DISCUSSION

Section:

ChooseTop of pageAbstractMETHODSRESULTSDISCUSSION <<ReferencesCITING ARTICLES

The main finding of our systematic review is that mortality due to ARDS has remained static at 44.0% for observational studies and 36.2% for RCTs since a standard definition was introduced in 1994, contrary to commonly held perceptions (1–3). Although mortality decreased over time in observational studies conducted from 1984 to 1993, it was consistently higher in observational studies than in RCTs. Of the variables studied, patient age was the only other factor explaining between-study variability in mortality, a finding consistent with previous multicenter epidemiologic studies (3, 26).

Based on the relatively low mortality observed in recent multicenter RCTs, it has been suggested that the current benchmark for mortality of ALI/ARDS in clinical practice and future clinical trials should be 25 to 30% (14). However, we found an overall mortality of 36.2% in RCTs conducted between 1994 and 2006, higher than this benchmark, and an even higher mortality in observational studies. Characteristics of RCTs that tend to reduce overall mortality compared with observational studies include performance in specialized centers, possibly with ventilation protocols shown to improve outcomes (7), and exclusion of patients with poor prognosis (15). For example, various ARDS Network trials had previously excluded 88% of patients meeting the inclusion criteria (7–9), and data from screening logs have shown a 10% absolute lower mortality in included compared with excluded patients (16). This is comparable to the mortality difference between observational studies and RCTs observed in our study.

Five observational studies of ARDS have shown a decrease in mortality over time (17–21). However, these studies were conducted from the 1980s to the mid-1990s in single centers and included many trauma patients, a subset of ALI/ARDS with a better prognosis (17–20). The extent to which their findings apply to patients with ALI/ARDS in general is therefore unclear. Our review excluded two of these studies: one included only trauma patients (19), and the other was retrospective (20). An analysis of death records in the United States also suggested that the number of deaths from ARDS, while increasing between 1979 and 1993, may have decreased slightly between 1993 and 1996 (27). However, this study used administrative datasets (using ICD-9 codes) to identify ARDS and only included nonsurvivors, so it is unclear whether this reduction is due to a change in mortality or in the number of ARDS cases identified. Indeed, recent data from multicenter observational studies suggest that ARDS mortality in unselected patients may range from 39% (28) to as high as 50 to 60% (26, 29, 30).

Other investigators have reviewed mortality trends in ARDS across centers over time. An older meta-analysis by Krafft and colleagues showed no relation between publication year and mortality for studies published between 1967 and 1994 (31). However, this meta-analysis is not directly comparable with ours, given the different time periods (many studies published before 1984) and criteria for included studies (many retrospective studies with very small sample sizes).

More importantly, our findings seem to conflict with a recent systematic review that suggested a decrease in ARDS mortality since 1994 (22). When two systematic reviews on the same topic arrive at opposite conclusions, it is important to evaluate their methodologic differences and compare their study selection processes (32). In contrast with this other review, our search strategy required reviewers to work in duplicate, used more search terms and multiple databases, allowed only prospective studies, excluded small studies with fewer than 50 patients with ALI/ARDS, and was not restricted to the English language. In addition, the other review considered only the last cohort in studies with multiple historical cohorts and only the control group in RCTs. Because that review used the last (instead of the median) year of enrollment, several studies conducted primarily in the 1980s and early 1990s were labeled as “post-1994 studies” in their analysis (33–38). In addition, the authors included several reports of large studies that appear to contain overlapping patient populations (7, 8, 29, 39–41), some of which had a low mortality (7, 8, 40, 41). The combination of coding some high-mortality studies that enrolled patients before 1994 as “post-1994 studies” and the duplicate counting of some patients from recent RCTs with low mortality may thus have contributed to the downward trend observed by these investigators. Finally, although both systematic reviews included 42 studies in common, ours included an additional 27 post-1994 studies.

