Published ahead of print on February 5, 2003, doi:10.1164/rccm.200205-478OC
© 2003 American Thoracic Society Impact of Randomized Trial Results on Acute Lung Injury Ventilator Therapy in Teaching HospitalsDivision of Pulmonary, Allergy, and Critical Care Medicine and Clinical Outcomes Research Center, University of Minnesota Medical School, Minneapolis, Minnesota; Department of Experimental and Clinical Pharmacology, College of Pharmacy, University of Minnesota, Minneapolis, Minnesota; and Hennepin County Medical Center, Minneapolis, Minnesota Correspondence and requests for reprints should be addressed to Craig R. Weinert, M.D., M.P.H., MMC 276, 420 Delaware Street SE, Minneapolis, MN 55455. E-mail: weine006{at}umn.edu
Reducing tidal volumes administered to patients with acute lung injury is the only intervention reported to decrease mortality resulting from this life-threatening condition. Whereas many medical advances are slowly brought into practice, clinicians in teaching hospitals are often assumed to be early adopters of new medical advances. Our objective was to examine trends in the ventilatory prescription for 398 patients with acute lung injury treated in three teaching hospitals from 1994 to 2001. There was no change in tidal volumes until mid to late 1998, when volumes started to slowly decline at the rate of 48.0 (95% confidence interval, 21.0 to 74.4) ml/year. In the 2 years after the results were released from a large trial that demonstrated the superiority of 6 ml/kg tidal volume therapy over 12 ml/kg, clinicians prescribed tidal volumes of 651 ± 128 ml or 10.1 ± 1.9 ml/kg. Tidal volumes after intubation were minimally reduced over the subsequent 2 days of mechanical ventilation (mean reduction, 33 ml). Hospital category, male sex, and disease onset before May 1999 were associated with higher volumes whereas lung injury severity was inversely associated. We conclude that clinicians practicing at these teaching hospitals have not rapidly adopted low tidal volume ventilation that may reduce mortality.
Key Words: physician's practice patterns • respiration, artificial • respiratory distress syndrome, adult Mechanical ventilation is a life-saving technology for patients who develop acute lung injury (ALI) and investigators have conducted numerous investigations to determine the optimal support method (1). Randomized ventilator trials were inconclusive until a single-center trial demonstrated a 33% absolute mortality reduction for patients treated with a "lung-protective" intervention that included reduced tidal volumes, pressure-cycled mode, and a positive end-expiratory pressure adjusted by pressure–volume curve analysis (2). One year later a multicenter clinical trial (hereafter named the ARDSNet trial) commenced enrollment with a ventilator protocol that could be applied to patients in any intensive care unit (ICU) (3). The main conclusion of the ARDSNet trial was that ventilation with tidal volumes delivered at 6 ml/kg ideal body weight via volume-cycled ventilators and a plateau (static) pressure goal of 30 cm H2O or less was associated with an absolute 8.8% decrease in mortality compared with 12 ml/kg. However, three other contemporaneous reduced-tidal volume trials did not demonstrate a decrease in mortality or other important end points (4–6). These studies enrolled substantially fewer subjects, and the tidal volume and plateau pressure differences between the study arms were smaller. The ARDSNet trial results were disseminated by a National Heart, Lung, and Blood Institute Alert on March 15, 1999, a major presentation at the International Conference of the American Lung Association and the American Thoracic Society on April 26, 1999, and an influential medical journal publication on May 4, 2000 (3). However, in a preliminary report from a university-affiliated hospital with extensive experience in ALI trials, Rubenfeld and colleagues demonstrated that clinicians infrequently treated their own patients with the very low tidal volume strategy, even after publication of the ARDSNet trial (7). Because their results may only be representative of ICUs where patients are recruited for clinical trials, a prospective study that enrolled subjects from ICUs that have not participated in ventilator trials may produce more generalizable conclusions. Such a situation exists at teaching hospitals affiliated with the University of Minnesota Medical School. Therefore we wished to analyze trend data since 1994 from our Specialized Center of Research ALI database. We also wanted to discover the clinical and institutional characteristics associated with receiving high tidal volume therapy and investigate how clinicians adjust the ventilatory prescription over the initial 3 days of acute lung injury. Some of the results of these studies have been previously reported in the form of abstracts (8, 9).
