Prospective multi-centre randomised, double-blind, equivalence study comparing clonidine and midazolam as intravenous sedative agents in critically ill children. The SLEEPS Study (Safety ProfiLe, Efficacy and Equivalence in Paediatric intensive care Sedation)

Benzodiazepines

Currently in the UK, midazolam is the most popular sedative used in critically ill children, usually given in combination with an opioid by intravenous (i.v.) infusion at doses between 50 and 300 µg/kg/hour. 5 Alternative agents to midazolam include diazepam, clonidine, chloral hydrate and promethazine. The limited data on midazolam suggest a high incidence of side effects: in two studies6,7 designed to observe adverse reactions (ARs) to sedative agents the reported incidence was as high as 35% for midazolam, and this was related to duration of the infusion and cumulative dose. The duration of abnormal behaviour after drug withdrawal was as long as 1 week. Limiting the benzodiazepine dose may delay the onset of tolerance but is often unobtainable because of the need to maintain adequate sedation. The frequency and severity of symptoms are related to the cumulative amount of drug given and the duration of the infusion and are commonly identified as agitation, prolonged crying, abnormal movements, vomiting and cardiovascular disturbance. 5 Of concern is that in a study on neonatal sedation and neurological outcome the use of midazolam appeared to have an adverse effect on outcome compared with morphine or placebo. 8 This has led to a significant reduction of midazolam use in the neonatal intensive care. Moreover, recent studies on neurodevelopment have raised serious concerns with regard to the effects of even relatively brief exposure of the young child to gamma-aminobutyric acid (GABA) agonists, including midazolam, in terms of long-term behavioural and intellectual development. 9,10 The intrinsic effects of midazolam and morphine on outcome have never been compared with other regimens.

Clonidine

In recent years, considerable interest has been shown in the use of α₂-agonist drugs as an alternative to benzodiazepines in intensive care sedation both in adults and children. 11 Clonidine is a lipid-soluble, partial α₂-agonist with antihypertensive, analgesic and sedative effects. Its primary antihypertensive action is attributed to its central α₂ effect on the sympathetic outflow, resulting in reduced heart rate, vasodilatation and lowered blood pressure (BP). 12,13 More recently it has gained recognition for its sedative and analgesic properties. The mechanism for the sedative and analgesic actions is not clear but is thought to be a combination of central effects that modulate descending inhibitory nociceptive mechanisms and spinal analgesia, acting on the dorsal horn of the spinal cord. 14 Elimination is through both hepatic metabolism to inactive metabolites and direct renal excretion. 12 It is a drug that may have a protective effect on the developing brain in that, unlike benzodiazepines, it is not associated with apoptotic changes on exposure to the drug. 15,16

Caudal epidural and spinal clonidine have been evaluated in paediatric anaesthesia. It has been shown to augment pain relief and increase the duration of postoperative analgesia with minimal side effects,17–19 and is now used routinely in paediatric practice. Given as an oral premedicant, clonidine has similar anxiolytic and improved sedative properties compared to preoperative benzodiazepines20 but, in addition, it can attenuate haemodynamic responses to nociception and provide postoperative analgesia. These effects on central sympathetic outflow and centrally based analgesia mechanisms reduces intraoperative anaesthetic requirements and metabolic responses to surgery. 21 In anaesthesia of the critically ill neonatal cardiac patient, this has shown to have improved outcome in terms of survival. 22

In the last 10 years, following experience with clonidine in paediatric anaesthesia and its use in adults withdrawing from alcohol and opioids, it has become increasingly used for sedation and analgesia in the critically ill child in the PICU. 5 However, despite its widespread use there are few data on effectiveness, dose requirement and safety. A limited dose-finding study in the PICU has demonstrated that it can provide dose-dependent sedation in place of morphine using an i.v. infusion rate of 1–2 µg/kg/hour without haemodynamic compromise in terms of heart rate, BP or cardiac output. 23 A small prospective study of critically ill children demonstrated that concomitant administration of oral clonidine significantly reduces morphine and lorazepam requirements without additional side effects. 24 Clonidine has a good safety profile in the general population, even in extreme overdose,25,26 although it can be associated with significant side effects that include bradycardia, hypotension and rebound hypertension. There remains an unmet need for improved sedation in the PICU, and although clonidine is being increasingly used in the clinical situation in the PICU, a formal objective evaluation of i.v. clonidine as an alternative to i.v. midazolam needs to be undertaken.

Primary objective

To determine whether or not i.v. clonidine can provide equivalent control of sedation in the critically ill child when compared with i.v. midazolam.

Secondary objectives

To determine whether or not clonidine reduces side effects associated with sedation practice in intensive care compared with midazolam at clinically appropriate dosing regimens. To determine if there are any benefits on clinical outcomes using clonidine compared with midazolam.

Inclusion criteria

1. Children aged 30 days (≥ 37 weeks’ gestation) to 15 years inclusive. Children born before 37 weeks’ gestation are eligible if they are a minimum of 30 days post delivery and their corrected gestation is ≥ 37 weeks.

2. Admitted to PICU, ventilated and likely to require ventilation for > 12 hours.*

3. Recruitment within 120 hours of arrival in the PICU/intensive care unit (ICU).*

4. Child is ≤ 50 kg in weight.

5. Able to perform a COMFORT score on the child.

6. Adequately sedated: COMFORT score within the range of ≥ 17 and ≤ 26.

7. Fully informed written proxy consent.

*Eligibility criteria amended during trial and summarised (see Table 3); details provided in Appendix 5.

Exclusion criteria

1. Those patients with open chests following cardiac surgery.

2. Those patients chronically treated for raised BP.

3. Current treatment with beta-blockers (if patients have not received beta-blockers for 24 hours prior to entry into the trial then they are eligible to participate).

4. Acute traumatic brain injury.

5. Status epilepticus or active fitting (two or more seizures regularly on a daily basis).

6. Those patients requiring haemodialysis or haemofiltration.

7. Those patients requiring extracorporeal membrane oxygenation (ECMO) treatment.

8. Those patients with severe neuromuscular problems/impairment on whom you cannot perform a COMFORT score.

9. Known allergy to either of the trial medications (clonidine, midazolam or morphine).

10. Current treatment with continuous or intermittent muscle relaxants.

11. Those patients known to be pregnant.

12. Currently participating in a conflicting clinical study or participation in a clinical study involving a medicinal product in the last month.

13. Previously participated in Safety profiLe, Efficacy and Equivalence in Paediatric intensive care Sedation (SLEEPS) trial.

The use of midazolam or clonidine to establish sedation did not preclude entry into the trial.

Loading dose of trial intervention and morphine

After consent and randomisation, and before starting the trial drugs, sedation with the pre-existing drugs were adjusted to ensure that the COMFORT score was within the desired range (17–26). This necessitated that the child was not on muscle relaxants and did not have suppressed motor function and, therefore, was evaluable by the COMFORT score. At this point the trial morphine and the study drug (either midazolam or clonidine) were then given in a standardised loading fashion, irrespective of the pre-existing drugs that had been used prior to study. Loading with i.v. morphine consisted of a dose of 100 µg/kg over 15 minutes. The trial drug was then administered, over 1 hour, from the syringe that had been made up in blinded fashion. For midazolam this corresponded to 200 µg/kg over 1 hour followed by an initial maintenance infusion rate of 100 µg/kg/hour. For clonidine this corresponded to a loading dose of 3 µg/kg over the first hour followed by an initial maintenance infusion rate of 1.5 µg/kg/hour. After this first hour, the pre-existing sedative/analgesic drugs were discontinued.

