Background

The acute respiratory distress syndrome (ARDS), first described in 1967,[1]is characterized by diffuse inflammation of the alveolar-capillary membrane in response to various pulmonary and extrapulmonary insults.[2]These insults cause pulmonary injury by direct (e.g. gastric aspiration, pneumonia, inhalational injury, pulmonary contusion) or indirect (e.g. sepsis, trauma, pancreatitis, multiple transfusions of blood products) mechanisms. An American-European Consensus Conference[3]formulated a widely cited definition of ARDS as follows: the acute onset of (i) hypoxemia, with a ratio of the partial pressure of arterial oxygen (PaO2) to the inspired fraction of oxygen (FiO2) of 200mm Hg or less; (ii) bilateral infiltrates on a frontal chest radiograph; and (iii) no clinical evidence of left atrial hypertension or a pulmonary artery occlusion pressure of 18mm Hg or less. Acute lung injury (ALI) includes a milder form of lung injury, with a PaO2/FiO2 ratio of ≤300mm Hg. Almost all patients with ARDS and most with ALI require mechanical ventilation to survive. The mortality of ARDS is high, estimated to be between 34% and 60%.[2]In addition, survivors have a prolonged stay in the intensive care unit (ICU) and demonstrate significant functional limitations, primarily fatigue and muscle weakness, that reduce quality of life and persist for at least 1 year after hospital discharge.[4]

Research on therapy for ARDS has focused on a variety of mechanical ventilation strategies and pharmacologic treatments. Animal and clinical studies, including randomized controlled trials, have demonstrated the role of mechanical ventilation in perpetuating lung injury,[5]leading to macroscopic damage (for example, pneumothorax), diffuse ventilator-induced lung injury, and systemic harm in the form of multiple organ system failure.[6,7]Several randomized trials have compared traditional mechanical ventilation to lung protective ventilation strategies (pressure and volume limited) strategies.[8–12]The largest and most recent study,[8]conducted by the ARDS Network, was stopped after a planned interim analysis showed a clinically and statistically significant reduction in hospital mortality from 39.8% to 31.0% (p = 0.007) using low tidal volume ventilation compared with traditional care. The results of this trial have convinced many clinicians and investigators that lung protective ventilation using tidal volume limitation is the most important therapeutic intervention available for patients with ARDS and ALI. More recently, a meta-analysis[13]of five randomized controlled trials (RCTs) found that lung protective ventilation decreased 28-day mortality in patients with ARDS and ALI.

The pathogenesis of ARDS (extensively reviewed elsewhere)[2,14]provides multiple potential targets for pharmacologic interventions. Regardless of the inciting cause of lung injury, the alveolar-capillary membrane is damaged with leakage of protein-rich edema fluid into alveoli. Alveolar epithelial damage involves the basement membrane and types I and II cells. Injury to type II alveolar epithelial cells leads to reduced levels and functionality of surfactant, causing increased surface tension, atelectasis, and decreased lung compliance. Endothelial damage is associated with numerous inflammatory events. These include neutrophil recruitment, sequestration and activation; formation of oxygen radicals; activation of the coagulation system, leading to microvascular thrombosis with platelet-fibrin thrombi; and recruitment of mesenchymal cells with the production of procollagen. Within the alveolar space, the balance between pro-inflammatory (e.g. tumor necrosis factor [TNF]-α and interleukins [IL]-1, -6, and -8) and anti-inflammatory mediators (e.g. IL-1 receptor antagonist and soluble TNF receptor) favors ongoing inflammation. In summary, the initial lung injury is followed by repair, remodeling, and fibrosing alveolitis.

