Alpha‐2 agonists for long‐term sedation during mechanical ventilation in critically ill patients
Abstract
Background
Sedation reduces patient levels of anxiety and stress, facilitates the delivery of care and ensures safety. Alpha‐2 agonists have a range of effects including sedation, analgesia and antianxiety. They sedate, but allow staff to interact with patients and do not suppress respiration. They are attractive alternatives for long‐term sedation during mechanical ventilation in critically ill patients.
Objectives
To assess the safety and efficacy of alpha‐2 agonists for sedation of more than 24 hours, compared with traditional sedatives, in mechanically‐ventilated critically ill patients.
Search methods
We searched the Cochrane Central Register of Controlled Trials (CENTRAL, Issue 10, 2014), MEDLINE (1946 to 9 October 2014), EMBASE (1980 to 9 October 2014), CINAHL (1982 to 9 October 2014), Latin American and Caribbean Health Sciences Literature (1982 to 9 October 2014), ISI Web of Science (1987 to 9 October 2014), Chinese Biological Medical Database (1978 to 9 October 2014) and China National Knowledge Infrastructure (1979 to 9 October 2014), the World Health Organization international clinical trials registry platform (to 9 October 2014), Current Controlled Trials metaRegister of controlled trials active registers (to 9 October 2014), the ClinicalTrials.gov database (to 9 October 2014), the conference proceedings citation index (to 9 October 2014) and the reference lists of included studies and previously published meta‐analyses and systematic reviews for relevant studies. We imposed no language restriction.
Selection criteria
We included all randomized and quasi‐randomized controlled trials comparing alpha‐2 agonists (clonidine or dexmedetomidine) versus alternative sedatives for long‐term sedation (more than 24 hours) during mechanical ventilation in critically ill patients.
Data collection and analysis
Two review authors independently assessed study quality and extracted data. We contacted study authors for additional information. We performed meta‐analyses when more than three studies were included, and selected a random‐effects model due to expected clinical heterogeneity. We calculated the geometric mean difference for continuous outcomes and the risk ratio for dichotomous outcomes. We described the effects by values and 95% confidence intervals (CIs). We considered two‐sided P < 0.05 to be statistically significant.
Main results
Seven studies, covering 1624 participants, met the inclusion criteria. All included studies investigated adults and compared dexmedetomidine with traditional sedatives, including propofol, midazolam and lorazepam. Compared with traditional sedatives, dexmedetomidine reduced the geometric mean duration of mechanical ventilation by 22% (95% CI 10% to 33%; four studies, 1120 participants, low quality evidence), and consequently the length of stay in the intensive care unit (ICU) by 14% (95% CI 1% to 24%; five studies, 1223 participants, very low quality evidence). There was no evidence that dexmedetomidine decreased the risk of delirium (RR 0.85; 95% CI 0.63 to 1.14; seven studies, 1624 participants, very low quality evidence) as results were consistent with both no effect and appreciable benefit. Only one study assessed the risk of coma, but lacked methodological reliability (RR 0.69; 95% CI 0.55 to 0.86, very low quality evidence). Of all the adverse events included, the most commonly reported one was bradycardia, and we observed a doubled (111%) increase in the incidence of bradycardia (RR 2.11; 95% CI 1.39 to 3.20; six studies, 1587 participants, very low quality evidence). Our meta‐analysis provided no evidence that dexmedetomidine had any impact on mortality (RR 0.99; 95% CI 0.79 to 1.24; six studies, 1584 participants, very low quality evidence). We observed high levels of heterogeneity in risk of delirium (I² = 70%), but due to the limited number of studies we were unable to determine the source of heterogeneity through subgroup analyses or meta‐regression. We judged six of the seven studies to be at high risk of bias.
Authors' conclusions
In this review, we found no eligible studies for children or for clonidine. Compared with traditional sedatives, long‐term sedation using dexmedetomidine in critically ill adults reduced the duration of mechanical ventilation and ICU length of stay. There was no evidence for a beneficial effect on risk of delirium and the heterogeneity was high. The evidence for risk of coma was inadequate. The most common adverse event was bradycardia. No evidence indicated that dexmedetomidine changed mortality. The general quality of evidence ranged from very low to low, due to high risks of bias, serious inconsistency and imprecision, and strongly suspected publication bias. Future studies could pay more attention to children and to using clonidine.
PICOs
Plain language summary
Dexmedetomidine and clonidine for long‐term sedation during mechanical ventilation in critically ill patients
Review question
We reviewed the evidence about the safety and efficacy of dexmedetomidine and clonidine (known as alpha‐2 agonists) for long‐term sedation during mechanical ventilation in critically ill patients in the intensive care unit (ICU).
Background
Sedation is an important treatment for critically ill patients who need a machine to support breathing, because it reduces anxiety and stress and facilitates the delivery of nursing care. However, some commonly‐used sedatives, such as propofol, midazolam and lorazepam, might decrease blood pressure, depress breathing, and delay awakening after a long‐term infusion. They may prolong breathing support time and length of stay in hospital. Dexmedetomidine and clonidine sedate but allow staff to interact with patients, and they ease pain but do not depress breathing. Those treated with them could be more easy to awake, and more able to communicate their discomfort and pain. These drugs are therefore attractive alternatives for long‐term sedation, and we planned to assess their efficacy and safety for long‐term (more than 24 hours) sedation, compared with traditional sedatives.
Study characteristics
We searched the databases until October 2014. We included seven randomized controlled trials, with a total of 1624 participants, comparing dexmedetomidine versus traditional sedatives. All the studies required participants to have an anticipated need for sedation of more than 24 hours. The alternative sedatives included propofol, midazolam and lorazepam. We found no eligible studies in children or for clonidine. Of the seven studies, six were funded by the drug manufacturer, and one did not state any conflict of interest.
Key results
Compared with traditional sedatives, dexmedetomidine reduced the breathing support time by approximately one‐fifth, and the length of stay time in ICU by one‐seventh. Dexmedetomidine was at least as effective as traditional sedatives for producing sedation and maintaining a light sedation level. There was no clear evidence in support of dexmedetomidine reducing the risk of delirium (a kind of acute confusion state), as results were consistent with both no effect and appreciable benefit. We had insufficient information to draw conclusions about reducing the risk of coma. Dexmedetomidine doubled the incidence of slow heartbeat, which was the most commonly reported adverse event. Our review provides no evidence that dexmedetomidine changed the overall death rate.
Quality of the evidence
The general quality of evidence ranged from very low to low, as most of the studies were at high risk of bias, serious inconsistency and imprecision, or strongly suspected publication bias.
Authors' conclusions
Summary of findings
| Dexmedetomidine compared to traditional sedative agents for long‐term sedation during mechanical ventilation in critically ill patients | ||||||
| Patient or population: Critically ill patients requiring long‐term sedation during mechanical ventilation | ||||||
| Outcomes | Illustrative comparative risks* (95% CI) | Relative effect | No of Participants | Quality of the evidence | Comments | |
| Assumed risk | Corresponding risk | |||||
| traditional sedative agents | Dexmedetomidine | |||||
| Duration of mechanical ventilation | Not estimable | Not estimable | Relative decrease 22% in the geometric mean (95% CI 10% to 33%) | 1120 | ⊕⊕⊝⊝ | Alpha‐2 agonists reduced geometric mean duration of mechanical ventilation by 0.25 (95% CI 0.10 to 0.40), corresponding to a reduction of 22% in the geometric mean (95% CI 10% to 33%). The relative change in geometric means usually produces similar results to the relative change in arithmetic means. |
| Risk of delirium | Study population | RR 0.85 | 1624 | ⊕⊝⊝⊝ | ‐ | |
| 266 per 1000 | 226 per 1000 | |||||
| Low | ||||||
| 76 per 1000 8 | 65 per 1000 | |||||
| High | ||||||
| 824 per 1000 8 | 700 per 1000 | |||||
| Risk of coma | Study population | RR 0.69 | 103 | ⊕⊝⊝⊝ | ‐ | |
| 922 per 1000 | 636 per 1000 | |||||
| Incidence of bradycardia | Study population | RR 2.11 | 1587 | ⊕⊝⊝⊝ | ‐ | |
| 90 per 1000 | 189 per 1000 | |||||
| Low | ||||||
| 39 per 1000 8 | 82 per 1000 | |||||
| High | ||||||
| 188 per 1000 8 | 397 per 1000 | |||||
| Duration of weaning | See comment | 85 | ‐ | Only one study assessed this outcome and only reported median and range. We were not able to estimate a relative effect. | ||
| ICU length of stay | Not estimable | Not estimable | Relative decrease 14% in the geometric mean (95% CI 0.01% to 24%) | 1223 | ⊕⊝⊝⊝ | Sedation using alpha‐2 agonists reduced geometric mean ICU LOS by 0.15 (95% CI 0.01 to 0.28), corresponding to a reduction of 14% in the geometric mean (95% CI 0.01% to 24%). Relative change in geometric means usually produces similar results to the relative change in arithmetic means. |
| Mortality | Study population | RR 0.99 | 1584 | ⊕⊝⊝⊝ | ‐ | |
| 205 per 1000 | 203 per 1000 | |||||
| Moderate | ||||||
| 192 per 1000 12 | 190 per 1000 | |||||
| *The basis for the assumed risk (e.g. the median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). | ||||||
| GRADE Working Group grades of evidence | ||||||
| 1Downgraded by 1 point due to all the studies being at high risk of bias, which was considered very serious. 2Downgraded by 1 point because all the studies received funding support from pharmaceutical firms and thus publication bias was strongly suspected. 3Downgraded by 1 point due to 6 of the 7 studies being at high risk of bias, which was considered serious. 4Downgraded by 1 point because the heterogeneity is substantial and thus inconsistency was serious. 5Downgraded by 1 point because results were consistent with both no effect and appreciable benefit, and thus imprecision was serious. 6Downgraded by 1 point because 6 of the 7 studies received funding support from pharmaceutical firms and thus publication bias was strongly suspected. 7Downgraded by 1 point because the sample size was small (less than 2000) and thus inconsistency was serious. 8The baseline risk varies broadly among studies, and thus we chose the lowest and highest control group risk in the included studies. 9Downgraded by 1 point due to 5 of the 6 studies being at high risk of bias, which was considered serious. 10Downgraded by 1 point because 5 of the 6 studies received funding support from pharmaceutical firms and thus publication bias was strongly suspected. 11Downgraded by 1 point due to wide confidence interval, although there may still be enough precision to make decisions about the utility of the intervention. 12There was little variation in the baseline risk across the studies, and thus we calculated the median control group risk. | ||||||
Background
Description of the condition
Sedation has become an important treatment in intensive care settings, especially for those undergoing mechanical ventilation. In the United States, half of patients with mechanical ventilation received intravenous sedation for more than 70% of their ventilation time (Wunsch 2009). Sedation could reduce levels of anxiety and stress, facilitate the delivery of nursing care, prevent self extubation and ensure synchrony with mechanical ventilation (Gehlbach 2002).