Maintenance rates of trial interventions

From this point onwards the study drug and, if necessary, morphine infusion rates were changed, in a formalised fashion, either upwards or downwards according to the objective measure of the COMFORT score. The incremental changes allowed five delivery options: for clonidine, 0.00, 0.75, 1.50, 2.25 and 3.00 µg/kg/hour; for midazolam 0, 50, 100, 150 and 200 µg/kg/hour.

Morphine usage

After the initial loading dose of morphine of 100 µg/kg over a 15-minute period, the initial maintenance infusion rate of morphine was set at 20 µg/kg/hour, with an option to increase the morphine infusion if the maximum treatment dose of the study drugs had been reached. In addition, provision was made for the bedside nurses to increase morphine infusion rates if they considered that the COMFORT score had risen through pain rather than sedation, allowing a morphine infusion of up to 60 µg/kg/hour. This was particularly relevant in the case of children who were in the PICU immediately after surgery. Multiple changes of the infusion rates were allowable within a 1-hour period, provided that there was an accompanying COMFORT record that supported a change in infusion rate.

The flow diagram used to direct infusion rates of morphine and trial drug in response to the COMFORT score are shown in Figure 1.

FIGURE 1.

Trial interventions and morphine. a, If a COMFORT score of < 17 is recorded for 2 consecutive hours then reduce morphine or trial sedation as clinically indicated incrementally down to 20 µg/kg/hour of morphine and to the minimum trial infusion rate for the weight group. If minimum infusion rate is administered and the morphine infusion rate is 20 µg/kg/hour and the child still has a COMFORT score below 17, then if there are no analgesic requirements, the morphine can be further decreased by an increment to 10 µg/kg/hour. If the COMFORT score is still below 17, the morphine can be stopped (providing there are no analgesic requirements). If the COMFORT score is still below 17, the trial sedation can be temporarily stopped until the COMFORT score rises to 17 or more.

Interventions

In addition to recording hourly COMFORT scores and modulation of the infusion rates required to maintain children within the desired 17–26 score range, events either related to PIC or additional sedative/analgesia control were documented during the study.

Children who became unsettled and were outside the ideal COMFORT score within each hourly period were brought back into the acceptable range with increase in trial drug or morphine according to the treatment protocol. If the change was deemed to be urgent or there was clinical need then sedation, anaesthesia and, if necessary, rescue muscle relaxant drugs could be given at any point. However, If this occurred three times in a 12-hour period, the trial treatments were deemed to have failed and the study drugs replaced with conventional medication according to individual PICU practice.

Children who required invasive procedures or investigations, such as magnetic resonance imaging (MRI) scan or computed tomography scan were allowed to remain in the study provided that the study drugs could be continued throughout. In this situation, anaesthetic, analgesic or muscle relaxant drugs could be administered to ensure appropriate unconsciousness, pain relief and, if necessary, immobilisation. If muscle relaxants were administered, this temporarily caused some of the behavioural measures of the COMFORT score to be invalid, and therefore the BP and heart rate became the sole measures of the COMFORT score until muscle function returned.

The study drugs were continued until the patient had recovered sufficiently to allow extubation, or had completed 7 days of the study drug, or had failed treatment due to reaching the maximum allowed dose of study drug and morphine and still inadequate after an hour, or had required more than two rescue treatments in any 12-hour period. In addition, patients requiring advanced organ support, such as haemofiltration or extracorporeal life support, did not continue to receive study drugs, although they continued to be monitored as part of the study.

Primary outcome

Adequate sedation defined as at least 80% of total evaluated time spent sedated within a COMFORT score range of 17 to 26.

Secondary outcomes

Sample size calculations

Sample size calculations were undertaken using NQuery Advisor software version 4.0 (Statistical Solutions, Saugus, MA, USA).

The trial was originally designed with an equivalence margin of ± 0.10. During teleconferences with PIs at sites it was suggested that this margin was too narrow. Recruitment into the trial was challenging, and consideration was given to widening the margin of equivalence as suggested by site PIs. The revision to the sample size calculation was submitted to ethics and the Medicines and Healthcare products Regulatory Agency (MHRA) on 18 April 2011 and accepted by the MHRA on 26 May 2011 and ethics on 15 June 2011. The original and revised sample size calculations are presented in full in the next two sections below.

Original sample size calculations

The proportion of children adequately sedated on midazolam is reported to be 0.65,45 with an expected proportion of 0.66 on clonidine. A two-group, large-sample normal approximation test of proportions with a two-sided 5% significance level to have 80% power to reject the null hypothesis that midazolam and clonidine are not equivalent (with margin of equivalence ± 0.10) would require 440 children in each group. The trial would therefore aim to recruit a total of 1000 children across both treatment groups to allow for approximately 10% loss to follow-up.

Sample size calculation revision

The revised sample size calculation uses a 15% margin, as agreed by the PIs and Trial Steering Committee (TSC) members, and indicates the statistical power that could be achieved with expected recruitment rate. Owing to observed completeness of the data collected at the time of the sample size revision, the 10% loss to follow-up adjustment was removed.

The proportion of children adequately sedated on midazolam is reported to be 0.65,45 with an expected proportion of 0.66 on clonidine. When the sample size in each group is 125, a two-group large-sample normal approximation test of proportions with a one-sided 0.025 significance level will have 64% power to reject the null hypothesis that the test and the standard are not equivalent (the difference in proportions, pT – pS, is 0.150 or farther from zero in the same direction) in favour of the alternative hypothesis that the proportions in the two groups are equivalent.

Interim monitoring

Safety profiLe, Efficacy and Equivalence in Paediatric intensive care Sedation was monitored by an Independent Data and Safety Monitoring Committee (IDSMC), having agreed procedures based on a Charter. 46 The IDSMC was responsible for reviewing and assessing recruitment, interim monitoring of safety and effectiveness, trial conduct and external data. The extent and type of missing data were monitored, and strategies were developed to minimise its occurrence.

All interim analysis results were confidential to the IDSMC members. The IDSMC considered patient safety alongside treatment efficacy when making recommendations regarding continuation, amendment or discontinuation of the trial. In order to estimate the effect of clonidine and midazolam for the primary outcome it was planned that the Haybittle–Peto approach would be used for requested interim analyses considering superiority, with 99.9% confidence intervals (CIs) calculated for the effect estimate. This method was chosen to ensure that interim efficacy results would have to be extreme before recommending early termination in order to be convincing to the clinical community.

Health economic methods

The economic evaluation aimed to assess the cost-effectiveness of clonidine compared with midazolam in the treatment of critically ill children using clinical data from the SLEEPS trial. The cost-effectiveness analysis focused on the short-term costs and consequences of the two trial drug interventions. The primary analysis (base case) used cost data from the point of randomisation until 14 days post-treatment cessation, and was carried out from a UK NHS hospital services perspective. We used a cut-off of 14 days post-treatment cessation as the time horizon of interest for the base case analysis. This decision was based on discussions with clinical experts and published evidence of clinical effectiveness describing the use of sedative agents in a PIC context. 12,23 The measure of benefit used in the economic evaluation mirrored that adopted for SLEEPS as a whole, namely an additional case of adequate sedation defined within the primary outcome as ‘at least 80% of total time spent sedated within a COMFORT score of 17 to 26’. Given the methodological limitations surrounding preference-based outcomes measurement in young children, outcomes were not expressed in terms of preference-based metrics, such as the quality-adjusted life-year (QALY).

