ARDS

In depth

The one-page summary above is what the PDF prints. These sections go further.

Pathophysiology and phases

ARDS is diffuse inflammatory injury of the alveolar-capillary membrane. A direct insult (pneumonia, aspiration, inhalation, contusion) or an indirect one carried by the blood (sepsis, pancreatitis, transfusion, trauma) activates neutrophils and endothelium; the barrier leaks protein-rich fluid into the alveoli, surfactant is inactivated, and alveoli flood or collapse. The result is shunt, so PaO₂ responds poorly to FiO₂ alone, plus dead space from microthrombi that raises PaCO₂ for any minute ventilation.

The exudative phase occupies about the first 7 days: edema, hyaline membranes, neutrophil infiltration and the lowest compliance. The proliferative phase follows, lasting more than 3 weeks in survivors, with type II pneumocyte growth, fluid clearance and repair; most survivors improve here. A minority develop fibrosis within that phase, with collagen deposition, persistent low compliance, dead space and ventilator dependence (Matthay 2019).

CT scans reshaped the model. Gattinoni and Pesenti (2005) showed that the normally aerated tissue in most patients weighs about 300–500 g, the size of a 5- to 6-year-old's lung, and that compliance tracks the size of this 'baby lung' rather than the stiffness of the tissue. The ARDS lung is small rather than stiff, so the risk of a breath depends on tidal volume relative to the aerated lung, not on milliliters per kilogram. The baby lung is functional: densities redistribute when the patient is turned prone.

Definitions and phenotypes

Berlin (2012) required onset within one week of a known insult or new or worsening symptoms, bilateral opacities not fully explained by effusion, collapse or nodules, respiratory failure not fully explained by cardiac failure or fluid overload, and a P/F ≤300 mm Hg on PEEP or CPAP ≥5 cm H₂O. Its severity bands predicted mortality: mild 27%, moderate 32%, severe 45%, with median ventilation in survivors of 5, 7 and 9 days. Draft ancillary variables such as PEEP ≥10 for severe disease were dropped as unhelpful.

The 2024 Global Definition (Matthay and the ARDS Definition Task Force) kept the timing, imaging and edema criteria and extended the rest. Bilateral B-lines or consolidation on ultrasound by a trained operator now count as imaging. Patients on high-flow nasal oxygen at 30 L/min or more, or on NIV or CPAP with at least 5 cm H₂O, can have 'non-intubated ARDS' with P/F ≤300 or SpO₂/FiO₂ ≤315 (valid only when SpO₂ ≤97%). Intubated ARDS keeps a minimum PEEP of 5 with mild P/F 200–300 (S/F 235–315), moderate 100–200 (S/F 148–235) and severe ≤100 (S/F ≤148). Where blood gases, high-flow oxygen or ventilators are not routinely available, S/F ≤315 alone qualifies, with no PEEP or flow requirement.

The task force's reasons: high-flow oxygen use grew after FLORALI and during COVID-19, and in one COVID cohort 93% of patients intubated after high-flow therapy still met ARDS oxygenation criteria on the ventilator; blood gases are drawn less often; S/F imputes P/F well when SpO₂ ≤97%; and Berlin could not be applied where NIV or ventilators did not exist. They also note poor interrater reliability of chest radiographs, oximeters that miss hypoxemia in darker skin or shock, and under-recognition: in LUNG SAFE (Bellani 2016; 459 ICUs, 50 countries) clinicians recognized 51.3% of mild and 78.5% of severe ARDS, plateau was measured in 40.1%, and 16.3% of severe cases were proned. ARDS was 10.4% of ICU admissions and 23.4% of ventilated patients, with hospital mortality 34.9%, 40.3% and 46.1% by severity. ESICM 2023 kept the Berlin definition.

