Randomized and Nonrandomized Controlled Trials

We identified 46 randomized trials from which we abstracted data. All were English except one, which was published in French. Evidence Table 5 summarizes the methodologic characteristics of the studies. Over one-half did not describe the method of randomization, and most did not address issues of concealment and were unblinded.

We identified an additional 25 nonrandomized controlled trials in 23 unique publications, the methodologic characteristics of which are presented in Evidence Table 9. All were English language publications except one, which was Japanese. Most were retrospective reviews of patient outcomes before and after the implementation of managed approaches to weaning interventions. Many of these studies were characterized by common limitations of observational studies: dissimilar use of cointerventions between experimental and control groups and a lack of explicit definitions of study measurements (in this case, weaning criteria and extubation criteria).

Evidence Table 5 also presents a classification of the randomized trials we have included. (We use the same classification for nonrandomized controlled trials.) In the first clinical context we identified, the clinician believes the patient is likely to tolerate a test of spontaneous breathing (usually implying extubation) but has some uncertainty. The clinician has a variety of modes of ventilation available to test the patient's ability to tolerate unassisted breathing. We refer to this group of studies as "studies comparing alternative discontinuation assessment strategies."

In the second clinical context we identified, the clinician faces a patient who is improving but who is unlikely to tolerate unassisted ventilation for at least 24 hours and perhaps up to several days. The clinician wishes to progressively decrease the level of mechanical support and has a number of modes available to achieve this goal. We refer to this group of trials as "studies comparing alternative ventilation modes for stepwise reductions in mechanical support."

We identified three studies that addressed related issues but did not fit clearly into either one of these categories. These studies examined the impact of stepwise reductions in mechanical support in patients over periods of up to 48 hours.

We identified three RCTs that compared physician-directed weaning with weaning directed by a protocol with major involvement from respiratory therapists and nurses or a computer-driven protocol.

Another set of randomized trials dealt with the timing of extubation in patients who had undergone cardiac surgery. The management issue is whether clinicians can achieve early extubation without deleterious consequences for patients. Investigators have addressed two approaches to early extubation in such patients: alternative anesthesia regimens and alternative strategies once the patient reaches the ICU.

Investigators have tested, in both children and adults, whether parenteral steroid administration could reduce postextubation edema and thus prevent stridor or distress requiring reintubation.

The balance of fat and carbohydrate that patients receive determines their respiratory quotient and the amount of carbon dioxide they produce. Theoretically, this may influence the burden on their respiratory system during the weaning process. Two RCTs have tested the hypothesis that weaning could be expedited by use of low carbohydrate feeds.

We also found a number of studies in which the interventions do not fit into any of the categories listed above. Evidence Table 5 summarizes the miscellaneous interventions tested in these studies.

Evidence Table 5 lists the outcomes measured in these RCTs. Of these outcomes, there are a number we consider much more important than the others. In general, these outcomes are those that we believe patients would consider important. They include mortality, successful spontaneous breathing trial, reintubation rate, signs of impending airway obstruction postextubation, duration of intubation and mechanical intervention, length of ICU and hospital stay, and major morbidity.

We will now summarize the results of each of the studies in each of the categories we have identified above. Relevant tables for each category of randomized studies include Evidence Table 6 that describes the characteristics of randomized trials, Evidence Table 7 that describes RCT results, and Evidence Table 8 that summarizes the results of pooling across RCTs for those groups of studies in which pooling proved appropriate. The characteristics of the nonrandomized controlled trials are found in Evidence Table 9; the results of the nonrandomized controlled trials are summarized in Evidence Table 10. These results are not included in pooled analyses.

Controlled Trials of Discontinuation Assessment Strategies

The sample size of the eight randomized studies of discontinuation assessment strategies modes varied from 18 to 526; three of the studies enrolled more than 100 patients (Evidence Table 6). The largest studies (Esteban, Alia, Gordo, et al., 1997; Esteban, Alia, Tobin, et al, 1999) were methodologically strong, reporting the method of randomization and weaning, extubation, and reintubation criteria. Lack of concealment of randomization could easily bias the results of these studies, and although none of the studies used the optimal method of ensuring concealment (randomization by an independent methods center), the Esteban studies used the next best approach for a study such as this-opaque sealed envelopes. Other randomized trials were methodologically not as strong (Evidence Table 6).

The first Esteban study (Esteban, Alia, Gordo, et al., 1997) compared 2-hour trials of unassisted breathing using pressure support of 7 cm H₂O versus a T-piece trial. A smaller proportion of those in the pressure support group, 14 percent, failed to tolerate the wean and achieve extubation at the end of the 2-hour trial than in the T-piece group, 22 percent (relative risk 0.64, 95 percent CI 0.43 to 0.94) (Evidence Table 7). Of those extubated, 38 patients in the pressure support group and 36 in the T-piece group required reintubation (Evidence Table 7).

The second Esteban study (Esteban, Alia, Tobin, et al., 1999) compared a 30-minute with a 120-minute T-piece trial of spontaneous breathing prior to extubation. There was no reported difference in the rate of reintubation between groups, and patients randomized to the shorter T-piece trial benefited from statistically significant reductions in ICU and hospital length of stay (2 days and 5 days shorter, respectively).

The other randomized studies, all of which compared T-piece trials with alternative strategies usually including some form of pressure support, had much smaller sample sizes and generally had lower event rates. Our judgment was that we could pool only across two trials that compared T-piece to CPAP and even after pooling, the number of events is so low that the 95 percent CIs are extremely wide: relative risk for nonextubation in CPAP versus T-piece 1.66 (95 percent CI 0.60 to 4.64); relative risk for reintubation 1.61 (95 percent CI 0.39 to 6.59) (Evidence Table 8).

There were no nonrandomized trials in this category of weaning interventions.