A number of sensitivity analyses were planned to test the robustness of the base case economic evaluation results. In addition, a scenario analysis was planned, adopting a wider perspective and including additional direct NHS economic costs [e.g. those costs attributable to GP visits and accident and emergency (A&E) attendances] utilising data from the point of randomisation to 14 days post-treatment cessation.

Collection of resource-use data

The SLEEPS study CRFs captured all resource use related to the child’s primary hospital admission, including trial drug treatments as well any transfers between wards and hospitals. Wider NHS resource use (e.g. GP visits, A&E visits, and readmissions to hospital) that took place within 14 days of treatment cessation were also recorded.

Specifically, individualised hospital services resource use was estimated for trial drug interventions, including pharmaceuticals and consumables [e.g. clonidine, midazolam, morphine, needle, syringe, line extension kit, line filter), duration of primary hospital admission in PICUs, ward transfers, length of stay (LoS) in any ward post PICU [e.g. stays in high-dependency units (HDUs) and/or general medical paediatric wards (GMs)], hospital–hospital transfers, and theatre time incurred during the treatment of SAEs.

All resource-use data were entered directly from the SLEEPS CRFs into the MACRO (InferMed Ltd, London, UK) trial database, with in-built safeguards against inconsistent entries.

Valuation of resource-use cost data

Unit costs for resources used by children who participated in the study were obtained from a variety of secondary sources. All unit costs utilised followed recent guidelines on costing health-care services as part of an economic evaluation [National Institute for Health and Care Excellence (NICE47)]. Where necessary, secondary information was obtained from ad hoc studies reported in the literature. Unit costs of hospital and community health-care costs were largely derived from national sources and took account of the cost of the health professionals’ qualifications (Curtis48). All PICU and HDU costs were valued using the NHS Reference Costs,49 a catalogue of costs compiled by the Department of Health in England (Department of Health49).

The main cost driver in the economic evaluation was the cost of critical care. The 24-hour critical care (per diem) NHS reference cost was calculated on a full absorption costing basis and included hotel services, nursing/medical and other clinical staff, therapy services and staff, ward consumables, blood and blood products, drugs, diagnostics, and medical and surgical equipment. Stays in critical care were valued using a half-a-day PICU cost for periods of < 12 hours and a full-day PICU cost for periods of between 12 and 24 hours.

Drug costs were obtained from the British National Formulary (BNF 201250) and Monthly Index of Medical Specialities (MIMS 201351). Consumables were costed using data from NHS Supply Chain catalogue. 52 All costs were expressed in pound sterling and valued at 2011–12 prices (with the exception of a small number of drug costs; see Appendix 10). None of the costs were inflated or deflated for use in the economic evaluation. For the base case analysis, unit costs were combined with resource volumes to obtain a net cost per child covering all categories of hospital costs. A range of sensitivity analyses also explored the implications of uncertainty surrounding the values of key cost parameters (described below). Further details on the methods used to value resource use are provided in Appendix 10.

Cost-effectiveness analytic models

As described above, the primary measure was an additional case of adequate sedation. The results of the economic evaluation were restricted to the patients for whom the primary outcome in the SLEEPS trial was available. In the cost-effectiveness analysis, the incremental cost-effectiveness ratio (ICER) was calculated as the difference in average costs (ΔC) divided by the difference in average effects (ΔE) between the clonidine and midazolam groups and expressed as the incremental cost per case of adequate sedation. No discounting of costs or benefits to present values was necessary as the time horizon of the economic evaluation (period of follow-up) was < 12 months.

Independent-sample t-tests were used to test for differences in resource use, costs and primary clinical outcomes between treatment groups. All statistical tests were two-tailed. Differences in resource use, costs and effects between the comparator groups were considered significant if two-tailed p-values were ≤ 0.05.

In common with many trial-based economic evaluations, the distributions for costs were skewed. Consequently, non-parametric bootstrap estimation was used to derive 95% CIs for mean cost differences between the comparator groups. 53 Each of these CIs was calculated using 1000 bias-corrected bootstrap replications. Non-parametric bootstrap simulation of the cost–effect pairs was also performed to generate 1000 replications of the ICER; these were subsequently represented graphically on a four-quadrant cost-effectiveness plane as described by Black et al. 54 As illustrated in the paper by Stinnett and Mullahy,55 mean net benefits, defined as R_c.ΔE – ΔC, were estimated for alternative values of R_c, the maximum acceptable ICER or cost-effectiveness threshold for the primary outcome, namely each additional case of adequate sedation. Although both stated and revealed that preference techniques have been used to estimate maximum acceptable ICERs or cost-effectiveness thresholds for generic measures of health outcome, such as the QALY (Gray et al. 56), no comparative data are available for the primary health outcome for this study. Consequently, the cost-effectiveness threshold was varied in our analyses between hypothetical values of £0 and £5000 per additional case of adequate sedation. A value of £1000 per additional case of adequate sedation was selected as the primary cost-effectiveness threshold for statements about cost-effectiveness and mean net benefits.

Cost-effectiveness acceptability curves (CEACs) showing the probability that clonidine is cost-effective relative to midazolam across a range of cost-effectiveness thresholds were also generated based on the proportion of bootstrap replicates with positive incremental net benefits. 55,57 The probability that clonidine is less costly or more effective than midazolam was based on the proportion of bootstrap replicates that had negative incremental costs or positive incremental health benefits, respectively.

Sensitivity analyses

Several sensitivity analyses were undertaken to assess the impact on cost-effectiveness results of areas of uncertainty surrounding components of the base case economic evaluation. All of these sensitivity analyses comprised complete data for 120 children (mirroring the strategy adopted for the primary efficacy assessments):

• Sensitivity analysis (1) We varied the cost of higher-level inpatient care (stay in the PICU or HDU) by applying upper-quartile NHS Reference Costs49 across trusts.

• Sensitivity analysis (2) We varied the cost of higher-level inpatient care (stay in the PICU or HDU) by applying lower quartile NHS Reference Costs49 across trusts.

• Sensitivity analysis (3) We used exact proportions of 24-hour periods to value total lengths of stay rather than apply either one half day or full day per diems to 0–12 hour or 12–24 periods for costing purposes.

• Sensitivity analysis (4) We extended the time horizon of the economic evaluation to cover the period between randomisation and 14 days postventilation cessation.

• Sensitivity analysis (5) We widened the primary outcome definition to ‘. . . at least 75% of total time spent sedated within a COMFORT score of 17 to 26’.

• Sensitivity analysis (6) We narrowed the primary outcome definition to ‘at least 85% of total time spent sedated within a COMFORT score of 17 to 26’.

In addition, a scenario analysis was performed, which comprised 106 children for whom complete data were available, and included wider NHS costs incurred within 14 days post-treatment cessation. These wider costs included costs associated with GP visits, A&E attendances and hospital readmissions. These additional costs are unlikely to be attributable to choice of sedative agents used in PICUs; hence their relegation to a scenario analysis.

Recruitment rate targets

The initial target sample size of the trial (1000 participants) was expected to be achieved within a 2-year recruitment period. This was based on average accrual of one patient per week at each of 12 sites. The proposal was presented to the Paediatric Intensive Care Society Study Group (PICSSG) and each site agreed with the target recruitment rates and considered them achievable at the outset.