The diversity of approaches to pharmacologic therapy for ARDS and ALI reflects the complex pathophysiology. Therapies evaluated in randomized trials in humans include corticosteroids, other anti-inflammatory agents, immunomodulating agents, pulmonary vasodilators, antioxidants, and surfactants. The evidence for inhaled nitric oxide (NO), a selective pulmonary vasodilator, was appraised in a recent systematic review,[15]which included five (RCTs) of inhaled NO for acute hypoxemic respiratory failure (including ARDS, ALI, and other diagnoses) in adults and children. The authors concluded that inhaled NO improved oxygenation for up to 72 hours without evidence of effect on the duration of mechanical ventilation or mortality (relative risk [RR] 0.98; 95% confidence interval [CI] 0.66, 1.44). A multicenter RCT published after the systematic review confirmed these clinical findings.[16]

The objective of our systematic review was to evaluate the effects of pharmacologic treatments on important clinical outcomes in patients with established ARDS and ALI. A version of this review is published and will be updated in the Cochrane Database of Systematic Reviews.[17]

Methods

Study Eligibility

We selected studies using the following inclusion criteria: RCT design; adult patients admitted to an ICU with established ARDS or ALI; any pharmacologic treatment compared with no therapy or placebo; measurement of early all-cause mortality (at or before 3 months after randomization), late all-cause mortality (beyond 3 months after randomization), duration of ventilation, number of ventilator-free days to day 28,[18]incidence of non-pulmonary organ dysfunction, or adverse events. The primary outcome was early mortality. We accepted the definitions of adult, ARDS, and ALI used by the primary study authors. Unless otherwise noted, we defined adverse events as those leading to discontinuation of the study medication. In studies without a placebo control, they were defined as ‘serious adverse events’ using authors’ definitions.

We excluded any pharmacologic therapy administered as prophylaxis for ARDS or ALI and studies using nitric oxide, partial liquid ventilation (because this intervention uses a pharmacologic therapy as part of a strategy of mechanical ventilation), fluid management and nutritional therapies. We also excluded subgroups of patients with ARDS or ALI reported in RCTs of interventions for other populations, because of the methodologic limitations of subgroup analyses.[19]

Search Strategy

With the assistance of a professional librarian, we searched OVID versions of CENTRAL (The Cochrane Library Issue 3, 2003), MEDLINE (1966 - week 2, January 2004), EMBASE (1980 - week 4, 2004), CINAHL (1982 - week 2, January 2004), and HEALTHSTAR (1995 - December 2003). We used the MeSH term ‘adult respiratory distress syndrome’ and the text words ‘acute lung injury, shock lung, ARDS’, and (‘acute’ or ‘adult’) ‘respiratory distress’, without applying language restrictions. We used a highly sensitive strategy to retrieve RCTs from MEDLINE[20]and modified it for other databases. Full details of the search strategies are available from the authors. We searched conference proceedings (1994-2003) published in American Journal of Respiratory and Critical Care Medicine, Chest, Critical Care Medicine and Intensive Care Medicine and screened the bibliographies of all retrieved studies and recent review papers[21–38]for additional references.

Data Extraction

Two of the authors independently screened articles and abstracts for inclusion and retrieved all potentially relevant studies. For each study we independently extracted data on population, intervention, outcomes and study methods. We considered the following methodologic features: (i) concealment of allocation; (ii) baseline similarity of treatment and control groups (with respect to age, severity of illness, non-pulmonary organ failures, presence of sepsis, and duration of hospitalization, ICU stay, or mechanical ventilation); (iii) level of blinding (caregivers, data collectors, outcomes assessors); (iv) placebo characteristics that potentially enhanced blinding; (v) cointerventions (standardization or documentation of positive end-expiratory pressure, a lung protective ventilation strategy, weaning protocols, corticosteroids, or other cointerventions); (vi) intention-to-treat analysis (patients analyzed by assigned group and not withdrawn post randomization); and (vii) completeness of follow-up for the outcome of early mortality. We contacted trial authors when we required clarification of the primary outcomes data; four authors[39–42]provided additional information. Disagreements between reviewers were resolved by consensus and in consultation with the third author of this manuscript. We did not systematically evaluate the mechanistic quality of the pharmacologic therapies studied (e.g. whether adequate amounts of the therapeutic agent were delivered to the presumed target for sufficient time to achieve the desired biologic response).