We examined data from the initial 3 days of mechanical ventilation for 412 subjects who developed acute lung injury (American–European Consensus Criteria) (10) from May 1994 through May 2001. The protocol initially enrolled subjects 18 years of age or older from medical and surgical ICUs in three major teaching hospitals; however, because of low enrollment, Regions Hospital (St. Paul, MN) ended recruitment in late 1995. The two hospitals with complete enrollment are Fairview-University Medical Center (Minneapolis, MN; inpatient census of about 300) and Hennepin County Medical Center (Minneapolis, MN; census of about 340). See the online supplement for additional detail about facilities, subject recruitment, variable definitions, and statistical analysis. The Institutional Review Boards at all facilities approved the study. We defined tidal volume as the clinician-set tidal volume recorded in proximity to 8 AM on the ventilator flowsheet if the patient was receiving assist-control or synchronous intermittent mandatory ventilation (SIMV) modes. Volumes were not corrected for ventilator tube compliance. If the patient was receiving solely pressure-cycled ventilation, then the exhaled tidal volume was recorded. If the patient received SIMV/pressure support, we recorded the set tidal volume and the (usually smaller) exhaled volume from the pressure support breaths was disregarded. Not all cases contributed 3 days of ventilator data for several possible reasons: delayed intubation after ALI onset or early extubation, death before Day 3, or missing data. Therefore, to describe the ventilatory prescription for as many subjects as possible, an aggregate measure of tidal volume over the initial 3 days of ALI was calculated. For individuals who had at least one value of set or exhaled tidal volume for any of the first three ALI days, we calculated the mean tidal volumes (excluding pressure support in SIMV breaths) for up to 3 days. Three hundred fifty-seven subjects had tidal volume data for Day 1, 370 had tidal volume data for Day 2, and 363 had tidal volume data for Day 3. The aggregate tidal volume variable was calculated from three ALI days in 321 cases, two ALI days in 50 cases, and one ALI day in 27 cases. A primary ALI risk factor was defined as pneumonia, aspiration, or toxic inhalation (n = 254) and remaining subjects had a secondary risk factor (n = 144).
Statistical Analysis To describe the evolution of the ventilatory prescription during the early ALI period, we compared the Day 1 positive end-expiratory pressure (PEEP) with Day 3 PEEP values and similarly compared Day 1 and 3 clinician-set tidal volume values. We excluded cases with tidal volume values obtained during pressure-cycled ventilation because of the inevitable breath-to-breath variation. We performed analyses with SPSS version 10.1 (SPSS, Chicago, IL).
Four hundred twelve subjects were enrolled over a 7-year period. An additional 185 patients with lung injury had exclusion criteria. For our tidal volume analysis, we excluded 12 additional patients because 8 were never intubated, 3 were not intubated until at least Day 4 of ALI, and 1 was managed with a high-frequency ventilator only. After inspection of the remaining data representing 400 subjects, 2 subjects were excluded as extreme outliers because their height (less than 121 cm) was more than 3 standard deviations below the sample mean. The sample mean measured body weight (MBW) was 78.3 ± 24.4 kg (n = 396) and the calculated mean ideal body weight (IBW) was 64.0 ± 11.3 kg (n = 222). Table 1 demonstrates that the majority of patients were managed with volume-cycled ventilation, with the clinician directly prescribing the tidal volume. The overall sample had severe illness with elevated lung injury and severity of illness scores, and a median duration of ventilator support of almost 2 weeks. Subject 30-day mortality for each year (year intervals starting in May 1994) was 40, 39, 40, 37, 47, 52, and 57%.