Screening

A total of 10,023 participants were screened and assessed for eligibility to be randomised, of whom 9196 did not meet the inclusion criteria, leaving 827 who were eligible. A summary of the screening results by site is provided in Table 4, with the reasons for ineligibility provided in Table 5 and 6. The most common reasons for ineligibility were that the patient was aged < 30 days; did not require sedation; was not intubated; was not expected to be ventilated for sufficient time; or was on muscle relaxants.

Screening logs were monitored throughout the trial. Teleconferences and a face-to-face meeting were held with PIs and research nurses to learn from the processes and experiences at sites with the highest recruitment rates, and to identify and resolve barriers to recruitment across sites. This led to changes to the inclusion criteria to increase the time allowed for the child to be on PICU prior to randomisation; decrease the time for which the child was expected to be ventilated (key amendments summarised within Table 3 and details provided within Appendix 5); and increase the flexibility in the administration of the interventions and concomitant medications, including morphine.

Eligible patients

The number of eligible participants was lower than anticipated and the percentage converted to randomised participants much lower with considerable variation across sites. Figure 2 displays predicted recruitment compared with actual recruitment across sites during the trial recruitment period. The recruitment period was extended, and the original and revised recruitment curves are displayed. Of the 827 eligible participants, 698 (84%) participants were eligible and not randomised. Key reasons for eligible patients not being randomised are provided in Table 7 and show that 14% patients were missed, consent was not obtained for 23% (194/827) and various other reasons for 41%. A breakdown of ‘other reasons’ provided are given in Appendix 8, Table 40.

FIGURE 2.

Expected versus actual recruitment rates.

Discussions and monitoring across sites also led to suggestions that consent rates (Table 8) were higher the earlier that parents were first approached to discuss the trial. This was supported by data recording times of first contact and consent. The importance of an early approach to inform parents that the hospital was participating in the trial and that they maybe approached at a later time for consent was stressed to sites. However, all sites recognised that parents were more reluctant to enter the trial than expected when the trial was initially planned. Parents entering into PICU with a critically ill child were reluctant to give consent once the child had stabilised on the standard sedation medication, even although the agents used were often the same drugs as used in the study. This reflects both the fear of changing from something that was perceived as being one stable aspect of their care to an unknown, and the societal change to be less willing to participate in paediatric research studies (consent rates for drug based/intervention paediatric studies have fallen in recent years). In addition, pressures on PICU beds, the need for reduced patient day occupancy on PICU, and techniques in non-invasive ventilation have driven forward clinical techniques of early extubation. Although this may have been beneficial for patient care, a side effect of this has been the reduction in available ventilated clinical cases that could be entered into the study. Specifically, children undergoing cardiac surgery that would have been ventilated for several days 5 years ago are now being extubated the same day or even in the operating theatre (fast-track and ultra-fast-track surgery). 58 At the planning stage of the study it had been envisaged that a significant number of the patient recruitment would come from the postoperative cardiac patients, but owing to the above issue the best recruitment came from non-cardiac intensive care patients and only four patients were entered into the trial post cardiac surgery.

Additional feedback from the sites was that the inclusion criteria requiring the child to be adequately sedated at the time of randomisation were a barrier to participation. This appeared within the ‘other reasons’ provided across sites and was discussed extensively among the trial management team, PIs and research nurses, but the eligibility criteria could not be amended. It was felt by participating centres that commencing the study when baseline control of sedation had not been obtained would be unacceptable for clinical management, even although it was acknowledged that loading of sedative and analgesia drugs with the commencement of the study would improve the COMFORT scores towards the target range.

The reasons why the eligible 698 patients were not randomised; more than one reason per patient could be recorded are summarised above (see Table 7).

During trial treatment

The percentages of time spent adequately sedated (Table 16) per group are similar (73.86% clonidine, 72.73% midazolam). The lower limits of the IQRs demonstrate the difficulty in maintaining adequate sedation, with one-quarter of participants being adequately sedated only ≤ 58% of the time across both groups.

Competing risk analyses were conducted for time to maximum permitted dose of sedative and time to maximum permitted dose of morphine instead of the pre-planned Kaplan–Meier analyses. Kaplan–Meier analyses were not carried out because participants withdrew from treatment due to reasons such as experiencing an AE, receiving muscle relaxants, etc. (all reasons listed in Table 34) before the events of interest could be observed. The treatment withdrawals for other reasons constitutes a competing risk, and a cumulative incidence analysis for time to events is more appropriate.

Of the 125 participants (64 clonidine, 61 midazolam) that received at least one dose of trial treatment, a total of 19 participants (15.2%) were a treatment failure, with 12 (18.8%) in the clonidine group and seven (11.5%) in the midazolam group. As the midazolam participants tended to be on treatment for longer, a post hoc ‘time to treatment failure’ analysis was conducted.

Table 17 shows the categorisation of participants for these competing risks analyses. Those participants who withdrew from treatment due to sedation no longer being required or those who completed the full 7 days of trial treatment are censored. The cumulative incidence curves for competing reasons of reaching maximum permitted dose of sedative/morphine and treatment withdrawal for each treatment group are shown in Figures 5–7.

FIGURE 5.

Time to reach the maximum permitted dose of sedation: cumulative incidence plot.

FIGURE 6.

Time to reach the maximum permitted dose of morphine: cumulative incidence plot.

FIGURE 7.

Cumulative incidence curves for competing reasons of treatment withdrawal for clonidine and midazolam.

For time to maximum permitted dose of sedative (Table 18), the Gray’s test60 indicates no difference detected between the treatments for either competing risks [p-values of 0.75 (reaching maximum dose of sedative), 0.07 (treatment failure) and 0.11 (other reasons)].

For time to maximum permitted dose of morphine, the Gray’s test60 indicates no difference detected between the treatments for either competing risks [p-values of 0.88 (reaching maximum dose of morphine), 0.10 (treatment failure) and 0.15 (other reasons)].

Therefore, overall there were no differences detected in the time taken to reach the maximum permitted doses of allocated sedative or morphine. Similarly, there were no differences in the proportions who reached the maximum dose (see Table 20).

For time to treatment failure, the Gray’s test60 indicates no difference detected between the treatments for either competing risks [p-values of 0.12 (treatment failure) and 0.19 (other reasons)]. Similarly, there were no differences in the proportion of patients experiencing treatment failure (see Table 20).

The mean profile rise of daily cumulative sedative infusion is given in Figure 8 and cumulative morphine infusions in Figure 9. The numbers contributing data at each time point for each analysis are given in Table 19. The figures show large overlapping regions of the standard error bars at each time point in the cumulative mean profile plots. The increasing widths of the standard error bars within the figure demonstrate increasing uncertainty as the number of participants contributing data decrease with increasing time according to Table 19. Note that post 80 hours there is only one patient in the clonidine group and hence the standard error bar is 0.

FIGURE 8.

Mean profile plots for cumulative sedative infusion. Note: the bars on the plot are one standard error bars. Where there are no standard error bars on the clonidine profile there is only one patient providing data.

FIGURE 9.

Mean profile plots for cumulative morphine infusion. Note: the bars on the plot are one standard error bars. Where there are no standard error bars on the clonidine profile there is only one patient providing data.

