ARDS: challenges in patient care and frontiers in research

Noninvasive versus invasive mechanical ventilation in patients with ARDS

The main advantages of noninvasive over invasive mechanical ventilatory support are the reduced need for sedation and prevention of ventilator-associated pneumonia. Noninvasive mechanical ventilation can be interrupted easily, and thus has the potential to shorten the time the lungs are subjected to the harmful effects of mechanical support.

While noninvasive mechanical ventilatory support by face mask was found to affect neither intubation rates nor clinical outcomes in a randomised controlled trial (RCT) of 123 ARDS patients [5], helmet ventilation was found to reduce the need for intubation when compared to oxygen therapy in a multicentre RCT of 81 ARDS patients [6], and to reduce intubation rates and even to improve survival when compared to a face mask in a single-centre randomised clinical trial of 83 ARDS patients [7].

Oxygen administered through a high-flow nasal cannula is suggested to be an attractive strategy as a first-line therapy to avoid intubation, seen in the results of a multicentre RCT in which 301 hypoxaemic patients without hypercapnia received either high-flow oxygen therapy, standard oxygen therapy delivered through a face mask or noninvasive positive-pressure ventilation [8], but the beneficial effects found thus far need to be confirmed.

When noninvasive mechanical ventilatory support is used, patients should be monitored closely [9], as it can fail, for example because of patient nonacceptance and technical issues, but also when the disease progresses with a worsening of gas exchange. Substantial clinical (respiratory rate) and laboratory (blood gas analyses) improvements must be seen within the first few hours, otherwise it is necessary to proceed with invasive ventilation. A prospective observational trial of 101 patients suggested that a rapid shallow breathing index >105 breaths·min⁻¹·L⁻¹ [10] could be used to decide to intubate the patient. A prospective observational trial of 62 patients suggested that expired tidal volumes persistently >9.5 mL·kg⁻¹ predicted body weight (PBW) predicts noninvasive ventilation failure [11].

Modes of invasive mechanical ventilatory support

Volume-controlled modes of mechanical ventilation allow for better measurement of respiratory mechanics and driving pressures and may have a potential advantage with regard to the prevention of volutrauma. In addition, in contrast to pressure-controlled modes of mechanical ventilation, volume-controlled modes do guarantee a predefined tidal volume. Conversely, pressure-controlled modes have the alleged advantage of a decelerating flow curve during inspiration. Despite these potential physiological considerations, there are no outcome advantages of using a pressure-controlled mode over a volume-controlled mode, or vice versa. Indeed, one systematic review and meta-analysis of 34 studies did not suggest differences in any clinical outcome [12]. However, the studies were small and varied considerably in quality, and were at high risk of bias.

Tracheotomy

Tracheotomy may increase patient comfort. Several RCTs, including a large trial of 909 patients who did not necessarily have ARDS showed that early tracheostomy (within 4 days of admission to critical care) was not associated with an improvement in mortality or morbidity [13]. However, one systematic review and meta-analysis of 12 investigations suggested that early tracheotomy could lead to more ventilator-free days, shorter stays in intensive care, shorter duration of sedation and reduced long-term mortality [14]. The optimal timing of tracheotomy in critically ill patients remains a topic of debate. Presently, tracheotomy is recommended only when prolonged mechanical ventilation is expected.

Tidal volume size and plateau pressure

One key causal factor of so-called ventilator-induced lung injury is damage to the lungs due to overdistension from the use of high tidal volumes and inappropriately high airway pressures.

Mechanical ventilation targeting a tidal volume of 6 mL·kg⁻¹ PBW and plateau pressure <30 cmH₂O was found to reduce mortality and morbidity in the pivotal ARDS Network “ARMA” trial of 861 ARDS patients [15]. One important notion is that tidal volume size should be titrated according to PBW (based on height and sex) and not to actual body weight, a frequently missed issue [16]. However, setting a low tidal volume could still result in overdistension and lung stress if the amount of aerated lung tissue is or remains low, as shown in an observational study in 30 patients [17]. In this elegant study, pulmonary computed tomography scanning at end-expiration showed that limiting tidal volume to 6 mL·kg⁻¹ PBW and plateau pressure to 30 cmH₂O may not be sufficient in patients characterised by large nonaerated parts. Either tidal volumes of <6 mL·kg⁻¹ PBW [18] or higher positive end-expiratory pressures (PEEP) to recruit more lung tissue could be used [19]. Another approach is to adjust tidal volume to compliance by using the driving pressure (plateau pressure minus PEEP), aiming at a value <15 cmH₂O [20]. Notice that this last approach is based on evidence coming from an individual patient data meta-analysis of RCTs that did not test the role of driving pressure. Direct RCT evidence for this approach is lacking.