Data Analysis

For each pharmacologic treatment, we pooled the results of studies where permitted by the available data. For studies reporting more than one treatment arm differing in medication dose, we combined data from all doses to determine an overall outcome measure for the treatment arm. We considered ICU and hospital mortality to be early mortality. We used Review Manager software, version 4.2 (Cochrane Collaboration, Oxford, UK) to aggregate data using a random-effects model, calculating an overall RR and a 95% CI for pooled categorical data. We tested the significance of the overall treatment effect using the method of DerSimonian and Laird.[43]For each pooled comparison, we used the Q statistic to test for homogeneity.[44]We considered p = 0.05 to be statistically significant for the test of overall treatment effect and p = 0.10 to be statistically significant for the test of homogeneity. No subgroup analyses or exploratory analyses to explain heterogeneity were planned. We anticipated variability in the reporting of adverse events and therefore did not pool these events across studies.

When pooled analyses included at least five studies, we constructed funnel plots of study precision versus treatment effect, in order to explore the presence of bias arising from the non-publication of small non-beneficial trials.

Results

Overall Description of Included Studies

Searches of electronic databases yielded 5705 citations. From these citations, conference abstracts, and bibliographies of retrieved studies, we identified 75 potentially relevant publications. We excluded 42 articles and abstracts for the following reasons: non-randomized;[45–51]cost-effectiveness data from a previously published RCT;[52]no placebo or standard care control group;[53]alternative diagnosis in the study population (at risk for rather than established ARDS or ALI,[54–65]sepsis,[66–70]ventilated patients in surgical ICU,[71]malaria,[72]pancreatitis,[73,74]unilateral lung injury after thoracic aneurysm surgery[75]); outcomes of interest not reported;[76,77]subgroup of patients with ARDS or ALI reported in RCT in another population;[78–83]and duplicate publication.[84–86]

Thirty-three publications describing 33 studies met our inclusion criteria. One paper[39]described two similar but separate studies of surfactant therapy. We counted this paper as two studies. One study of alprostadil (prostaglandin E1) reported mortality data in 147 randomized patients,[87]of whom clinical outcomes in 101 patients had been previously published.[88]We counted these two publications as one study and extracted data on early mortality from the later publication[87]and data on late mortality and adverse events from the earlier publication.[88]One publication[89]reported a trial of patients randomized to receive acetylcysteine, oxothiazolidine carboxylate (procysteine), or placebo. We included the patients randomized to acetylcysteine and placebo in a pooled analysis of the effect of acetylcysteine on early mortality and described the procysteine arm separately. Another study randomized patients to acetylcysteine, acetylcysteine plus rutin (a flavanoid antioxidant), or control.[90]We combined data from both acetycysteine groups in our pooled analysis and did not analyze the effects of rutin, which we classified as a nutritional supplement.

Studies meeting our inclusion criteria randomized 3272 patients and evaluated a wide range of pharmacologic treatments with differing mechanisms of action (table I). They include seven studies of alprostadil (table II),[41,87,88,91–95]five studies of acetylcysteine (table III),[89,90,96–98]three studies of corticosteroids (table IV),[99–101]nine studies of surfactant (table V),[39,40,42,102–106]and single studies (table VI) of dazoxiben,[107]acyclovir,[108]indomethacin,[109]pentoxifylline,[110]neutrophil elastase inhibitor (ICI 200,880),[111]oxothiazolidine carboxylate,[89]interleukin-10,[112]ketoconazole,[113]lisofylline,[114]and granulocyte macrophage colony-stimulating factor (GM-CSF).[115]

Table I
figure Tab1

Proposed mechanisms of action of pharmacotherapies for acute respiratory distress syndrome and acute lung injury evaluated in randomized trials

