Supported by   

Pijitra Phomkham | Shutterstock

The Trial

In critically ill patients receiving invasive mechanical ventilation for acute hypoxaemic respiratory failure, would extracorporeal carbon dioxide removal (ECCO₂R ), with the aim of facilitating a reduction of tidal volumes from 6 ml/kg predicted body weight (PBW) to ≤ 3 ml/kg PBW, reduce 90-day mortality by 9%, from 41% to 32%. 1120 patients would provide 90% power to identify such an effect size at the 5% significance level.

The trial was terminated for futility during a planned interim analysis. 412 patients had been recruited. There was no significant difference in the primary outcome: 41.5% in the intervention group vs 39.5% in the standard care group (risk ratio, 1.05; 95% CI, 0.83 to 1.33; difference, 2.0%; 95% CI, −7.6% to 11.5%).

The Question

Why did the REST trial¹ fail to clearly demonstrate benefit with the addition of an ECCO₂R  device in invasively mechanically ventilated patients with acute hypoxaemic respiratory failure?


The Answer


The design of the trial had two potential limitations. The first was the underlying premise, that an an ultra-low tidal volume ventilatory strategy could reduce ventilator-induced lung injury (VILI) sufficiently to affect mortality. The second was an optimistic power calculation. Both issues were understandable, given this was the first major trial in this field, and therefore was designed with limited prior data.


The underlying premise of REST was that reducing tidal volume from 6 ml/kg BPW to ≤ 3 ml/kg PBW would decrease mechanical ventilator-induced lung inflammation sufficiently to improve mortality. In the Xtravent study², which also compared  tidal volumes of 6 ml/kg PBW with 3 ml/kg BPW supported by ECCO₂R, there was no clear effect on the inflammatory biomarkers TNF-a, IL-6, and IL-8, when measured in blood. Similarly, a recent study in patients undergoing ultra-low tidal volume ventilation (≈ 3 ml/kg PBW) facilitated by ECMO also failed to show a reduction in pulmonary inflammation.³ The data published thus far from REST do not allow conclusions to be drawn on whether additional lung protection was achieved by this strategy, but it may be that ultra-low tidal ventilation does not reduce VILI much beyond standard low tidal volume ventilation. However, it is only by running a trial such as REST that this information may be learned.

The extreme ventilatory management required to treat refractory hypoxaemia during the COVID-19 pandemic raises another possibility. Anecdotally, the provision of protective ventilation in desperate measures has required the reduction of tidal volume to 3-4 ml/kg BPW without the use of ECCO₂R. If renal buffering of hypercarbia is maintained, this approach appears to be possibly life sustaining. In the PREVENT trial, comparing low (4-6 ml/kg PBW) with intermediate (8-10 ml/kg PBW) tidal volumes in invasively ventilated patients in the ICU without ARDS, in those receiving volume controlled ventilation, lower tidal volumes (~5.5 vs 9.5 ml/kg) on the day of randomisation and day 1 both translated to more patients having a pH < 7.25 at these time points. This pH difference was lost by day 2. There was no difference between the two groups in the primary outcome of the number of ventilator-free days and alive at day 28 (21 vs 21). 

In a French prospective before-and-after study of 35 ARDS patients within 24 hours of ARDS diagnosis and with PaO₂/FiO₂ ≤ 150 mmHg, it was feasible to decrease tidal volumes from 6 ml/kg PBW to 4 ml/kg PBW without the use of ECCO₂R. One third of patients developed a transient pH < 7.15. Overall, at 41%, the 28 day mortality of this cohort was similiar to the hospital mortality of 40% of the moderate ARDS group (PaO₂/FiO₂ 100 to 200 mmHg) of the global epidemiology study LUNG-SAFE. Whilst speculative, if this held true, then the addition of an ECCO₂R machine may add little in terms of benefit, but potentially subject a patient to uncommon, but dangerous, harms from the insertion or presence of the device. Prior to the pandemic the routine occurance of such profound hypoxia was rare and typically managed with ECMO; as such, this (largely untested) information was not available to the investigators during the design or early running of the trial.

Power Calculation

The second potential limitation was the power calculation, seeking a large effect size, of a 9% absolute mortality reduction. Few interventions in critical care produce an effect of this magnitude.To identify a more realistic effect size of 5%, from 41% to 36%, with the same alpha and beta, just under 3976 patients would be required. To put that potential sample size in perspective, the two randomised controlled ECMO trials completed to date, CSEAR and EOLIA¹⁰, recruited 180 and 249, respectively. The contemporary ECCO₂R trial Xtravent² recruited 79, and the feasibility study SUPERNOVA¹¹ enrolled 95. REST had recruited 412 patients at the time of termination.

