© 2003 American Thoracic Society
Meta-analysis of tidal volumes in ardsTo the Editor:The paper by Eichacker and colleagues (1) discusses important issues in weighing the design, monitoring, and interpretation of landmark clinical trials of patients with ARDS (2, 3). Significant differences between groups in a randomized trial inform only about the relevant merits of the tested therapies, and in isolation cannot differentiate between more benefit in the "superior" group versus more harm in the "inferior" group. An analysis by Eichacker and colleagues (1) indicating that inflation pressures were increased in the control group after randomization, amplifies these concerns. Similar issues about the validity of the control groups apply to both of the two "beneficial" ARDS studies cited (2, 3), and raise two important questions. First, are the control groups representative of the standard of care? Second, does entry into the trial confer a disadvantage to patients randomized to the control group (versus not being in the trial)? Interval data from ongoing studies—not usually available to reviewers or readers—is available from the two publications by Amato and colleagues (3, 4). Their final publication (3) describing 53 patients, represents the addition of 25 patients to the earlier publication (4) that described 28 patients. Simple subtraction allows calculation of the mortality in "conventional ventilation" and "protective ventilation" groups for the interval between the two publications (see Table 1) .
Two points are important. First, the internal validity of such a study is questionable, because the likelihood of the interval and original "conventional" cohorts being drawn from the same population is only 0.078 (Fisher's Exact test). Second, the "external" validity is doubtful given that the mortality of 91% far exceeds contemporaneous experience. Finally, these data are used to support the notion that protective mechanical ventilation reduces mortality in ARDS. However, the interval data suggest to us that the correct interpretation may be that harm occurred in the "conventional" ventilation group, rather than protection occurring in the "protective" ventilation group. In the future, such concepts should be borne in mind in study interpretation, but also when obtaining informed consent. We all need to be clear as to whether trials are comparisons against "current standard of care," as opposed to comparisons of two nonstandard interventions, especially if we claim that patients benefit from entry into clinical trials, as has been suggested in other areas of medicine (5).
Hospital for Sick Children and University of Toronto Toronto, Ontario, Canada REFERENCES
From the Authors:We concur with Drs. Parshuram and Kavanagh that the difference in the ventilatory strategies for the control patients as compared with current practice may have been responsible for much of the reported treatment effect in the study by Amato and coworkers (1).There are two pieces of data (2, 3) not originally reported by ARDSNet investigators (4), which help us understand how treatment in controls differed from current practice and how this difference may have influenced outcome in their trial. First, in their commentary (2) on our metaanalysis (5), the ARDSNet investigators presented the tidal volumes and airway pressures in patients receiving volume-cycled ventilation before randomization. We have replotted these tidal volumes (Figure 1) . It is clear that the tidal volumes received by patients before randomization were normally distributed in a bell-shaped curve. The "traditional" tidal volume of 12 mL · kg-1 predicted body weight (PBW) that controls were randomized to in the ARDSNetwork trial represented the 80th percentile of current practice according to their data (Figure 1B). Therefore, of patients assigned to the control arm of the study, nearly 80% required an increase over their usual tidal volume to receive 12 mL · kg-1 PBW (Figure 1).
Second, the effect of this intervention can be estimated by comparing outcome of the control group with data obtained in eligible critically ill patients who declined to participate in ARDSNetwork trials (3). ARDSNetwork investigators followed the 28-day survival rate of these eligible nonparticipants receiving routine care (3) in the same way they followed enrolled patients (4) (Table 1) . Control patients in the ARDSNetwork trial (n = 429) (4) had a significantly higher mortality rate compared with these eligible nonparticipants (n = 260) (40 versus 31%, p = 0.02). Importantly, the low mortality rate in the nonparticipants is equivalent to the mortality that was reported in ARDS patients in 1996, the year the ARDSNetwork trial began (6) (Figure 2) . Examining the mortality rate of trial subjects (controls combined with the low tidal volume group) compared with nonparticipants, the ARDSNetwork investigators themselves concluded that, "... critically ill patients who participated in this trial did not fare better overall than eligible patients who declined to participate" (Table 1) (3).
Figure 1 also shows that less than 3% of patients receiving volume-control ventilation at the time of the ARDSNetwork trial were routinely receiving tidal volumes less than or equal to the low tidal volume they tested (i.e., 6 mL · kg-1 PBW). Therefore, this trial compared approximately the 3rd to the 80th percentile of current practice without assessing the most commonly used level of care (i.e.,  10 mL · kg-1 PBW, the 50th percentile of care) (Figure 1). In our meta-analysis (5), we show that tidal volumes somewhere between the 3rd (6 mL · kg-1 PBW) and 80th percentile (12 mL · kg-1 PBW) (Figure 1) could possibly produce a better outcome than either of these extremes of routine care. Therefore, the ARDSNetwork trial does not appear to answer the most important question: whether 6 mL · kg-1 PBW was better, worse, or no different than routine care given to most patients with ARDS (Figure 1). Ventilatory care for individual patients with ARDS varies based on the severity of the syndrome and the presence of other complicating conditions. Alternative study designs based on data such as that shown in Figure 1 might have better minimized changes in ventilatory care for controls while still effectively testing lowered tidal volumes. For example, if routine care as shown in Figure 1 had been used for controls, there would have been less than 3% overlap with the tidal volume of 6 mL · kg-1 PBW that was tested. Such a design would have provided a definitive comparison between the tested tidal volume (6 mL · kg-1 PBW) and the routine care that patients were receiving. Most importantly, such a design would not have subjected control patients to increases in tidal volume independent of the need. To achieve greater differentiation from the treatment (low tidal volume) group, the controls might have been subjected to a range of tidal volumes between 9 and 12 mL · kg-1 PBW instead of current practice. This range would have encompassed the majority (> 60%) of current practice (Figure 1A). With this design, however, patients receiving less than 9 mL · kg-1 PBW before randomization would have required an increase in tidal volume for enrollment as a control subject. These patients could have been excluded before randomization or randomized to a third arm without a change in care. Each of these designs would have a control arm that was representative of current practice and therefore most effective for the monitoring of safety in a rapidly lethal disease with a high mortality rate (7). Furthermore, these designs would have avoided increases in tidal volumes and airway pressures in most control patients. If the ARDSNetwork investigators believed that increasing tidal volumes in controls was critical for their evaluation, pilot studies could have been done to ensure the safety of the practice, particularly before starting a large Phase III clinical trial (4). The ARDSNetwork has attempted to address important questions critical to the improvement of patient care. However, in response (2) to our meta-analysis (5) the ARDSNetwork investigators indicated that current practice at the time of their low tidal volume trial was divided into "... two general approaches" employing either "... generous tidal volumes with relatively high airway pressures" or "... lower tidal volumes and inspiratory airway pressures." If this were the case, the data available from the ARDSNetwork trial, shown in Figure 1, would show two peaks (i.e., a bimodal distribution), one centered around 6 and the other around 12 mL · kg-1 PBW. Based on this original assumption, the ARDSNetwork trial incorporated two study arms, neither of which represented predominant physician practice. Understanding the range and frequency of current practice under question at participating centers would likely improve the design and safety of future trials.
National Institutes of Health Bethesda, Maryland REFERENCES
|
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
