© 2003 American Thoracic Society
Tidal volumes in ards and meta-analysisTo the Editor:As author of a publication cited by Dr. Eichacker and colleagues, I read their article with interest (1). In contrast to their claim that a "best" or "current" standard practice existed, our worldwide survey of ventilation provides data to the contrary (2). Variability in reported practice was profound and included multiple modes of ventilation and a wide disparity in the selection of tidal volumes, oxygen, positive end-expiratory pressure (PEEP), and use of hemodynamic monitoring. Unsolicited comments by respondents indicated great passion and certainty for their practices, but curiously, the comments of equally enthusiastic respondents were often contradictory. Dr. Eichacker and colleagues use our survey to support their claim that the acute respiratory distress syndrome (ARDS) Network study (3) of 12 ml/kg tidal volume did not represent practice, but Dr. Brower (4) is correct when he points out that more than half of respondents reported using a tidal volume larger than 10 ml/kg. I suspect few physicians were calculating predicted body weight yielding an even larger effective tidal volume. Years from now, we may learn a better way to treat acute lung injury (ALI)/ARDS patients than the 6 ml/kg predicted body weight ARDS Network protocol. However, currently the lower tidal volume ARDS Network strategy is the only one proven beneficial after undergoing rigorous human testing against representative community practice. After examining Dr. Brower's evidence, I conclude claims that lower tidal volumes are only beneficial because they were compared with an artificially high tidal volume untenable. I agree with Dr. Stewart (5) in his conclusion that the ARDS Network protocol is now the standard against which all other strategies should be tested, and based on the evidence, it is the one I use for my patients.
Pulmonary and Critical Care Medicine, Sleep Disorders Sheffield, Alabama REFERENCES
To the Editor:The recent meta-analysis by Eichacker and colleagues (1) argues that the ARDS Network trial did not have a control arm that was consistent with the standard of care (2). They state "further clinical trials are necessary to determine whether lowered tidal volumes produce a survival benefit when compared with the intermediate tidal volumes (8–9 ml/kg) routinely used by participating physicians at the time of these trials."I don't believe that this is true, based upon my own observations while visiting many intensive care units around the country and in Europe over many years, including the decade of the nineties when the ARDS trial was done. Most units employed high tidal volumes according to our original recommendations. In a state-of-the-art review in 1990 (3), I recommended a tidal volume of 10–15 ml/kg, as per the table in the online supplement (4). This was the tidal volume setting that evolved during the time that the Denver group was first developing new concepts in ventilator management aimed at how best to support patients with ARDS (4, 5). Although it would be presumptuous to argue that the Denver group set the standard of care, this is what we were advising before the ARDS Net controlled clinical trial. The ARDS Net used a tidal volume of 12 ml/kg, which was consistent with our recommendations (3). Now the meticulously well-designed and conducted ARDS Network trial shows that we were wrong! Their study now provides a sound scientific basis for the present standard of care.
