perspective [10] and CO2 removal should in theory be highly predictable. Future studies should focus on developing and validating reliable models to predict CO2 removal under specified device settings for each of the devices on the market.

Despite the imperfect reliability of predicted treatment response, we show that predicting treatment effect would nevertheless substantially reduce sample size requirements, as suggested by the trial simulation results. Even when incorporating a random error in the predicted change in ΔP consistent with the limits of agreement between predicted and observed effects obtained in this study, simulated median changes in ΔP were substantially higher in predicted responders.

The estimated decrease in mortality using ECCO2R was computed from the predicted decrease in ΔP based on two crucial assumptions. First, the computation assumes that the association between ΔP and mortality reported in a previous mediation analysis is entirely causal [11]. In a sensitivity analysis, the use of a less optimistic hazard ratio for treatment effect on mortality increased sample size requirements. Second, this analysis assumes that the effect of reducing ΔP on mortality is independent of the baseline ΔP (i.e. there is no threshold effect). The benefit of reducing driving pressure even when plateau pressure is not elevated has been suggested in secondary analyses of clinical trials [11, 19] but remains unconfirmed. These assumptions are considered and discussed in detail elsewhere [6]. A future clinical trial is required to address these uncertainties. Assumptions about treatment effect are unavoidable in trial design, particularly in sample size computations. Adaptive designs can help to mitigate such assumptions by permitting ongoing enrollment if the treatment effect is weaker than expected [20]. An adaptive design might also initially employ broad inclusion criteria and adopt increasingly restrictive criteria if evidence accumulates that predicted responders experience greater benefit than predicted non-responders.

The main limitation of this analysis, aside from those reviewed above, is that the SUPERNOVA study was planned and carried out before the theoretical analysis linking changes in ΔP obtained with ECCO2R to individual patient physiological characteristics was published. For this reason, the measurements and procedures required to optimize predictive accuracy were not performed and adjustments were required to address confounding factors (i.e. concomitant changes in respiratory rate and PaCO2 target). A future ECCO2R trial designed to maximize the reduction in ΔP obtained from ECCO2R could address these limitations by directly measuring all relevant baseline variables prospectively (ADF, Crs, CO2 elimination) and computing the predicted change in ΔP prior to initiation of ECCO2R.

Conclusions

The effect of ECCO2R on Vt, ΔP, and PowerRS varies widely in patients with moderate ARDS as substantially determined by ADF and Crs. These findings suggest that patients with higher ADF or lower Crs and patients treated with higher CO2 extraction devices are most likely to benefit from ECCO2R (Fig. 5). Incorporating predicted treatment response and higher CO2 removal rates as factors in trial design might substantially reduce screening and sample size requirements in a future trial of ECCO2R-facilitated ultra-protective ventilation.