Most physicians choose propofol and benzodiazepines for long‐term sedation (Mehta 2011). Propofol is fast onset and offset, but decreases blood pressure, and may cause lipid metabolism disorder and fatal propofol infusion syndrome after long‐term use (Kam 2007; Miller 1998). Although benzodiazepines have little effect on blood pressure, they may cause apnoea and respiratory depression (Watling 1996). Benzodiazepines tend to be accumulated in body tissues and thus lead to delayed awakening (Shelly 1991). Furthermore, benzodiazepines might be associated with high risks of delirium and coma (Agarwal 2010; Pandharipande 2006; Pandharipande 2008;Watson 2008), which increase ventilation time, mortality and healthcare costs (Dubois 2001; Ely 2004).
Increasingly, many institutions use sedation strategies, such as nursing‐implemented sedation protocols and daily interruption to minimize the use of sedatives and limit the adverse events. These strategies reduce the dosage of sedatives, ventilation time and intensive care unit length of stay (ICU LOS) (Aitken 2012; Brook 1999; Burry 2014; Girard 2008), but lack of physicians' specific order and nursing acceptance have prevented them from being widely adopted. A survey of members of the Society of Critical Care Medicine showed that only 64% of respondents used sedation protocols, and 40% used interruption of sedation every day (Tanios 2009). Moreover, even when physicians applied sedation strategies, the choice of sedatives ‐ for example, using short‐acting, non‐benzodiazepine sedatives rather than benzodiazepines ‐ could still affect the ventilation time and hospital stay (Carson 2006; Pandharipande 2010). This is probably because these sedatives are easy to titrate and rapid to wake from, which in turn helps to finish sedation protocols and daily interruption.
Results from recent studies suggest that a strategy of light sedation could reduce ventilation time and ICU LOS (Shehabi 2012; Treggiari 2009). In addition, many investigators also recommend that any sedation strategy should incorporate analgesia, due to the difficulty of determining pain in critically ill patients (Fraser 2007) and the favourable results in shorter ventilation time, weaning time and ICU LOS (Breen 2005). A randomized controlled trial (RCT) found that when all the participants received intravenous analgesia (morphine), those with no infusion of sedatives had shorter ventilation time and ICU LOS, although more occurrences of delirium, than those with protocolized sedation and daily interruption (Strom 2010). The 'protocol of no sedation' in Strom 2010 was not really a protocol without any sedation, since morphine has a mild sedative effect, but it explored new challenges for sedation management: whenever possible, using the minimum level of sedation, providing adequate pain control and making a careful trade‐off between the level of sedation and the patient's comfort and safety. A sedative agent that provided lighter sedation, less pain and fewer adverse events would therefore be an attractive alternative.
Description of the intervention
Alpha‐2 agonists have a wide range of effects including sedation, analgesia and relief of anxiety (Mantz 2011; Pichot 2012). These effects are mainly mediated though adrenoceptors of the alpha‐2a subtype, which are distributed in the locus coeruleus (Gregoretti 2009; Mantz 2011). Unlike benzodiazepines and propofol, which inhibit the neurons in the central nervous system, alpha‐2 agonists reduce the activity but preserve the reactivity of neurons in the locus coeruleus. Alpha‐2 agonists therefore sedate but allow staff to interact with patients and do not suppress the respiratory drive. Alpha‐2 agonists might also be involved in neuroprotection and inflammatory responses, which may potentially help protect against the occurrence of delirium or coma (Mantz 2011). The adverse events of alpha‐2 agonists, such as bradycardia and hypotension, are mediated via adrenoceptors of the alpha‐2 in the medullary dorsal motor nucleus and motor complex, and are thus independent of sedative effect (Gregoretti 2009; Pichot 2012).
Clonidine and dexmedetomidine are two extensively‐studied alpha‐2 agonists, and both have similar mechanisms and adverse events. Although physicians have used clonidine to treat hypertension and acute withdrawal syndrome (Bohrer 1990; Gregoretti 2009), or have used it as an anaesthetic adjuvant (Dahmani 2010; Gregoretti 2009; Lambert 2014), its use for sedation remains 'off label' in many countries. Dexmedetomidine has a higher alpha‐2/alpha‐1 selectivity ratio (dexmedetomidine 1620:1, clonidine 220:1) (Virtanen 1988). The US Food and Drug Administration (FDA) authorized it for sedation during mechanical ventilation in 1999, and it can also be used as an analgesic adjuvant for acute postoperative pain control (Aho 1991; Jessen 2013). Dexmedetomidine is now available in the United States, Asia, the Middle East, Australia and Europe (Mantz 2011; Pichot 2012).
Clonidine is usually administered to adults intravenously at a dose of 2 mcg/kg per hour, or given as a bolus of 50 mcg to 150 mcg every eight hours (Ise 2002; Jamadarkhana 2010). The oral dose of clonidine is 50 mcg to 600 mcg every eight hours (Jamadarkhana 2010). The maintenance infusion rate of dexmedetomidine is 0.2 mcg/kg per hour to 0.7 mcg/kg per hour, for less than 24 hours (Precedex Prescribing Information).
How the intervention might work
Several RCTs showed alpha‐2 agonists for long‐term sedation reduced the duration of mechanical ventilation compared with traditional sedatives (Riker 2009; Ruokonen 2009; Spies 2003). Two aspects of alpha‐2 agonists might be related to this reduction: one could be that people treated with alpha‐2 agonists were more arousable, easier to communicate with and more able to express pain and needs (Martin 2003; Ruokonen 2009). The analgesia‐based sedation retained spontaneous respiration, and facilitated the delivery of care and the implementation of spontaneous breathing trials (Bailey 1991; Hall 2000; Hall 2001; Hsu 2004). A second aspect could be that alpha‐2 agonists were associated with fewer occurrence of delirium and coma (Pandharipande 2007; Riker 2009; Rubino 2010), probably because they work differently to propofol and benzodiazepines, and provide sleep‐like sedation (Huupponen 2008; Mason 2009).
Why it is important to do this review
Although some RCTs of long‐term sedation suggest that alpha‐2 agonists could reduce the duration of mechanical ventilation and delirium, other RCTs disagree: they found no improvement in ventilation time (Pandharipande 2007) and even a higher incidence of delirium (Ruokonen 2009). The FDA did not approve clonidine for sedation and only authorized dexmedetomidine infusions for less than 24 hours. There was therefore no way to know whether they were safe and effective for longer sedation. A previous meta‐analysis assessed the effects of dexmedetomidine in critically ill patients and found no difference in ventilation time or incidence of delirium compared with traditional sedatives or placebo (Tan 2010). But a number of RCTs have been published since its last search in 2009 (Jakob 2012 MIDEX; Jakob 2012 PRODEX; Shehabi 2013), and a comment on that meta‐analysis pointed out flaws in methodology (for example, unit‐of‐analyses error, confusion of agitation with delirium, and inadequate investigation of heterogeneity), which weaken the value of its results (Tejani 2010). They included RCTs investigating both short‐term and long‐term sedation, but most were short‐term, and thus provided limited information on long‐term safety and efficacy.
Despite the theoretical advantages of alpha‐2 agonists, their safety and efficacy for long‐term use remain controversial. We also consider that people undergoing long‐term ventilation and sedation are substantially different from those receiving short‐term treatment; they have a different disease spectrum, are more critically ill, need more dosage of sedatives, are more exposed to the adverse events of sedatives, and are more likely to wake up later. A comprehensive synthesis is therefore needed to assess whether alpha‐2 agonists are safe and whether they have advantages over traditional sedatives for long‐term sedation, by reducing the duration of mechanical ventilation and the risks of delirium and coma.
Objectives
To assess the safety and efficacy of alpha‐2 agonists for sedation of more than 24 hours, compared with traditional sedatives, in mechanically‐ventilated critically ill patients.
Methods
Criteria for considering studies for this review
Types of studies
We include all randomized and quasi‐randomized controlled trials. Quasi‐randomized controlled trials include those trials that use inappropriate randomization strategies (e.g. alternation, birth date, hospital registration number, etc.) (Higgins 2011a).
We exclude cluster‐randomized, cross‐over and non‐randomized trials.
Types of participants
We include randomized controlled trials (RCTs) of critically ill participants of any age, except neonates, who required intensive care and invasive mechanical ventilation when recruited.
We exclude RCTs in which participants only had non‐invasive mechanical ventilation.
Types of interventions
We include studies comparing clonidine or dexmedetomidine versus alternative sedative agents. The alternative sedatives included benzodiazepines, such as midazolam or diazepam, propofol and other sedatives.
We include studies which explicitly report that each participant was anticipated to need sedation for more than 24 hours. There was no limitation for dose, frequency or route of administration of alpha‐2 agonists.
We exclude studies in which all participants were sedated for less than 24 hours or studies without an explicit statement that they required or anticipated participants to have sedation for more than 24 hours . We exclude studies that compare one alpha‐2 agonist with another.
Types of outcome measures
Primary outcomes
-
Duration of mechanical ventilation (as defined by authors of the studies).
-
Risk of delirium, measured with any diagnostic criteria, such as the confusion assessment method for the intensive care unit (CAM‐ICU).
-
Risk of coma (as defined by authors of the studies).
Secondary outcomes
-
Adverse events:
-
incidence of bradycardia (as defined by authors of the studies);
-
incidence of hypotension (as defined by authors of the studies);
-
incidence of hypertension (as defined by authors of the studies);
-
any other adverse events.
-
-
Proportion of sedation time spent at target sedation level (as defined by authors of the studies).
-
Duration of weaning (time from weaning to extubation as defined by authors of the studies).
-
intensive care unit length of stay (ICU LOS).
-
Mortality; if mortality is assessed at various follow‐up times (e.g. ICU mortality, 30‐day mortality, one‐year mortality), we will include only the one that most closely approximates to that used in the other included studies.
Search methods for identification of studies
Electronic searches
We searched the Cochrane Central Register of Controlled Trials (CENTRAL, The Cochrane Library, Issue 10, 2014; see Appendix 1 for detailed search strategy), MEDLINE (Ovid SP, 1946 to 9 October 2014; see Appendix 2), EMBASE (Ovid SP, 1980 to 9 October 2014; see Appendix 3), CINAHL (EBSCO host, 1982 to 9 October 2014; see Appendix 4), Latin American and Caribbean Health Sciences Literature (LILACS, via BIREME interface, 1982 to 9 October 2014; see Appendix 5), ISI Web of Science (1987 to 9 October 2014; see Appendix 6), Chinese Biological Medical Database (CBM, 1978 to 9 October 2014; see Appendix 7) and China National Knowledge Infrastructure (CNKI, 1979 to 9 October 2014; see Appendix 8) for relevant studies.
We imposed no language and publication status restrictions. However, we excluded studies without detailed methodological information. Detailed methodological information means that the information obtained from full copies, abstracts or study authors should be sufficient to enable us to make judgements on the overall study quality.
Searching other resources
We searched for unpublished or ongoing trials on the World Health Organization international clinical trials registry platform WHO ICTRP (www.who.int/ictrp/search/en/), Current Controlled Trials metaRegister of controlled trials (www.isrctn.com/page/mrct) active registers, and the clinicaltrials.gov database.
We also searched the conference proceedings citation index.
