This chapter reports the evidence of the clinical effectiveness of dexmedetomidine compared with clonidine and of dexmedetomidine or clonidine compared with propofol or benzodiazepines (midazolam or lorazepam) in mechanically ventilated adults admitted to ICUs.
Methods for assessing the outcomes arising from the use of the intervention
The methods for this assessment were prespecified in a research protocol (www.crd.york.ac.uk/PROSPERO/display_record.asp?ID=CRD42014014101).105
Identification of studies (search strategy and information sources/dates)
Highly sensitive literature searches, using an appropriate combination of controlled vocabulary and text word terms, were developed to identify reports of published, ongoing and unpublished studies reporting the clinical effectiveness of dexmedetomidine or clonidine in comparison with propofol and benzodiazepines (e.g. midazolam, lorazepam and diazepam) in mechanically ventilated adults admitted to ICUs. Literature searches were carried out from 12 to 15 November 2014 for publications from 1999 onwards. Details of the search strategies are reported in Appendix 1. Major electronic databases were searched including MEDLINE without revisions, MEDLINE In-Process & Other Non-Indexed Citations, EMBASE, Science Citation Index, Bioscience Information Service and the Cochrane Central Register of Controlled Trials. Reports of relevant evidence synthesis were sought from the Cochrane Database of Systematic Reviews and Database of Abstracts of Reviews of Effects. The World Health Organization International Clinical Trials Registry Platform, metaRegister of Controlled Trials and ClinicalTrials.gov were searched for evidence of ongoing studies.
Websites of regulatory bodies and Health Technology Assessment agencies were checked for relevant unpublished reports, while websites of relevant pharmaceutical companies and professional organisations were searched for further pertinent information and reports.
In addition, reference lists of all included studies were perused for further citations.
Inclusion and exclusion criteria
Types of studies
Evidence was considered from RCTs comparing dexmedetomidine with clonidine or dexmedetomidine or clonidine with propofol or benzodiazepines such as midazolam, lorazepam and diazepam.
The following types of reports were excluded:
narrative reviews, editorials and opinions
case reports
conference abstracts for which a full publication or further methodological information could not be found
non-English-language reports for which a translation could not be organised
studies that focused predominantly on people with primary brain injuries.
Types of participants
The types of participants considered were critically ill adults in ICUs who required MV. We did not prespecify definitions for ‘critically ill’ or ‘adults’, so any study population described as such was deemed suitable for inclusion.
Interventions
The sedative interventions considered were dexmedetomidine and clonidine.
Comparator interventions
The comparator interventions assessed were propofol and benzodiazepines such as midazolam, lorazepam and diazepam.
Outcomes
The following primary outcomes were considered:
mortality
duration of MV
ventilator-free days
length of ICU stay
adverse events as reported by trial investigators and including the rate of:
unpleasant side effects as reported by trial investigators (e.g. unpleasant memories, constipation or diarrhoea).
Secondary outcomes considered were:
Data extraction strategy (study selection and data collection)
One reviewer (MC) screened all titles and abstracts identified by the search strategies. A second reviewer (MB) independently double-screened the first 100 abstracts and titles of the 2011–14 list. Agreement between the two reviewers was 100%.
All potentially relevant reports were retrieved in full and assessed independently by one reviewer (MC). A total of 40 reports were double-assessed by a second reviewer (Pawana Sharma or MB). Any disagreements were resolved by consensus. The full-text screening form is presented in Appendix 2. A data extraction spreadsheet (Microsoft Excel®, 2013; Microsoft Corporation, Redmond, WA, USA) was developed specifically for the purpose of this assessment, piloted and amended as necessary. From each study, one reviewer (MC) extracted information on geographical location, sponsor, study design, participants’ characteristics, setting and characteristics of ICU practice, characteristics of sedative intervention and outcome measures. Data extraction was double-checked by a second reviewer (MB). Any disagreements were resolved by discussion.
Critical appraisal strategy
The risk of bias of included RCTs was initially assessed by one reviewer (MC) using Cochrane’s risk-of-bias tool106 and, subsequently, cross-checked by a second reviewer (MB). The following domains were assessed: sequence generation, allocation concealment, blinding of participants and medical personnel, blinding of outcome assessors, incomplete outcome data and selective outcome reporting. Assessment of ‘other bias’ was based on the funding source, and a study was judged to be at high risk of bias if it was funded by the manufacturer(s) of the sedative agent(s) under investigation. Individual outcomes were judged as being at ‘high’, ‘low’ or ‘unclear’ risk of bias. Overall, risk of bias for each study was based on the findings of three key domains: sequence generation, allocation concealment and blinding of outcome assessor.