Phenotypes are the next step. Calfee (2014) applied latent class analysis to 473 ARMA and 549 ALVEOLI patients and found two classes in both: a hyperinflammatory phenotype (33% and 26%) with higher IL-6, IL-8, soluble TNF receptor-1 and PAI-1, more vasopressor use, lower bicarbonate, more sepsis and higher 90-day mortality (44% vs 23%, 51% vs 19%), and a hypoinflammatory one. In ALVEOLI the effect of higher PEEP differed by phenotype (interaction p = 0.049). Famous (2017) found the same split in FACTT (27% hyperinflammatory): the hypoinflammatory group did better with liberal fluids (mortality 18% vs 26% conservative) and the hyperinflammatory group with conservative fluids (40% vs 50% liberal; interaction p = 0.0039), and a three-variable model (IL-8, bicarbonate, soluble TNF receptor-1) classified patients. The Global Definition left phenotypes out until point-of-care assays validate them prospectively.

Mechanics and ventilator-induced lung injury

Ventilator-induced lung injury (VILI) has four mechanisms: volutrauma (overstretch of the baby lung), barotrauma (air leak), atelectrauma (cyclic opening and closing of unstable units) and biotrauma (the cytokine release that follows and injures other organs). Stress is the transpulmonary pressure applied to the lung; strain is the volume change relative to resting lung volume. A normal breath delivered to a small baby lung produces a large strain, which is why ARMA (2000; 861 patients) cut mortality from 39.8% to 31.0% with 6 mL/kg PBW and plateau ≤30 against 12 mL/kg and plateau ≤50, with 12 vs 10 ventilator-free days.

Driving pressure (plateau − PEEP) is tidal volume divided by respiratory-system compliance, so it scales the breath to the lung that is open. In Amato's 2015 analysis of 3,562 patients from nine trials, each standard deviation of driving pressure (about 7 cm H₂O) raised the relative risk of death by 1.41 (95% CI 1.31–1.51), even among patients already on protective Vt and plateau (RR 1.36), and changes in Vt or PEEP improved survival only when they lowered driving pressure (mediation p = 0.004 and 0.001). No trial has yet set ventilation by driving pressure, so 'under about 15' remains an observational target that ATS 2024 and ESICM 2023 do not endorse; ESICM lists it as a research gap.

Mechanical power (Gattinoni 2016) sums the energy per minute delivered by tidal volume, driving pressure, flow, PEEP and rate, computed from the equation of motion; it is a research tool without an agreed threshold. Plateau pressure is a poor stress estimate when the chest wall is stiff (obesity, ascites, burns), where a plateau above 30 may be safe and esophageal manometry (transpulmonary pressure = airway − esophageal pressure) separates lung from chest-wall load.

PEEP titration: methods and what the trials showed

PEEP keeps recruited alveoli open and improves oxygenation, but too much overdistends the open units, raises dead space and loads the right ventricle. The simplest method is the ARDSNet PEEP/FiO₂ tables (lower from ARMA, higher from ALVEOLI; see Ventilator Modes), stepped to SpO₂ 88–95% or PaO₂ 55–80 mm Hg. Alternatives titrate PEEP to the best compliance or lowest driving pressure on a decremental trial, to the highest PEEP that keeps plateau at 28–30 (ExPress), to a positive end-expiratory transpulmonary pressure by esophageal balloon (EPVent), by the recruitment-to-inflation ratio (a single-breath drop from PEEP 15 to 5 that estimates how much volume went to newly recruited lung), or by electrical impedance tomography, which maps regional ventilation and shows collapse against overdistension as PEEP changes. The last two have physiological but no outcome data.

The guidelines read this evidence differently. ATS 2024 suggests higher PEEP without recruitment maneuvers rather than lower PEEP in moderate–severe ARDS (conditional, low to moderate certainty) and recommends against prolonged maneuvers of PEEP ≥35 cm H₂O for more than 60 s (strong, moderate), with no consensus on brief maneuvers or esophageal-pressure titration. ESICM 2023 makes no recommendation for or against a higher PEEP/FiO₂ strategy or mechanics-guided PEEP (high certainty of no effect), recommends against prolonged high-pressure maneuvers (strong) and suggests against routine brief ones (weak). SSC 2026 suggests higher PEEP in moderate–severe ARDS (conditional, moderate) and recommends against incremental PEEP titration (strong, moderate). ART's 6-month mortality was also worse (65.3% vs 59.9%).