For all subjects the mean ± SD tidal volume was 694 ± 123 ml. The tidal volume for the first 5 years was greater, 711 ± 118 ml (n = 288), than the tidal volume, 651 ± 128 ml (n = 110), for the final 2 years (difference of -60 ml; 95% confidence interval [CI], -87 to -32 ml). There were also statistically significant decreases in weight-adjusted tidal volume between the two time periods: 9.5 ml/kg MBW for the first 5 years versus 8.9 ml/kg MBW for the final 2 years (difference of -0.6 ml/kg; 95% CI, -1.2 to -0.17 ml/kg). There was no trend in MBW over the 7-year period except an isolated slight increase for the year May 1997 to May 1998. For IBW the mean tidal volume was 11.2 ml/kg for the first 5 years (n = 143) versus 10.1 ml/kg (n = 79) for the final 2 years (difference of -1.1 ml/kg; 95% CI, -1.7 to -0.6 ml/kg). In comparison, the achieved Day 3 tidal volumes in the ARDSNet trial intervention arm was 6.2 ± 1.1 ml/kg (IBW) (3). In our sample, 0.9% (2 of 222) subjects received volumes of 6.2 ml/kg or less. The location of the breakpoint in Model 1 (the A3 coefficient described in the online supplement) was at time block 26 representing July–August 1998, with a 95% confidence interval from September–October 1997 to July–August 1999. The model based only on significant parameters (Model 2) estimates the breakpoint at time block 28, representing November–December 1998 with 95% confidence interval from January–February 1998 to September–October 1999. Confidence intervals from both models include the ARDSNet trial release date of March–April 1999. There was no trend in tidal volume prescription before the breakpoint (the confidence interval for the prebreakpoint slope parameter, A11, contains zero) but a decline occurred after the breakpoint (the confidence interval for the slope parameter after the breakpoint, A21, does not contain zero and the sign is negative). Parameters were included in Model 1 to adjust for differences in tidal volume according to the LIS. The average tidal volume setting was related to the LIS, as shown by a significant, negative intercept-adjustment factor (A02). However, neither pre- nor postbreakpoint trends were related to the LIS, as slope adjustment parameters (A12 and A22) were not statistically different from zero. These results indicate that the LIS is inversely associated with tidal volume, and the effect of milder lung injury is to shift upward the tidal volume trend line without changing the slope or relationship of the lines over time; that is, the lines are parallel. Tidal volumes after the breakpoint declined at 8.0 ml per 2-month interval or 48.0 ml (95% CI, 21.0 to 74.4 ml) per year. Figure 1 shows that mean tidal volumes linearly decline from ALI Day 1 through Day 3 consistently throughout the 7-year period. For the 321 cases with complete Day 1, 2, and 3 tidal volume data, there were differences in the mean tidal volumes: 714 ± 134, 695 ± 135, and 681 ± 133 ml, respectively. Wilks' lambda (a multivariate F test for repeated measures analysis of variance) = 15.1, p < 0.001, and the test for linear trend was significant at p < 0.0001. However, the mean decrease in tidal volume from ALI Day 1 to 3 was a modest 33 ml. An interaction term for tidal volume and date of ALI onset was not significant, indicating there was no change in the slope of the tidal volume decline from Day 1 to 3 over the 7-year interval.
We next determined whether demographic, pulmonary, or institutional characteristics were associated with the distribution of tidal volume per kilogram (MBW) analyzed in aggregate over 7 years. We chose weight-adjusted tidal volume because characteristics such as sex and age are associated with body weight. The mean tidal volume per kilogram for males was 9.63 ml/kg, and for females it was 8.97 ml/kg (difference of -0.66 ml/kg; 95% CI, -1.1 to -0.23 ml/kg). Older subjects (divided at the median age of 48 years) received a greater weight-adjusted tidal volume of 9.56 ml/kg compared with 9.13 ml/kg in the younger group (difference of 0.43 ml/kg; 95% CI, 0.10 to 0.86 ml/kg). To confirm that clinicians treated more severe lung injury with smaller tidal volumes, we compared the distribution of tidal volume over quartiles of the maximum LIS (Figure 2) . The tidal volume decreases from the first to the third LIS quartile (higher quartiles represent increasing severity) but the only significantly different comparison is between the first quartile and the remaining three (one-way analysis of variance, global F = 11.5, d.f. 3, p < 0.001), suggesting that clinicians are less likely to prescribe low-tidal volume therapy for milder lung injury.