We handsearched the reference lists of included studies and previously published meta‐analyses and systematic reviews for relevant studies (Lin 2012; Tan 2010; Xia 2013; Zhuo 2012). We contacted the first author and the corresponding author of the included studies for information on other published and unpublished studies.
Data collection and analysis
Selection of studies
Two review authors (KC and YCX) independently scanned each title and abstract retrieved by the search strategy. We excluded irrelevant studies in this step. We excluded studies in abstract only and where no detailed methodological information could be obtained from the study authors. We obtained full copies of the remaining studies for further assessment. Our methodologist (YIC) resolved any disagreements unresolved by consensus. We used Endnote X7 to manage the search results and the articles. We noted the reasons for exclusion in Endnote X7. We were not blinded to the study authors, institutions or journals.
Data extraction and management
Two review authors (KC and YCX) independently extracted data from the articles using a modified version of the Cochrane Anaesthesia Review Group's (CARG) data extraction form (Appendix 9). We selected three articles randomly to test our form before formal extraction. We exported data from these articles to doc.google.com. Our statistician (YC) resolved inconsistent data extraction between the two review authors.
Assessment of risk of bias in included studies
Two review authors (KC and YCX) independently assessed the risk of bias using the Cochrane Collaboration’s tool (Higgins 2011b) in the following seven key domains:
-
Random sequence generation;
-
Allocation concealment;
-
Blinding of participants and personnel;
-
Blinding of outcome assessment;
-
Incomplete outcome data;
-
Selective reporting;
-
Other bias (whether the groups were balanced at baseline, whether the co‐interventions were similar, etc.).
We assessed blinding for each outcome as recommended. We judged each domain as 'low risk', 'high risk' or 'unclear risk'. We noted the summary of known facts and the source of this information for each domain. We asked the authors of the study for clarification if any details were unclear.
We tested the 'Risk of bias' assessment tool on some similar studies. When KC and YCX disagreed, we consulted our methodologist (YIC) and content specialists (SMP and ZJL). After reaching consensus, we classified each included study into one of three categories, as follows.
-
Low risk of bias: low risk of bias for all key domains.
-
Unclear risk of bias: unclear risk of bias for one or more key domains.
-
High risk of bias: high risk of bias for one or more key domains.
We generated a 'Risk of bias' graph and a 'Risk of bias' summary. We also reported the risk of selective outcome reporting in the Results section under Assessment of risk of bias in included studies.
We performed the kappa test to evaluate the interobserver reliability between two review authors (KC and YCX).
Measures of treatment effect
We calculated the geometric mean difference (GMD) for continuous outcomes and the risk ratio (RR) for dichotomous outcomes. We describe the effects by their values and 95% confidence intervals (CIs). We considered two‐sided P < 0.05 to be statistically significant.
Unit of analysis issues
We expected all the trials to be simple parallel‐group designs. The unit of analysis was a single measurement for each outcome from each participant.
Dealing with missing data
We tried to contact the study authors for missing data and further information twice by e‐mail. For those missing data that could not be obtained from authors, we did the following:
-
If we judged missing data to be 'missing at random', we only analysed the available data. We categorized the data as 'missing at random' if the loss was unrelated to the actual values of the missing data (Higgins 2011a). For example, if participants died before receiving any study drugs.
-
If we judged missing data to be 'not missing at random' and meta‐analysis was not feasible, we performed a qualitative synthesis.
-
If we could systematically meta‐analyse missing data, we imputed average measurements for continuous outcomes and we inferred bad outcomes for categorical outcomes.
Assessment of heterogeneity
We measured heterogeneity using both the Chi² statistic and the I² statistic. Values for I² of 25%, 50% and 75% corresponded to low, moderate and high degrees of heterogeneity, respectively (Higgins 2003). We considered there to be substantial heterogeneity when a P value of Chi² < 0.10 or I² > 50%, or both (moderate and high degrees of heterogeneity). If we identified substantial heterogeneity, we checked the data entered into Review Manager 5 (RevMan 5.3). If I² exceeded 75%, we did not perform meta‐analyses.
Assessment of reporting biases
We had planned to generate funnel plots if there were more than 10 studies included in the meta‐analyses of primary outcomes (Higgins 2011a). For continuous outcomes, we would have conducted the Egger test to assess funnel plot asymmetry; for dichotomous outcomes, we would have conducted the arcsine test (Sterne 2011). We would have performed all tests using STATA/SE 12.0.
Data synthesis
We performed meta‐analyses only if we could include more than three studies. Otherwise, we undertook a qualitative synthesis. We selected a random‐effects model due to expected clinical heterogeneity. We used the Mantel‐Haenszel method to estimate dichotomous outcomes and the inverse variance method to estimate continuous outcomes. We did not perform a meta‐analysis for 'Proportion of sedation time spent at target sedation level' because there was no consistent outcome measure and statistical methods for pooling such percentage rates are not well developed (Higgins 2011a). We would have synthesized the data of adult studies and paediatric studies separately if both data were available. We performed all statistical analyses using Review Manager 5 software (RevMan 5.3).
The data for the durations of mechanical ventilation, weaning and ICU LOS are always skewed. We performed meta‐analyses on log‐transformed data. If we could obtain means and standard deviations of raw data without log‐transformation, we conducted a transformation using Method 1 mentioned in Higgins 2008. If means and standard deviations of raw data were not available, we estimated log‐transformed means using log‐transformed medians (Hozo 2005) and calculated approximate standard deviations by dividing the log‐transformed interquartile range (log‐transformed third quartile minus first quartile) by 1.35 (Higgins 2011b).
We did not make any assumptions about the distribution of censored data from time‐to‐event data (duration of mechanical ventilation and ICU LOS). Some of the studies analysed time‐to‐event data using, for instance, Cox's proportional‐hazards regression model and Kaplan‐Meier survival analyses. We undertook a narrative synthesis for these adjusted outcomes.
Subgroup analysis and investigation of heterogeneity
If there were adequate data we performed subgroup analyses as follows.
For duration of mechanical ventilation and duration of weaning
-
Dexmedetomidine compared with clonidine.
-
Class of alternative agents, classified according to the 'Anatomical Therapeutic Chemical' (ATC) classification system advocated by the World Health Organization’s collaborating centre for drugs statistics methodology.
-
Sedation with or without analgesics/pain control.
For risk of delirium
-
Different delirium scores used.
-
Class of alternative agents.
For risk of coma and adverse events
-
Class of alternative agents.
-
Dexmedetomidine compared with clonidine.
-
Rapid‐dosing regimens compared with slow‐dosing regimens, where rapid‐dosing regimens include intravenous bolus and continuous intravenous infusion with initial loading infusion, and where slow‐dosing regimens include continuous intravenous infusion without initial loading infusion and oral or nasogastric administration.
For ICU LOS and mortality
-
Class of alternative agents.
-
Dexmedetomidine compared with clonidine.
-
Type of ICU (e.g. mixed ICU, medical ICU, surgical ICU, coronary care unit (CCU), respiratory care unit, burn unit).
We determined whether there were differences of effect size among subgroups based on the P values from tests for subgroup differences (Higgins 2011a). If differences existed, we compared the direction and magnitude of effect size in these subgroups with each other.
If there were adequate data, we performed random‐effects meta‐regression to investigate how mean ventilation time (log‐transformed) of the included studies was associated with the risk of delirium (Thompson 2002). We performed meta‐regression using R 3.0.2 with 'Metafor' package (version 1.9‐2). We used a restricted maximum‐likelihood estimator method to calculate the amount of heterogeneity (Viechtbauer 2010). We determined whether this difference was statistically significant based on the P value of the coefficients.
Sensitivity analysis
If there were sufficient studies, we undertook sensitivity analyses by excluding studies as follows:
-
Studies at high or unclear risk of bias;
-
Studies of a non‐standardized design: RCTs without a target sedation level or using non‐intravenous administration.
We also undertook sensitivity analyses by pooling observed data only, without any imputation.
Summary of findings
We used the principles of the GRADE system (Guyatt 2008) to assess the quality of the body of evidence associated with specific outcomes: duration of mechanical ventilation, rate of delirium, rate of coma, rate of bradycardia, duration of weaning, ICU length of stay and mortality. We constructed a 'Summary of findings' table using GRADE software. The GRADE approach appraised the quality of a body of evidence based on the extent to which one can be confident that an estimate of effect or association reflects the item being assessed. The quality of a body of evidence is derived from within‐study risk of bias (methodological quality), the directness of the evidence, heterogeneity of the data, precision of effect estimates, and risk of publication bias.
Results
Description of studies
Results of the search
Running the search strategy of electronic databases yielded 3009 citations: 1348 citations from western databases and 1661 citations from Chinese databases. We excluded 2956 citations based on abstract content alone. Three citations were in abstract only, and we failed to retrieve further information or to trace the original investigators. We performed full‐text review of 53 citations, and finally included six citations in our data synthesis (and one citation in Studies awaiting classification). One included citation reported on two independently‐conducted multi‐centre, randomized, double‐blind studies, the MIDEX trial (clinicaltrials.gov Identifier: NCT00481312) and the PRODEX trial (clinicaltrials.gov Identifier: NCT00479661). We considered them as two discrete studies during data synthesis (Jakob 2012 MIDEX: MIDEX trial; Jakob 2012 PRODEX: PRODEX trial). We found two potentially relevant unpublished/ongoing studies from clinical trial registries and emailed the contact investigators (NCT01059929; NCT01760967). One replied (NCT01760967), but provided no useful information (Characteristics of ongoing studies table shows detailed information about the ongoing studies). We did not obtain any information about unpublished or ongoing randomized controlled trials (RCTs) from authors of the included studies. Figure 1 presents the entire study selection process.
Selection process for studies included in data synthesis.
Included studies
See Characteristics of included studies.
We include seven studies in our review. Six studies were of a two‐arm design. One used a four‐arm design that compared dexmedetomidine infusion alone, midazolam infusion alone, dexmedetomidine infusion combined with midazolam infusion and dexmedetomidine infusion combined with midazolam bolus (Xu 2012). We extracted the data only from the dexmedetomidine‐infusion‐alone group and the midazolam‐infusion‐alone group. Five of the included studies were international multi‐centre (Jakob 2012 MIDEX; Jakob 2012 PRODEX; Riker 2009; Ruokonen 2009; Shehabi 2013); one was multi‐centre, conducted in the United States (Pandharipande 2007); and one was single‐centre, conducted in China (Xu 2012). Six studies received funding support from pharmaceutical firms which manufacture dexmedetomidine (Jakob 2012 MIDEX; Jakob 2012 PRODEX; Pandharipande 2007; Riker 2009; Ruokonen 2009; Shehabi 2013), and one did not state any potential conflict of interest (Xu 2012).
Types of participants
There were 1624 participants meeting the inclusion criteria. The sample sizes of studies ranged from 37 to 501. All the studies recruited adults from mixed intensive care units (ICUs) which may include medical, surgical and trauma patients. All the studies required participants to have an anticipated need for sedation of at least 24 to 36 hours. The main exclusion criteria of the studies include conditions as follows: neurological disease, severe bradycardia, second or third degree atrioventricular conduction block, hepatic impairment and pregnancy.
