© 2002 American Thoracic Society
Meta-analysis of acute lung injury and acute respiratory distress syndrome trialsTo the Editor:Members of the National Institutes of Health Acute Respiratory Distress Syndrome (ARDS) Network carefully read "Meta-analysis of acute lung injury and acute respiratory distress syndrome trials testing low tidal volumes" by Eichacker and colleagues published in this issue of AJRCCM (1). We believe their analysis is flawed and their conclusions are incorrect. Was there a "standard care" for mechanical ventilation in acute lung injury (ALI) and ARDS? Eichacker and colleagues contend that there was a standard for mechanical ventilation in ALI and ARDS patients in the 1990s. In the context of their argument, this standard must have consisted of clearly understood principles for adjusting tidal volumes and inspiratory pressures to prevent lung injury from over distention. However, Eichacker and colleagues provide no evidence that such a standard existed or that physicians adhered to it. There were two general approaches to setting tidal volumes and inspiratory pressures that assigned different priorities to competing clinical objectives. One approach used generous tidal volumes with relatively high airway pressures (2, 3). This was useful for maintaining gas exchange and breathing comfort, especially in patients with elevated dead space and intrapulmonary shunt (4, 5). However, some clinicians knew from studies in experimental models that this approach might cause lung injury from overdistention (6, 7). An alternative approach used lower tidal volumes and inspiratory pressures to prevent overdistention, but this approach frequently compromised gas exchange and breathing comfort (8, 9). In the absence of good evidence for the superiority of either approach, physicians' interpretations of the preclinical data and their resulting clinical practices were highly variable. In a survey conducted in 1992, physicians reported using initial tidal volumes in ARDS patients of 5–17 ml/kg measured body weight (10). In the ARDS Network trial of tidal volume reduction (11), 95% of the tidal volumes prescribed by physicians before patients were enrolled were between 6 and 14 ml/kg predicted body weight (PBW). More importantly, there was no apparent plateau pressure threshold that triggered tidal volume reduction (Figure 1). Thus, the broad ranges of tidal volumes indicate the disparity in physicians' estimations of the risks and benefits of the two approaches to setting tidal volumes and inspiratory pressures. The mean values for tidal volumes and plateau pressures simply represent the means of these ranges, not "standard care."
Referring to the survey of physicians' practices (10), Eichacker and colleagues stated: "... nearly half of the ... physicians surveyed reported using tidal volumes ... that were similar to the ones patients received prerandomization in the ARDSNet trial..." This statement supported their contention that "standard care" tidal volumes were lower than those received in our traditional study group. They failed to mention that over half of the respondents in the survey reported using tidal volumes that were as high or higher than those received in our traditional group. Eichacker and colleagues also stated: "Importantly, 96% of all respondents in this survey said that the level of airway pressure would influence their choice of tidal volume, suggesting that most clinicians ... were ... decreasing tidal volumes if airway pressures were high." The specific pressure limits used for this purpose were not included in the survey. Referring to the mean prerandomization plateau pressures in the two beneficial studies, Eichacker and colleagues stated: "Thus ... clinical practice was to ventilate patients with tidal volumes that produced plateau airway pressures averaging 29 to 31 cm H2O." However, these were simply the mean values of broad ranges of the prerandomization plateau pressures, and plateau pressures were not used to adjust tidal volumes in any systematic fashion (Figure 1). It is notable also that four of the five trials of tidal volume reduction reviewed by Eichacker and colleagues included pressure limits in the higher tidal volume study group protocols: peak inspiratory pressures of 60 (12) and 50 cm H2O (13) and plateau pressures of 45–55 (15) and 50 cm H2O (11). Thus, four independent groups of investigators selected very similar inspiratory pressure limits for their higher tidal volume protocols. These limits were carefully selected to represent a mainstream clinical approach in which the clinical objectives of