Controlled Trials of Progressive Reduction in Mechanical Support

Five RCTs compared alternative methods of decreasing ventilatory support in patients in whom clinicians thought that extubation was still several days away. Their sample size varied from 19 to 130; two trials enrolled more than 100 patients (Evidence Table 6). Three of the trials, including the two largest, used sealed opaque envelopes to protect concealment and described criteria for weaning and extubation. The most informative results come from the two largest studies (Brochard, Rauss, Benito, et al., 1994; Esteban, Frutos, Tobin, et al., 1995). These two studies compared three modes that were delivered in similar ways in the two studies: multiple daily T-piece, pressure support, and SIMV. The Esteban trial also included a fourth arm, once daily T-piece, which demonstrated results very similar to those from the multiple daily T-piece (Evidence Table 7).

In both studies, the investigators recruited patients who had failed a T-piece trial. Esteban, Frutos, Tobin, et al. (1995) conducted their T-piece trial in 546 patients, only 130 of whom had respiratory distress during a 2-hour T-piece trial. Brochard, Rauss, Benito, et al. (1994) found a similar strikingly high proportion of patients who tolerated their 2-hour T-piece trial: of 456 patients in who underwent the T-piece trial, only 109 were unable to tolerate spontaneous breathing and were therefore randomized.

Of those randomized, patients in the Esteban trial had a prior mean duration of mechanical ventilation of approximately 9.3 days, with a minimum of 24 hours. Brochard's patients also had a minimum duration of ventilation of 24 hours. The approximate mean duration of ventilation in the Brochard patients was 14 days.

Because the interventions of the two studies are reasonably similar, we pooled the results for the outcomes that were measured in similar ways (Evidence Table 8). We present two sets of Evidence Tables 8: one deals with duration of ventilation, and the other the relative risk of the combined endpoint of nonextubation in 2 to 3 weeks and the need for reintubation. In the comparison of T-piece to pressure support, the pooled results showed no difference in duration of ventilation, the trends going in opposite directions in the two studies (Evidence Table 8): The results from the Esteban, Frutos, Tobin, et al. (1995) study favored T-piece and those from the Brochard, Rauss, Benito, et al. (1994) study favored pressure support. As a result, the CI around the pooled estimates for both duration of ventilation and relative risk of nonextubation or reintubation is extremely wide.

In the comparison of T-piece with SIMV, the two trials showed similar trends in favor of T-piece in the duration of ventilation. The CI around the pooled results comes close to excluding no difference, with the magnitude of the pooled estimate being a difference of over 40 hours additional duration of ventilation in favor of the T-piece wean, or a relative risk of the combined endpoint of 1.48 in the SIMV group (Evidence Table 8).

In the comparison of pressure support to SIMV on duration of weaning, both studies found trends in favor of pressure support; the effect in the Brochard, Rauss, Benito, et al. (1994) study was much larger. The magnitude of the trend in the pooled result of duration of ventilation is over 60 hours, and the CI comes close to excluding no effect. The pooled results of pressure support versus SIMV on the combined endpoint show an extremely wide CI.

Jounieaux, Duran, and Levi-Valensi (1994) randomized 19 patients to SIMV with pressure support versus SIMV without pressure support. Neither group received CPAP. The duration of the wean was approximately 1 day shorter in the group that received pressure support, with the lower boundary of the CI being approximately 7 hours (Evidence Table 7). Two patients in the SIMV group, and none in the group that also received pressure support, required reintubation.

Two groups of investigators (Girault, Daudenthun, Chevron, et al., 1999; Nava, Ambrosino, Clini, et al., 1998) evaluated NPPV as a mode for stepwise reductions in mechanical support for patients admitted with COPD exacerbation who had failed a 2-hour T-piece trial. The control strategies in the two studies included pressure support ventilation, with or without CPAP. In the larger study, Nava, Ambrosino, Clini, et al. (1998) found a reduction in duration of mechanical ventilation associated with a reduction in ICU stay of almost 9 days associated with NPPV (Evidence Table 7). When the results of these studies were pooled, the reduction in ICU length of stay was 5 days (95 percent CI -12.2 days to +1.9 days) (Evidence Table 8). Pooling also indicated favorable trends in mortality (relative risk 0.30, 95 percent CI 0.09 to 1.02) and in the incidence of nosocomial pneumonia (relative risk 0.29, 95% CI 0.02 to 3.88).

Two additional nonrandomized trials evaluated the use of NPPV in weaning. Patel, Petrini, and Dwyer (1999) compared 24 hours of intermittent nasal CPAP with 24 hours of invasive CPAP in patients who were weaned for 24 hours to PS 5 cm H₂O plus positive end-expiratory pressure (PEEP) 5 cm H₂O following prolonged ventilation. There was a marginal increase in the reintubation rate in the patients receiving invasive CPAP versus noninvasive CPAP. Hilbert, Gruson, Portel, et al. (1998a) evaluated NPPV in COPD patients who developed hypercapnic respiratory insufficiency after extubation. Comparing patients managed with NPPV with historical controls, there was a statistically significant reduction in the rate of reintubation with NPPV and a nonsignificant survival benefit.

Controlled Trials Comparing Alternative Ventilation Modes for Weans Lasting Less Than 48 Hours

Three randomized trials addressed an intermediate group of patients not yet ready for discontinuation assessment, but likely to be ready within 48 hours. These trials were methodologically relatively weak and included sample sizes of less than 50 patients (Evidence Table 6). Chopin, Chambrin, Mangalaboyi, et al. (1989) showed a trend in favor of CO₂ mandatory ventilation over intermittent mandatory ventilation (IMV) and multiple daily T-piece in the proportion of patients extubated at 24 hours. However, there were only 14 patients in each of the three study groups.