Types of interventions
We found no eligible studies for clonidine. All the included studies compared dexmedetomidine with traditional sedatives. The traditional sedatives could be midazolam (Jakob 2012 MIDEX; Riker 2009; Xu 2012), lorazepam (Pandharipande 2007), propofol (Jakob 2012 PRODEX), or standard care (either propofol or midazolam) (Ruokonen 2009; Shehabi 2013). A total of 871 participants (53.6%) received dexmedetomidine, 393 (24.2%) midazolam, 51 (3.1%) lorazepam, 249 (15.3%) propofol or 60 (3.7%) standard care. All studies administered sedative agents intravenously. The infusion rate of dexmedetomidine varied among trials, ranging from 0.15 to 1.5 mcg/kg per hour. Only one study routinely administered loading doses of dexmedetomidine (Xu 2012).
Only four studies explicitly stated that participants might receive sedation prior to the study treatment (Jakob 2012 MIDEX; Jakob 2012 PRODEX; Pandharipande 2007; Riker 2009): three reported that the sedation scores at the time of study treatment initiation were similar between groups (Jakob 2012 MIDEX; Jakob 2012 PRODEX; Pandharipande 2007), and one started treatment only when the participants were within a prespecified range of sedation score (Riker 2009). Other studies reported no detailed information about sedation prior to the study treatment (Ruokonen 2009; Shehabi 2013; Xu 2012). All the studies stated that they titrated sedatives to a target sedation level: Jakob 2012 MIDEX and Jakob 2012 PRODEX set the target as a Richmond Agitation‐Sedation Scale (RASS) score of 0 to ‐3 and assessed it every two hours; Pandharipande 2007 set it as a "clinically individualized target" based on RASS and assessed it twice daily; Riker 2009 set it as a RASS score of ‐2 to 1 and assessed it every four hours; Ruokonen 2009 set it as a RASS score of ‐4 to 0 and assessed it every four hours; Shehabi 2013 set it as a RASS score of ‐2 to 1 and assessed it every four hours; Xu 2012 set it as a Motor Activity Assessment Scale (MAAS) score of 3 and assessed it every two hours. Three studies performed daily interruption as part of the study protocol (Jakob 2012 MIDEX; Jakob 2012 PRODEX; Ruokonen 2009). Six studies routinely gave opioids to treat pain during sedation (Jakob 2012 MIDEX; Jakob 2012 PRODEX; Pandharipande 2007; Riker 2009; Ruokonen 2009; Shehabi 2013) and one study did not report that they used analgesics (Xu 2012). Two studies used intravenous haloperidol to control delirium (Riker 2009; Shehabi 2013), and the others did not report any information about delirium treatment. Four studies stopped sedation after extubation (Jakob 2012 MIDEX; Jakob 2012 PRODEX; Pandharipande 2007; Riker 2009); two studies stopped it when the investigators thought it was no longer required (Riker 2009; Shehabi 2013); one study did not explicitly state the indication for sedation discontinuation (Xu 2012).
Types of outcome measures
Five studies reported on duration of mechanical ventilation (Jakob 2012 MIDEX; Jakob 2012 PRODEX; Riker 2009; Ruokonen 2009; Shehabi 2013). All the studies reported on the risk of delirium, based on the confusion assessment method for the intensive care unit (CAM‐ICU). Three studies explicitly described the frequency of delirium assessment: Riker 2009 assessed it daily during sedation interruption; Ruokonen 2009 assessed it daily; Shehabi 2013 assessed it when the RASS score was more than ‐3. Only two studies explicitly described the follow‐up period of delirium assessment: Pandharipande 2007 assessed it until hospital discharge or for 12 days; Riker 2009 assessed it during study treatment. One study reported on the risk of coma (Pandharipande 2007). The investigators defined coma as a RASS score of ‐4 or ‐5 and assessed it until hospital discharge or for 12 days.
Excluded studies
Of the 46 studies excluded during full‐text review, 44 were because they did not explicitly report the required or anticipated duration of sedation or did not meet the criteria, one because it was not an RCT and one because not all the participants were mechanically ventilated. The Characteristics of excluded studies table provides detailed reasons for the exclusions.
Risk of bias in included studies
Allocation
Five studies had adequate sequence generation (Jakob 2012 MIDEX; Jakob 2012 PRODEX; Pandharipande 2007; Riker 2009; Shehabi 2013). Two studies reported they used stratified randomization, but did not state in detail how they randomized the participants (Ruokonen 2009; Xu 2012). We contacted the authors of Ruokonen 2009 and Xu 2012 for additional methodological information, but no‐one replied. We therefore judged them to be at 'unclear risk of bias for random sequence generation. Five studies had adequate allocation concealment (Jakob 2012 MIDEX; Jakob 2012 PRODEX; Pandharipande 2007; Riker 2009; Shehabi 2013). Two studies did not state that they concealed the randomization sequence (Ruokonen 2009; Xu 2012), and we therefore judged them to be at 'high' risk of bias for allocation concealment. The two review authors were in complete agreement for this domain.
Figure 2 and Figure 3 illustrate the proportion of studies with those judgements and the judgement made for each study. See the Characteristics of included studies for details of those judgements. .
Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.
Risk of bias summary: review authors' judgements about each risk of bias item for each included study.
Blinding
Four studies used an adequate double‐blinded design and described the details of the blinding method (Jakob 2012 MIDEX; Jakob 2012 PRODEX; Pandharipande 2007; Riker 2009). One study used a double‐dummy design, but the blinding method was brittle and we believe that the study investigators could easily have broken the blinding (Ruokonen 2009). One study did not mention that the participants or the investigators were blinded, and therefore probably did not perform blinding (Xu 2012). One study stated that they were "unblinded" (Shehabi 2013). We believed blinding had little impact on these objective outcomes, and we therefore judged Ruokonen 2009, Xu 2012 and Shehabi 2013 to be at 'low' risk for blinding of outcome assessment in objective outcomes, but at 'high' risk for blinding of participants and personnel and blinding of outcome assessment in subjective outcomes. Agreement between the two review authors for blinding of participants and personnel was at 86% (kappa = 0.70). The two review authors were in complete agreement for blinding of outcome assessment, both subjective and objective outcomes.
Figure 2 and Figure 3 illustrate the proportion of studies with those judgements and the judgement made for each study. Characteristics of included studies showed the details for the judgements.
Incomplete outcome data
Five studies had low withdrawal rates, and reasons for withdrawal were explicitly reported and balanced across groups (Pandharipande 2007; Riker 2009; Ruokonen 2009; Shehabi 2013; Xu 2012). However, in Jakob 2012 MIDEX and Jakob 2012 PRODEX, withdrawal due to lack of efficacy was significantly more frequent in dexmedetomidine participants and would therefore inevitably bias the results. We judged them to be at 'high' risk of bias. The two review authors were in complete agreement on this domain.
Figure 2 and Figure 3 illustrate the proportion of studies with those judgements and the judgement made for each study. Characteristics of included studies showed the details for the judgements.
Selective reporting
There was no indications of selective reporting in any of the included studies. The two review authors were in complete agreement for this domain.
Other potential sources of bias
Six studies appeared to be free of other sources of bias (Jakob 2012 MIDEX; Jakob 2012 PRODEX; Riker 2009; Ruokonen 2009; Shehabi 2013; Xu 2012). One study used continuous infusion of lorazepam as a comparison intervention (Pandharipande 2007). Because of the long half‐life of lorazepam, such a study design might increase the risk of coma in the control group, and we suspected it to be a potential source of bias. Agreement between the two review authors for other potential sources of bias was at 86% (kappa = 0.70).
Figure 2 and Figure 3 illustrate the proportion of studies with those judgements and the judgement made for each study. Characteristics of included studies showed the details for the judgements.
Effects of interventions
Duration of mechanical ventilation (Analysis 1.1; Analysis 1.2)
Four studies (1448 participants) compared the duration of mechanical ventilation (Jakob 2012 MIDEX; Jakob 2012 PRODEX; Riker 2009; Ruokonen 2009), and showed the median time of ventilation in the dexmedetomidine group was less than control. Jakob 2012 MIDEX, Jakob 2012 PRODEX and Ruokonen 2009 analysed the data using Cox’s proportional‐hazards regression and Riker 2009 analysed it using Kaplan‐Meier survival analysis. Three of them were statistically significant (Jakob 2012 MIDEX; Jakob 2012 PRODEX; Riker 2009), but one was not (Ruokonen 2009) (calculations from median/interquartile range to geometric mean/standard deviation (SD) are shown in Table 1). Riker 2009 reported the median but no interquartile range. We were unable to obtain the raw data, and thus could not estimate the standard deviation and add the data into our meta‐analysis. We obtained data for Shehabi 2013 from the authors and pooled it with the other three studies (Jakob 2012 MIDEX; Jakob 2012 PRODEX; Ruokonen 2009). Meta‐analysis of 1119 participants yielded a random‐effects estimate of the geometric mean difference (MD) of ‐0.25 log hours (95% confidence interval (CI) ‐0.40 to ‐0.10, P = 0.001), corresponding to a reduction of 22% in the geometric mean (95% CI 10% to 33%). Neither the I² nor the Chi² statistics indicated evidence of statistical heterogeneity across the trials, (I² = 0%; Chi² = 0.42, P = 0.94).
| Study ID | Outcome | Study group | Median | Interquartile range | Geometric mean | Geometric SD |
| Duration of mechanical ventilation | Intervention | 123 hours | 67 ‐ 337 hours | 4.812 | 1.197 | |
| Control | 164 hours | 92 ‐ 380 hours | 5.1 | 1.051 | ||
| Duration of mechanical ventilation | Intervention | 97 hours | 45 ‐ 257 hours | 4.575 | 1.291 | |
| Control | 118 hours | 48 ‐ 327 hours | 4.771 | 1.421 | ||
| Duration of mechanical ventilation | Intervention | 77.2 hours | 17.5 – 338.8 hours | 4.346 | 2.195 | |
| Control | 110.6 hours | 20.1 – 675.0 hours | 4.706 | 2.603 | ||
| ICU LOS | Intervention | 8.8 days (211 hours) | 4.8 ‐ 34.6 days (115 ‐ 831 hours) | 2.174 | 1.465 | |
| Control | 10.1 days (243 hours) | 5.8 ‐ 26.2 days (140 ‐ 630 hours) | 2.315 | 1.114 | ||
| ICU LOS | Intervention | 6.8 days (164 hours) | 3.8 ‐ 20.0 days (90 ‐ 480 hours) | 1.922 | 1.24 | |
| Control | 7.7 days (185 hours) | 3.9 ‐ 21.7 days (93 ‐ 520 hours) | 2.042 | 1.275 | ||
| ICU LOS | Intervention | 7.5 days | 5 ‐ 19 days | 2.015 | 0.989 | |
| Control | 9 days | 6 ‐ 15 days | 2.197 | 0.679 | ||
| ICU LOS | Intervention | 5.5 days | 1.7 – 19.5 days | 1.705 | 1.807 | |
| Control | 5.7 days | 1.7 – 29.0 days | 1.74 | 2.101 |
ICU: intensive care unit
LOS: length of stay
We performed subgroup analysis for different classes of alternatives. The subgroups were small, with one or two studies in each. The results from all subgroups favoured dexmedetomidine, and we detected no significant difference between them (P = 0.84). We did not perform subgroup analyses for dexmedetomidine compared with clonidine or sedation with or without pain control, because of inadequate data.