Prone positioning: physiology and practice

Supine, the heart and abdomen compress the dorsal lung, where most lung mass and perfusion sit, so dorsal units collapse while ventral units overdistend. Prone, ventilation becomes more homogeneous from front to back, perfusion barely changes, and shunt falls. The mortality benefit comes from more uniform stress and strain (less VILI) rather than from the oxygenation gain, which is why patients whose P/F barely improves still benefit.

PROSEVA (Guérin 2013; 466 patients) enrolled patients with P/F <150 on FiO₂ ≥0.6, PEEP ≥5 and Vt near 6 mL/kg PBW after a stabilization period, and proned them in sessions of at least 16 h. 28-day mortality was 16.0% vs 32.8% (hazard ratio 0.39, 95% CI 0.25–0.63) and 90-day mortality 23.6% vs 41.0% (HR 0.44). Complications did not differ except cardiac arrests, more frequent in the supine group. Earlier trials proned less severe patients for fewer hours and found no benefit.

ESICM 2023 recommends prone over supine positioning for moderate–severe ARDS, defined as P/F <150 on PEEP ≥5 despite optimized settings, started early after intubation once a stabilization period confirms P/F is still <150, in sessions of 16 consecutive hours or more (strong, high certainty), and extends the recommendation to COVID-19 ARDS. ATS 2024 keeps its 2017 wording: prone positioning for more than 12 h per day in severe ARDS (strong, moderate certainty). SSC 2026 suggests more than 12 h daily in moderate–severe ARDS (conditional, moderate). In LUNG SAFE only 16.3% of severe ARDS patients were proned, so under-use, not the evidence, is the gap.

• Before: one person at the airway plus 4–5, preoxygenate on FiO₂ 1.0, suction, tube and lines secured with depth noted, ECG leads to the back, eyes lubricated and closed, padding for face, chest, pelvis and knees
• After: tube depth and EtCO₂ checked, ABG at 1 h, head and swimmer arms alternated every 2 h, no pressure on the eyes, feeds restarted
• Stop when P/F stays ≥150 on PEEP ≤10 and FiO₂ ≤0.6 for 4 h after turning supine (PROSEVA criterion)
• Contraindications vary by protocol: unstable spine or pelvis, open chest or abdomen, raised intracranial pressure, massive hemoptysis
• Arrest while prone: compress between the scapulae until a turn is possible

Neuromuscular blockade and sedation

Paralysis removes dyssynchrony, double-triggering and the large spontaneous efforts that add to transpulmonary pressure (patient self-inflicted lung injury) and lowers oxygen consumption. ACURASYS (Papazian 2010; 340 patients, P/F <150 within 48 h of onset) gave cisatracurium for 48 h: adjusted 90-day hazard ratio for death 0.68 (95% CI 0.48–0.98), crude 90-day mortality 31.6% vs 40.7% (p = 0.08), 28-day mortality 23.7% vs 33.3%, no excess ICU-acquired weakness. ROSE (Moss 2019; 1,006 patients, P/F <150 on PEEP ≥8, enrolled a median of 7.6 h after onset) repeated the regimen with deep sedation against usual care with lighter sedation and stopped for futility: 90-day mortality 42.5% vs 42.8%, with less physical activity and more cardiovascular adverse events when paralyzed.

The difference was the comparator: ACURASYS compared paralysis with deep sedation, ROSE with light sedation. ATS 2024 draws that conclusion: its meta-analysis (7 trials, 1,598 patients; RR 0.74) found the benefit against deep sedation (RR 0.72) and none against light (RR 0.99), so it suggests neuromuscular blockade in early (<48 h) severe (P/F ≤100) ARDS for patients already deeply sedated, or lightly sedated patients with dyssynchrony and deterioration that ventilator and sedation changes do not fix, limited to 48 h, with cisatracurium as used in the two largest trials (conditional, low certainty). ESICM 2023 recommends against routine continuous infusions in moderate–severe ARDS (strong, moderate); SSC 2026 suggests intermittent boluses over an infusion (conditional, moderate).