Full size image
Table II
figure Tab2

A summary of randomized controlled trials of alprostadil for acute lung injury/acute respiratory distress syndrome (ARDS)

Full size image
Table III
figure Tab3

A summary of randomized controlled trials of acetylcysteine and oxothiazolidine carboxylate for acute lung injury (ALI)/acute respiratory distress syndrome (ARDS)

Full size image
Table IV
figure Tab4

A summary of randomized controlled trials of corticosteroids for acute lung injury /acute respiratory distress syndrome (ARDS)

Full size image
Table V
figure Tab5

A summary of randomized controlled trials of surfactant for acute lung injury (ALI)/acute respiratory distress syndrome (ARDS)

Full size image
Table VI
figure Tab6

A summary of randomized controlled trials of additional therapies for acute lung injury (ALI)/acute respiratory distress syndrome (ARDS)

Full size image

Studies of the same pharmacologic therapy differed with respect to patient populations and drug administration as discussed in the following sections.

Studies of Aprostadil

All seven studies of alprostadil [41,87,88,91–95]included patients with predominantly ARDS, although three studies restricted enrollment to those with trauma, surgery, or sepsis as a risk factor.[87,88,92,93]They varied with respect to method of medication administration (continuous infusion[87,88,91–93]versus intermittent boluses[41,94,95]), formulation (liposomal in three trials[41,94,95]) and dose (ranging from 7.2 μg/kg/day[41]to 43.2 μg/kg/day[87,88,91–93]for 7 days).

Studies of Acetylcysteine

The five studies of acetylcysteine [89,90,96–98]included four enrolling primarily patients with ARDS[89,90,96,98]and one that restricted enrollment to those with mild ALI.[97]Two trials used continuous infusions[96,97]and the other three used intermittent intravenous boluses.[89,90,98]The total dose of acetyleysteine varied considerably, from 120 mg/kg[97]to 3510 mg/kg,[96]delivered over 3–10 days. The total dose delivered in one trial[90]was unclear.

Studies of Corticosteroids

The three corticosteroid studies included two trials of high dose, short course (≤48 hours) methylprednisolone early in the course of ALI[99]and ARDS.[100]The duration of ARDS prior to randomization was less than 7 days in both trials. One trial studied a lower dose of methylprednisolone for a longer duration (1 month) in patients with non-resolving ARDS who had been mechanically ventilated for at least 7 days.[101]

Studies of Surfactant

The nine studies of surfactant[39,40,42,102–106]enrolled primarily patients with ARDS. Three restricted eligibility to patients with sepsis-induced ALI or ARDS.[40,102,106]The surfactant preparations were variable: synthetic surfactant with phospholipids,[40,102,106]synthetic surfactant with phospholipids and protein,[39,103,104]bovine surfactant,[42]and porcine surfactant.[105]The method of delivery (continuous aerosolization in early trials[40,102,106]and intratracheal instillation in later trials[39,42,103–105]) and the delivered doses (when measured) were also variable.