The results of the longitudinal mixed model for mean profile, rise in daily cumulative sedation gave least-squares mean estimates of 3.28 and 3.14, with standard errors of 0.17 and 0.17, respectively, for clonidine and midazolam. The difference in least-squares means and 95% CI is 0.14 (–0.33 to 0.61), with a p-value of 0.56.

From the longitudinal mixed model for mean profile, rise in daily cumulative morphine infusions gave least-squares mean estimates of 33.70 and 33.86, with standard errors of 1.80 and 1.80, respectively, for clonidine and midazolam. The difference in least-squares means and 95% CI is –0.16 (–5.20 to 4.88), with a p-value of 0.95.

A post hoc analysis of cumulative sedative/morphine infusion data has been summarised in 5-hour intervals with medians and IQRs split by patients who achieved the primary outcome of ≥ 80% adequately sedated in Appendix 11, Tables 46 and 48, and those patients who did not achieve the primary outcome of ≥ 80% adequately sedated (see Appendix 11, Tables 47 and 49). This post hoc summary aims to provide information for clinicians on dose response for those who were adequately sedated and those who were not.

There were no differences identified in the proportion of participants who had at least 1 day with a fall in BP requiring intervention (Table 20) or the number of days with a fall in BP judged by the clinician to require intervention.

A higher proportion of participants on clonidine (5/64, 11.1%) than on midazolam (3/61, 5.8%) required increased inotropic support in the first 12 hours after randomisation (see Table 20); however, the number of events is small and the CI width is wide. This variable was added to data collection part of the way through the trial and is available for 77.6% of participants. In addition, at the time of consent a greater proportion of participants on clonidine were on inotropic support (30.3% clonidine vs. 14.8% midazolam). This baseline variable was collected throughout the trial but incompletely recorded, being available in 46.5% of participants. Of the five who required increased inotropic support in the first 12 hours on clonidine, only one participant was known to be receiving inotropic support at baseline. The status of the three participants on midazolam at baseline is unknown.

Similarly, there were no differences in the proportion of patients with at least one instance requiring supplementary analgesia during sedation (Table 20).

Table 21 shows the number of instances when participants required supplementary analgesia/sedation. Participants randomised to midazolam received supplementary analgesia with greater frequency than those receiving clonidine (p = 0.01). Table 21 also shows the median and IQR for time on treatment for each number of instances participants required supplementary analgesia/sedation. This shows that the number of instances increases with the length of time on sedation, with Table 13 showing those allocated to midazolam were sedated for longer. Details of the supplementary analgesia used are provided in Table 22 with reasons for use in Appendix 9.

There were no differences between the groups for daily urine output (Table 23).

There were insufficient data to be able to analyse the urinalysis outcomes, as this was not required to be routinely measured.

No discernible differences were noted for any of the blood biochemistry variables (Table 24). The numbers to have at least one abnormal result not expected for the patients’ condition were low (just 0–4 for each group). No patients in either of the two groups had any abnormal results that were not expected for the patient’s condition for chloride or creatinine.

Secondary outcomes post-trial treatment phase

Results for the outcome time from stopping sedation to being fully awake are available in Table 25 and Figure 10. Being fully awake was determined by a score of 4 or 5 on the alertness category of the COMFORT score sustained for ≥ 2 hours or more. The Kaplan–Meier curve and the hazard ratio suggest that children who had been allocated to midazolam tended to take less time to be fully awake than those allocated to clonidine (see Table 25). This is also supported by the results for the outcome ‘fully awake within 24 hours’ provided in Table 26; however, results are not statistically significant.

FIGURE 10.

Time from stopping all sedation to being fully awake: Kaplan–Meier plot. Note: ‘+’ = censored; log-rank p = 0.0898.

There were no differences in the proportions experiencing withdrawal symptoms or the ‘average total score per day’ (Table 27, not significant); however, a higher proportion of participants allocated to midazolam required clinical intervention for those symptoms (see Table 26).

Signs of withdrawal were measured using an 11-point assessment for abnormal behaviour and were recorded until 5 days following trial treatment cessation or until discharge, whichever was soonest. The 11 descriptors that make up this assessment are scored 0 = none, 1 = mild (does not interfere with routine activities), 2 = moderate (interferes with routine activities) and 3 = severe (impossible to perform routine activities). Therefore, higher scores indicate worse withdrawal symptoms. Table 28 gives details of the text descriptors provided when the ‘Other’ category was selected.

There was just one case of rebound hypertension in the study. This was a mild case for a patient in the clonidine group.

Adverse reactions

Adverse reactions by severity

Serious adverse events

Serious adverse events by severity

There was one patient who had multiple SAEs, the first being ‘Failed extubation requiring reintubation’ with a severity of ‘severe’, and the other being ‘Postoperative wound infection’ with a ‘mild’ severity.

Line listings for SAE data are provided in Appendix 7, Table 50.

Two patients in the study had SAEs that were assessed as sudden unexpected serious adverse reactions (SUSARs). Both occurred in patients receiving clonidine. In one patient (as discussed previously), the heart rate fell to a low point of 64 beats per minute (bpm), 2 hours after completing the loading dose and during the maintenance phase. Although other patients receiving clonidine did have either significant heart rate reduction or BP that required intervention, this was the only event that resulted in withdrawal from study, prompted a discussion with the principal investigator (PI) at the local centre and was reported to the IDSMC. In normal clinical practice without blinding, clinicians would be aware that clonidine was being used and there would be more confidence in either treating the problem or reducing the dosage as it occurred. The second SUSAR involved a failed extubation after the use of clonidine. Although this was reported as a SUSAR, this was in a complex post cardiac patient and failure of extubation in these circumstances is not uncommon. However, the team in the local centre had not expected the child to fail extubation and the clinical features were of pulmonary oedema. It was therefore reported as a SUSAR and reviewed by the IDSMC.

Completeness of follow-up

There were two phases to the trial: during treatment phase and post-treatment follow-up. All 125 participants who received at least one dose of trial treatment have a reason for the end of the treatment phase (Table 34; for ‘other’ reasons, see Table 35). Multiple reasons were indicated for cessation of treatment in four participants (one clonidine, three midazolam), with details provided in Table 36. Of note, the number of treatment failures and AEs were small but a higher proportion occurred in the clonidine group.

Post-treatment follow-up consent was withdrawn for three participants (two clonidine, one midazolam). At least one post-treatment withdrawal assessment was carried out on the two clonidine participants prior to withdrawal of consent. The midazolam participant withdrew consent at the time of treatment cessation so had no post-treatment withdrawal assessments.

Six participants (four clonidine, two midazolam) withdrew from study at treatment cessation for reasons other than withdrawal of consent, so no post-treatment follow-up data were collected. The reasons why these seven participants came off treatment are as follows:

• Three participants (two clonidine, one midazolam) required continuous muscle relaxation.

• One clonidine participant was extubated but then needed reintubating, as he/she needed to be paralysed and sedated.

• One clonidine participant was lost to follow-up.

• One midazolam participant was a treatment failure and no post-treatment data were recorded.

In addition to the clonidine patient who was lost to follow-up with no post-treatment data, there were four more participants (three clonidine, one midazolam) who were lost to follow-up, who had some post-treatment follow-up data collected. They were all transferred without any further data collection. All four participants had a reason for treatment cessation of ‘sedation no longer required’.

Only two participants, both on midazolam, required sedating for > 7 days.