Jakob 2012 MIDEX and Jakob 2012 PRODEX reported not only 'duration of mechanical ventilation', which included non‐invasive ventilation time, but also 'time to extubation' which also matched our definition. We performed a post hoc sensitivity analysis to assess how the result would be changed if we extracted the data from 'time to extubation' instead of 'duration of mechanical ventilation'. The sensitivity analysis yielded a geometric mean difference of ‐0.34 log hours (95% CI ‐0.49 to ‐0.20, P < 0.00001), corresponding to a reduction of 29% in the geometric mean (95% CI 18% to 39%). We did not undertake sensitivity analyses by excluding studies at high risk of bias or studies with non‐standardized design because of the small number of studies.
We were not able to conduct the Egger test because of the small number of studies, but we suspected reporting bias for this outcome, because these four studies received funding support from pharmaceutical firms which manufacture dexmedetomidine.
Risk of delirium (Analysis 1.3; Analysis 1.4)
The data for meta‐analysis were available from all the studies, with 1624 participants (Jakob 2012 MIDEX; Jakob 2012 PRODEX; Pandharipande 2007; Riker 2009; Ruokonen 2009; Shehabi 2013; Xu 2012). The risk of delirium was numerically lower in the dexmedetomidine group than in the control group (risk ratio (RR) 0.85; 95% CI 0.63 to 1.14), but was not statistically significant (P = 0.27). The heterogeneity was high across the trials (I² = 70%).
We performed subgroup analyses for different classes of alternative agents. The number of studies within subgroups ranged from one to four, and the difference in effect size was significant among subgroups (P = 0.02). Pooled results from subgroups showed the risk of delirium was lower in participants with dexmedetomidine than in those with benzodiazepine derivatives (RR 0.81; 95% CI 0.59 to 1.09; 1007 participants) or propofol (RR 0.37; 95% CI 0.16 to 0.87; 495 participants), but higher than in participants with standard care (RR 1.44; 95% CI 0.86 to 2.41; 122 participants). There was a high level of heterogeneity in the subgroup of benzodiazepine derivatives (I² = 72%), but a low level in the subgroup of standard care (I² = 6%). Because all the studies assessed delirium based on the confusion assessment method for the intensive care unit (CAM‐ICU), subgroup analyses for different delirium scores were not possible.
We performed meta‐regression on log‐transformed mean ventilation time using data from six studies (Jakob 2012 MIDEX; Jakob 2012 PRODEX; Pandharipande 2007; Riker 2009; Ruokonen 2009; Shehabi 2013) (Table 2). The meta‐regression did not show that the log‐transformed mean ventilation time was associated with the risk of delirium (P = 0.94) and was unable to explain the heterogeneity (R² = 0.00%).
| tau² (estimated amount of residual heterogeneity) | 0.1460 (SE = 0.1596) |
| tau (square root of estimated tau² value) | 0.3820 |
| I² (residual heterogeneity / unaccounted variability) | 70.09% |
| H² (unaccounted variability / sampling variability) | 3.34 |
| R² (amount of heterogeneity accounted for) | 0.00% |
| Intercept | 0.0895 (95% CI ‐5.4218 to 5.6007, P = 0.9746) |
| Coefficient for log‐transformed mean ventilation time | ‐0.0442 (95% CI ‐1.1989 to 1.1105, P = 0.9403) |
We undertook a sensitivity analysis by pooling observed data without any imputation, and confirmed a nearly identical result (RR 0.84; 95% CI 0.63 to 1.14, P = 0.27). We did not undertake sensitivity analyses by excluding studies at high risk of bias or studies with non‐standardized design, due to the small number of studies.
Two studies also reported on the number of delirium‐free days during study treatment (Riker 2009) or a 12‐day period (Pandharipande 2007). Riker 2009 showed participants treated with dexmedetomidine had more delirium‐free days than those treated with midazolam (2.5 days vs 1.7 days, P = 0.02), but Pandharipande 2007 did not (dexmedetomidine: 9 days, lorazepam: 7 days, P = 0.09).
We were not able to conduct the arcsine test because of the small number of studies, but we suspected reporting bias for this outcome because six studies received funding support from pharmaceutical firms which manufacture dexmedetomidine (Jakob 2012 MIDEX; Jakob 2012 PRODEX; Pandharipande 2007; Riker 2009; Ruokonen 2009; Shehabi 2013).
Risk of coma
Because only one study investigated the risk of coma (Pandharipande 2007; 103 participants), meta‐analysis was impossible. In Pandharipande 2007, the risk of coma was 63% in the dexmedetomidine group and 92% in the lorazepam group (P < 0.001). Participants with dexmedetomidine also had more coma‐free days than those in the lorazepam group (10 days vs 8 days, P < 0.001).
We were not able to conduct the arcsine test, but we suspected reporting bias for this outcome because Pandharipande 2007 received funding support from pharmaceutical firms which manufacture dexmedetomidine.
Adverse events
Incidence of bradycardia (Analysis 1.5; Analysis 1.6; Analysis 1.7)
Six studies, covering 1587 participants, reported on the incidence of bradycardia (Jakob 2012 MIDEX; Jakob 2012 PRODEX; Pandharipande 2007; Riker 2009; Ruokonen 2009; Xu 2012). Compared with traditional sedatives, those participants allocated to dexmedetomidine had a 111% significantly higher incidence of bradycardia (RR 2.11; 95% CI 1.39 to 3.20, P = 0.0004). The statistical heterogeneity was moderate across the trials (Chi² = 8.28, P = 0.14; I² = 40%).
We conducted two subgroup analyses, one for the class of alternative agents and one for the dosing regimens:
For subgroup analyses of different classes of alternative agents, the difference in effect size was significant (P = 0.04). Although all the pooled results from subgroups showed a higher incidence associated with dexmedetomidine, the standard care group had the highest risk ratio (RR 7.50; 95% CI 0.40 to 140.91; 85 participants) and the propofol group had the lowest (RR 1.20; 95% CI 0.74 to 1.94; 495 participants).
For subgroup analyses of different dosing regimens (rapid or slow), no between‐groups difference was detected (P = 0.32).
We undertook a sensitivity analysis by pooling the observed data without any imputation, and confirmed almost the same result (RR 2.11; 95% CI 1.39 to 3.21). We did not undertake sensitivity analyses by excluding studies at high risk of bias or studies with non‐standardized designs, due to the small number of studies.
Incidence of hypotension (Analysis 1.8; Analysis 1.9)
Six studies, covering 1587 participants, reported on the incidence of hypotension (Jakob 2012 MIDEX; Jakob 2012 PRODEX; Pandharipande 2007; Riker 2009; Ruokonen 2009; Xu 2012). The risk of hypotension was higher with dexmedetomidine, but this difference was not statistically significant (RR 1.22; 95% CI 0.86 to 1.74, P = 0.26). The level of heterogeneity was high (I² = 53%, P = 0.07).
We conducted two subgroup analyses, one for the class of alternative agents and one for the dosing regimens, but none of them indicated differences in effect size between subgroups (class of alternative agents, P = 0.57; dosing regimens, P = 0.17).
We undertook a sensitivity analysis by pooling the observed data without any imputation, and confirmed almost the same result (RR 1.22; 95% CI 0.86 to 1.73, P = 0.27). We did not undertake sensitivity analyses by excluding trials at high risk of bias or trials with non‐standardized designs, due to the small number of trials.
Incidence of hypertension
Three trials, covering 1356 participants, reported on the incidence of hypertension (Jakob 2012 MIDEX; Jakob 2012 PRODEX; Riker 2009). None of the trials reported a difference between dexmedetomidine and alternatives: in Jakob 2012 MIDEX, the incidence of hypertension was 21.5% in the dexmedetomidine group and 20.8% in the midazolam group (P = 0.913). In Jakob 2012 PRODEX, the incidence of hypertension was 21.1% in the dexmedetomidine group and 15.0% in the propofol group (P = 0.08). In Riker 2009, the incidence of hypertension was 18.9% in the dexmedetomidine group and 29.5% in the midazolam group (P = 0.91).
Other adverse events possibly caused by alpha‐2 agonist
We added other alpha‐2‐associated adverse events that were reported to be significant in at least one study. Other adverse events included tachycardia, first‐degree atrioventricular block, hyperglycaemia and hypoglycaemia.
1. Tachycardia
Jakob 2012 PRODEX reported tachycardia as a significant adverse event, and we therefore include it in our assessment. Four studies, covering 1462 participants, reported on the incidence of tachycardia (Jakob 2012 MIDEX; Jakob 2012 PRODEX; Pandharipande 2007; Riker 2009). One of them indicated that dexmedetomidine was more likely to induce tachycardia than propofol (19.5% vs 11.3%, respectively; P = 0.02) (Jakob 2012 PRODEX). Two studies showed that participants allocated to dexmedetomidine compared with midazolam had a statistically significantly lower risk of tachycardia (Jakob 2012 MIDEX: dexmedetomidine 13.8%, midazolam 21.6%, P = 0.025; Riker 2009: dexmedetomidine 25.4%, midazolam 44.3%, P < 0.001), but one study comparing dexmedetomidine with lorazepam did not find a significant difference (dexmedetomidine 69%, lorazepam 73%, P = 0.71) (Pandharipande 2007). When we tried to pool these data, we detected an extremely high level of statistical heterogeneity (I² = 84%), and therefore abandoned the analysis.
2. First‐degree atrioventricular block
Jakob 2012 PRODEX reported first‐degree atrioventricular block as a significant adverse event. Two studies, covering 993 participants, reported on the incidence of first‐degree atrioventricular block (Jakob 2012 MIDEX; Jakob 2012 PRODEX). In Jakob 2012 PRODEX, the incidence was significantly higher in the dexmedetomidine group than in the propofol group (0.8% vs 3.7%, respectively; P = 0.04), but in Jakob 2012 MIDEX, comparing dexmedetomidine with midazolam (1.2% in each group; P = 0.99), we did not find any difference.
3. Hyperglycaemia
Riker 2009 reported hyperglycaemia as a significant adverse event. Three studies, totaling 1359 participants, reported on the incidence of hyperglycaemia (Jakob 2012 MIDEX; Jakob 2012 PRODEX; Riker 2009). Only in Riker 2009 was the incidence higher in the dexmedetomidine group than in the midazolam group (56.6% vs 42.6%, respectively; P = 0.02). In the other studies, the incidence was low or absent, and did not reach statistical significance (Jakob 2012 MIDEX: 2% in dexmedetomidine group and 2% in midazolam group; Jakob 2012 PRODEX: 0.8% in dexmedetomidine group and 0% in propofol group).