The mean tidal volume per kilogram (MBW) was slightly greater in the primary risk factor group: 9.5 ml/kg compared with 9.0 ml/kg in patients with secondary risk factors (difference of 0.5 ml/kg; 95% CI, 0.02 to 0.91 ml/kg). Excluding the 12 subjects from the hospital that ended enrollment in 1995, Table 2 shows that modest but consistent and statistically significant smaller tidal volumes were administered to subjects at Hospital A compared with Hospital B. These differences were present whether the data were analyzed by tidal volume, tidal volume per kilogram (MBW), or for volumes analyzed separately for ALI Day 1, 2, or 3.
To evaluate the independent contribution of the preceding five variables and ARDSNet trial impact in a linear model, we regressed tidal volume per kilogram (MBW) on age and LIS, sex, risk factor, and institution and enrollment before or after May 1999. A total of 384 subjects with complete data was analyzed. Male sex, treatment at Hospital B, and ALI onset before May 1999 were independently associated with receiving higher weight-adjusted tidal volumes. LIS and age showed a trend toward statistical significance (global F = 6.0, 6 df, p < 0.0001, R2 = 0.087). See online supplement for details. We examined plateau pressure trends to determine whether clinicians were modifying other elements of the ventilatory prescription besides tidal volume. Although not set directly by the clinician, plateau pressure estimates end-inspiratory alveolar pressure; which, if elevated, contributes to ventilator-induced lung injury (13). For comparison, the ARDSNet trial 6-ml/kg arm had a maximum plateau pressure allowed by the protocol of 30 cm H2O with an actual measured pressure for enrolled subjects of 26–27 ± 7 cm H2O over the initial 3 days (3). Figure 3 shows that in the last year there was a decline in the mean plateau pressure to a Day 3 value of 27.7 ± 6.8 cm H2O. There was a slight rise in the PEEP with a decrease in the spread of mean values between Days 1 and 3. Figure 4 presents the plateau pressure data as the proportion of values greater than 30 cm H2O (reflecting the ARDSNet protocol). More than 25% of plateau pressures for Day 1, 2, or 3 (994 subject-days with plateau pressures) were greater than 30 cm H2O with a steep decline for Day 3 in the final year.
During early ALI, more subjects (58.5%) had a different PEEP value on Days 1 and 3 compared with the proportion of subjects with different tidal volumes on Days 1 and 3 (44.9%). This difference (McNemar test, p < 0.001) suggests that clinicians treating early ALI adjust the PEEP more often than the tidal volume. This difference occurred despite the much greater tidal volume range (range, 1,000 ml) delivered to subjects compared with PEEP (range, 22 cm H2O).