Resource use and costs

Table 37 provides a summary of the key resource-use values for each arm of the SLEEPS trial; results are presented separately for the clonidine and midazolam groups. There were no statistically significant differences between the trial arms in any category of resource use, with the exception of length of time on treatment; patients in the midazolam group had a statistically significant longer time on treatment than patients in the clonidine group. There were no deaths during the two time horizons considered in the economic evaluation.

Table 38 shows clearly that the most costly resource category was LoS in hospital. Three specific categories of LoS were estimated: LoS in admitting ward (PICU), LoS in any ward after PICU (this may have included stays in HDUs, GM wards or a return to PICUs) and LoS in GM wards in a different hospital. All other costs (drugs, consumables, SAEs and transfers) were relatively inexpensive compared with the per diem costs associated with hospital admissions.

Statistical analysis revealed that there were no statistical differences, at the 5% level, between the two trial groups in any cost category when all 120 children considered in the primary efficacy assessments were included in the economic analyses and the perspective was restricted to NHS hospital costs only. The mean total NHS hospital service cost in the clonidine group (n = 61) was £11,445 and the mean total NHS hospital service cost in the midazolam group (n = 59) was £12,276, generating a mean difference in costs of –£831 (p = 0.494).

Results of the cost-effectiveness analysis

The economic evaluation assessed the cost-effectiveness of clonidine compared with midazolam in terms of natural units of health gain, expressed as the incremental cost per additional case of adequate sedation. The time horizon in the base case analysis covered the period from randomisation to 14 days post-treatment cessation. The incremental cost-effectiveness of clonidine is shown in Appendix 7, Table 51. For the 120 children receiving clonidine (n = 61) or receiving midazolam (n = 59), we had complete cost and outcomes data. Within the base case analysis, the average cost was £11,445 in the clonidine group compared with £12,276 in the midazolam group, generating a mean cost saving of £831 (p = 0.494). There was no statistically significant difference in total costs between the two groups, with 71% of bootstrap replicates suggesting that clonidine is, on average, less costly than midazolam in terms of hospital service costs. However, the results of the cost-effectiveness analysis should be interpreted with caution, as the SLEEPS trial did not have sufficient power to identify any difference in the primary outcome between children in the trial arms should there have been one.

In the base case analysis, the incremental cost-effectiveness was estimated at –£21,216 per additional case of adequate sedation. However, there was substantial uncertainty around this finding. The variability around the base case estimate of cost-effectiveness is evident in the cost-effectiveness plane shown in Figure 11.

FIGURE 11.

Cost-effectiveness plane for base case cost-effectiveness analysis.

As the bootstrapped replications fall across all four quadrants of the cost-effectiveness plane, the CI around the mean ICER is difficult to interpret. For example, a negative ICER might represent lower costs and improved outcomes attributable to clonidine (south-east quadrant of cost-effectiveness plane) or higher costs and worse outcomes (north-west quadrant). Similarly, a positive ICER might represent higher costs and improved outcomes attributable to clonidine (north-east quadrant of cost-effectiveness plane) or lower costs and worse outcomes (south-west quadrant). As a result, a meaningful ordering of the bootstrapped replicates required to make the CI surrounding the mean ICER interpretable is very difficult. Under these circumstances, CEACs provide an appropriate approach to representing the uncertainty surrounding the mean ICER.

The CEAC for the primary clinical outcome measure is shown in Figure 12, and indicates that, despite being relatively flat, the higher the value that decision-makers place on an additional case of adequate sedation, the slightly higher the probability that clonidine will be cost-effective. At the notional cost-effectiveness threshold (or ceiling ratio) of £1000 per additional case of adequate sedation, the probability that clonidine is cost-effective compared with midazolam is 73%. Although no previous research has shown how much society or the NHS may or should be willing to pay for an additional case of adequate sedation for this group of children experiencing intensive care, the economic burden of not being able to adequately sedate a child is likely to be significant. Indeed, a recent NICE clinical guideline61 that focuses on sedation in children and young people states that ‘sedation failure is both distressing for the child and has major NHS cost implications’. If decision-makers are willing to pay as much as £5000 per additional case of adequate sedation then the probability that clonidine is cost-effective compared with midazolam increases to 76%.

FIGURE 12.

Cost-effectiveness acceptability curve for base case cost-effectiveness analysis.

Mean net benefits were estimated for alternative cost-effectiveness thresholds per additional case of adequate sedation (see Appendix 7, Table 52). Assuming that the cost-effectiveness threshold equals £1000 per additional case of adequate sedation generates a mean net benefit to the health service attributable to each additional use of clonidine of £679 (i.e. on average, there is a net gain to the health service in monetary terms). This is analogous to stating that if the actual benefit of clonidine, in terms of additional cases of adequate sedation, is multiplied by a willingness to pay of £1000 per additional case of adequate sedation, and the net cost is subtracted, then the benefit to the NHS of adopting clonidine is, on average, positive in monetary terms. Note, however, that the 95% CI surrounding the mean net benefit (–1818 to 3086) includes negative values, i.e. there is a possibility of a net monetary loss associated with clonidine (see Appendix 7, Table 52). If the cost-effectiveness threshold is increased as high as £5000 per additional case of adequate sedation, the mean net benefit increases to £932 (95% CI –£1799 to £3212).

Sensitivity analyses were conducted to determine the impact of changing particular parameter values or assumptions on the magnitude of the mean ICER (see Appendix 7, Table 52). Assuming that higher-level inpatient care (e.g. stays in PICUs or HDUs) is valued using upper quartile NHS Reference Costs49 results in the mean cost difference between the trial arms increasing to –£997, with a corresponding ICER of –£24,933. Assuming that higher level inpatient care (e.g. stays in PICUs or HDUs) is valued using lower quartile NHS Reference Costs49 results in the mean cost difference between the trial arms falling to –£716 with a corresponding ICER of –£18,299. In both sensitivity analyses, the probability of clonidine being cost-effective compared with midazolam at a £1000 cost-effectiveness threshold increases from baseline. Assuming that part of a day spent by a child in an inpatient ward equates to a proportional period for costing purposes and, that, consequently, the vacated inpatient bed would be filled immediately, reduces the mean cost difference between the trial arms to –£753, with a corresponding ICER of –£19,224; under this assumption, there is no change from baseline in terms of probability of cost-effectiveness. Varying the costs of care associated with hospital admissions does not have a substantial effect on the magnitude of the base case ICER.

Extending the time horizon of the economic evaluation results in a mean cost difference of –£809, with a corresponding ICER of –£20,651. Even although extending the time horizon meant that for some children a slightly longer length of hospital stay was captured by the economic evaluation, and for one child the additional cost of a SAE was also captured, the size of the mean ICER does not vary substantially. The probability of clonidine being cost-effective in this sensitivity analysis is higher (76%) than the baseline value (73%).

The primary clinical outcome in the trial was framed around a case of adequate sedation; the definition of adequate sedation was ‘at least 80% of total time sedated within a COMFORT score range of 17 to 26’. In post hoc sensitivity analyses, we increased this proportion to 85% and also reduced this proportion to 75%. With a narrower definition of adequate sedation (85%), there was a reduction in the mean effect size in both groups and an increase in the mean difference in effect size (0.06); the corresponding ICER was –£13,979. With a broader definition of adequate sedation (75%), there was an increase in the mean effect size in both groups and an increase in the mean difference in effect size (0.07); the corresponding ICER was –£12,111.