This 9% effect size was based on the landmark ARMA trial¹², which ran from 1996 to 1999  and was published in the New England Journal of Medicine in 2000. It reported a mortality reduction from 39.8% to 31.0%. However, this mortality benefit was seen with reducing tidal volumes from 12 ml/kg PBW to 6 ml/kg BPW. It is unlikely this 6 ml reduction in tidal volume, from a harmful value to a more physiological value, is comparable to a 2–3 ml reduction in tidal volume, starting from a safer physiological tidal volume. Three additional points are worth considering. Firstly, the tidal volume reduction in the ARMA trial was just one of a bundle of interventions, which also included controlling peak airway pressures to < 50 cmH₂O in the control group and < 30 cmH₂0 in the intervention group, and also the use of bicarbonate infusions to control mild and moderate acidosis in the low tidal volume group. Secondly, this tidal volume reduction was not achieved in the setting of an ECCO₂R  device. Thirdly, critical care practice has changed significantly over the past 25 years. Thus, both the intervention and population were possibly sufficiently different to doubt the direct imputation of a potential 9% mortality reduction, forcing the question as to whether this was the best trial to base a power calculation on.

Regardless of any design questions about the power calculation, sample size or effect size, the actual results in over 400 patients point towards a lack of benefit with ECCO₂R, with the point estimate for the primary outcome non-significantly favouring standard care, making it less likely that completing the trial would have changed the final result.


This population had a high severity of illness, as described by their median APACHE II score at ICU admission of 19-20 and overall 90-day mortality of about 40%. As such, if a beneficial effect from ECCO₂R existed, a sick cohort should as this would be ideal to identify it.

Although not specifically an ARDS trial, an issue noted in ARDS trials has been the recruitment of patients with mild ARDS, which resolves within 24 hours of the initiation of mechanical ventilation.¹³ In one small study, almost 60% of patients meeting ARDS criteria, and placed on standardised ventilator settings including a tidal volume of 7-8 ml/kg PBW, PEEP of 10 cmH₂O and FiO₂ 1.0, demonstrated a transient nature to their ARDS. Such patients had a much lower mortality than those with persisting ARDS (12.5% vs. 52%, P=0.001).¹⁴ Studies such as ART¹⁵ and Xtravent² have used an enrichment strategy, including high PEEP, to identify and exclude those likely to do well regardless, and in whom their trial continuation would otherwise fail to contribute an outcome event and lead to a loss of power. REST included patients within 48 hours of the onset of hypoxaemia, defined as a PaO₂/FiO₂ < 20 kPa on ≥ 5 cmH₂O of PEEP.  Approximately 60% of patients in REST met criteria for ARDS. The inclusion of patients less likely to benefit should, with randomisation, balance across groups and lead to a loss of power to identify a true beneficial effect, should one exist.


Many randomised controlled trials of procedures and devices are not simple head-to-head comparisons of intervention vs control, but are complex multi-faceted contrasts between an intervention, and the associated alterations in management it brings, and the control group. Many such trials are necessarily open-label, which also introduces biases, both overt and covert. As would be expected, REST exhibits several of these issues.

The REST trial tested ultra-low tidal volume ventilation enabled by ECCO₂R. The intervention group was adequately exposed to the ECCO₂R device, with 186/202 (92%) patients in the intervention group receiving it, and achieving CO₂ removal of between 46 and 85 ml/minute. This allowed tidal volumes of 4.4 to 6.3 ml/kg PBW in the intervention group, which was consistently lower than the standard care group. The per protocol population in the intervention group had tidal volumes of 4.2 and 3.8 ml/kg PBW on days 2 and 3. As such, the trial did not quite achieve the desired tidal volume in the intervention group of ≤ 3 ml/kg. It is not clear why this happened, but it only minorly reduces the internal validity of the trial. REST ultimately tested 6-7 ml/kg PBW against a lower tidal volume of 4-6 ml/kg (difference approximately 2 ml on days 1 – 3 and 1 ml on days 5 to 7), rather than against ≤ 3 ml/kg tidal volme, as intended.

Table 1: Mean (SD) Tidal Volumes on Days 2 to 7

Naturally, the results of a trial are determined by the care the patients actually receive. Although an intention-to-treat analysis reduces bias from non-compliance with the protocol, it tests the intention to use the intervention as stated in the trial hypothesis (i.e. ≤ 3 ml/kg). As such, even though the target tidal volume wasn’t achieved, the intention to achieve this was adhered to, permitting the trial to retain internal validity, although at a slightly lower value than hoped for. In practice, clinicians using ECCO₂R may intend to achieve tidal volumes of ≤ 3 ml/kg, but may not do so for a multitude of reasons.