Methodologic Assessment of Studies

The methodologic quality of the included studies was variable (table VII). Ten studies[40,87–89,93,100,103,106,113–115]described adequate concealment of allocation. Only four studies did not report at least one prognostically important baseline characteristic,[102,104,105,107]of which three were reported in abstract form.[102,104,105]Of the remaining 29 studies, seven had at least one clinically important imbalance at baseline.[41,93–95,108,112,115]Six studies[39,89,101,106,113]conducted additional analyses adjusting for baseline imbalances. Twenty-six studies reported blinding of at least caregivers or used the term ‘double-blind’. Four of the remaining studies were not placebo-controlled and did not report blinding of any study personnel.[42,103,105,110]One study used a placebo and was described as ‘single-blind’.[107]Two studies explicitly reported no blinding.[90,104]Potential unblinding features associated with study medications used in trials describing caregiver blinding included hypotension with alprostadil[41,87,88,91–95]and dazoxiben,[107]hyperglycemia with corticosteroids,[99–101]and reflux of surfactant into the ventilator tubing.[39,40,102,106]Fourteen studies standardized at least one cointervention or documented its application in the treatment and control groups.[39–42,91,92,95,96,101,103,113–115]The remaining studies did not describe cointerventions. Two ARDS Network trials investigated tidal volume limitation combined with ketoconazole[113]and lisofylline[114]in a factorial design. Of the six other trials[39,41,103,105,115]published after the low tidal volume ventilation study,[8]two[39]documented encouragement of a lung protective ventilation strategy. No study reported imbalances in the application of cointerventions. All studies analyzed patients according to their assigned treatment group; however, 11 reported at least one post-randomization withdrawal[39,42,87–89,95,98,99,108]or did not clearly report this information.[40,107]Twenty-eight studies reported no losses to follow-up; the remaining studies had incomplete follow-up data on fewer than 1% of randomized patients[87,88,95]or did not fully describe the extent of patient follow-up.[40,102,104,114]

Table VII
figure Tab7

Methodologic quality of studies included in the meta-analysis

Full size image

Effects of Pharmacotherapies on Clinical Outcomes

We conducted pooled analyses of the effect of alprostadil, acetylcysteine, early corticosteroids and surfactant therapy on early mortality (figure 1, figure 2, figure 3, and figure 4). In all analyses the test for homogeneity was not significant. We were unable to perform pooled analyses for the secondary outcomes (late mortality, duration of mechanical ventilation, ventilator-free days, incidence of non-pulmonary organ dysfunction, adverse events) because of insufficient data and variability in definitions and reporting. Individual trials of these treatments demonstrated no consistent effect on secondary outcomes (table VIII). Effects of additional therapies are summarized separately (table IX).

Fig. 1
figure 1

Effect of alprostadil on mortality in acute respiratory distress syndrome and acute lung injury. The test for homogeneity was nonsignificant (p = 0.27). CI = confidence interval; N = number of patients randomized; n = number of deaths; random = random effects model; RR = relative risk. (See table II for relevant bibliographic references.)

Full size image
Fig. 2
figure 2

Effect of acetylcysteine on mortality in acute respiratory distress syndrome and acute lung injury. The test for homogeneity was nonsignificant (p = 0.63). CI = confidence interval; N = number of patients randomized; n = number of deaths; random = random effects model; RR = relative risk. (See table III for relevant bibliographic references.)

Full size image
Fig. 3
figure 3

Effect of early high-dose corticosteroid therapy on mortality in acute respiratory distress syndrome and acute lung injury. The test for homogeneity was nonsignificant (p = 0.16). CI = confidence interval; N = number of patients randomized; n = number of deaths; random = random effects model; RR = relative risk. (See table IV for relevant bibliographic references.)

Full size image
Fig. 4
figure 4

Effect of surfactant on mortality in acute respiratory distress syndrome and acute lung injury. The test for homogeneity was nonsignificant (p = 0.28). Spragg 2004a (North America) and 2004b (Europe/South Africa) are reported in reference 39. CI = confidence interval; N = number of patients randomized; n = number of deaths; random = random effects model; RR = relative risk. (See table V for relevant bibliographic references.)

Full size image
Table VIII
figure Tab8

Effects of alprostadil, early high-dose corticosteroids, and surfactant therapy on other outcome measures in patients with acute lung injury/acute respiratory distress syndrome

Full size image
Table IX
figure Tab9

Effects of additional miscellaneous therapies on outcomes in acute respiratory distress syndrome and acute lung injury

Full size image

Alprostadil

We pooled data from seven trials (figure 1) and found no effect on early mortality (693 patients with outcomes data; RR 0.95; 95% CI, 0.77, 1.17). One study[94]showed a decreased need for ventilation at day 8 but did not use a weaning protocol. Adverse events (figure 5) led to discontinuation of study medication in five trials.[41,91,92,94,95]They occurred more frequently in the alprostadil arm and included cardiopulmonary (hypotension, dysrhythmias, hypoxia) and central nervous system (agitation) events. Data from Bone et al.[88]precluded calculating the total number of patients with adverse events; instead, all patients with hypotension are shown.