University of Colorado Health Sciences Center Denver, Colorado Rush-Presbyterian St. Luke's Medical Center Chicago, Illinois REFERENCES
From the Authors:We have read the comments of Drs. Carmichael and Petty. We believe they do not reflect the tidal volumes most commonly used by clinicians for managing patients with ARDS in 1996, the time that the ARDS Net trial of low tidal volume began. However, the combination of Dr. Petty's review article, the survey of Carmichael and coworkers, and textbooks current at the time of the ARDS Network trial do indicate that substantial changes were occurring in the field of mechanical ventilation in the early 1990s. Dr. Petty, a pioneer and leader in the management of ARDS, points out that he recommended tidal volumes of 10 to 15 ml/kg actual body weight for patients with ARDS in 1990 (1). Only two years later, 1992, Carmichael and coworkers (2) undertook a survey of the Critical Care Assembly of the American Thoracic Society. Of the more than 1,000 physicians answering, 45% indicated that they would restrict tidal volumes to a range of 5 to 9 ml/kg actual body weight (2). Furthermore, 96% indicated that level of airway pressure would influence their choice of tidal volume.In 1994, 2 years before the initiation of the ARDS Network trial of low tidal volume, an authoritative textbook in the field, "Principles and Practice of Mechanical Ventilation," included the following passage about the selection of tidal volume in ARDS patients. For many years, physicians have chosen ventilator VT [tidal volume] between 10 and 15 cm3/kg [actual body weight]. This recommendation can be traced back to the early days of positive pressure ventilation when this therapy was reserved for the treatment of neuromuscular diseases, such as polymyelitis... However, in patients with lung injury, whose inspiratory capacity and total lung capacity (TLC) are substantially reduced and whose end-expiratory volume has been raised with PEEP, a VT of at least 10 cm3/kg [actual body weight] can have devastating effects on lung structure and function ... Assuming that inflating the lungs to volumes above TLC is unsafe, it has become common practice to reduce VT to no more than 7 cm3/kg [actual body weight] in the management of ARDS. (3) Notably, the mean tidal volume received by patients in the ARDS Network trial before they were randomized to the two arms of the trial was similar (10 ml/kg predicted body weight or approximately 8 ml/kg actual body weight) (4, 5) to the recommendation in this textbook (3). The "traditional tidal volume" of 12 ml/kg predicted body weight assigned to controls in the ARDS Network trial was significantly greater than the volume routinely used by physicians at the participating centers when managing patients with ARDS who were subsequently randomized into the trial (5). The control tidal volume in the ARDS Network trial was based on a survey that had been conducted 4 years earlier (2) and on expert opinion (6, 7). Accordingly, the control tidal volume may not have reflected the rapid changes that were occurring in this field of medicine in the early 1990s. The composition of the control group in any trial is critical for interpreting study results and for safety monitoring. Use of routine care as a control should automatically incorporate recent changes in clinical practice, and therefore provide a strong basis for determining whether an investigational therapy can improve or worsen current clinical practice. If routine care is not the comparison group, documenting the applicability and ensuring the safety of the proposed control treatment is necessary. Two possible methods for meeting this obligation are the performance of pilot studies and recent chart reviews that summarize current daily practice at participating centers.
National Institutes of Health Bethesda, Maryland REFERENCES
To the Editor:Eichacker and colleagues (1) omit a number of facts when comparing beneficial (2, 3) and nonbeneficial trials (4–6) in their meta-analysis of trials testing low tidal volumes.The PaO2/FIO2 in the control groups of the two beneficial trials are lower (both 134) than the ones reported for the control groups of the three nonbeneficial trials (145, 150 and 155). APACHE II scores were reported in one beneficial trial (2) and two nonbeneficial trials (4, 5), and were markedly higher in the control group of the former (27 versus 17 and 21.5). Two of the nonbeneficial trials show baseline differences between the groups that favor a worse outcome in the low tidal volume group. In the study by Stewart and colleagues (4), the low tidal volume group has a significantly lower PaO2/FIO2 at baseline compared with the control group (123 versus 145; p < 0.05). Brower and colleagues (6) find no difference in outcome between the low tidal volume (PaO2/FIO2 129 at baseline) and control group (PaO2/FIO2 150 at baseline), have a 50% higher prevalence of comorbid conditions (AIDS, bone marrow transplantation, cancer) in the low tidal volume group, and do not report a statistical analysis of these baseline differences. The remaining nonbeneficial trial by Brochard and colleagues (5) has the highest reported PaO2/FIO2 and the lowest APACHE II values of all analyzed trials; inclusion of a limited number of moderately ill patients may have contributed to the failure to demonstrate any difference between the treatment groups in this trial. Would the conclusions by Eichacker and colleagues (1) be the same had they excluded some or all of the nonbeneficial trials from their meta-analysis?