The trial protocol required ECCO₂R  to be used for at least 48 hours. However, patients required significant sedation, and possibly neuromuscular blockade, to tolerate the ultra-low tidal volumes of 3-4 ml/kg. Once a patient entered the trial and was allocated to the intervention arm, they were more deeply sedated, and possibly curarised, for at least 48 hours. While curarisation rates were similar between the groups on day 1, from days 2 to day 7 they were increased in the intervention group (day 2, 25% higher; day 3, 66% higher; and from days 3 to 7, approximately 100% higher). This increase in neuromuscular blockade, and implementation of ECCO₂R, was associated with lower tidal volumes (approximately 2-3 ml/kg difference per day) and lower plateau pressures (approximately 2-3 cmH₂O difference per day).  In the ACCURSY trial,¹⁶ investigating neuromuscular blockade in ARDS, the implementation of paralysis did not result in differences in tidal volume, plateau pressure or respiratory system compliance on days 1, 3 or 7, suggesting it was indeed the ECCO₂R device which facilitated the observed decrease in ventilation intensity in the intervention group.

Whilst the ECCO₂R machine facilitated the reduction in tidal volume and airway pressures, ironically, it may have caused ventilatory weaning to effectively stop. The intervention could be applied for up to 7 days. The mean duration of ECCO₂R  in the trial was 4 days (standard deviation, 2 days). Day 1 was largely consumed with obtaining consent and setting up the ECCO₂R device and transitioning to ultra-low tidal volume ventilation. Between days 2 and 6, PaO₂/FiO₂ was consistently better in the control group and a greater proportion progressed to spontaneous modes of ventilation, equating to between 20% and 25% more per day between days 3 and 7. Once the ECCO₂R machine was withdrawn, a patient in the intervention group was required to wake from the additional sedative load and catch-up the weaning which had occurred in the standard care group. This effective halt on ventilator weaning may have contributed to the two  less ventilator-free days in the intervention group.

Unfortunately, sedation use and depth of sedation were not recorded in the trial, but with more patients in the intervention group requiring neuromuscular blockade, it is logical to conclude this group may have received more sedatives. Additional sedative use,¹⁷ and deeper depths of sedation,¹⁸ have been associated with worse outcomes. Interestingly, the opposite was found in the Xtravent study,² with the interventional group requiring less sedoanalgesia (sufentanil plus midazolam), although with no change in the RAAS scores.

Proning reduces mortality in severe hypoxaemic respiratory failure. The intervention group in REST had lower rates of proning on days one (5.5% vs 13.9%) and two (9.1% and 13.0%), which largely balanced out for the remaining 5 days of the intervention period. However, these percentages translate into small numbers and would be unlikely to impact the overall trial result. Inclusion criteria were similar to that of the landmark PROSEVA trial,¹⁹ which reported a mortality reduction from 32.8% to 16% with proning in 566 patients with moderate to severe ARDS. Common inclusion criteria were a PaO₂/FiO₂ ≤150 mmHg (20 kPa) and PEEP ≥ 5 cmH₂O, with patients in PROSEVA and the control arm of REST receiving a tidal volume ~ 6 ml/kg PBW. The main difference between the REST and PROSEVA inclusion criteria was the requirement for a FiO₂  ≥ 0.6 in PROSEVA, which was absent in REST, although the baseline PaO₂/FiO₂  was 118 mmHg in the intervention arm. Whilst curious to note the low rates of prone positioning, this difference is unlikely to explain the lack of clear benefit seen in the trial.

ECCO₂R was successfully deployed and resulted in the intended seperation of the two groups in their mechanical ventilatory load. Whilst a post-hoc evaluation²⁰ of 5 randomised controlled trials suggested there was no safe lower limit to plateau pressures, the 2 cmH₂O reduction in plateau pressures in REST did not convert into an outcome benefit. Similarly, patients in the ECCO₂R group had driving pressure approximately 2-3 cmH₂O lower (~11 vs 14 cmH₂O) between days 2 to 5. It will be intriguing to see additional data on pulmonary inflammatory markers in subsequent substudies, although this will be somewhat confounded by the effect of neuromuscular blockade, which itself reduces lung inflammation. The collorary of this, is that the intervention group received two mechanisms capable of lessing lung inflammation and potentially improving outcome, neither of which ultimately resulted in a clearly superior outcome.