Fig. 5
figure 5

Adverse events due to alprostadil, early corticosteroids, and surfactant. Adverse reactions included cardiopulmonary (hypotension, dysrhythmias, hypoxia) and neurological (agitation) events in alprostadil trials; infections in trials of early corticosteroids; and cardiopulmonary events (exhalation valve occlusion, barotrauma, hypoxemia, respiratory arrest, increased peak airway pressure, hypotension, bradycardia) in surfactant trials. CI = confidence interval; N = number of patients randomized; n = number of adverse events; RR = relative risk. (See table II, table IV, and table V for relevant bibliographic references.)

Full size image

Acetylcysteine

Aggregate analysis of five studies (figure 2) showed no effect on early mortality (235 patients with outcomes data; RR 0.89; 95% CI, 0.65, 1.21). One study [97]showed a decreased need for ventilation on day 3 but did not use a weaning protocol. The five studies reported no adverse events in the placebo group and only one (rash) in the acetylcysteine group,[96]which resolved after acetylcysteine was stopped.

Early High-Dose Corticosteroids

The meta-analysis of two studies (figure 3) showed no effect on early mortality (180 patients with outcomes data; RR 1.12; 95% CI, 0.72, 1.74). The major adverse effect (figure 5) was the development of new infection during the first 7 days. The definition of new infection differed. Bernard et al.[100]excluded positive urine and sputum cultures from catheterized and intubated patients whereas Weigelt et al.[99]did not report excluding them. Both studies demonstrated a statistically significant increase in hyperglycemia with corticosteroid therapy (data not able to be pooled). These adverse effects did not lead to discontinuation of study medications.

Surfactant Therapy

Pooling nine studies (figure 4) demonstrated no effect on early mortality (1418 patients with outcomes data; RR 0.93; 95% CI, 0.77, 1.12). One study assessed the incidence of non-pulmonary organ dysfunction. Gregory et al.[42]followed patients for the development of organ failures (not defined) for 5 days after randomization to surfactant therapy or standard therapy and reported greater cardiovascular failure (not quantified) on day 2 in the high-dose arm compared with controls. Adverse events (figure 5) were not reported to have resulted in discontinuation of the intervention and were primarily cardiopulmonary (for example, exhalation valve occlusion, barotrauma, hypoxemia, respiratory arrest, increased peak airway pressure, hypertension, trachycardia). One trial[103]reported no adverse events leading to discontinuation of therapy.

Evaluation of Publication Bias

Funnel plots of study precision versus treatment effect (not shown) for trials of acetylcysteine, alprostadil, and surfactant therapy suggested the presence of bias arising from non-publication of small non-beneficial trials.

Additional Therapies

One study enrolling 24 patients[101]meeting consensus conference criteria for ARDS[3]showed that administration of corticosteroids for late phase non-resolving ARDS reduced mortality in the ICU (RR 0.05; 95% CI, 0.00, 0.78) and hospital (RR 0.20; 95% CI 0.05, 0.81) and reduced duration of ventilation and organ failure-free days, with no increase in infectious complications. One study of 30 patients[110]with metastatic cancer and ARDS (as defined by the authors) found that pentoxifylline reduced one month mortality (RR 0.67; 95% CI 0.47, 0.95) with no adverse events leading to discontinuation of therapy. There was no evidence of effect of any other intervention (table IX) on prespecified outcomes.