In summary, all of the mean ICERs generated in the base case analysis and sensitivity analyses are negative, suggesting that clonidine is, on average, more effective and cheaper than midazolam. However, none of the differences in mean costs or consequences between the comparison groups was statistically significant, regardless of assumptions surrounding key parameters of the economic evaluation over which there was a degree of uncertainty. Under these circumstances, it is important to assess the likelihood that clonidine is cost-effective, primarily through the use of CEACs, rather than testing any particular hypothesis concerning its cost-effectiveness.

Cost-effectiveness acceptability curves generated following each sensitivity analysis are shown in Figure 13. Estimates of net monetary benefits for notional cost-effectiveness thresholds for an additional case of adequate sedation are shown in Appendix 7, Tables 52 and 53. For example, assuming that the cost-effectiveness threshold equals £1000 per additional case of adequate sedation, adopting a broad definition of adequate sedation (at least 75% of total time spent sedated within a COMFORT range of 17–26) generates a mean net benefit to the health service of £933, attributable to clonidine (i.e. there is a net gain to the health service in monetary terms). Note, however, that the 95% CI surrounding the mean net benefit (95% CI –£1414 to £3426) includes negative values, i.e. there is a possibility of a net monetary loss associated with clonidine (see Appendix 7, Table 53).

FIGURE 13.

Cost-effectiveness acceptability curves for base case analysis and sensitivity analyses.

In addition to sensitivity analyses, we also conducted a scenario analysis using data from a sample of children (n = 106) for whom complete data were available on wider NHS resource use. We were able to collect data describing wider NHS resource use (e.g. GP visits, A&E visits and hospital readmissions) experienced by this group of children for 14 days after their treatment had ceased. At this time point, 29% (15/52) of patients in the clonidine arm were still in hospital and 31% (17/54) of patients in the midazolam arm were still in hospital. Clearly, not all of the patients included in this wider analysis incurred additional costs; this wider NHS resource use could have taken place only if the child had been discharged from hospital during this time period. In this analysis, the mean cost of care in the clonidine group is still cheaper than in the midazolam group, with a mean cost difference of –£552. However, midazolam is now slightly more effective than clonidine, with a mean effect difference of –0.006. With an ICER of £86,102 per additional case of adequate sedation there is a 48% probability of clonidine being more effective than midazolam, a 67% probability of clonidine being less costly than midazolam, and a 63% probability of clonidine being cost-effective compared with midazolam (see Appendix 7, Table 52). The variability around the base case estimate of cost-effectiveness is evident in the cost-effectiveness plane shown in Figure 14. The CEAC for the scenario analysis is shown in Figure 15, and indicates that the probability that clonidine is cost-effective declines slightly with increasing values that decision-makers place on an additional case of adequate sedation. Again, there were no statistically significant differences in costs or health consequences between the comparison groups. Estimates of net monetary benefits attributable to clonidine across alternative notional cost-effectiveness thresholds are shown in Appendix 7, Table 54. Assuming that the cost-effectiveness threshold equals £1000 per additional case of adequate sedation, including wider NHS costs generates a mean net benefit of £485 to the health service, attributable to clonidine (i.e. there is a net gain to the health service in monetary terms). Note, however, that the 95% CI surrounding the mean net benefit (–£2080 to £3174) includes negative values, i.e. there is again a possibility of a net monetary loss associated with clonidine (see Appendix 7, Table 54).

FIGURE 14.

Cost-effectiveness plane for cost-effectiveness scenario.

FIGURE 15.

Cost-effectiveness acceptability curve for cost-effectiveness scenario analysis.

Primary outcome

The trial did not recruit to target and was substantially underpowered in its objective to demonstrate equivalence. Equivalence between the treatment arms (± 0.15) for the proportion of children who were adequately sedated for ≥ 80% or more of the time was not demonstrated: 21 of 61 (34.4%) clonidine; 18 of 59 (30.5%) midazolam; difference in proportions 0.04 (95% CI –0.13 to 0.21). Non-inferiority of clonidine to midazolam was supported. However, this should be interpreted cautiously, as the wider CI that included values from outside the equivalence range favouring clonidine could have been due to the reduced numbers in the trial rather than the better performance of clonidine.

Other outcomes

Participants in the midazolam group were sedated for longer than those receiving clonidine (38.25 hours vs. 22.83 hours), but also took less time to become fully awake once sedation was stopped (medians 11.17 hours vs. 6.22 hours).

Fewer treatment failures were observed on midazolam [12/64 (18.8%) clonidine, 7/61 (11.5%) midazolam]. Only one case of rebound hypertension was observed (clonidine group). There were no discernible differences in the urine analysis or blood biochemistry results, and no differences in the proportions experiencing withdrawal symptoms; however, a higher proportion of participants allocated to midazolam required clinical intervention for those symptoms [11/60 (18.3%) clonidine, 16/58 (27.6%) midazolam].

Clonidine is an α₂-agonist of a different pharmacological group to midazolam, which acts as a GABA agonist. Although clonidine affects sympathetic outflow from the brain (thereby reducing BP and heart rate), provides some analgesia and has a calming effect, the major actions of benzodiazepines are to provide both sleep/anaesthesia and amnesia. Although both drugs are used in the PICU, and can provide reasonable sedation in conjunction with morphine, it is clear that the drugs are different in their characteristics. Both agents provided reasonable sedation, and the amounts of time adequately sedated during the treatment phase were similar (73.8% clonidine, 72.8% midazolam). The data do also show that sedation in the PICU is far from perfect, in that 25% of patients are adequately sedated for only 58% of the time. This clearly indicates that the regimens currently used in the PICU with morphine and a second sedative drug (either clonidine or midazolam) remain suboptimal, regardless of choice of current agents, and strongly indicates that a third-line drug may be needed to approach the goal of adequate sedation for ≥ 80% of the time in the PICU.

The study has been able to confirm, but also quantifies, the risks associated with individual side effects of the two drugs and this will be helpful in informing the clinician on the selection of agent to use.

One of the barriers to increasing the use of clonidine has been concern about the potential cardiovascular side effects in the unstable and critically ill child. There have been very few data on incidence of side effects of hypotension, bradycardia or other dysrhythmias prior to the SLEEPS study other than anecdotal evidence or case reports. The study demonstrated that use of clonidine in comparison with midazolam was associated with increased inotrope delivery and/or fluid administration during the loading phase and in the first 12 hours. These effects were classified by the observers as mild and did not result in withdrawal from study. One subject did develop significant bradycardia without hypotension 4 hours after commencing clonidine, prompting the investigator to abandon the study and report a SUSAR. Recovery was spontaneous and required no intervention. The implication for clinical practice is that specific attention needs to be taken during the loading and early infusion phase when clonidine is used, and anticipation of fluid or drug intervention should be made. The other concern with clonidine has been the fear of rebound hypertension on abrupt withdrawal of the drug, which, although a known feature of adult sedation with clonidine, has not been observed in children. Nevertheless, the fear of this has led to a clinician practice that is not based on evidence of reducing clonidine dosage very gradually after even short exposure in the PICU. The SLEEPS study identified only a single case of increased BP after withdrawal of the sedative agents. It was categorised as mild and required no intervention. Although future use of clonidine must still be aware of the possibility of rebound hypertension, it would appear that it is not common and that the practice of tailoring the dosage of clonidine downwards over days and weeks for fear of this effect (which, in itself, has led to delayed discharge from hospital) should now be reviewed.