4. Hypoglycaemia
Jakob 2012 MIDEX reported hypoglycaemia as a significant adverse event. Two studies, covering 993 participants, reported on the incidence of hypoglycaemia (Jakob 2012 MIDEX; Jakob 2012 PRODEX). Jakob 2012 MIDEX, comparing dexmedetomidine with midazolam, showed that the incidence in the dexmedetomidine group was higher (4% vs 0.8%, respectively; P = 0.02), while Jakob 2012 PRODEX compared dexmedetomidine with propofol, and showed the same incidence in both groups (1.2%; P = 0.99)
Proportion of sedation time spent at target sedation level (Analysis 1.11)
All the studies evaluated the proportion of sedation time spent at target sedation level. Pandharipande 2007 reported that participants treated with dexmedetomidine spent a higher proportion of time at the target sedation level than did those treated with lorazepam. Both physicians and nurses assessed the sedation level in this study: the proportion assessed by nurses was 80% for the dexmedetomidine group and 67% for the lorazepam group (P = 0.04); the proportion assessed by physicians was 67% for the dexmedetomidine group and 55% for the lorazepam group (P = 0.008). Similarly, in Shehabi 2013, there were more Richmond Agitation‐Sedation Scale (RASS) assessment scores in the target range in the dexmedetomidine group (66%) than in the standard sedation group (38%) during the first 48 hours of the trial (P = 0.01). The other five trials showed similar proportions of time between study groups: in Jakob 2012 MIDEX, 60.7% for the dexmedetomidine group and 56.6% for the midazolam group (P = 0.15); in Jakob 2012 PRODEX, 64.6% for the dexmedetomidine group and 64.7% for the propofol group (P = 0.97); in Riker 2009, 77.3% for the dexmedetomidine group and 75.1% for the midazolam group (P = 0.18); in Ruokonen 2009, 64% for the dexmedetomidine group and 63% for the standard care group; in Xu 2012, 52.8% for the dexmedetomidine group and 48.6% for the midazolam group. Ruokonen 2009 also separately calculated the proportion of participants with a different sedation target: RASS ‐4 and RASS ‐3 to 0. Of those with RASS ‐4, participants on dexmedetomidine spent 42% of the time at the target level, compared with 62% for participants on standard care; of those with RASS ‐3 to 0, participants on dexmedetomidine spent 74% of the time at the target level, compared with 64% for participants on standard care.
Duration of weaning
One study reported on the duration of weaning (Ruokonen 2009; 85 participants). The data for Ruokonen 2009, sourced from the electronic supplementary material which was available online, show that the median weaning times were 59.4 hours for the dexmedetomidine group (range 0 ‐ 259 hours) and 78.0 hours for the standard care group (either propofol or midazolam) (range 0 ‐ 565 hours), but the difference was not statistically significant (P = 0.270).
Intensive care unit length of stay (ICU LOS) (Analysis 1.12)
Six trials, covering 1589 participants, compared intensive care unit (ICU) lengths of stay (Jakob 2012 MIDEX; Jakob 2012 PRODEX; Pandharipande 2007; Riker 2009; Ruokonen 2009; Shehabi 2013). Jakob 2012 MIDEX, Jakob 2012 PRODEX and Ruokonen 2009 analysed the data using Cox’s proportional‐hazards regression, Riker 2009 and Pandharipande 2007 analysed it using Kaplan‐Meier survival analysis and Shehabi 2013 analysed it using Wilcoxon rank‐sum tests. All the studies showed that the median ICU length of stay was shorter in the dexmedetomidine groups than in the control groups, but none of the studies showed statistically significant results. Since Riker 2009 reported only the median without the interquartile range, we carried out meta‐analysis only on the other five trials (calculations from median/interquartile range to geometric mean/SD are shown in Table 1). Meta‐analysis yielded a random‐effects estimate of the geometric mean difference of ‐0.15 log days (95% CI ‐0.28 to ‐0.01, P = 0.04), a statistically significant finding corresponding to a reduction of 14% in the geometric mean (95% CI 1% to 24%). Neither the I² test (0%) nor the Chi² test (0.28, P = 0.99) indicate the presence of statistical heterogeneity. Subgroup analyses for different classes of alternative agents did not find any significant difference in effect size between subgroups (P = 0.93). We did not undertake sensitivity analyses by excluding studies at high risk of bias or studies with non‐standardized designs, due to the small number of studies.
Mortality (Analysis 1.13; Analysis 1.14)
Six studies, covering 1584 participants, assessed mortality (Jakob 2012 MIDEX; Jakob 2012 PRODEX; Pandharipande 2007; Riker 2009; Ruokonen 2009; Shehabi 2013). We did not observed a significant difference between the dexmedetomidine and traditional sedative agents (RR 0.99; 95% CI 0.79 to 1.24, P = 0.92). The level of heterogeneity was low (Chi² = 6.05, P = 0.30; I² = 17%). We undertook subgroup analyses of different classes of alternative agents and found no significant differences in effect size among subgroups (P = 0.42). A sensitivity analysis pooling observed data only, without any imputation, did not change the result (RR 0.99; 95% CI 0.78 to 1.24, P = 0.90) (Analysis 1.14). We did not undertake sensitivity analyses by excluding studies at high risk of bias or studies with non‐standardized designs, due to the small number of studies.
Discussion
Summary of main results
In this systematic review, we assessed evidence from seven studies, involving 1624 critically ill participants, to investigate the effectiveness and safety of alpha‐2 agonists for long‐term sedation during mechanical ventilation. All the included studies investigated adults, and compared dexmedetomidine with traditional sedatives. Our meta‐analyses show that the use of dexmedetomidine in people who need more than 24 hours sedation was associated with a reduction of 22% in the geometric mean for ventilation time (95% confidence interval (CI) 10% to 33%). Because the relative change in geometric means usually produces results similar to the relative change in arithmetic means (Friedrich 2012), dexmedetomidine could cut approximately one‐fifth of the ventilation time. Our meta‐analysis provides no clear evidence that dexmedetomidine could reduce the risk of delirium. We found only one study reporting the risk of coma, but some methodological flaws prevented us from drawing a robust conclusion (Pandharipande 2007).
The results of our meta‐analyses, along with subgroup analyses for different classes of alternatives, showed a consistent trend across all the studies indicating a reduction in ventilation time. Two studies, Jakob 2012 MIDEX and Jakob 2012 PRODEX, which contributed most weight to the meta‐analysis, assessed both invasive and noninvasive ventilation time, and thus offer a wider application for this outcome. As a likely consequence of less ventilation time, we also observed a 14% reduction in intensive care unit (ICU) length of stay. However, the reason for this effect was unclear. One potential explanation might be that dexmedetomidine could provide adequate pain control. Early studies in rats and healthy volunteers suggested that dexmedetomidine offered additive or synergist analgesic effects (Hall 2000; Jaakola 1991; Meert 1994; Tham 2005). Results from previous meta‐analyses on perioperative use of dexmedetomidine for postoperative pain also found that dexmedetomidine decreased pain intensity and opioid consumption (Blaudszun 2012; Schnabel 2013). The importance of analgesia in reducing ventilation time and hospital stay has been confirmed by other randomized controlled trials (RCTs) (Breen 2005; Strom 2010). In our review, Ruokonen 2009 compared the visual analogue scale (VAS) scores of nurses’ assessments in those patients who were able to communicate, and found that the mean VAS scores in the dexmedetomidine group were lower than in the standard care group (30.6 vs 47.5; P < 0.001). Both Jakob 2012 MIDEX and Jakob 2012 PRODEX confirmed better pain scores in the dexmedetomidine group, although Ruokonen 2009 showed no difference in the cumulative dose of fentanyl for pain control between both groups, probably because the dexmedetomidine participants might request more additional analgesics. In Jakob 2012 MIDEX and Jakob 2012 PRODEX, participants on dexmedetomidine were more able to communicate pain, more arousable and more co‐operative. This characteristic of dexmedetomidine might promote the rational use of opioids and facilitate the delivery of patient care and medical examination. Another candidate explanation was suggested by a subgroup analysis of Pandharipande 2007. This post hoc analysis revealed that dexmedetomidine provided greater improvement in ventilation time among septic participants than among non‐septic participants, with the investigators attributing this effect to the anti‐inflammatory, organ‐protective and anti‐sympathetic properties of dexmedetomidine.
We found no evidence for the risk of delirium, with a high level of statistical heterogeneity. Two studies compared the number of ICU days without delirium but demonstrated inconsistent results (Pandharipande 2007; Riker 2009). Subgroup analysis of different classes of alternative agents found that both the benzodiazepine derivatives subgroup and the propofol subgroup favoured dexmedetomidine, while the standard care subgroup favoured the control condition. Because standard care participants used midazolam or propofol (both drugs were analysed alone in other subgroups), we were unable to attribute this difference to the characteristics of the control drugs. One possible explanation could be that in standard care both studies were at high risk for blinding of personnel and outcome assessment, which may have biased the results. Meta‐regression on geometric mean ventilation time found little statistical evidence to explain the heterogeneity, probably due to the small number of studies. We noted considerable variation among the studies for risk of delirium, which might be another important source of heterogeneity. There were several reasons for this: firstly, the time in ICU prior to study treatment ranged from 12 hours to 96 hours between the studies, which therefore altered the baseline risk of delirium; secondly, although all the studies used the confusion assessment method for the intensive care unit (CAM‐ICU), the frequency of assessments and the follow‐up periods were different; thirdly, even within same study the frequency might vary between groups, because a successful CAM‐ICU assessment was highly dependent on the participant's sedation level (within a RASS score of ‐3 or ehighermrhigher), which may explain in Ruokonen 2009 why more CAM‐ICU assessments were performed in the dexmedetomidine group than in the standard care group. Furthermore, because the sensitivity of the CAM‐ICU is poor (Van Eijk 2011), interpreting a positive CAM‐ICU assessment as an occurrence of delirium might overestimate the incidence and increase the uncertainty of the intervention effects.
Pandharipande 2007 showed that dexmedetomidine reduced the risk of coma and increased the number of ICU days without coma. However, because of its long half‐life, continuous infusion of lorazepam, which was used in the control group, could result in higher concentrations and could induce deeper sedation (De Wit 2006). On the one hand, such methodological limitation might increase the risk of iatrogenic coma (oversedation) in the control group, especially where the investigators assessed the sedation level infrequently (twice daily), but on the other hand dexmedetomidine caused less oversedation, even without frequent adjustment.
Bradycardia was the most commonly‐reported adverse event. Our meta‐analysis found that dexmedetomidine doubled the incidence of bradycardia. Subgroup analysis indicated that the risk ratio might differ depending on which alternatives were used as the comparator. For example, the risk ratio in the propofol subgroup was lower than in the benzodiazepine derivatives subgroup, probably because propofol itself could also induce bradycardia (Hashiba 2003). Riker 2009 showed the risk of bradycardia requiring an intervention (including readjusting or interruption of study drugs and use of atropine) in the dexmedetomidine group was low (4.9%) and not significantly increased. The bradycardiac effect of dexmedetomidine might therefore be mild. We did not find an increase in the risk of hypotension or hypertension. We included tachycardia, first‐degree atrioventricular block, hyperglycaemia and hypoglycaemia, which were reported to be significant in at least one study, as outcomes of adverse events post hoc. The results of the post hoc adverse events were inconsistent among studies, probably due to the different comparators. Because of the small number of studies, there was insufficient power to investigate any impact of these rare adverse events.