Memorial Hospital of Rhode Island Pawtucket, Rhode Island REFERENCES
From the Authors:Dr. tefanec raises an important question as to whether there is a sounder explanation than plateau pressure for the differences in treatment effects among the five trials testing low tidal volumes in patients with ARDS (1–6). He wonders whether baseline differences in PaO2/FIO2 and APACHE scores in the control and treatment groups might explain the contrasting outcomes we noted in our comparison of beneficial and nonbeneficial trials. The differences in PaO2/FIO2 noted by Dr. tefanec in some of the trials support that possibility, but the pattern in other trials, not reported by Dr tefanec, makes the possibility unlikely (Table 1)
. The three nonbeneficial trials (3–5) had higher PaO2/FIO2 values in the control groups than in the treatment arms; however, this was also true for one of the beneficial trials (2). In fact, this difference in PaO2/FIO2 in the beneficial trial by Amato and colleagues was one of the greatest differences among all five trials (Table 1). Furthermore, there was no clear relationship between the magnitude of difference in PaO2/FIO2 at baseline and outcome comparing control and treatment groups for the five trials (2–6). This difference in PaO2/FIO2 was greatest in one (2) of the two beneficial trials, and smallest in the other (6). Furthermore, the trial with the smallest difference in PaO2/FIO2 at baseline (6) had one of the largest differences in survival rates (Table 1). Likewise, baseline differences in disease severity scores between study groups do not consistently predict the effect of treatment in these studies. In contrast, as our analysis showed, plateau airway pressure after randomization to high tidal volumes were consistently and significantly greater in controls in the beneficial trials than in the nonbeneficial trials. However, we hope that continued scrutiny of these five trials and analysis of the findings will enhance understanding and improve the conduct of future trials in this field.
National Institutes of Health Bethesda, Maryland REFERENCES
To the Editor:I read with interest the paper by Eichacker and coworkers (1). I believe their meta-analysis is flawed and an incorrect method is applied.The authors stated that a standard meta-analysis reporting an overall effect of estimate of treatment was not appropriate due to heterogeneity in patients outcome (p = 0.06, Breslow-Day test). Therefore, they spilt the five studies into two groups: beneficial and non-beneficial trials. After that, the authors tried to explain the resultant different mortality rates comparing tidal volumes and plateau pressures within and between the two groups. They concluded that significant differences in the control arms provided a basis for the contradictory results of these five trials. Actually, the method used in this paper to perform a meta-analysis deserves some comments:
Azienda Ospedaliera Desenzano Desenzano, Italy REFERENCES
From the Authors:Dr. Petrucci raises some methodological questions regarding our meta-analysis (1). As reported originally (1), a Breslow-Day test was used to assess the homogeneity of the odds ratios across the five studies. Using the SAS statistical analysis package, the resulting chi-square statistic was 8.8 with 4 degrees of freedom, yielding the reported p value of 0.06. Petrucci asserts that our analysis "did not take into account the different sample sizes of the five trials." However, this is not correct, because the Breslow-Day test depends not only on the observed odds ratios across the five studies, but also their sample sizes.As a consequence of classifying trials into beneficial and non-beneficial, the Breslow-Day chi-square may be decomposed into three independent chi-squares, representing the heterogeneity among the beneficial trials, the heterogeneity among the non-beneficial trials, and the magnitude of the differences between the average odds ratio for the beneficial and non-beneficial trials. The chi-square for the difference between the odds ratios is 6.08, with 1 degree of freedom (p = 0.014). The combined chi-square for the heterogeneity in both the beneficial and non-beneficial subgroups is 2.75, with 3 degrees of freedom (p = 0.43), although more heterogeneity was observed in the beneficial than non-beneficial studies (chi-squares 2.73 and 0.2, respectively, with 1 and 2 degrees of freedom, respectively). Petrucci also notes that the 95% confidence intervals for the odds ratios in the five studies overlap. Although this is true, this point, in and of itself, does not provide evidence of homogeneity of the odds ratios. The observation that overlapping confidence intervals may yield statistically significant differences is best illustrated using a t test for mean differences. Assume that two groups have a similar number of subjects, that one group has a mean of 15 with a standard error of the mean of 1, and that the second group has a mean of 18 and also a standard error of 1. Although the 95% confidence intervals of the two groups overlap, a t test for difference between the means yields a value of 2.12 with a p value of 0.03.
Finally, Petrucci notes that the observed heterogeneity among the studies should be explored to identify alternative explanations. As we have noted in response to Dr.
National Institutes of Health Bethesda, Maryland REFERENCES
This article has been cited by other articles:
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