PEEP

Three large RCTs comparing high with low PEEP in ARDS patients, the ALVEOLI trial in 549 ARDS patients [21], the LOVS trial in 983 ARDS patients [22] and the EXPRESS trial in 767 ARDS patients [23] showed that a ventilation strategy using a high PEEP does not reduce mortality, despite improved lung function. One meta-analysis using individual patient data from these three RCTs showed ventilation with high PEEP to be associated with better survival [24]. However, it should be acknowledged that the mortality benefit of high PEEP was restricted to patients with moderate or severe ARDS, and ventilation with high PEEP was associated with prolonged duration of ventilation in patients with mild ARDS.

Ventilation with high PEEP balances between potential benefit (i.e. recruitment) and harm (i.e. overdistension), and individual responses to PEEP are very variable [25]. A test comparing several PEEP levels could help select the best PEEP level for an individual patient: PEEP titrations may be based on the extent of lung recruitability [26], changes in transpulmonary pressures [27] or more-easily captured changes in driving pressures [20].

Rescue therapies

Prone positioning results in redistribution of lung densities, with recruitment of the well perfused dorsal regions. Thus, prone positioning could result in a reduction in alveolar shunt and improved ventilation–perfusion matching. While several RCTs of prone positioning in ARDS patients failed to show improvements in outcome, one systematic review and meta-analysis of nine RCTs showed an association between prone positioning and mortality, although only in patients with an arterial oxygen tension (P_aO2)/inspiratory oxygen fraction (F_IO2) ratio <100 [28]. One multicentre, prospective RCT in 466 ARDS patients with a P_aO2/F_IO2 <150, F_IO2 ≥0.6 and PEEP ≥5 cmH₂O showed that early application of prolonged prone positioning sessions significantly decreased 28-day and 90-day mortality [29]. Of note, not all patients may benefit from prone positioning, but it should always be considered in the early phase of severe ARDS [29]. Relative contraindications for prone positioning include open abdominal wounds, unstable pelvic fractures, instable spinal lesions and brain injury and the absence of staff well-trained in prone positioning.

Muscle paralysis with neuromuscular blocking agents may avoid dyssynchrony, prevent excessive transpulmonary pressures and breath stacking by inspiratory efforts [30] and derecruitment by expiratory efforts [31]. Neuromuscular blocking agents have been shown to benefit patients with ARDS [32], confirmed in a meta-analysis of three RCTs [33]. Muscle paralysis with neuromuscular blocking agents should probably be reserved for patients with the most severe forms of ARDS, and restricted to the acute phase and during the first 48 h.

Extracorporeal membrane oxygenation provides different degrees of carbon dioxide removal plus oxygenation, thus preventing in part harm caused by mechanical ventilatory support. This technique should only be used in the most severe cases of ARDS [34], and only by well-trained staff [35]. Extracorporeal carbon dioxide techniques have been used to allow mechanical ventilatory support at very low tidal volumes (3–4 mL·kg⁻¹ PBW) [36, 37]. However, convincing evidence of the benefit of both techniques is absent and eagerly awaited. Trained staff and sufficient exposure could be critical for its safe use [35].

Glucocorticosteroids

Glucocorticosteroids are effective in numerous inflammatory diseases across many fields in medicine. It seems plausible that high-dose glucocorticosteroids could improve outcomes in ARDS because of their pleiotropic anti-inflammatory action [42]. However, the evidence for benefit of glucocorticosteroids in ARDS is conflicting and the use of glucocorticosteroids in patients with ARDS remains controversial.

The latest individual data meta-analysis indicates an accelerated resolution of ARDS when patients are treated with a high dosage of glucocorticosteroids for a prolonged period of time [43]. Furthermore, methylprednisolone treatment has been found to be associated with shorter periods of need for invasive mechanical ventilation and lower mortality rates. Glucocorticosteroids have not been found to increase the risk of infections. These results are very encouraging, but should be seen in the light of the limitations of the original studies. There is clear need for additional clinical trials of glucocorticosteroids.