Mortality trends over time offer important insights to the state of medical care for a particular disease. Similar to other investigators who have found stable mortality risks for diseases such as stroke, schizophrenia, chronic obstructive pulmonary disease, and diabetes (42–44), our findings raise the question of why mortality for patients with ARDS has not improved. Several plausible explanations exist. The number of effective therapeutic interventions for ARDS continues to pale in comparison to the number of failed interventions (1, 2). Even positive results from RCTs generated in selected populations may be significantly diluted and have limited generalizability when applied to unselected ICU populations. Furthermore, the extent to which effective therapies from RCTs are adapted into clinical practice is variable. For example, the use of tidal volume limitation—arguably the only mortality-reducing intervention specifically for ARDS— is in evolution but is not universally practiced (30, 45). Because ALI/ARDS has been defined as a syndrome that includes patients with multiple pathophysiologic mechanisms, therapeutic progress may not be related to specific disease-related interventions but rather to the optimization of general supportive care. In this way the consensus definition of ALI/ARDS may itself have contributed to the plateau in mortality seen after 1994 because it captures an extremely heterogeneous patient population that may be unlikely to respond to a given intervention. Meanwhile, examples of supportive measures that could have improved outcome (in addition to avoidance of ventilator-induced lung injury) include early aggressive resuscitation and appropriate antibiotics in sepsis and sedation protocols (46–48). Although guidelines exist to promote these practices (49), variable adherence may contribute to the lack of improvement in outcomes (50). It has also been suggested that standardized disease definitions and improved coding practices may produce samples with lower disease severity and hence lower mortality (51). It is tempting to speculate that the decreases in mortality before 1994 reflect differences and changes in the definitions of ARDS before the current consensus definition (4). However, our study is unable to confirm this hypothesis because the definition of ARDS was not an independent predictor of mortality.

Strengths of our study include a comprehensive search of the international literature, duplicate data extraction, and predefined criteria for methodologic assessment and analysis. This systematic review also has limitations. First, the outcome of ICU mortality, although feasible and reliably reported, is dependent on the ICU discharge decision, which may vary across institutions and time. Twenty-four included studies reported ICU mortality, and we therefore substituted other mortality rates (28/30 days, hospital or other mortality as available). However, it is reassuring to note that the need for these substitutions did not show any association with the year of study conduct, and a sensitivity analysis using hospital mortality as the preferred primary outcome did not change our findings. Second, differences in case-mix and/or the availability of ICU beds in the various countries during the study period may have influenced the types of patients admitted to the ICU and their outcomes over time. We made every possible effort to adjust for such baseline differences but were limited by the data provided in the individual studies. Multiple imputation of data was required to generate values for age and illness severity score for studies that did not provide such information (in 10 and 23 studies, respectively). However, because data on baseline Pa_O2/Fi_O2 ratios and delivered tidal volumes were not available from 25 and 50 studies, respectively, we could not include these variables in the meta-regression analysis and are thus unable to comment on the impact of these care practices on mortality. Third, we used the median year of conduct for each study as the independent variable in analyses of the effect of year on mortality. This was necessary because many studies were conducted over several years. Fourth, unpublished data and studies that were reported only in abstract form were not included, thus potentially leading to publication bias, particularly for RCTs. However, the effect of any small unpublished RCTs showing high overall mortality would be unlikely to change our main findings. Fifth, although we evaluated only ARDS whenever possible, 14 studies grouped data for all patients with ALI (including both ARDS and mild ALI). Inclusion of such studies with a proportion of patients with mild ALI and less severe oxygenation impairment may have generated some bias toward better prognosis after 1994. Despite this potential bias, our results indicate that mortality has remained relatively constant since 1994. Sixth, given the heterogeneous nature of ALI/ARDS and our exclusion of studies that enrolled patients with specific risk factors for ALI/ARDS, our findings may not be applicable to specific subgroups, such as trauma populations (17–20). Correspondingly, although the pooled mortality estimates reported in our study reflect international trends in ARDS outcomes, they must not be used for prognostication for the individual patient.

In conclusion, our systematic review has shown that mortality due to ARDS has remained relatively unchanged since 1994, coincident with the publication of the current syndrome definition. Higher mortality was associated with observational study design and increased patient age. Based on our findings, clinicians and investigators should expect a baseline mortality risk from ARDS of 40 to 45% for observational studies and 35 to 40% for RCTs. Most importantly, our results highlight the need for future effective therapeutic interventions for this highly lethal syndrome. They also compel us to revisit the manner in which studies of these interventions are conducted and their results implemented to ensure their applicability in the “real world.”