The study by Esen, Denkel, Telci, et al. (1992) showed trends in favor of pressure support over IMV in both duration of ventilation and successful extubation in 48 hours (Evidence Table 7).

Davis, Potgieter, and Linton (1989) compared an IMV wean (18 patients) to a wean based on setting the mandatory minute volume (MMV) to 75 percent of the minute volume prior to beginning the wean. The MMV was achieved by decreasing the frequency and maintaining the tidal volume. Patients who weaned quickly did so in less than 5 hours in the MMV group and in over 30 hours in the IMV group (Evidence Table 7). Five patients in each group failed to wean quickly, and the authors do not tell us more about these patients.

Two nonrandomized controlled trials also examined alternative weaning modes for patients expected to wean very quickly from mechanical ventilation (Rathgeber, Schorn, Falk, et al., 1997; Tomlinson, Miller, Lorch, et al., 1989), with findings similar to those of the RCTs (Evidence Table 10). Tomlinson found no difference in the duration of mechanical ventilation or the duration of weaning in medical-surgical ICU patients weaning over a period of 2 hours to (rarely) 3 days using IMV (without CPAP) versus multiple daily T-piece trials. In contrast, Rathgeber, Schorn, Falk, et al. (1997) compared the use of (1) T-piece trials to (2) SIMV or (3) invasive biphasic positive airway pressure (BiPAP) weans in 586 patients following cardiac surgery. The methods in this large study were relatively strong. The results of this study were consistent with the results of related RCTs discussed above, suggesting superiority of BiPAP (CPAP plus pressure support) over T-piece weans and both modes over SIMV weans with respect to the duration of mechanical ventilation.

Controlled Trials Comparing Weaning Protocols to Physician-Directed Weaning

Three RCTs have compared protocolized to conventional weans. One very small trial (15 patients) compared a computer-directed wean with a physician-directed wean and found trends in favor of the computer-directed wean in both nonextubation and reintubation rates (Strickland and Hasson, 1993) (Evidence Table 7).

Two RCTs compared weaning protocols that were largely implemented by respiratory therapists and nurses with conventional physician-directed weaning (Ely, Baker, Dunagan, et al., 1996; Kolef, Shapiro, Silver, et al., 1997). These trials were both methodologically strong (Evidence Table 6) and, for this area of investigation, very large (300 and 357 patients). Both studies enrolled virtually all the patients in their units receiving mechanical ventilation during the study periods.

Ely, Baker, Dunagan, et al. (1996) studied a different group of patients: median durations of mechanical ventilation were 4.5 and 6 days in the protocol- and physician-directed groups respectively. The relative risk of successful extubation in the protocol-directed group was 2.13 (95 percent CI 1.55 to 2.92, p<0.001), indicating that mechanical ventilation was discontinued sooner than in the control group. The largest separation between groups was at approximately 5 days and differences disappeared by about 15 days. Patients in the physician-directed group spent a day longer in the intensive care unit and 1.5 days longer in the hospital; neither of these differences reached statistical significance.

Kollef, Shapiro, Silver, et al. (1997) conducted their study in four intensive care units using three different weaning protocols that had been developed and tested by the ICU staff prior to the start of the study. Despite the large sample size, the power of their study to detect differences in key endpoints was limited, since most patients spent a relatively short period of time on the ventilator. In the protocol- and physician-directed groups, respectively, 25 percent of the patients were extubated by 15 and 21 hours, 50 percent by 35 and 44 hours, and 75 percent by 114 and 209 hours. Only 12% and 17%, respectively, of the patients spent more than 7 days on the ventilator. The authors used a number of sophisticated statistical survival and regression analyses that favored the intervention group and showed borderline statistical significance. More simple tests also favored the protocol-directed group, but failed to reach statistical significance (Evidence Table 7).

In addition to these RCTs, 11 nonrandomized controlled clinical trials have examined the impact of, largely, respiratory therapist- or nursing-directed weans, compared with physician-directed weans, on weaning outcomes in critically ill patients (Evidence Tables 9 and 10). These studies, conducted in a variety of patient populations, are generally much larger than the corresponding RCTs but are more prone to biased results (Evidence Table 9d). Their results are generally consistent with the results of the RCTs, demonstrating statistically significant reductions (Foster, Conway, Pamulkov, et al., 1984; Horst, Mouro, Hall, et al., 1998; Rotello, Warren, Jastremski, et al., 1992) or trends toward reductions (Burns, Marshall, Burns, et al., 1998; Kollef, Horst, Prang, et al., 1998; Saura, Blanch, Mestre, et al., 1996a; Wood, MacLeod, and Moffatt, 1995) in the duration of mechanical ventilation and ICU length of stay, as well as benefits to protocolized weans with respect to a variety of other study outcomes (number of arterial blood gases required). Mortality and reintubation rates did not appear to differ between experimental and control groups among these nonrandomized studies, nor were other complications associated with protocolized weaning reported.

Controlled Trials of Early Versus Late Extubation Following Cardiac Surgery

Investigators have tested two types of interventions to try to reduce the duration of mechanical ventilation following cardiac surgery. One strategy involves modification of anesthesia, in particular reduction in fentanyl dose or substitution of fentanyl for propofol, and the other involves different approaches to care once patients reach the ICU. In addition to the methodologic limitations of randomized trials shown in Evidence Table 6, in particular the consistent lack of information concerning concealment, these studies failed to conduct intention-to-treat analyses. Some restricted their analysis to patients who achieved early extubation or to the extubation goals of the study arm to which the patients were allocated. We have chosen to report only outcomes in which at least 80 percent of the randomized patients in both groups are included in the analysis.