maintaining gas exchange and breathing comfort have higher priority than the objective of preventing overdistention. The disparity in physician-selected tidal volumes and the absence of a clear plateau pressure limit (Figure 1) indicate that the intensive care community as a whole was in equipoise regarding the approach to mechanical ventilation that would yield better clinical outcomes. The introductions to the reports of all five trials of tidal volume reduction (11–15) express this opinion. To identify the prioritization scheme that would yield better outcomes, each of the investigator groups designed protocols that were consistent with commonly held opinions and practices. How high were the tidal volumes in the ARDS Network traditional group? There has been considerable confusion about tidal volumes because four different units were used to report tidal volumes in the five trials. The ARDS Network trial and one of the nonbeneficial trials (15) set tidal volumes according to predicted body weight. The other three studies set tidal volumes according to ideal body weight (13), dry body weight (12), and measured body weight (14). In the ARDS Network study, predicted body weight was approximately 20% lower than measured body weight. The mean tidal volume in our traditional group after randomization was 10 ml/kg measured body weight. This was close to the middle of the range of tidal volumes reported in the survey of physicians' ventilator practices (10). In a sepsis trial in which most patients had acute lung injury (16), the mean physician-prescribed tidal volume on the first study day was 10 ml/kg measured body weight. In a large study of surfactant in ARDS (17), the mean physician-prescribed tidal volume on the first study day was 11 ml/kg measured body weight. In large surveys of mechanical ventilation practices in ALI patients conducted in the mid–late 1990s (18, 19), the interquartile range (25th–75th percentile) for physician-prescribed tidal volumes included 10 ml/kg measured body weight. Thus, our traditional group tidal volumes were consistent with prevailing practices in the 1990s. What are the effects of lower tidal volume ventilation on mortality? Eichacker and colleagues contend that ventilation with lower tidal volumes is detrimental. This opinion is based on the observation that there were trends towards higher mortality in the lower tidal volume groups in the three nonbeneficial studies (12, 13, 15). However, these studies were too small (underpowered) to provide convincing evidence for lack of efficacy. Even if these three studies are combined (total of 288 patients), the difference in mortality between higher and lower tidal volume groups is not statistically significant. Moreover, tidal volumes in the control group of one of the nonbeneficial trials (13) were similar or slightly higher than those used in the ARDS Network traditional study group. For example, the tidal volume for a female with a height of 1.65 meters would have been 729 ml in the control group of this nonbeneficial study, and 673 ml in the traditional group of the ARDS Network trial. Despite the relatively high tidal volumes in the control group of this nonbeneficial trial, mortality was slightly lower than in the lower tidal volume group, the opposite of what Eichacker and colleagues would have predicted. Thus, there were only two nonbenefical studies in which the higher tidal volume groups received lower tidal volumes than those used in the ARDS Network traditional group. In one of these two studies, the difference in mortality between groups (4%) was from just one more death in the lower tidal volume group than in the traditional group (15). This study does not provide reliable information regarding the effects of tidal volume reduction on mortality. The evidence for a detrimental effect of lower tidal volume ventilation is nonexistent. To address this issue, we constructed a smoothed plot of mortality versus plateau pressures on the first day after randomization in our trial (Figure 2). The positive slope of this relationship primarily reflects greater disease severity in patients with lower respiratory system compliances. However, in a logistic regression analysis, plateau pressure independently predicted mortality when APACHE III (20) and tidal volume were held constant. Most importantly, there was no suggestion of a detrimental effect of low plateau pressures in the relationship shown in Figure 2. This graph and analysis, which are based on evidence, differ from the hypothetical relationship shown in Figure 3 of the commentary by Eichacker and colleagues.