The results from this 7-year study of clinician ventilatory prescription for patients with acute lung injury indicate that in mid to late 1998 there was a detectable but modest decrease in tidal volumes in the direction supported by a major clinical trial released in the spring of 1999 and published in May 2000 (3). The change in therapy was confirmed by using two distinct methods: by visual inspection of a smoothed line and by statistically demonstrating a breakpoint and subsequent decline in tidal volume. However, this decline is slight and in the last 9 months of enrollment the mean tidal volume was 607 ml, 8.14 ml/kg (MBW), or, for subjects with known height, 9.9 ml/kg (IBW). The absence of a consistent decrease in the proportion of patients with elevated plateau pressures does not support the hypothesis that, during the 7-year study interval, clinicians have rapidly reduced an alternative measure of lung overinflation such as plateau pressure. Substantial reductions in plateau pressure may have already occurred before 1994, and the mean plateau pressure in more recent years is comparable to pressures achieved in the low tidal volume arm of the ARDSNet trial. Whether clinicians chose tidal volumes solely by monitoring plateau pressure cannot be determined although Figure 1 would suggest that is not the case, as there is only a small decrease in tidal volume from Day 1 to Day 3 of lung injury. It is possible that the tidal volume prescription was influenced by plateau pressure within the first few hours of mechanical ventilation and the daily data collection protocol was insufficiently sensitive to detect this relationship. Two other studies examined changes in tidal volume prescription delivered to patients with ALI. In a 4-month period after the ARDSNet trial publication date, Rubenfeld and coworkers did not detect any change in the proportion of patients with ALI receiving a tidal volume of 6 ml/kg or less (IBW) and a plateau pressure of 30 cm H2O or less (7). Thompson and colleagues analyzed ventilator settings from ARDSNet patients before randomization and were unable to detect a linear trend in tidal volume over 3 years. However, after the ARDSNet trial was stopped, the last 5% of enrolled subjects had a mean decrease of 1.3 ml/kg (IBW) compared with the previous subjects (14). Their data set included only randomized subjects and the ventilator data were abstracted at only a single time point corresponding to the time of protocol initiation rather than the onset of ALI. In none of these studies, including this one, is there proof that changes in ventilator therapy are the direct result of a specific trial result or publication. We hypothesized that there would be a breakpoint in proximity to the ending of the ARDSNet trial and the April 1999 conference presentation because another study had demonstrated a distinct increase in the rate of carotid endarterectomy procedures immediately after the results of long-awaited trials were released but before the full article appeared in press (15). Because ventilator data are not routinely abstracted into administrative data sets, community-based, longitudinal studies of changes in ALI therapy are not available. Therefore we cannot determine whether the rate of change in ventilator practice described in this study is different from that in other academic or community facilities. A reasonable single time-point comparison is from the study by Esteban and coworkers, who enrolled 5,183 ventilated patients in March 1998. Of these patients, 4.4% had acute lung injury and were ventilated on ALI Day 3 with tidal volumes of 607 ± 131 ml and a tidal volume per kilogram (MBW) of 8.5 ± 2.0 ml/kg (16). Another study spanning 18 years showed that clinicians decreased tidal volume in patients with ALI from 13 ml/kg (1978–1981) to 9 ml/kg (1993–1996) (17). The variables in our data set explain only a small amount of the tidal volume variance, suggesting other unmeasured factors influence treatment. For instance, we did not design the protocol to identify which individual(s) were setting the ventilator, their reasoning or training, or the degree of influence by other practitioners that may have led to a choice of tidal volume or PEEP. Clearly, physicians frequently adjust the ventilator in early ALI; however, the tidal volume remains unchanged from Day 1 and 3 more often than the PEEP and the decrease in tidal volume from Day 1 to Day 3 is a modest 33 ml. We hypothesize that these small intrasubject changes in tidal volume are due to clinician reluctance to change an initial volume setting if the patient appears stable or if other ventilator parameters are interpreted as acceptable, such as FIO2 or plateau pressure. For instance, the editorial accompanying the ARDSNet trial publication concluded that although tidal volumes should be reduced to maintain a plateau pressure of less than 32 cm H2O, further volume reductions may be unnecessary as three clinical trials with a control arm plateau pressure goal of less than 32 cm H2O did not demonstrate a survival advantage (18). An analysis of five ventilator trials suggested that the two trials that showed a survival benefit of low tidal volumes did so because the control arm subjects were ventilated with increased tidal volumes relative to the standard of practice and that low tidal volume therapy should not replace tidal volumes of 9–10 ml/kg (19). Studies of physician behavior and medical therapy have identified predictors of slow adoption (20–22). If