Studies were classified as follows: (1) high risk of bias if one or more key domains were at high risk; (2) unclear risk of bias if one or more key domains were judged to be at unclear risk; and (3) low risk of bias if all key domains were judged to be at low risk. Any disagreements between reviewers were resolved by discussion.
Method of analysis/synthesis
The general approach recommended by Cochrane was used for data analysis and synthesis.106 For binary outcomes, the Mantel–Haenszel approach was used to pool risk ratios (RRs) derived from each study. A random-effects model was used to calculate the pooled estimates of effect. For continuous outcomes (duration of MV, ICU length of stay, hospital length of stay, time to extubation, time in target sedation range and ventilator-free days), mean differences between groups were pooled when possible using the inverse variance weighted mean difference method and a random-effects model. Random-effects methods, rather than fixed-effects methods, as outlined in the original protocol, were chosen because of the clinical and statistical heterogeneity observed among included studies.
For each continuous outcome, an initial analysis was conducted using only studies where the mean and standard deviation (SD) were provided. In studies that did not report a mean and SD [and we could not derive these summary measures from reported p-values, standard errors or confidence intervals (CIs)], we tried to impute these from the data reported. The imputation strategy was as follows:
Where the median, range and
n for each group were available, we used the formulae reported by Hozo and colleagues
107 to estimate the mean and SD.
Where this method proved unfeasible, we imputed a SD from the available data using the methods outlined by Furukawa and colleagues.
108In studies where a median and interquartile range were reported, we used two methods to calculate the mean. If the sample size was < 25, then first the median was used and second the value midway between the lower quartile and upper quartile was used. If the two methods yielded results that reversed the direction of treatment effect for a certain outcome within a study, then the study was excluded from the pooled analysis of that outcome.
For each outcome where the above provided extra data, a second analysis was done using the imputed data.
Heterogeneity across studies was explored by visual inspection of forest plots and using the chi-squared test and I2-statistics.
When data were available, subgroup analyses were performed according to type of comparator intervention.
Results of the evidence synthesis
Quantity of the evidence (studies included and excluded)
The literature searches identified 1182 potentially relevant citations, of which 83 were selected for full-text assessment and 107 for background information. Of these, 59 were subsequently excluded because the patient population, study design, outcomes reported or publication type were not eligible. A total of 18 RCTs published in 24 papers with a total of 2489 people were included in this assessment.5,52,69–72,102,109–125 It is worth noting that the results of the two large multicentre trials of PROpofol compared with DEXmedetomidine (the PRODEX trial) and of MIdazolam compared with DEXmedetomidine (the MIDEX trial) were published in a single report by Jakob and colleagues70 for the Dexmedetomidine for Long-Term Sedation investigators. presents the flow chart of the selection process. Appendix 3 provides the details of the 18 included trials and related secondary publications. Appendix 4 categorises the excluded studies according to the main reasons for their exclusion.
Flow chart of the study selection process.
Study characteristics
Appendix 5 details the study characteristics of the 18 included trials. All 18 trials were published in full. Four different comparators were assessed. One trial, with a total of 70 randomised patients, compared the effects and safety of dexmedetomidine with clonidine;55 nine trials, with a total of 1134 randomised patients, compared dexmedetomidine with propofol;70,109–111,114,117,120,122,123 four trials, with a total of 939 randomised patients, compared dexmedetomidine with midazolam;70,71,112,116 one trial, with a total of 118 randomised patients, compared dexmedetomidine with propofol or midazolam (three arms);72 two trials, with a total of 122 randomised patients, compared dexmedetomidine with ‘standard care’ (i.e. propofol and/or midazolam);69,121 and one trial with a total of 106 randomised patients, compared dexmedetomidine with lorazepam.102 A total of 2446 patients were analysed in the 18 included trials.
Six trials assessed dexmedetomidine in patients admitted to ICUs following elective surgery,72,110,111,114,120,123 whereas the remaining trials included general ICU patients.
Four trials were conducted in the USA,72,102,110,116 two in India,55,120 three in Turkey,112,117,122 two in Egypt,109,111 one in the UK,123 one in North America (USA and Canada),114 one in Finland and Switzerland,69 and one in Australia and New Zealand.121 The MIDEX multicentre trial70 was conducted in nine European countries; the PRODEX multicentre trial70 was conducted in six European countries and in Russia; and the SEDCOM71 (Safety and Efficacy of Dexmedetomidine COmpared with Midazolam) multicentre trial was conducted in the USA, Argentina, Brazil, Australia and New Zealand. All included trials involved prospective collection of data.