Four RCTs that tested lower doses of fentanyl in patients after coronary artery bypass surgery enrolled between 85 and 144 patients (Berry, Thomas, Mahon, et al., 1998; Cheng, Karski, Peniston, et al., 1996b; Michalopoulos, Nikolaides, Antzaka, et al., 1998; Silbert, Santamaria, O'Brien, et al., 1998) and a fifth tested fentanyl versus proprofol in 70 patients (Evidence Table 6). All five RCTs suggested a reduction in the duration of mechanical ventilation with the lower anaesthetic doses. No other outcomes consistently differed between the early and late intervention groups in the five trials (Evidence Table 7), although Cheng, Karski, Peniston, et al. (1996b) demonstrated an increase in early ischemia with a trend toward an increase in myocardial infarction.

The pooled results (Evidence Table 8) confirm a reduction in the duration of mechanical ventilation, with a mean effect of approximately 7 hours. Although there is considerable heterogeneity between studies, both the smallest mean effect seen in an individual study and the lower boundary of the CI are approximately 1 hour. The pooled results also show a difference in reduction of hospital stay of 1 day with early extubation, with a very narrow CI.

With respect to mortality and important morbidity, even after pooling across the four studies there were very few events, and as a result CIs are so wide as to be uninformative (Evidence Table 8). Except for the outcome of reintubation, what trends there were favored the early extubation strategy.

Five RCTs that used other approaches to achieve early extubation included more varied populations: one trial in elderly patients undergoing elective abdominal aortic reconstruction (Shackford, Virgilio, and Peters, 1981), one trial in patients undergoing mitral valvulotomy (Tempe, Cooper, Mohan, et al., 1995), and three trials in patients undergoing coronary artery bypass surgery (Dumas, Dupuis, Searle, et al., 1999; Quasha, Loeber, Feeley, et al., 1980; Reyes, Vega, Blancas, et al., 1997). Two RCTs reversed neuromuscular blockade to achieve early extubation (Tempe, Cooper, Mohan, et al., 1995; Quasha, Loeber, Feeley, et al., 1980); one discontinued sedation at an earlier point (Dumas, Dupuis, Searle, et al., 1999) whereas two simply instituted early efforts at extubation (Shackford, Virgilio, and Peters, 1981; Reyes, Vega, Blancas, et al., 1997). Sample sizes varied from 35 to 404; only Reyes, Vega, Blancas, et al. (1997) recruited more than 100 patients.

Results of all five trials suggested they achieved, on average, a shorter duration of ventilation in the early extubation group (Evidence Table 7). Morbid events were rare and similar in the 2 groups. The pooled analysis confirms these findings and suggests, in addition, a decrease of one-half day in ICU stay in the early extubation group (Evidence Table 8).

There are an additional eight nonrandomized controlled studies of early versus late extubation following cardiac surgery (Evidence Tables 9 and 10). These were large studies conducted primarily in adults, and all studies evaluated a combination of altered anesthetic techniques and altered ICU care to achieve early extubation. The results were very similar to the RCT results. Duration of intubation was reduced with the implementation of early extubation strategies by 1 to 28 hours, though associated reductions in ICU and hospital lengths of stay were relatively small (though inconsistent), ranging from 1 to 53 hours, and 0.3 to 2.6 days, respectively. Complication rates varied across studies in early versus late extubation groups, and these event rates were rather small.

Controlled Trials of Corticosteroids to Prevent Post-Extubation Airway Complications

Three RCTs have addressed whether preextubation steroid administration can reduce postextubation stridor and the necessity for reintubation in children (Anene, Meert, Uy, et al., 1996; Harel, Vardi, Quigley, et al. 1997; Tellez, Galvis, Storgion, et al., 1991). In all three studies, patients, caregivers, and those assessing outcome were blind to allocation, patients having received dexamethasone or a matched placebo. Two trials (sample sizes of 66 and 153 children) enrolled patients who had not previously been extubated (Anene, Meert, Uy, et al., 1996; Tellez, Galvis, Storgion, et al., 1991). A smaller trial enrolled 23 children who had been reintubated for postextubation stridor, who received two doses of dexamethasone or placebo over 6 hours and were then extubated.

Both of the trials of primary extubation examined both stridor scores and reintubation (Evidence Table 7). Early stridor was present more frequently in both trials in the group that did not receive steroids and the differences persisted until 12 hours in the one trial that measured stridor sequentially (Anene, Meert, Uy, et al., 1996; Evidence Table 7). Reintubation occurred more frequently in the steroid group in one study (Telez, Galvis, Storgion, et al., 1991) and more frequently in the no steroid group in the other study (Anene, Meert, Uy, et al., 1996). In the trial of secondary extubation, the stridor score was slightly and not significantly greater in the no steroid group, which also had a higher incidence of reintubation (5 of 11 versus 3 of 12). This difference did not approach statistical significance.

One nonrandomized controlled study evaluated corticosteroids to prevent postextubation airway complications in children. In a study evaluating steroids in patients who had failed extubation the first time, Freezer, Butt, and Phelan (1990) reported a statistically significant reduction in prolonged reintubation (>6 days) and in failed reextubations.

The pooled analysis of the two randomized trials of primary extubation demonstrated a substantial reduction in the frequency of stridor with a relatively narrow CI (relative risk 0.57, 95 percent CI 0.40 to 0.81) (Evidence Table 8). The pooled analysis also suggested a reduction in reintubation with steroids; but partly because of the trends in different directions in the two studies, the CI is extremely wide (relative risk 0.50, 95 percent CI 0.02 to 13.87) (Evidence Table 8).