the intervention is not under the clinician's control, requires repeated effort, penalizes the busy clinician (e.g., substantially prolonging the patient encounter), inconveniences the patient, or has an insignificant effect on patient outcome, then the intervention is less likely to be implemented. Some of these reasons may apply to academic ICUs staffed by multiple physician trainees and respiratory care practitioners who all may contribute to adjusting the ventilator. Nevertheless, there is a paradox that these teaching hospitals are only slowly adopting a potentially beneficial intervention while simultaneously conducting basic and clinical ALI research. How can this be explained? The diffusion model of technology adoption suggests that increasing the availability of scientific information leads to behavioral change, but evidence suggests this is an oversimplification (22, 23). Current strategies advocate "repackaging" of biomedical information to make interventions appear more clinically applicable (24) or evaluating the triggers that motivate change, such as feedback, prompts, incentives, or administrative rules (20, 22). A systematic review of randomized trials that tested whether audit and feedback changed physician behavior concluded these interventions are inconsistently effective and that the benefits are only small to moderate (25). In addition, medical therapy change may be determined more by geographic place than by individual practitioner, which may explain our findings of institutional differences in tidal volume prescription (21). One general explanation for our findings is that ALI ventilatory strategies are a "dynamic" (still developing) as opposed to a "formed" (complete) technology (21). Over the next several years, trials will support or refute the benefits of specific PEEP titration end points, ventilator adjustment based on pressure–volume curve analysis (26), and recruitment maneuvers (27) as well as determine which therapies are more effective in primary or secondary lung injury (28). The rate of change in clinician behavior may accelerate when a more complete picture of the optimal strategy is established. On the other hand, experienced practitioners may be aware of current research but have defensible objections to treating patients with a protocol that minimizes their ability to adjust therapy based on an individual's physiology, believe their patient's condition is not representative of those in clinical trials, are unconvinced of the efficacy of low tidal volume ventilation, or may be using alternative measurements (e.g., plateau pressure) to guide therapy. For instance, if the goal of reduced tidal volume therapy is to achieve a plateau pressure under 32–35 cm H2O, then most patients in this study have reached that goal and further declines in tidal volumes or plateau pressure are unlikely to occur. Similarly, in our study, the slight reduction in tidal volumes that commenced before the release of the ARDSNet trial results could be explained by clinicians ventilating patient with ALI using a physiological approach while awaiting randomized trial results (29). In summary, there are numerous possible explanations that may account for the tidal volume trend in these teaching hospitals. With our study design, a cause-and-effect relationship cannot be established for any of them. Future research that directly determines clinician reasoning and motivation for choosing a ventilatory strategy would be useful. The strengths of this study include 7 years of data collection under the same protocol, multiple days of ventilator data, and lack of contamination by clinical trial interventions. However, our reliance on hospital record abstraction increases missing data such as height. For our analysis, we used the set tidal volume for subjects on volume-controlled modes. These values were not corrected for tube compliance and therefore slightly overestimate the volume actually delivered to the subject's lungs. Although the exclusion of 185 patients with ALI decreases the generalizability of our findings, it increases the likelihood that subjects had relatively normal lung function before lung injury. In conclusion, this report and one other (7) demonstrate that academically oriented hospitals that are centers of ALI research are not rapidly adopting a seemingly simple intervention that may improve the outcome of patients with ALI. The prevalence of this surprising finding among other facilities is unknown as are the mechanisms that account for this paradox. Because additional advances in ALI therapy are inevitable, we believe that further research is warranted that tests interventions that accelerate transfer of scientific information to the bedside.
The authors thank Caron Bury, R.N., R.R.T. for subject enrollment and data abstraction; Carol Albright, M.S. for database management, Boris Bershadsky, Ph.D. for statistical advice; and Peter Bitterman, M.D. and David Ingbar, M.D. for review of the manuscript.
Supported by NHLBI SCOR grant P50HL50152. Versions of this research were published in abstract form in Am J Respir Crit Care Med 1997;155:A950 and Am J Respir Crit Care Med 2002;165:A220. This article has an online supplement, which is accessible from this issue's table of contents online at www.atsjournals.org Received in original form May 28, 2002; accepted in final form January 31, 2003
This article has been cited by other articles:
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
TOP
ABSTRACT
METHODS