Discussion

The principal finding of our systematic review is that no pharmacotherapy convincingly improves survival in patients with ARDS and ALI. We identified 33 randomized trials of pharmacologic treatments of established ARDS and ALI enrolling 3272 patients, with the first trials being published in 1985.[99,107]The median number of patients randomized per trial was low (45, range 9–725). Our pooled analyses had sample sizes of 180–1418 patients (early high-dose cotricosteroids and surfactant therapy respectively). These analyses showed that alprostadil, acetylcysteine, early administration of high-dose corticosteroids, and surfactant therapy, had no effect on early mortality. With respect to adverse effects, all trials of alprostadil and early high-dose corticosteroids and six out of seven surfactant trials with safety data showed more adverse events with active therapy. However, the potential difficulty of achieving effective blinding for alprostadil and surfactant may have biased adverse event reporting in the active treatment arms. Of the miscellaneous interventions evaluated in single small trials, corticosteroids given late in the course of ARDS decreased ICU and hospital mortality,[101]and pentoxifylline decreased mortality in patients with metastatic cancer and ARDS.[110]

Because the confidence intervals around pooled relative risks for alprostadil, acetylcysteine, early high-dose corticosteroids, and surfactant therapy are wide, we cannot exclude the possibility of a modest benefit or harm to patients. However, in each case the overall best estimate of relative risk suggested no effect. Clinical application of these therapies is therefore not currently justified. Any additional large trials should await pre-clinical and preliminary clinical studies to identify variations of dose, timing of administration, and formulation that may lead to clinical benefit. For example, a recently completed trial of HL 10[122]evaluated the only surfactant preparation shown to be potentially beneficial for ALI/ARDS,[105]a porcine surfactant with a high phospholipid concentration delivered by intratracheal installation to maximize distal airway deposition.

We found that the scientific quality[123]of included trials was variable. Methodologic strengths of a clear majority of studies included caregiver blinding or ‘double-blinding’,[124]analysis of early mortality by strict intention-to-treat criteria (analysis according to assigned group and zero withdrawals), and complete patient follow-up. No study documented differential application of potentially important cointerventions, although the majority provided no information about cointerventions. About one-half of the trials documented similarity between treatment groups for at least one prognostically important baseline characteristic. However, only a minority of studies described adequate allocation concealment. Our assessment of methodologic features was restricted to published descriptions, and we may therefore have underestimated the methodologic strength of some studies, especially those reported in abstract form.[102,104,105,111,112]

Strengths of this systematic review include the use of strategies to minimize bias in the selection and reporting of studies: (i) extensive literature search; (ii) duplicate independent screening of articles and data abstraction; and (iii) explicit criteria for methodologic assessment. [125]We used clinical judgment to decide a priori to combine studies for which a similar direction and magnitude of treatment effect could reasonably be expected. However, aggregated studies differed with respect to eligibility criteria, study drug formulations, methods of administration, and doses. We used random effects models to aggregate data and generate conservative confidence limits for the point estimate of the pooled treatment effect.[126]

Another limitation of this systematic review is the identification of a small number of trials of each pharmacologic therapy, most of which randomized a small number of patients. Funnel plots for trials of acetylcysteine, alprostadil and surfactant therapy suggested the existence of small, unpublished, non-beneficial trials. However, the inclusion of any such trials would not qualitatively change our findings or conclusions.[127]We did not search for RCTs of pharmacologic therapies for other related critically ill populations (such as sepsis) reporting outcomes in subgroups of patients with ALI and ARDS because of limitations in the interpretation of subgroup analyses.