The four trials of steroids in adult patients used different medications (methylprednisolone, dexamethasone, and hydrocortisone); three (Chaney, Nikolov, Blakeman, et al., 1999; Darmon, Rauss, Dreyfuss, et al., 1992; Ho, Harn, Lien, et al., 1996) were placebo-controlled. Only one of the studies assessed post-extubation stridor (Ho, Harn, Lien, et al., 1996) and found little difference between the two groups (Evidence Table 7). The need for reintubation was very infrequent in all 4 studies. As a result, even the pooled analysis demonstrates extremely wide CIs around the pooled estimate of steroid impact on reintubation (Evidence Table 8).

Controlled Trials of Enteral Nutrition

High carbohydrate loads can markedly increase carbon dioxide production, resulting in a respiratory quotient of >1.0. The biologic rationale for the high fat, low carbohydrate intervention tested in the following two trials is that the lower respiratory quotient might improve gas exchange and facilitate weaning from mechanical ventilation in patients with limited ventilatory reserve.

In a randomized, double-blind trial (al-Saady, Blackmore, and Bennett, 1989), 20 medical ICU patients were allocated to: (1) a high fat, low carbohydrate enteral feeding solution (Pulmocare: 17 percent protein, 55 percent fat, 28 percent carbohydrates) or (2) isocaloric feeds (Ensure Plus: 17 percent protein, 30 percent fat, 53 percent carbohydrates). Nutritional requirements were calculated at 1.5 times basal metabolic rate, delivered through a nasogastric tube. Patients were eligible if they had either COPD, asthma, pneumonia, or neurologic disease and if they could tolerate enteral nutrition. Exclusion criteria were nephrotic syndrome, hepatic failure, or diabetes mellitus. Patients were mechanically ventilated using intermittent positive pressure ventilation. Weaning was started when patients had a respiratory rate <30 breaths/minute, minute ventilation was <12 L/minute, if arterial partial pressure of oxygen (PaO₂) at fractional inspired concentration of oxygen (FiO₂ ) was >60 mmHg, when arterial partial pressure of carbon dioxide (PaCO₂) was 38-45 mmHg, and when pH was >7.3. T-piece weaning continued and was considered successful at 24 hours if the following were maintained: clinically and hemodynamically stable, respiratory rate <30 breaths/minute, minute ventilation <10 L/minute, if PaO₂ at FiO₂ was >60 mmHg, when PaCO₂ was 38-45 mmHg, and when pH was >7.3. The study was terminated as soon as patients were able to tolerate 3 hours of spontaneous breathing.

Randomization methods and concealment of allocation were not reported. The study was doubleblinded. Patients were comparable at baseline with respect to basic demographics, and preintervention duration of ventilation (approximately 64 and 70 hours, respectively). Cointervening carbohydrate loading in intravenous dextrose or oral medication was avoided.

Only one patient developed delayed gastric emptying in the high fat group; feeds were held for 2 hours but recommenced with no further problem. The PaCO₂ decreased significantly in the high fat feeding group just prior to weaning but increased slightly in patients receiving the isocaloric feed (p=0.003), whereas there was no difference in PaO₂ and tidal volume or respiratory rate. The time from feeding commencement to successful weaning was significantly shorter in the high fat group than in the isocaloric feeding group (86.1±17.8 versus 148.7±36.7 hours).

In a second randomized unblinded enteral nutrition trial (van den Berg, Bogaard, and Hop, 1994), 32 medical ICU patients were allocated to: (1) a high fat, low carbohydrate enteral feeding solution (Pulmocare: 17 percent protein, 55 percent fat, 28 percent carbohydrates) or (2) isocaloric feeds (Ensure Plus: 17 percent protein, 30 percent fat, 53 percent carbohydrates). Nutritional requirements were calculated at 1.5 times basal metabolic rate, delivered through a nasogastric tube. Patients were eligible if they had either COPD, neurologic disease, or pneumonia without COPD and if they could tolerate enteral nutrition. Exclusion criteria were renal failure, hepatic failure, diabetes mellitus, or respiratory failure "without a prospect of weaning from the ventilator." Patients were mechanically ventilated with a volume controlled mode. Weaning was started using CPAP when patients were afebrile, hemodynamically stable, required PEEP <10 cm H₂O on FiO₂ and when serum bicarbonate was <28 mmol/L. CPAP continued for a maximum of 3 hours until patients were too dyspneic or tired too continue; specific failure criteria were not reported. Rest periods lasted 4 to 6 hours and CPAP weaning trials occurred between rest periods. The study was terminated as soon as patients were able to tolerate 3 hours of spontaneous breathing.

Randomization was stratified based on the presence or absence of COPD. Concealment and allocation methods were not reported. Adherence to the feeding regimens was successful in all but one patient in each group, whose feeding was discontinued because of gastric distension; the outcome of these patients is not reported. Patients were comparable at baseline with respect to illness severity and nutritional status. Although the stratification of randomization based on COPD resulted in similar numbers of COPD patients in each arm, the distribution of cases of acute or chronic COPD as a reason for respiratory failure was not similar. In the high fat group, 10/11 patients with COPD were ventilated for acute or chronic respiratory failure versus 5/13 in the isocaloric feeding group. The former group also had severe hypercapnia during ventilation and during the weaning process; since this imbalance affected interpretation of the trial, two-way analysis of variance was used to evaluate whether the results obtained were related to the respiratory failure rather than the nutritional intervention.

The respiratory quotient was significantly lower in patients receiving the high fat, low carbohydrate feed compared with the isocaloric feed (0.72±0.02 versus 0.86±0.02, p<0.01). The minute ventilation during weaning was also lower (8.8±0.9 versus 10.5±0.8, p<0.01). A similar proportion of patients in both arms had a successful 3-hour trial of spontaneous breathing on CPAP (12/14 versus 13/16, p=0.74).

In a related nonrandomized controlled trial, Bassili and Dietel (1981) evaluated the effect of any (enteral or parenteral) nutritional support on weaning patients from mechanical ventilation. Patients comprising the control group received intravenous D5W (glucose solution), only. Though there was no difference in duration of mechanical ventilation between the two groups, there was a very large excess of nonextubations in the control patients. Possible causes of nonextubation are not reported (e.g., death, need for permanent tracheostomy, etc.).