Sensitivity analyses were undertaken to explore other possible reasons for a lack of clear benefit from ECCO₂R. Analysis excluding the first two patients recruited at each centre, to minimise a learning effect, showed no sizable difference from the main trial result. Equally, excluding the 8% of patients allocated to the ECCO₂R group but who did not receive the intervention, again did not change the outcome. Interestingly, just 1 patient in the control arm received ECCO₂R. When compared with a massive 28% crossover from control to intervention in the EOLIA trial,¹⁰ investigating ECMO in severe ARDS, this may reflect a lack of conviction of local investigators in the efficacy of ECCO₂R, especially as 41% of this group died, with just 12 patients (7%) converting to ECMO.


The question of harm (adverse events) is confounded by the open-label nature of the trial, and a propensity to report less events in the standard care arm. A key finding was 9 serious intracranial haemorrhages in the intervention group and none in the control group. Systemic anticoagulation with heparin was used to prevent thrombosis in the extracorporeal circuit, targeting an activated partial thromboplastin time ratio of 1.5 to 2.0 (aPPT 45 to 65 seconds), which is similar to the degree of anticoagulation typically provided when undergoing systemic heparinisation for renal replacement therapy. Two recent renal replacement therapy trials did not appear to report any cases of intracerebral haemorrhage. Although STARRT-AKI²¹ predominantly used citrate anticoagulation, out of 667 patients receiving systemic heparin anticoagulation, no cases of intracerebral stroke associated with RRT were reported. Similarly, in the AKIKI2 trial²², there were no reported cases of intracerebral haemorrhage associated with RRT use, although the modalities of anticoagulation were not reported.

In a small feasibility trial of 20 patients with mild to moderate ARDS, there were no cases of intracerebral haemorrhage when ECCO₂R was administered via a renal replacement therapy platform.²³ These patients received systemic heparin, aiming for an aPTTr 1.5 to 2.0. PaCO₂ peaked when ventilated at 4 ml/kg, at 53 ± 9 mmHg. As cerebral vasodilation is influenced by PaCO₂, it may be that in the REST trial, over the week of intervention, the higher PaCO₂ in the ECCO₂R group could have contributed, although at its maximum on day 3, this difference was just 6.7 mmHg (0.9 kPa; difference 60.8 mmHg vs 54.2 mmHg). The rate of intracranial haemorrhage seen in REST is similar to previously reported values in vv-ECMO of about 5%.²⁴ In comparison to ECCO₂R, ECMO requires larger cannulae, faster blood flows and the patients are typically sicker.

Two single-centre retrospective studies report a prevalence of intracranial haemorrhage of 10.7%²⁵ and 16.7%,²⁶ respectively, when CTs were performed within the first 24 hours of the initiation of ECMO. As many ECMO patients are extremely unwell and unable to undergo scanning prior to cannulation and implementation of ECMO, these data may not be easily improved upon. Perhaps the high rates of intracranial haemorrhage seen in REST reflect a vulnerability of a relatively hypoxic brain to intracerebral haemorrhage, which is then exacerbated by systemic anticoagulation. Although approximately 75% of patients in STARRT-AKI were receiving invasive mechanical ventilation, their PaO₂/FiO₂ were not reported. The REST and STARRT-AKI populations may have been sufficiently different for such an underlying propensity to intracranial haemorrhage to exist in one group but not the other.

Implications of REST

The clinical implications of the REST trial are that ECCO₂R presently offers little in the routine management of patients with acute hypoxaemic respiratory failure. However, this landmark trial, the first major RCT in the field and the largest yet of an extra-corporeal respiratory support device, provides much valuable information for the design of future studies. Improving tolerance to ultra-low tidal volume ventilation and avoiding the need for deeper sedation and neuromuscular blockade would be a significant improvement. Science is an iterative process, and this story, which began in 1984²⁷ is doubtless advanced by the findings of this trial.


In summary, the REST trial has high internal validity and adds substantial knowledge to this relatively new field. It identified a sick population, executed the intervention with good fidelity and had an appropriate control arm. Sources of bias were minimised as much as is possible in an open label trial. It achieved what is set out to achieve, and adequately tested the research hypothesis. In this trial, ECCO₂R was not clearly beneficial, although the inconclusive nature of the result, incorporating a range of outcomes including a 17% mortality improvement to a 33% mortality excess, does not exclude this possibility.


Promoting Open Access to Scientific Advancement
Supported by   

© All rights reserved.

Back to Top