Salbutamol

Salbutamol was suggested to reduce pulmonary oedema through stimulation of the sodium–potassium pump (Na/K/ATPase) but the largest trial to date was stopped after the inclusion of 326 patients because an interim analysis suggested that the intervention might cause harm [44].

Surfactant

Administration of exogenous surfactants was found to improve oxygenation when results of all studies were pooled in a systematic review, but did not influence mortality and should therefore not be used in patients with ARDS [45].

N-acetylcysteine

N-acetylcysteine is the best-studied antioxidant in ARDS. It increases glutathione levels and was thought to reduce the oxidative pulmonary damage seen in patients with ARDS. A systematic review of several smaller clinical trials in patients with ARDS showed that infusion of N-acetylcysteine did not affect mortality [46].

Neutrophil esterase inhibitor

Neutrophil esterase seems to play a central role in the activation of neutrophils in the lung during the development of ARDS [47]. Sivelestat, a neutrophil esterase inhibitor, is used in the treatment of ARDS in Japan. However, pooled analysis of all available data showed no reduction in mortality, even when only studies performed in Japan were included [48].

Granulocyte-macrophage colony-stimulating factor

Granulocyte-macrophage colony-stimulating factor was tested as a treatment to promote resolution of lung injury, 3 days after patients fulfilled the ARDS criteria. A phase II study with 130 patients did not show any effect on length of stay or mortality [49].

Ineffective interventions

Intervention efficacy is tested frequently in small-animal models for acute lung injury. Several factors, besides the fact that rodents may inherently respond differently to treatment, can prevent successful drug selection in preclinical studies.

Lung injury is experimentally evocated through “hits”, which correspond to the many risk factors for ARDS (such as sepsis, trauma, pneumonia and aspiration). However, none of the small-animal models provide all the hallmarks of human ARDS [52, 53] and therefore, by definition, fail to represent the clinical scenario of ARDS. Besides, most experiments use young male rodents that almost never have comorbidities for the evaluation of new treatments for ARDS, while patient populations seen in intensive care units are elderly and frequently have concomitant diseases such as diabetes mellitus, chronic obstructive pulmonary disease or active cancer [54]. The few studies that compared elderly rodents to younger ones indeed found a difference in the biological response between these groups [55, 56].

The timing of the start of treatment is another factor that may limit the translation of positive results from preclinical experiments to clinical trials in humans. Interventions may only be effective when given preventively or early in the course of disease; ARDS patients have to be recognised and informed consent must be obtained from the legal representative before a drug can be administrated in a clinical trial. Together, these factors may, at least in part, explain the failed translation of preclinical results and we should put more effort in creating models that better reflect the clinical situation.

Balance between benefit and harm

A second challenge in clinical trials of ARDS treatments is the inherent biological heterogeneity. The biological heterogeneity directly relates to the ARDS definition, which is based on clinical criteria without a central, singular underlying pathophysiological mechanism [57]. A drug targeting a specific molecular mechanism that is effective in part of the population might harm patients with a separate biological response. Therefore, the outcome of the study is highly dependent on the prevalence of patients that are responsive (e.g. in whom benefit outweighs harm) to the treatment in the included population. It is the premise of precision medicine to enrich this group as much as possible with patients with a responsive phenotype (also called predictive enrichment) [58]. The selection of the right patient for the right treatment is one the largest challenges for the coming decade [59, 60].

Exclusion of patients with a negative balance between benefit and harm

Patients with a particular phenotype that is responsive to a treatment can be identified through pathophysiological reasoning. An illustrative example is the application of prone positioning in patients with a P_aO2/F_IO2 ratio <150 [61]. The physiological consequences of laying a patient in prone position include the recruitment of the dorsal parts of the lungs and derecruitment of the ventral parts of the lung [62, 63]. Because of the ventro–dorsal asymmetry of the thorax, prone positioning a patient with ARDS frequently results in an overall recruitment effect [63]. These phenomena can be utilised in patients with ARDS (see the section on mechanical ventilation) because it will improve gas exchange and reduce ventilator-induced lung injury [62]. However, this procedure is not without complications such as hypotension, pressure sores, accidental extubation and the complications of more sedation [64]. With the background knowledge that patients with lower P_aO2/F_IO2 scores have an increased heterogeneity of ventilation in the lung and the most problems with gas exchange, this is the phenotype that is most likely to benefit from the intervention.