With respect to other secondary outcomes, we found dexmedetomidine was at least as effective as traditional sedatives on maintaining the target sedation level. In Ruokonen 2009, among participants with a target RASS score of ‐4 compared with ‐3 to 0, those on dexmedetomidine spent less time at target level than did those on standard care, which implies that dexmedetomidine may be a poor sedative choice when deep sedation is needed. Additionally, in both Jakob 2012 MIDEX and Jakob 2012 PRODEX, approximately one in every eight to ten participants ceased participation due to lack of efficacy, even on the maximum dose, and hence these two studies may overestimate the effectiveness of dexmedetomidine. Lack of blinding in some studies (Ruokonen 2009; Shehabi 2013; Xu 2012) would also have an unpredictable impact on this subjective outcome. We failed to identify any evidence for reducing duration of weaning.
Our meta‐analysis provides no evidence that dexmedetomidine had any impact on overall mortality. On the one hand the adverse events which might potentially influence haemodynamic stability, such as bradycardia and hypotension, did not contribute to an increase in mortality, while on the other it would seem that choosing sedatives still occurs 'downstream' in a framework of intensive care, and has limited effect on the prognosis for survival.
When we compared dexmedetomidine sedation in our review to 'no sedation protocol' in Strom 2010, we found some common patterns which might result in reduced ventilation time: using analgesia‐based sedation, retaining some degree of participant's cognitive function, and providing on‐demand pain control. It is also clear that dexmedetomidine was a more 'professional' sedative than the morphine used in the 'no sedation protocol'. A cost‐minimization analysis of Riker 2009 showed that dexmedetomidine could reduce total intensive care unit costs by about USD 10,000, because of decreased mechanical ventilation costs and ICU stay costs (Dasta 2010). All these properties of dexmedetomidine seemed to make it a cost‐effective choice for ICU sedation. But we must also consider any negative aspects to offset the beneficial effects. Firstly, patients might request additional sedatives or analgesics because they could express discomfort (Corbett 2005; Jakob 2012 MIDEX; Jakob 2012 PRODEX; Maclaren 2013). Secondly, despite a theoretical advantage of dexmedetomidine in providing nature‐like sleep, patients on dexmedetomidine were more likely to have sleeping difficulties (Corbett 2005; Oto 2012). Thirdly, a recent study focusing on psychological outcomes showed that dexmedetomidine was associated with greater recall of ICU experiences and higher risk of acute stress disorder (Maclaren 2013). As more frequent assessment of discomfort or pain and appropriate supplemental sedatives or analgesic would be needed, dexmedetomidine for sedation might potentially increase the workload for medical staff.
Overall completeness and applicability of evidence
Although the results of our meta‐analyses seem to support beneficial outcomes for the use of alpha‐2 agonists for long‐term sedation, there were several limitations. Firstly, the included studies investigated only adults treated with dexmedetomidine. Our findings therefore should not be extrapolated to children or to patients treated with clonidine. Secondly, none of the studies described in sufficient detail how they predicted that a patient would require long‐term ventilation and sedation. We believe that clinicians in different healthcare system may have different opinions on this point, so that the participants in our review might be substantially different from those in other healthcare systems. Thirdly, most studies excluded people with severe bradycardia, second‐ or third‐degree atrioventricular conduction block or hepatic impairment, and used a dose of dexmedetomidine of no more than 1.5 mcg/kg per hour. Thus, it was not clear whether dexmedetomidine could be used safely in these excluded patients, or whether it could safely be used in higher doses. Fourthly, most studies incorporated new ICU sedation strategies which were recommended by recent guideline (Barr 2013): light level of sedation protocols, daily interruptions, analgesia‐based sedation and monitoring of delirium. It is unclear whether our findings can be generalized to other healthcare systems where these revised sedation strategies do not apply.
Quality of the evidence
In our systematic review, six of the seven studies were at high risk of bias across the six domains (Figure 3). As a consequence, the general quality of evidence ranged from very low to low, based on the GRADE system (summary of findings Table for the main comparison). We noted several methodological limitations in the included studies. Firstly, the risk of breaking blinding might be high, even when double‐blinding and a double‐dummy design were used. Wunsch 2012 has pointed out that when the comparator was propofol, even a single drop visible at the end of an infusion would break the blinding. We also suspect that a sophisticated clinician could easily guess the identity of the study drugs from the bradycardiac effect of dexmedetomidine. Secondly, high withdrawal rates in some studies might bias the overall results, especially in Jakob 2012 MIDEX and Jakob 2012 PRODEX, where more than 20% of participants withdrew from each study group. Most reasons for withdrawal were unbalanced between the groups and were related to outcomes, such as lack of efficacy and adverse events, which might lead to an underestimate of the adverse consequences and might have unpredictable impacts on the other outcomes. Thirdly, some participants from Jakob 2012 MIDEX, Jakob 2012 PRODEX and Ruokonen 2009 received continuous sedation prior to the study treatment. The carry‐over effect of preceding sedation might reduce the required dosages of dexmedetomidine and have unpredictable impacts on the intervention effect.
Potential biases in the review process
There were several potential biases in the review process which should be reiterated. Firstly, our systematic review includes only seven studies, and thus we were not able to generate funnel plots or to assess publication bias by using the Egger test and the arcsine test. Secondly, we emailed the first and corresponding authors of all seven included studies twice, but received only two responses (Pandharipande 2007; Shehabi 2013), with one providing study data (Shehabi 2013). In the absence of individual patient data or log‐transformed data, we estimated means using medians. We do not consider that this estimation would excessively influence the results of the meta‐analyses, since log‐transformed means will be approximately equal to log‐transformed median under the assumption that the log‐transformation of the raw data may substantially reduce skew. Thirdly, due to small numbers of included studies, our predefined subgroup analyses might be underpowered to explore and eliminate heterogeneity.
Agreements and disagreements with other studies or reviews
Previous meta‐analyses have investigated the effect of dexmedetomidine in an ICU setting. Tan 2010 reached similar conclusions to our findings for the ICU length of stay, risk of delirium and adverse events, but not for duration of mechanical ventilation. Two meta‐analyses comparing propofol head‐to‐head with dexmedetomidine both suggested that dexmedetomidine did not reduce ventilation time (Xia 2013; Zhuo 2012). One reported that dexmedetomidine reduced the risk of delirium (Zhuo 2012), which was confirmed in our subgroup analysis. Another meta‐analysis (Lin 2012) compared dexmedetomidine with placebo or traditional sedatives in the setting of a post‐cardiac surgery unit and found that dexmedetomidine reduced ventilation time and risk of delirium. However, this review included both randomized controlled trials and cohort studies, which may have increased heterogeneity and the risk of publication bias.
Selection process for studies included in data synthesis.
Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.
Risk of bias summary: review authors' judgements about each risk of bias item for each included study.
Comparison 1 Dexmedetomidine versus traditional sedative agents, Outcome 1 Duration of mechanical ventilation (subgroup: class of alternative agents).
Comparison 1 Dexmedetomidine versus traditional sedative agents, Outcome 2 Duration of mechanical ventilation (sensitivity analysis using data from 'time to extubation'").
Comparison 1 Dexmedetomidine versus traditional sedative agents, Outcome 3 Risk of delirium (subgroup: class of alternative agents).
Comparison 1 Dexmedetomidine versus traditional sedative agents, Outcome 4 Risk of delirium (sensitivity analysis for observed data only).
Comparison 1 Dexmedetomidine versus traditional sedative agents, Outcome 5 Incidence of bradycardia (subgroup: class of alternative agents).
Comparison 1 Dexmedetomidine versus traditional sedative agents, Outcome 6 Incidence of bradycardia (subgroup: dosing regimens).
Comparison 1 Dexmedetomidine versus traditional sedative agents, Outcome 7 Incidence of bradycardia (sensitivity analysis for observed data only).
Comparison 1 Dexmedetomidine versus traditional sedative agents, Outcome 8 Incidence of hypotension (subgroup: class of alternative agents).
Comparison 1 Dexmedetomidine versus traditional sedative agents, Outcome 9 Incidence of hypotension (subgroup: dosing regimens).
Comparison 1 Dexmedetomidine versus traditional sedative agents, Outcome 10 Incidence of hypotension (sensitivity analysis for observed data only).
| Study | Dexmedetomidine | Traditional sedatives |
| Jakob 2012 MIDEX | 60.7% | 56.6% |
| Jakob 2012 PRODEX | 64.6% | 64.7% |
| Pandharipande 2007 | 80% (nurse); 67% (physician) | 67% (nurse); 55% (physician) |
| Riker 2009 | 77.3% | 75.1% |
| Ruokonen 2009 | 64% | 63% |
| Shehabi 2013 | 66% | 38% |
| Xu 2012 | 52.8% | 48.6% |
Comparison 1 Dexmedetomidine versus traditional sedative agents, Outcome 11 Proportion of sedation time spent at target sedation level.
Comparison 1 Dexmedetomidine versus traditional sedative agents, Outcome 12 ICU length of stay (LOS) (subgroup: class of alternative agents).
Comparison 1 Dexmedetomidine versus traditional sedative agents, Outcome 13 Mortality (subgroup: class of alternative agents).
Comparison 1 Dexmedetomidine versus traditional sedative agents, Outcome 14 Mortality (sensitivity analysis for observed data only).