At the bedside, deep sedation must be confirmed before the first dose and maintained throughout, because the patient cannot signal distress. Monitor train-of-four, lubricate and close the eyes, keep the head of bed up and protect skin and heels. Once the blocker stops, return to light sedation (PADIS 2018: RASS −2 to +1, analgesia first; see Sedation Scales) and resume daily awakening and breathing trials.

Fluids, steroids and the pharmacologic trials

FACTT (2006; 1,000 patients) compared conservative and liberal fluid protocols for 7 days once shock had resolved. Cumulative balance was −136 vs +6,992 mL, 60-day mortality 25.5% vs 28.4% (not significant), ventilator-free days 14.6 vs 12.1 and ICU-free days 13.4 vs 11.2, without more shock or nonpulmonary organ failure. Conservative fluids after resuscitation are therefore standard, with the phenotype caveat from Famous above.

Corticosteroids blunt the inflammatory and fibroproliferative response. DEXA-ARDS (Villar 2020; 277 patients, P/F ≤200 on PEEP ≥10 and FiO₂ ≥0.5 at 24 h) gave dexamethasone 20 mg daily for 5 days then 10 mg for 5 days: 4.8 more ventilator-free days and 60-day mortality 21% vs 36%, hyperglycemia 76% vs 70%, no excess infection. In COVID-19, RECOVERY (2020; 2,104 vs 4,321) gave dexamethasone 6 mg daily for up to 10 days: 28-day mortality 22.9% vs 25.7% overall, 29.3% vs 41.4% in ventilated patients (rate ratio 0.64), 23.3% vs 26.2% on oxygen alone, and no benefit (17.8% vs 14.0%) without oxygen. LaSRS (Steinberg 2006; 180 patients, ARDS ≥7 days) found methylprednisolone did not change 60-day mortality (29.2% vs 28.6%) and raised it when started 14 days or more after onset, with more neuromuscular weakness.

ATS 2024 suggests corticosteroids for ARDS (conditional, moderate certainty; 19 trials, mortality RR 0.84) but names no agent or course, notes that starting more than 2 weeks after onset may harm and that many trials stopped the drug at extubation, and asks for surveillance for hyperglycemia (RR 1.11) and infection. ESICM 2023 did not address steroids; SSC 2026 suggests them for septic shock without an ARDS statement.

The failures are an exam theme. Simvastatin 80 mg (HARP-2 2014; 540 patients: 28-day mortality 22.0% vs 26.8%) and rosuvastatin (SAILS 2014; 745 patients: 28.5% vs 24.9%, with fewer hepatic- and renal-failure-free days) did not help. Aspirin did not prevent ARDS in at-risk emergency patients (LIPS-A 2016; 390 patients: 10.3% vs 8.7%). Inhaled nitric oxide raises P/F at 24 h but not survival (Cochrane 2016: mortality RR 1.04) and increases renal failure (RR 1.59), so inhaled NO and epoprostenol are rescue for oxygenation or RV failure. High-frequency oscillation raised mortality in OSCILLATE (2013; 548 patients: 47% vs 35%), and ATS 2024 keeps its strong recommendation against routine use. ARDS still has no drug.