If we apply this reasoning to pharmacological interventions, we could aim to identify key mediators in the pathophysiology of ARDS and design drugs to target these mediators. Studies can be enriched through the selection of patients that have a higher expression of the targeted molecules. For example, a study investigating an antibody against interleukin-8 could choose to select patients based on high level of interleukin-8 in plasma, or, even better, locally in the lung. Patient selection based on the treatment mechanism is still not utilised sufficiently in intensive care medicine research and may improve the chances on finding effective pharmacological interventions for ARDS.

Better selection of phenotypes through an unbiased, data-driven approach

A second approach towards the definition of ARDS phenotypes is an unbiased, data-driven analysis of clinical or biological data. Although the case for the application of prone positioning in a patients with a clinical phenotype of a low P_aO2/F_IO2 ratio seems logical with hindsight, almost all studies were designed to include all patients with ARDS. The low P_aO2/F_IO2 ratio phenotype was recognised only after several RCTs were completed [28].

In essence, the data-driven approach recognises that we are inherently biased in our recognition of subgroups and it faces the problem of phenotype selection from the other direction, not initiated by pathophysiology, but by empirical evidence, which is later translated back to pathophysiology.

If we recognise the complexity of the biological processes that contribute to ARDS it seems overly simplistic and therefore unlikely that we would identify a single marker such as P_aO2/F_IO2 ratio that is predictive of a specific, responsive phenotype. Over the past decades, vast improvements in bioanalytical assays and computing power have introduced the age of “omics” analyses. When more than one variable is considered to define a phenotype, a data-driven approach is even more warranted. Computer algorithms have no inherent struggles with highly dimensional datasets and we can utilise that capacity to reduce complex datasets to understandable proportions. The output of computer algorithms can serve as inputs for clinicians and can guide them in making the right clinical decision. With unbiased cluster analysis, one of the most widely used algorithms for identifying patterns, researchers can identify patients who show similar biological “fingerprints”, even though they might not look alike from a clinical perspective. This data-driven analysis limits the danger of self-fulfilling prophecies based on widely accepted paradigms and are hypothesis-generating.

Two landmark articles suggest that ARDS can be separated into at least two distinct phenotypes based on unbiased, data-driven analyses of clinical data and several plasma biomarkers of inflammation, coagulation and endothelial activation [65, 66]. One phenotype is characterised by a hyperinflammatory and hypercoagulatory state in the blood, while the other phenotype shows a hypo-inflammatory state and high bicarbonate levels [65, 66]. One of the major disadvantages of such a data-driven approach is the danger of over-fitting to the specific dataset. Importantly, very similar phenotypes were found independently in a group of ARDS patients that was separated in time and place, and analysed using an alternative approach [67]. Progress in the identification of specific phenotypes is not only seen in ARDS, but also in other problems frequently encountered in the intensive care unit, such as sepsis and failure to wean [68–70].

Implications for clinical practice in the future

Post hoc analysis of two multicentre RCTs indicate that the two above-described phenotypes show a differential response to a high (versus conventional) PEEP strategy and a conservative (versus liberal) fluid management. Fluid management was steered by a strict protocol relying on the central venous pressure or pulmonary artery occlusion pressure, need for vasopressors and an “effective circulation” (combination of cardiac index and physical examination) [71] resulting in the advice for vasopressor use, furosemide administration or fluid administration. The hyper-inflammatory phenotype seemed to benefit from a higher PEEP strategy and conservative fluid management, while the hypo-inflammatory phenotype did not seem to benefit and might even be harmed by the intervention (figure 1). Of course, post hoc analyses of RCTs do not provide definitive answers, as research questions are not stated beforehand and the randomisation process is disturbed, but it indicates that our simplistic views on the management of ARDS may need to be revised in the near future.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 1

Differential response to treatment between two phenotypes of acute respiratory distress syndrome. a) High and low positive end-expiratory pressure (PEEP); b) conservative and liberal fluid management.