Randomized Trials of Miscellaneous Studies

There are seven randomized trials grouped in this miscellaneous category. Outcomes are summarized in the Evidence Table 7 (Gust, Gottschalk, Schmidt, et al., 1996; Holliday and Hyers, 1990; Jiang, Kao, and Wang, 1999; Lee, Chien, Hsu, et al., 1998; Niehoff, DelGuercio, LaMorte, et al., 1988; Pichard, Kyle, Chevrolet, et al., 1996). An additional article did not report outcomes in a manner that allowed extraction for our purposes; therefore, it is not represented in the Evidence Table 6 or 7.

(1) The rationale for the first study (Niehoff, DelGuercio, LaMorte, et al., 1988) is that arterial blood gas analysis may not be needed as often if continuous monitoring of oxygenation and ventilation is provided during weaning. This trial was not designed to test the accuracy or utility of oximetry and capnography (which has been evaluated in technology assessment literature) but to evaluate its utility as a weaning adjunct.

In a randomized, unblinded trial, 24 postoperative cardiac patients were allocated to pulse oximetry and capnography or periodic arterial blood gases. IMV was used for weaning but stepwise decrements were not specified. The oximetry and capnography group had arterial blood gases on ICU admission, just before extubation, and if SaO₂ <95 percent or end-tidal CO₂ (PetCO₂) <26 or >40 mmHg, and as clinically indicated. The blood gas group was weaned if PaCO₂ was 35 to 45 mmHg and pH was 7.35 to 7.45, PaO₂>70 mmHg and respiratory rate <30 breaths/minute.

There were fewer blood gases performed in the oximetry and capnography group (5.9±2.7 versus 10.1±1.8, p<0.01). The duration of ventilation was similar (18.8±2.0 versus 19.7±1.9 hours). One patient in the blood gas group did not get extubated and was excluded from analysis. No patients required reintubation.

(2) To induce anxiolysis and minimize muscle fatigue, the effect of relaxation biofeedback on respiratory mechanics and weaning was tested (Holliday and Hyers, 1990).

In an unblinded randomized trial, 40 patients ventilated for >7 days were allocated to relaxation biofeedback or a control group. The biofeedback group received a multifaceted intervention for 30 to 50 minutes per day on CPAP 5 cm H₂O for 5 days per week until study termination, consisting of of communication (the patient was asked about feelings, breathing, and sleeping and was encouraged), tidal volume (V_t) feedback (auditory and visual feedback on a computer screen of the patients V_t compared with a threshold V_t) and computerized visual feedback of frontalis muscle tension by electromyography (EMG).

Cointerventions were not well described during the weaning process. Four patients randomized to the control group died and were not included in the analysis. Within-group changes in maximum inspiratory pressure (MIP), tidal volume, and minute ventilation were no different. The duration of ventilation was 12 days shorter in the biofeedback group (20.6±8.9 days versus 32.6±17.6 days, p=0.01). Nonextubation rates were the same. The undisclosed weaning using T-piece or IMV in this unblinded study that found a 12-day difference in duration of ventilation makes interpretation difficult; moreover, the generalizability of this technologically complex intervention is also limited.

(3) Functional residual capacity postspontaneous breathing on T-piece may be better restored with NPPV and CPAP than with spontaneous breathing and physiotherapy, thereby minimizing pulmonary edema and extubation failure.

In a randomized, unblinded trial of 75 cardiac surgery patients (Gust, Gottschalk, Schmidt, et al., 1996), three postextubation interventions were evaluated after 10 to 14 hours of controlled ventilation and 30 minutes of T-piece breathing. Patients were extubated, then randomized to either NPPV (n=25) involving BiPAP using the spontaneous timed mode (S/T) via nasal mask with an inspiratory positive airway pressure (IPAP) of 10 cm H₂O and an expiratory positive airway pressure (EPAP) of 5 cm H₂O and 10 L/minute of oxygen via nasal mask for 30 minutes; CPAP 7.5 cm H₂O and FiO₂ 0.5 for 30 minutes (n=25); or chest physiotherapy for 10 minutes and oxygen via nasal mask at 6 L/minute for 30 minutes (n=25).

Left ventricular function and inotropic support were comparable across groups. Cardiac surgical, anesthetic, and preextubation ICU management for the entire cohort are well described. No patients were lost to followup, and the analysis was intention-to-treat. All three groups had an increase in pulmonary blood volume index (PBVI) over time: 17 ml/m², 9 ml/m², and 17 ml/m² for BiPAP, CPAP, and chest physiotherapy, respectively. Following 30 minutes of each intervention, however, PBVI in the BiPAP group was significantly lower than the other two groups (p<0.05). Extravascular lung water (EVLW) increased significantly from extubation through the 30-minute intervention to 90 minutes following extubation in the chest physiotherapy group, compared with the other two groups (p<0.05). All patients in each group had sustained extubation.

(4) The rationale for this intervention is the catabolism of critical illness and functional and structural neuromuscular abnormalities in ventilated patients. Peripheral skeletal muscle function may improve following 1 week of postoperative growth hormone; however, a recent multicenter randomized trial showed that growth hormone was associated with increased ICU mortality (Takala, Ruokonen, Webster, et al., 1999). Nevertheless, this study of 12 days of growth hormone evaluated the duration of ventilation (Pichard, Kyle, Chevrolet, et al., 1996).