Effective pharmacologic therapy for established ARDS and ALI is therefore extremely limited. We identified two potentially beneficial interventions. First, prolonged corticosteroid administration for late ARDS, which may attenuate the fibroproliferative process, was evaluated in one trial. This study[101]randomized 24 patients with late-phase ARDS and showed an impressive but clinically implausible point estimate of benefit with a wide CI (absolute risk reduction in ICU mortality 0.62; 95% CI, 0.30, 0.95; number needed to treat to prevent one ICU death, 1.6). The authors did not describe concealment of allocation; methodologic strengths include blinding of caregivers, intention-to-treat analysis, complete follow-up, and adjustment for baseline imbalances. However, the major methodologic limitation relates to the data analysis. The investigators calculated a priori that 99 patients would be required to detect an absolute survival improvement of 30% in corticosteroid-treated patients. Enrollment was terminated prematurely using early stopping rules based on sequential clinical trials.[128]However, this approach may inflate the treatment effect in small trials[129]by increasing the probability of confounding by imbalances in baseline prognostic characteristics. Patients in the corticosteroid group had less organ dysfunction and pulmonary morbidity and thus may have had a better prognosis, a bias that may have persisted and confounded the results despite statistical adjustment. In addition, the crossover of four of the eight placebo-assigned patients to corticosteroids (as per protocol) precluded an accurate evaluation of the risk of infection associated with this treatment regimen. Therefore, this trial provides important, albeit preliminary, evidence for the efficacy of corticosteroids for late phase ARDS. The recently completed Late Steroid Rescue Study,[130]which randomized 180 patients, should provide more definitive data.

The second potentially beneficial intervention we identified was administration of pentoxifylline, a phosphodiesterase inhibitor that prevents neutrophil chemotaxis and activation. This therapy was evaluated in a randomized trial enrolling 30 patients with metastatic cancer.[110]Although unclear allocation concealment and lack of blinding threaten the internal validity of this trial, the most important limitation is lack of generalizability. Eighteen of 30 randomized patients had lymphangitic metastases that may have mimicked the characteristic radiographic changes of ARDS, resulting in diagnostic misclassification. The mortality in this trial was higher (25 of 30 patients died at 1 month) than generally observed in patients with ARDS,[2,131]which limits the applicability of the findings to lower-risk patients. Finally, the severity of patients’ lung injury as defined by the widely used American-European Consensus Conference definition[3]was not clear because the authors did not describe the oxygenation criteria used for the diagnosis of ARDS, or whether ventilatory support was required. Further research is required to assess the potential role of pentoxifylline in the treatment of ALI and ARDS in patients without cancer.

There are several potential reasons to explain the large number of non-beneficial trials of pharmacologic therapies for ALI and ARDS. These explanations overlap with insights into the design of RCTs in patients with severe sepsis.[132]First, randomized trials may have been conducted before early investigations had established the optimal dose and duration of the candidate therapy that would achieve adequate tissue levels and biologic response. Second, the definition of the clinical syndrome of ALI and ARDS, although feasible to apply in clinical settings and in research, does not have optimal reliability and validity.[131,133,134]Trials may therefore have enrolled patients unlikely to respond to candidate therapies. However, an inadequate definition of ALI and ARDS cannot be the exclusive explanation for non-beneficial trials given the recent positive RCT of low total volumes in ALI patients.[8]Third, ALI and ARDS are clinically heterogeneous syndromes with various causes, genetic susceptibilities[134]and clinical courses. Trials may thus have included responsive and non-responsive subgroups. Fourth, there are many important non-pulmonary determinants of the outcome for patients with ALI and ARDS. Fifth, investigators may have overestimated the instrinsic therapeutic potential of candidate therapies. The latter three points may have contributed to the design of RCTs that were inadequately powered to reliably detect moderate clinical benefits.

Conclusion

In summary, we found that alprostadil, acetylcysteine, early administration of high-dose corticosteroids, and surfactant therapy had no beneficial effect on survival in critically ill patients with ARDS or ALI. Prolonged corticosteroid administration was effective for late phase ARDS in one small study, and a multicenter RCT[130]of this therapy was recently completed. Pentoxifylline was effective in a small study of highly selected patients with metastatic cancer and ARDS. Given the complex pathophysiology of lung injury, the lack of effective pharmacologic therapies is not surprising. Future treatment may involve a combination of lung-protective ventilation, multi-modality pharmacologic therapy, and earlier identification and better treatment of precipitating factors.