ECMO and extracorporeal CO₂ removal

VV-ECMO oxygenates venous blood and removes CO₂ across a membrane so the ventilator can rest the lung (EOLIA: plateau ≤24, PEEP ≥10, rate 10–30, FiO₂ 0.3–0.5; see ECMO). CESAR (Peek 2009; 180 patients, Murray score >3.0 or pH <7.20) randomized transfer to one ECMO center against conventional care: survival without severe disability at 6 months 63% vs 47% (death or severe disability 37% vs 53%), though only 75% of the ECMO arm received ECMO and the control arm had no ventilation protocol. EOLIA (Combes 2018; 249 patients) enrolled P/F <50 for more than 3 h, P/F <80 for more than 6 h, or pH <7.25 with PaCO₂ ≥60 for more than 6 h despite optimized ventilation: 60-day mortality 35% vs 46% (RR 0.76, 95% CI 0.55–1.04, p = 0.09), with 28% of controls crossing over as rescue, of whom 57% died; ECMO caused more transfusion-requiring bleeding (46% vs 28%) but fewer ischemic strokes (0% vs 5%).

ATS 2024 suggests VV-ECMO in selected severe ARDS (conditional, low certainty): reversible disease, P/F <80 or pH <7.25 with PaCO₂ ≥60 despite optimal conventional management, less than 7 days of ARDS and few futility risk factors, after the less invasive therapies, in high-volume centers (pooled mortality RR 0.76). ESICM 2023 recommends ECMO for patients meeting EOLIA criteria in a center with defined organizational standards (strong, moderate; pooled RR 0.72). SSC 2026 suggests VV-ECMO when conventional ventilation fails, in experienced centers (conditional, low). EOLIA's crossovers teach the timing: call early, because transfer takes hours.

Extracorporeal CO₂ removal uses lower blood flows to clear CO₂ so tidal volume can fall below 6 mL/kg. REST (McNamee 2021; 412 of a planned 1,120 patients) stopped for futility: 90-day mortality 41.5% vs 39.5%, ventilator-free days 7.1 vs 9.2, serious adverse events 31% vs 9%, intracranial hemorrhage 9 vs 0. ESICM 2023 recommends against ECCO₂R outside trials (strong).

Outcomes and recovery

Mortality tracks severity (Berlin and LUNG SAFE figures above). Most deaths come from multiorgan failure and the underlying illness rather than refractory hypoxemia. Herridge (2011) followed 109 survivors for 5 years: the 6-minute walk distance was 436 m, 76% of predicted, and the SF-36 physical component score 41 against a matched norm of 50; younger patients recovered faster but neither group returned to predicted physical function, pulmonary function was normal to near-normal, and health-care use and costs stayed high. Physical and psychological problems persisted in patients and family caregivers.

Post-intensive care syndrome (PICS) names the new or worsened physical (ICU-acquired weakness, dyspnea), cognitive (memory, attention) and mental health (depression, anxiety, PTSD) problems that follow critical illness; PICS-family is the relatives' distress. Deep sedation, immobility, delirium and steroids plus paralysis are the modifiable risks; awakening and breathing trials, early mobility, family presence and ICU diaries are the prevention (see Sedation Scales). Follow-up clinics and rehabilitation belong in the discharge plan.

Exam pearls

• P/F ≤300 on PEEP ≥5 within a week, bilateral opacities, not explained by heart failure: ARDS. Mild 201–300, moderate 101–200, severe ≤100
• Vt 6 mL/kg PBW from height and sex; plateau ≤30; driving pressure under about 15; SpO₂ 88–95%; pH 7.30–7.45, rate to 35 first
• Shunt: PaO₂ does not rise with FiO₂; PEEP and proning recruit. Dead space: PaCO₂ rises; rate is the lever
• P/F <150: prone ≥16 h (PROSEVA 16.0% vs 32.8%). Early severe, deeply sedated or fighting: cisatracurium ≤48 h (ACURASYS vs ROSE)
• No prolonged recruitment maneuvers (ART), no HFOV (OSCILLATE), no ECCO₂R outside trials (REST); conservative fluids after shock (FACTT)
• Dexamethasone within 14 days (DEXA-ARDS 21% vs 36%); after 14 days harms (LaSRS)
• ECMO call: P/F <80 for 6 h or <50 for 3 h, or pH <7.25 with PaCO₂ ≥60, after proning and paralysis (EOLIA 35% vs 46%)