Three trials assessed patients up to 45 days69,70 and one up to 90 days.121 In one trial,102 participants were observed in the hospital from enrolment until discharge from hospital or death, and survivors were observed for vital status until 1 year after enrolment using hospitals’ electronic record systems and a commercial version of the Social Security Death Master File (http://ssdi.rootsweb.com). One trial114 followed up patients for 24 hours after discharge from ICUs and another trial71 for 48 hours after study drug cessation. One trial72 reported that patients were followed up for 3 days post operatively. In one trial,109 length of follow-up was reported to be 6 hours, in two trials55,120 it was 24 hours and in another trial123 it was 48–72 hours. Two trials reported follow-up in terms of time post extubation: one trial110 assessed patients at least 24 hours post extubation and another trial116 at least 72 hours post extubation. Length of follow-up was not reported in four trials.111,112,117,122
Appendix 6 presents details of dosage and route of administration of the respective sedative agents.
In general, dexmedetomidine was initiated with a loading dose of 1 µg/kg, administered intravenously over a period of 10–20 minutes.109,110,112,114,117,120,122 Some trials involved lower55,72,102,116 or higher111,123 loading doses, four trials did not use a loading dose69,70,121 and, in one trial, the loading dose was optional.71 Dexmedetomidine maintenance doses were fixed in two trials: 0.4 µg/kg/hour110 or 0.7 µg/kg/hour.112 The remaining trials specified lower and upper limits for maintenance doses, with lower limits ranging from 0 µg/kg/hour121 to 0.015 µg/kg/hour,102,116 0.2 µg/kg/hour,55,70,72,111,114,117,120,122,123 0.4 µg/kg/hour,110 0.5 µg/kg/hour109 and 0.7 µg/kg/hour.112 The maximum allowable dose was 2.5 µg/kg/hour.117,122,123
Clonidine was used in one trial. Patients received an infusion of clonidine at 1 µg/kg/hour. Titration was achieved with dosage increments up to 2 µg/kg/hour.55
Of the 12 trials that included a propofol arm, four trials reported a loading dose: an initial bolus dose of 1 mg/kg in one trial111 and 1 mg/kg over 10–15 minutes in three trials.117,122,123 Six trials did not use a loading dose69,70,72,109,110,120 and two trials did not provide information on dosage.114,121 Maintenance infusions of propofol ranged from 0.5–1 mg/kg/hour111 to 4 mg/kg/hour across trials.69,70
Out of the seven trials that included a midazolam arm, one trial reported a loading dose of 0.05 mg/kg112 and another trial reported an optional loading dose of the same level.71 The remaining trials did not use a loading dose. One trial did not specify dosage of midazolam.121 Maintenance doses of midazolam were between 0.03 mg/kg/hour70 and 10 mg/hour across trials.116
In one trial, lorazepam infusion started at 1 mg/hour and was titrated to a maximum of 10 mg/hour.102
All trials titrated sedatives to a target sedation level.55,69–72,102,109–112,114,116,117,120–123 Target sedation level was measured by means of the RSS score in 11 trials,55,72,109–112,114,117,120,122,123 the RASS in six trials69–71,102,121 and the Riker SAS score in one trial.116
The main characteristics of the 18 included studies are shown in .
Summary of the main features of the included studies
Participant characteristics
The 18 included trials randomised a total of 1283 participants to dexmedetomidine and 1206 participants to a control intervention. The sample sizes of included studies ranged from 23 to 501 participants.
There was some doubt whether or not the trial by Memis and colleagues (40 patients in total)117 and that by Tasdogan and colleagues (40 patients in total)122 were mutually exclusive with regard to participants. Even though a number of similarities between the two trials were observed, the characteristics of the two patient populations were clearly not identical and, therefore, we treated them as two separate trials. Correspondence with the trials investigators (Dr Dilek Memis named as corresponding author for both trials) proved unsuccessful and did not elicit any response.
The mean age was reported in 12 trials.71,72,109–112,114,116,117,120–122 With the exception of one trial112 that focused exclusively on young pregnant women (mean age 25.1 years in the dexmedetomidine group and 26.8 years in the control intervention group), the 11 remaining trials mean age ranged from 43 to 65 years for dexmedetomidine and from 40 to 67 years for the comparator interventions. The median age was reported in six trials55,69,70,102,123 and ranged from 49 to 65 years for dexmedetomidine and from 46 to 67 years for the comparator interventions.
Sixteen studies reported information regarding the sex of participants.55,69–72,102,109,110,112,114,116,117,120–122 Study populations tended to involve more men than women, with the exception of one trial that involved only pregnant women112 (see Appendix 5 for further details).