Eichacker and colleagues cited an observational study in which low tidal volumes were associated with increased mortality in ALI patients (19). They also claimed that two animal studies "further validate this relationship." In none of these studies were either tidal volumes or inspiratory airway pressure purposely adjusted to explore effects on lung function or outcomes. Therefore, these studies provide evidence for association only, not causality. Importantly, mortality before hospital discharge in the lower tidal volume group of the ARDS Network study (n = 432) was 31%. Moreover, lower tidal volumes were used in all patients (n = 550) in our recently completed trial testing effects of higher positive end-expiratory pressure (21). Overall, mortality before hospital discharge in this study was 26%. These mortality rates are low compared with those reported in other large groups of ALI patients. These results from multicenter clinical trials are contrary to the contention of Eichacker and colleagues that lower tidal volume ventilation is harmful. Were the statistical methods used in their analysis appropriate? Eichacker and colleagues separated the five studies into "beneficial" and "nonbeneficial" groups, depending on differences in mortality and associated confidence intervals between study groups. The basis for grouping the studies in this manner was not sound. The standard for reporting confidence intervals (CI) around an estimate of mortality is the 95% CI. Eichacker and colleagues used 68% CIs in their Figure 1. Hence, the CIs for the beneficial and nonbeneficial studies appear to be widely separated. Standard 95% CIs for the combined nonbeneficial (0.50–1.27) and beneficial studies (1.20–2.07) overlap. Four of the five studies stopped early for either efficacy or futility of the lower tidal volume approach (11, 12, 14, 15). Under these circumstances, estimated treatment effects are biased in favor of the early stopping rule (22). Thus, the early stopping rules alone could account for at least some of the discrepancy in the mortality estimates among the studies. The differences in tidal volumes between study groups in the nonbeneficial studies was considerably smaller than in the beneficial studies. When this difference is accounted for, the results of the nonbeneficial and beneficial studies are consistent (p = 0.27 for heterogeneity among the studies). A scientifically rigorous conclusion is that the absence of beneficial effects in the three nonbeneficial studies was from chance variation. Referring to a nonstandard 90% CI (0.54–1.18) for the combined odds ratio of survival with low tidal volume treatment in the nonbeneficial studies, Eichacker and colleagues state: "... there is only a 5% chance that the ... trials could produce a beneficial effect with and odds ratio of survival greater than 1.18." This is misleading. There is a 16% chance that the combined nonbeneficial studies missed a survival benefit (without first accounting for the bias introduced by the early stopping rules used in two of the nonbeneficial studies). What about protocols and trial design? Eichacker and colleagues state: "Definitive Phase III clinical trials ... need a control arm that represents what is believed by participating physicians to be the best current care." However, there was no agreement on "best current care" for ventilator management in ALI/ARDS in the 1990s. All the tidal volume reduction trials used explicit protocols for the groups that received higher tidal volumes. This approach to trial design is common in clinical research. A recent trial compared clinical outcomes in critically ill patients randomized to receive blood transfusions according to two different explicit strategies (23). In a recent trial comparing pulmonary embolism therapies, anticoagulants in both study groups were controlled by explicit protocol rules (24). In countless phase III trials of chemotherapeutic strategies, control study groups receive therapy according to study protocols. In each of these studies, the study groups that could be considered "control groups" received treatment according to protocols that represented mainstream clinical approaches. If treatment were not evaluated in this fashion, there would be countless variations in "best care" choices of medications, dose, and timing according to physicians' practice styles, preferences, and abilities. It would be impossible to characterize the control groups, and the results of the studies would be uninterpretable. Eichacker and colleagues assumed that there was a standard for mechanical ventilation, but they presented no evidence for this assumption. They conducted an analysis using nonstandard techniques, leading to a misleading conclusion that lower tidal volumes were harmful. They recognized many limitations of their argument, but they proceeded as if those limitations were unimportant. Published literature and abundant data from the ARDS Network database provide unequivocal evidence that lower tidal volumes are beneficial for reducing mortality in ALI and ARDS. The Eichacker hypotheses, erroneously stated as conclusions, are inconsistent with the evidence. For the NIH NHLBI ARDS Network:
a Johns Hopkins University Baltimore, Maryland REFERENCES
This article has been cited by other articles:
|
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