| Dexmedetomidine compared to traditional sedative agents for long‐term sedation during mechanical ventilation in critically ill patients | ||||||
| Patient or population: Critically ill patients requiring long‐term sedation during mechanical ventilation | ||||||
| Outcomes | Illustrative comparative risks* (95% CI) | Relative effect | No of Participants | Quality of the evidence | Comments | |
| Assumed risk | Corresponding risk | |||||
| traditional sedative agents | Dexmedetomidine | |||||
| Duration of mechanical ventilation | Not estimable | Not estimable | Relative decrease 22% in the geometric mean (95% CI 10% to 33%) | 1120 | ⊕⊕⊝⊝ | Alpha‐2 agonists reduced geometric mean duration of mechanical ventilation by 0.25 (95% CI 0.10 to 0.40), corresponding to a reduction of 22% in the geometric mean (95% CI 10% to 33%). The relative change in geometric means usually produces similar results to the relative change in arithmetic means. |
| Risk of delirium | Study population | RR 0.85 | 1624 | ⊕⊝⊝⊝ | ‐ | |
| 266 per 1000 | 226 per 1000 | |||||
| Low | ||||||
| 76 per 1000 8 | 65 per 1000 | |||||
| High | ||||||
| 824 per 1000 8 | 700 per 1000 | |||||
| Risk of coma | Study population | RR 0.69 | 103 | ⊕⊝⊝⊝ | ‐ | |
| 922 per 1000 | 636 per 1000 | |||||
| Incidence of bradycardia | Study population | RR 2.11 | 1587 | ⊕⊝⊝⊝ | ‐ | |
| 90 per 1000 | 189 per 1000 | |||||
| Low | ||||||
| 39 per 1000 8 | 82 per 1000 | |||||
| High | ||||||
| 188 per 1000 8 | 397 per 1000 | |||||
| Duration of weaning | See comment | 85 | ‐ | Only one study assessed this outcome and only reported median and range. We were not able to estimate a relative effect. | ||
| ICU length of stay | Not estimable | Not estimable | Relative decrease 14% in the geometric mean (95% CI 0.01% to 24%) | 1223 | ⊕⊝⊝⊝ | Sedation using alpha‐2 agonists reduced geometric mean ICU LOS by 0.15 (95% CI 0.01 to 0.28), corresponding to a reduction of 14% in the geometric mean (95% CI 0.01% to 24%). Relative change in geometric means usually produces similar results to the relative change in arithmetic means. |
| Mortality | Study population | RR 0.99 | 1584 | ⊕⊝⊝⊝ | ‐ | |
| 205 per 1000 | 203 per 1000 | |||||
| Moderate | ||||||
| 192 per 1000 12 | 190 per 1000 | |||||
| *The basis for the assumed risk (e.g. the median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). | ||||||
| GRADE Working Group grades of evidence | ||||||
| 1Downgraded by 1 point due to all the studies being at high risk of bias, which was considered very serious. 2Downgraded by 1 point because all the studies received funding support from pharmaceutical firms and thus publication bias was strongly suspected. 3Downgraded by 1 point due to 6 of the 7 studies being at high risk of bias, which was considered serious. 4Downgraded by 1 point because the heterogeneity is substantial and thus inconsistency was serious. 5Downgraded by 1 point because results were consistent with both no effect and appreciable benefit, and thus imprecision was serious. 6Downgraded by 1 point because 6 of the 7 studies received funding support from pharmaceutical firms and thus publication bias was strongly suspected. 7Downgraded by 1 point because the sample size was small (less than 2000) and thus inconsistency was serious. 8The baseline risk varies broadly among studies, and thus we chose the lowest and highest control group risk in the included studies. 9Downgraded by 1 point due to 5 of the 6 studies being at high risk of bias, which was considered serious. 10Downgraded by 1 point because 5 of the 6 studies received funding support from pharmaceutical firms and thus publication bias was strongly suspected. 11Downgraded by 1 point due to wide confidence interval, although there may still be enough precision to make decisions about the utility of the intervention. 12There was little variation in the baseline risk across the studies, and thus we calculated the median control group risk. | ||||||
| Study ID | Outcome | Study group | Median | Interquartile range | Geometric mean | Geometric SD |
| Duration of mechanical ventilation | Intervention | 123 hours | 67 ‐ 337 hours | 4.812 | 1.197 | |
| Control | 164 hours | 92 ‐ 380 hours | 5.1 | 1.051 | ||
| Duration of mechanical ventilation | Intervention | 97 hours | 45 ‐ 257 hours | 4.575 | 1.291 | |
| Control | 118 hours | 48 ‐ 327 hours | 4.771 | 1.421 | ||
| Duration of mechanical ventilation | Intervention | 77.2 hours | 17.5 – 338.8 hours | 4.346 | 2.195 | |
| Control | 110.6 hours | 20.1 – 675.0 hours | 4.706 | 2.603 | ||
| ICU LOS | Intervention | 8.8 days (211 hours) | 4.8 ‐ 34.6 days (115 ‐ 831 hours) | 2.174 | 1.465 | |
| Control | 10.1 days (243 hours) | 5.8 ‐ 26.2 days (140 ‐ 630 hours) | 2.315 | 1.114 | ||
| ICU LOS | Intervention | 6.8 days (164 hours) | 3.8 ‐ 20.0 days (90 ‐ 480 hours) | 1.922 | 1.24 | |
| Control | 7.7 days (185 hours) | 3.9 ‐ 21.7 days (93 ‐ 520 hours) | 2.042 | 1.275 | ||
| ICU LOS | Intervention | 7.5 days | 5 ‐ 19 days | 2.015 | 0.989 | |
| Control | 9 days | 6 ‐ 15 days | 2.197 | 0.679 | ||
| ICU LOS | Intervention | 5.5 days | 1.7 – 19.5 days | 1.705 | 1.807 | |
| Control | 5.7 days | 1.7 – 29.0 days | 1.74 | 2.101 | ||
| ICU: intensive care unit | ||||||
| tau² (estimated amount of residual heterogeneity) | 0.1460 (SE = 0.1596) |
| tau (square root of estimated tau² value) | 0.3820 |
| I² (residual heterogeneity / unaccounted variability) | 70.09% |
| H² (unaccounted variability / sampling variability) | 3.34 |
| R² (amount of heterogeneity accounted for) | 0.00% |
| Intercept | 0.0895 (95% CI ‐5.4218 to 5.6007, P = 0.9746) |
| Coefficient for log‐transformed mean ventilation time | ‐0.0442 (95% CI ‐1.1989 to 1.1105, P = 0.9403) |
| Outcome or subgroup title | No. of studies | No. of participants | Statistical method | Effect size |
| 1 Duration of mechanical ventilation (subgroup: class of alternative agents) Show forest plot | 4 | 1120 | Mean Difference (IV, Random, 95% CI) | ‐0.25 [‐0.40, ‐0.10] |
| 1.1 Benzodiazepine derivatives | 1 | 500 | Mean Difference (IV, Random, 95% CI) | ‐0.29 [‐0.49, ‐0.09] |
| 1.2 Propofol | 1 | 498 | Mean Difference (IV, Random, 95% CI) | ‐0.20 [‐0.43, 0.04] |
| 1.3 Standard care (propofol and midazolam) | 2 | 122 | Mean Difference (IV, Random, 95% CI) | ‐0.31 [‐1.26, 0.64] |
| 2 Duration of mechanical ventilation (sensitivity analysis using data from 'time to extubation'") Show forest plot | 4 | 1120 | Mean Difference (IV, Random, 95% CI) | ‐0.34 [‐0.49, ‐0.20] |
| 3 Risk of delirium (subgroup: class of alternative agents) Show forest plot | 7 | 1624 | Risk Ratio (M‐H, Random, 95% CI) | 0.85 [0.63, 1.14] |
| 3.1 Benzodiazepine derivatives | 4 | 1007 | Risk Ratio (M‐H, Random, 95% CI) | 0.81 [0.59, 1.09] |
| 3.2 Propofol | 1 | 495 | Risk Ratio (M‐H, Random, 95% CI) | 0.37 [0.16, 0.87] |
| 3.3 Standard care (propofol and midazolam) | 2 | 122 | Risk Ratio (M‐H, Random, 95% CI) | 1.44 [0.86, 2.41] |
| 4 Risk of delirium (sensitivity analysis for observed data only) Show forest plot | 7 | 1621 | Risk Ratio (M‐H, Random, 95% CI) | 0.84 [0.63, 1.14] |
| 5 Incidence of bradycardia (subgroup: class of alternative agents) Show forest plot | 6 | 1587 | Risk Ratio (M‐H, Random, 95% CI) | 2.11 [1.39, 3.20] |
| 5.1 Benzodiazepine derivatives | 4 | 1007 | Risk Ratio (M‐H, Random, 95% CI) | 2.44 [1.77, 3.36] |
| 5.2 Propofol | 1 | 495 | Risk Ratio (M‐H, Random, 95% CI) | 1.20 [0.74, 1.94] |
| 5.3 Standard care (propofol and midazolam) | 1 | 85 | Risk Ratio (M‐H, Random, 95% CI) | 7.50 [0.40, 140.91] |
| 6 Incidence of bradycardia (subgroup: dosing regimens) Show forest plot | 6 | 1587 | Risk Ratio (M‐H, Random, 95% CI) | 2.11 [1.39, 3.20] |
| 6.1 Rapid dosing regimens | 1 | 40 | Risk Ratio (M‐H, Random, 95% CI) | 9.00 [0.52, 156.91] |
| 6.2 Slow dosing regimens | 5 | 1547 | Risk Ratio (M‐H, Random, 95% CI) | 2.05 [1.35, 3.11] |
| 7 Incidence of bradycardia (sensitivity analysis for observed data only) Show forest plot | 6 | 1584 | Risk Ratio (M‐H, Random, 95% CI) | 2.11 [1.39, 3.21] |
| 8 Incidence of hypotension (subgroup: class of alternative agents) Show forest plot | 6 | 1587 | Risk Ratio (M‐H, Random, 95% CI) | 1.22 [0.86, 1.74] |
| 8.1 Benzodiazepine derivatives | 4 | 1007 | Risk Ratio (M‐H, Random, 95% CI) | 1.39 [0.77, 2.53] |
| 8.2 Propofol | 1 | 495 | Risk Ratio (M‐H, Random, 95% CI) | 0.98 [0.62, 1.54] |
| 8.3 Standard care (propofol and midazolam) | 1 | 85 | Risk Ratio (M‐H, Random, 95% CI) | 2.15 [0.20, 22.79] |
| 9 Incidence of hypotension (subgroup: dosing regimens) Show forest plot | 6 | 1587 | Risk Ratio (M‐H, Random, 95% CI) | 1.22 [0.86, 1.74] |
| 9.1 Rapid dosing regimens | 1 | 40 | Risk Ratio (M‐H, Random, 95% CI) | 9.00 [0.52, 156.91] |
| 9.2 Slow dosing regimens | 5 | 1547 | Risk Ratio (M‐H, Random, 95% CI) | 1.18 [0.86, 1.62] |
| 10 Incidence of hypotension (sensitivity analysis for observed data only) Show forest plot | 6 | 1584 | Risk Ratio (M‐H, Random, 95% CI) | 1.22 [0.86, 1.73] |
| 11 Proportion of sedation time spent at target sedation level Show forest plot | Other data | No numeric data | ||
| 12 ICU length of stay (LOS) (subgroup: class of alternative agents) Show forest plot | 5 | 1223 | Mean Difference (IV, Random, 95% CI) | ‐0.15 [‐0.28, ‐0.01] |
| 12.1 Benzodiazepine derivatives | 2 | 603 | Mean Difference (IV, Random, 95% CI) | ‐0.15 [‐0.34, 0.03] |
| 12.2 Propofol | 1 | 498 | Mean Difference (IV, Random, 95% CI) | ‐0.12 [‐0.34, 0.10] |
| 12.3 Standard care (propofol and midazolam) | 2 | 122 | Mean Difference (IV, Random, 95% CI) | ‐0.21 [‐0.69, 0.26] |
| 13 Mortality (subgroup: class of alternative agents) Show forest plot | 6 | 1584 | Risk Ratio (M‐H, Random, 95% CI) | 0.99 [0.79, 1.24] |
| 13.1 Benzodiazepine derivatives | 3 | 967 | Risk Ratio (M‐H, Random, 95% CI) | 0.98 [0.68, 1.41] |
| 13.2 Propofol | 1 | 495 | Risk Ratio (M‐H, Random, 95% CI) | 0.81 [0.55, 1.21] |
| 13.3 Standard care (propofol and midazolam) | 2 | 122 | Risk Ratio (M‐H, Random, 95% CI) | 1.44 [0.67, 3.11] |
| 14 Mortality (sensitivity analysis for observed data only) Show forest plot | 6 | 1581 | Risk Ratio (M‐H, Random, 95% CI) | 0.99 [0.78, 1.24] |