The severity of illness at baseline was reported in eight trials55,71,102,112,117,121–123 by means of the APACHE II scores or APACHE III scores (one trial).116 The APACHE II scores have a possible range of 0–71, whereas the APACHE III scores can range from 0 to 299. In both cases, higher scores indicate more severe disease and a higher risk of death.104 Across the eight trials that used APACHE II, scores ranged from a mean of 5.1112 to a mean of 22 for dexmedetomidine117 and a mean of 6112 to a mean of 20117 for the control sedative intervention. One trial102 reported a median APACHE II score of 29 for dexmedetomidine and of 27 for the control sedative intervention. The trial116 that assessed severity of disease using the APACHE III scores reported mean scores of 74.1 for dexmedetomidine and of 70.4 for midazolam.
presents an overview of the participants’ characteristics of the 18 included trials. It is worth noting that not all trials provided the same participant details or used the same measures to assess them.
Summary of main participants’ characteristics (for trials that reported this information)
Risk-of-bias assessment of included studies
presents the summary of the risk-of-bias assessments for all included trials. The risk of bias of individual studies is presented in .
Summary of risk-of-bias assessments of all included trials.
Risk-of-bias assessments of individual studies. CLON, clonidine compared with dexmedetomidine, LORAZ, lorazepam compared with dexmedetomidine; MIDAZOLAM, midazolam compared with dexmedetomidine; PROPOFOL, propofol compared with dexmedetomidine; SC, standard (more...)
Overall, out of the 18 included trials, four were judged to be at low risk of bias70,71,102 and seven at high risk of bias.72,111,114,117,120–122 For the remaining seven trials, there was not sufficient information to make an overall judgement.55,69,109,110,112,116,123
With regard to the assessment of selection bias, around half of the trials were judged to be at low risk (i.e. adequate sequence generation and allocation concealment),55,70,71,102,110,112,116,117,122 whereas the remaining eight trials did not provide sufficient information to formulate a proper judgement.69,72,109,111,114,120,121,123
In eight of the included trials, participants were reported to be blinded to the intervention received,69–71,102,109,111,116 whereas in six trials they were not.72,114,117,120–122 The remaining four trials did not report information on blinding of participants.55,110,112,123 Blinding of outcome assessor was addressed adequately in five trials,69–71,102 not adequately in seven trials72,111,114,117,120–122 and not reported in six trials.55,109,110,112,116,123
With regard to ‘incomplete outcome data’,10 trials had low withdrawal/discontinuation rates, which were balanced between intervention groups and, therefore, judged to be at low risk of bias.69,71,72,102,112,116,117,120–122 Two trials reported significantly higher discontinuation rates, owing to lack of efficacy, among people treated with dexmedetomidine, and were judged to be at high risk of bias.70 The remaining six trials did not provide sufficient information on which to make a definitive judgement.55,109–111,114,123
There was no evidence of selective reporting in any of the included trials, with the exception of one trial111 in which data on hypotension and bradycardia were mentioned only in the discussion section of the published paper and not properly reported in the results section. For this reason, the study was judged to be at high risk of selective reporting.
With regard to ‘other sources of bias’, nine trials declared financial support by manufacturers of sedative agents and were, therefore, judged to be at high risk of bias.69–71,102,116,120,121,123 One trial was judged at low risk of bias, as the authors clearly stated that no funding was received from manufacturers.55 The remaining eight studies were judged to be at unclear risk of bias, as the authors did not explicitly report their source of funding.72,109–112,114,117,122
Summary of clinical effectiveness
Random-effects meta-analyses of relevant clinical outcomes were performed when appropriate.
We had initially planned to perform subgroup analyses according to the type of clinical setting (patients admitted to ICUs following elective surgery compared with general ICU patients) if enough data had been available. However, only 6 of the 18 studies included patients who were admitted to the ICU after elective surgery, and not all of them provided data for all efficacy outcomes. Therefore, because of the dearth of suitable data, subgroup analyses according to the type of clinical setting were deemed unfeasible. As patients admitted to the ICU after elective surgery represent a distinct type of patient population (short duration of sedation and MV, and lower mortality rate), we deemed it inappropriate to combine trials that included patients after elective surgery with those that enrolled more general, critically ill ICU patients. The results of trials that enrolled patients after elective surgery were instead summarised narratively.
It is worth pointing out that there was considerable variation among included trials in the choice, definitions and measurements of outcomes, especially with regard to measures of ventilator dependence such as duration of MV, ventilator-free days, time to extubation or duration of weaning. Often, trials that assessed duration of MV did not report ventilator-free days as an outcome. The number of ventilator-free days was available from three trials, but details on measurement were lacking.69,118,121 Information on time to extubation was reported in six trials,70,71,111,114,123 but definition and criteria for extubation were not consistent across trials. Two large trials (MIDEX and PRODEX)70 reported both duration of MV and time to extubation, but did not provide a clear definition or measurement criteria for time to extubation and failed to discuss the clinical difference between the two measures. Similarly, duration of weaning was reported by two trials,69,114 but only one provided a proper outcome definition and a description of the measurement criteria.114