In a randomized, double-blinded trial, 20 patients requiring ventilation for >7 days were allocated to either 0.43 IU of recombinant growth hormone/kg/day administered subcutaneously for 12 days or normal saline. Patients were excluded if they had known myopathy, neuropathy, or a risk factor for neuromuscular abnormalities. Weaning began for all patients when: minute ventilation <10 L/minute, vital capacity >1 L, PaO₂>60 with FiO₂<0.4, or if a T-piece trial was tolerated for 30 minutes. Weaning began with SIMV; PS was added when spontaneous breathing developed and was gradually lowered. At PS of 10 cm H₂O, patients underwent a T-piece trial; after 12 hours of spontaneous breathing, patients were extubated. Parenteral nutrition was provided for the first 48 hours, and enteral nutrition was instituted.

After 12 days, the growth hormone group had higher growth hormone, insulin-like growth hormone factor-1, and insulin levels. Fat-free mass was increased in the treated compared with the untreated group. The cumulative duration of weaning over 12 days was similar (235.6±17.6 versus 245.4±14.7 hours). Patients similarly remained mechanically ventilated at 12 days (7/10 and 7/10, respectively).

(5) In 21 patients requiring ventilation for at least 3 days, SIMV was compared with pressure support for patient comfort. Patients were randomized to SIMV or PS and underwent a sequential 20 percent reduction in support at timed intervals. Then patients crossed over to the other arm after a 1- to 3-hour rest. Dyspnea and anxiety remained stable over time in each mode and were no different between groups. Of the 21 patients, 10 were weaned; the attribution of success to one or the other modes is not possible given the crossover design of the study. Potentially important cointerventions such as verbal feedback and touch are not described.

(6) Acupuncture has been found to relieve sore throats; its potential benefit averting largyngospasm was tested in 76 postoperative children randomized to receive either acupuncture with bloodletting at the Shao Shang acupoint on both thumbs just prior to extubation, or to a control group (Lee, Chien, Hsu, et al., 1998). Patients undergoing oropharyngeal surgery were not enrolled. Laryngospasm was defined as occuring within 2 minutes of extubation, characterized by stridor, silence resulting from total closure of the vocal cords, and cyanosis. In patients in the acupuncture group, 2/38 (5.3 percent) developed laryngospasm, whereas 9/38 (23.7 percent) did in the control group. No patients required reintubation.

(7) This study (Jiang, Kao, and Wang, 1999) tested the use of NIPPV postextubation. Enrolled patients were 93 extubated individuals, 56 of whom were electively extubated and 37 of whom had unplanned extubations. Patients were randomized to either BiPAP delivering IPAP 12 cm H₂O and EPAP 4 cm H₂O, by face mask for up to 72 hours, temporarily removed for suctioning and eating, or to oxygen therapy. Patients in both arms had arterial blood gases measured 1 to 3 hours postextubation. BiPAP was terminated and patients were intubated if arterial blood gases deteriorated, or if labored breathing or hemodynamic stability developed. Extubation failure was defined as the need for reintubation as judged by the attending physician. Patients had similar preextubation blood gases. The oxygen group had 7/46 reintubations, whereas the BiPAP group had 13/47 reintubations (not significantly different). The postextubation management with or without NIPPV therefore did not influence outcome, however, compared with the elective extubation patients (6/56), the unplanned extubation patients were more likely to be reintubated (14/37).

Observational Studies Addressing Prediction of Successful Weaning and Duration of Ventilation

Summary ROC Curves (Figures 1-8)

Summary ROC curves deal with the problem of different thresholds among studies. We show the summary ROC curves for several predictors of two main outcomes: successful extubation (Figures 1-3) and successful discontinuation assessment and extubation (Figures 4-8).

Figure

Figure 1. Weaning predictor analysis: summary receiver operating characteristics curve Predictor: minute ventilation; outcome: successful extubation.

Figure

Figure 3. Weaning predictor analysis: summary receiver operating characteristics curve Predictor: rapid shallow breathing index (RSBI); outcome: successful extubation.

Figure

Figure 4. Weaning predictor analysis: summary receiver operating characteristics curve Predictor: minute ventilation; outcome: successful discontinuation assessment and extubation.

Figure

Figure 8. Weaning predictor analysis: summary receiver operating characteristics curve Predictor: maximal inspiratory pressure; outcome: successful discontinuation assessment and extubation.

Figure 1 displays the summary ROC curve for minute ventilation as a predictor of successful extubation. Figure 2 displays the ROC curve for respiratory rate as a predictor of successful extubation. Figure 3 displays the summary ROC curve for the rapid shallow breathing index as a predictor of successful extubation.

Figure

Figure 2. Weaning predictor analysis: summary receiver operating characteristics curve Predictor: respiratory rate; outcome: successful extubation.

Figures 4-8 show the strength of several variables predicting successful discontinuation assessment: minute ventilation (Figure 4), respiratory rate (Figure 5), tidal volume (Figure 6), rapid shallow breathing index (Figure 7), and PI_max (Figure 8).

Figure

Figure 5. Weaning predictor analysis: summary receiver operating characteristics curve Predictor: respiratory rate; outcome: successful discontinuation assessment and extubation.

Figure

Figure 6. Weaning predictor analysis: summary receiver operating characteristics curve Predictor: tidal volume; outcome: successful discontinuation assessment and extubation.

Figure

Figure 7. Weaning predictor analysis: summary receiver operating characteristics curve Predictor: rapid shallow breathing index (RSBI); outcome: successful discontinuation assessment and extubation.

Testing for the presence of a threshold for Figure 1, the regression coefficient is 0.23 (p value 0.38), indicateing that there is no evidence of a threshold effect for minute ventilation as a predictor of successful extubation. Similarly, testing for a threshold effect for Figure 2, the regression coefficient is 0.45 (p value 0.19), indicating that there is no evidence of a threshold effect for respiratory rate. The same holds for rapid shallow breathing index, for which the coefficient is -0.30 (p value 0.29).

There is no evidence of a threshold effect for any of the predictors of successful discontinuation assessment and extubation: minute ventilation (coefficient 0.09, p value 0.89), respiratory rate (coefficient 0.11, p value 0.66), tidal volume (coefficient 0.30, p value 0.38), rapid shallow breathing index (coefficient 0.18, p value 0.48), and PI_max (coefficient 0.02, p value 0.93).

ROC curves can be used to compare two or more different signs or tests. The one lying furthest to the top left corner is the most accurate; this value is where sensitivity and specificity are highest. Of these eight summary ROC curves, none appears to be much more powerful than the others. For the three predictors that are graphed twice (minute ventilation, respiratory rate, and rapid shallow breathing index, which predict both successful extubation and predict successful spontaneous breathing discontinuation and extubation), we did not test for significant differences between the summary ROC curves, since we did not have the recommended 10 studies represented in each ROC curve (Moses, Shapiro, and Littenberg, 1993). This rationale precluded our testing for significant differences across all summary ROC curves.

Qualitative Studies of Weaning From Mechanical Ventilation

In this section, we describe the results of the qualitative studies in this field, which offer insight into social, emotional, and experiential phenomena. In contrast, quantitative studies (such as epidemiologic investigations and clinical trials) aim to test well-specified hypotheses concerning a few predetermined variables. The first study in this section on qualitative research will be described in detail to explain how specific critical appraisal criteria can be used to interpret and understand qualitative research reports.

Observational Studies Describing Patient Experience During Weaning

We identified four observational studies of patient experience during weaning that did not directly address our research questions and did not allow data extraction in tables. However, given the importance of understanding patient experience during weaning, we summarize them in narrative form here. These studies provide quantitative information; in the prior section on qualitative studies, we have described complementary results from studies of patient experience conducted using qualitative methodology.

(1) The objective of the first study (Bouley, Froman, and Shah, 1992) was to compare patient experience during SIMV, T-piece, and pressure support weaning and relate these to physiologic variables. Nine COPD patients ventilated for 4 to 18 days were included in a nonrandomized crossover study; all patients were observed for four weaning trials using two weaning modes. Dyspnea was measured on a visual analog scale; and its properties of sensitivity, validity, and reliability in ventilated patients were reported. In addition, 20 observations each of heart rate, respiratory rate, minute ventilation, and oxygen saturation were made. Patients had both SIMV 4 and T-piece (N=6) or SIMV 8 and pressure support (N=3) periods.

Dyspnea was no different between modes. Physiologic measures and type of weaning did not predict dyspnea ratings using regression analysis. Individual patient regression analyses differed, however, such that different variables predicted the dyspnea ratings in different patients. Physiologic responses and subjective experience of dyspnea therefore appear to have distinct patterns across patients in this small study.

(2) The objective of the second study (Lowry and Anderson, 1993) was to learn about patients' feelings, their perceptions of hope and social support, and their notion of the locus of control during weaning, including how these change over the weaning process. Ten alert and oriented mechanically ventilated patients who were physiologically ready to wean and who had undergone two to five failed weaning attempts underwent extensive testing using several instruments. The instruments used included the Multidimensional Health Locus of Control Scales Form which uses Likert scales to measure issues of power and control, a Hope Scale measured using a visual analog scale, the Norbeck Social Support Questionnaire which contains a 9-item 5-point scale about support structures, and the Anderson-Lowry Ventilation Scale with 10 items about fear and other responses to mechanical ventilation. The validity and reliability data for each instrument were provided when possible. There were 2 interviews per patient, but only 4 of 10 patients completed both.

Mechanical ventilation was reported as a moderately fearful experience, and fear decreased over time. During ventilation, patients felt as though the locus of control was external to themselves, reflecting the intense dependence they have on the ICU team and family members. Hope increased as time passed since successful weaning, and hopelessness predominated for patients who continued to require mechanical ventilation. This study reported properties of the instruments used and captured some important experiences of patients while weaning-lack of a sense of mastery, hopelessness, and fear. As time passed and patients were weaned, the locus of control was internalized, and patients were more hopeful and less fearful.

(3) The objective of the third study (Pochard, Lanore, Bellivier, et al., 1995) was to evaluate psychological status in 43 consecutive patients who had undergone successful weaning 48 to 96 hours earlier using an interviewer-administered 32-item questionnaire with visual analog scales. Almost all patients had received opiates, benzodiazepines, or neuromuscular blockers during their ICU stay but had none for 48 hours before the interview. Of the 43 patients, 9 had suffered a cardiac arrest during their ICU stay and 9 had a prior psychiatric history. Interviewer training, validation, and technique are not reported. Results are reported in text format without means or standard deviations. Myriad difficult experiences were recorded, including an inability to communicate, sleep disorders, dreaming, diffuse anxiety, fear of abandonment by staff, and depression. Patients could not recall distinguishing between night and day (N=23), reported being confused during weaning (N=15), and had hallucinations (N=9).

(4) The objective of this study (Menzel, 1997) was to examine patients' responses to communication during intubation and 7 days afterwards, following extubation, and to relate these to situational and demographic variables. Among 29 oriented patients from 4 ICUs who were intubated for at least 24 hours, questions were asked using the Emotion Scale and the Ease of Communication Scale, for a total of 23 items each using a Likert scale. Validity and reliability data are reported on these scales when possible.

There were no significant differences between the intubation and postextubation data overall, but one-third of patients reported differences of 20 percent or more between time periods. There were no correlations with situational and demographic variables. Women tended to report more fear of being unable to speak postextubation than during intubation; men reported less fear. Patients recalling less difficulty with communication had shorter periods of intubation. The emphasis on this report was on correlations and pre-post comparisons; relatively sparse data describe patients' perceptions.