The key clinical concerns with midazolam prior to study were the rise in infusion requirements due to tolerance of the drug and the subsequent high incidence of withdrawal phenomena after the drug was stopped. These two phenomena are interrelated in that previous studies have shown that the incidence and severity of withdrawal phenomena with midazolam depend on the infusion rate of the drug, but also the duration and hence cumulative amount of drug received. The results of the SLEEPS study showed that there were no differences in times to achieve maximum sedation/analgesia in the groups, indicating similar tolerance/tachyphylaxis with midazolam compared to clonidine. There were fewer treatment failures in the midazolam group but this was associated with an increased requirement for supplementary analgesia compared with the clonidine group (p = 0.01), however patients were sedated for longer on midazolam. In terms of withdrawal associated with midazolam usage the SLEEPS study identified similar withdrawal symptoms for both groups. However, the proportion of patients requiring clinical intervention was higher in the midazolam group and this is in keeping with the known high instance of significant withdrawal side effects previously attributed to midazolam usage in the PICU. However, there may be some confounding with the greater length of time midazolam patients were sedated for and this is known to be a risk factor for increasing severity of withdrawal symptoms.

Design

Despite the much smaller size of this study than that anticipated, SLEEPS is the largest trial comparing sedative agents in a PICU setting. SLEEPS has provided robust data targeting the concerns around the use of clonidine and cardiovascular instability and those around developing tolerance and withdrawal side effects of midazolam. We have demonstrated that the conventional approach of providing just midazolam and morphine in an unparalysed patient results in breakthrough sedation at a high frequency that is unacceptable. In conventional practice, many of the more sick children are given neuromuscular blocking agents with morphine midazolam combinations, which may mask the inadequacy of the sedation quality. This study should provoke solutions to this problem, which will involve the use of higher-efficacy opioids, such as fentanyl or alfentanil, and the more routine use of an additional sedative agent. Although clonidine performed no better than midazolam, it also was broadly similar in efficacy and had an acceptable safety profile. The trial therefore confirms that this is a viable alternative to midazolam. The rigour of the double-blind design and the thorough exploration of the data will expand the evidence base for these sedatives and provide useful data to inform clinical practice. Future study would require the results of the SLEEPS data to be taken into account. A new study would necessitate a more relaxed protocol, which would allow greater numbers of children with a greater variety of conditions to be recruited, more effective drug combinations that could include three drugs and a higher-efficacy opioid, and earlier enrolment to allow sicker children to be entered into the study regardless of their current sedation regimen or the use of muscle relaxants.

There is a tendency to oversedate children on PICU using multiple drugs at high doses and the SLEEPS trial protocol entwined the scoring system with the increases/decreases in sedative and analgesia. From the data presented it is clear that when children were outside the adequate sedation score range they were more likely to be oversedated than undersedated. However, both undersedation and oversedation are harmful. The scoring system used was systematic in its application; however, a consequence of the criteria used meant that it was not possible to use muscle relaxants. Although this impacted on numbers recruited, it may also impact on the generalisability of results.

The definition of the primary outcome, which used an 80% cut-off point for the proportion of time spent adequately sedated, as defined by the COMFORT score range, may be considered to be somewhat arbitrary. The ideal would have been an ED of 95%, as in many drug studies, but clearly this study was well short of this target.

A large number of protocol deviations were observed. It had been intended to conduct a per-protocol analysis alongside the ITT analysis, which may not be conservative within an equivalence trial. Further, the high degree of non-compliance may have increased the type I error rate in this study. The volume of protocol deviations shows the difficulty in applying a sedation protocol within a PICU.

Studies on sedation require intensive recorded monitoring, evaluation by an observer who has been trained and validated in the use of a sedation score, and rapid manipulation and documentation of changes in the infusion rates. This is in addition to the large administrative load in surveillance, recording, storing and processing of data. This has huge resource implications for any participating unit and, practically, this study was only possible by using the bedside nurses to run the evaluations and sedation changes. Individual training of large numbers of nursing staff to this level (180 staff on largest unit) and ensuring that their COMFORT scoring was standardised took considerable time, and the provision at each centre of just one part-time research nurse was inadequate for this trial. Future studies will need to address this requirement at the outset or relax the stringent inclusion criteria for nurse observers. The COMFORT score itself is relatively cumbersome but it is one of the few validated tools for evaluation of sedation in a PICU. The COMFORT scale, which eliminates the heart rate and BP observations, is simpler but has been only partially validated.

Patients in a PICU represent a heterogeneous mixture of ages and pathologies. This presents a difficult challenge in the study of these patient groups: on one hand there is a desire to have the conformity to achieve valid and reproducible results with a relatively small homogeneous group, but on the other hand there is a need to represent the entire PICU population. Despite considerable planning, training and meetings with the participating centres, protocol deviations were inevitable, representing the changing needs of the patients and demonstrating the difficulties experienced with adhering to differences from standard practice. During the set-up of the trial, one unit that had originally expressed an interest in participating then decided to decline. The unit had recently implemented a change in the sedation practices and reported the difficulties in adherence as the reason against participation, as SLEEPS would have necessitated further changes.

Examination of the protocol deviations indicate that these were most commonly related to failure to adjust sedation/analgesia to a change in COMFORT score, primary outcome data missing for ≥ 1 hour or delays in timing of action in response to COMFORT score-directed changes. Although these are not ideal, this represents what actually happens in the PICU in the management of the critically ill patient in an environment that is changing constantly and is difficult to control. The SLEEPS study represents the largest randomised sedation trial of PICU children and, as such, the lessons learnt from this are clear in terms of setting a more relaxed treatment protocol, which would allow increased participation and maintain more patients in the study. This is already helping to inform a following trial that has recently received funding and the senior author (ARW) is actively involved in an advisory capacity in this work. Relaxing the study protocol to accept more variability in the drug administration, and thereby obtaining more patient numbers without excessive protocol deviation, would appear to be a potential way forward in future work.

Considerations for dose and concomitant medications

Standard analgesic regimens used in the PICU and for postoperative analgesia describe doses of morphine infusions of between 0 and 60 µg/kg/hour. Correspondingly, the dose of midazolam used in the PICU is generally 0–200 µg/kg/hour. Previous studies have shown that midazolam side effects, including withdrawal phenomena, increase once the dose exceeds 100 µg/kg/hour, although some PICU continue to use doses of up to 300 µg/kg/hour. These doses were chosen for the SLEEPS study because they were applicable and relevant to clinical practice. Similarly, the limited data on i.v. clonidine suggested that a dose infusion in the order of 0–2.5 µg/kg/hour would provide reasonable analgesia. However, in clinical practice it is not uncommon to provide muscle relaxants for a limited period, particularly in the seriously sick infant, and also to prescribe pro re nata doses of additional drugs, whether oral or i.v. In addition to the obvious effects of muscle relaxant drugs, they may also have an intrinsic deafferenting effect, which can reduce sedation requirements in themselves. The use of muscle relaxants may be necessary in the sick infant with cardiac or airway disease but the consequence of their use is that COMFORT score cannot be assessed. The SLEEPS study needed to recognise this, and therefore it was a contraindication to entry and contributed to the low recruitment rates throughout the study. The requirement for additional as required i.v. sedation drugs on top of the two agent regimens was built into the protocol to allow for fluctuations in conscious level and sudden arousal. It was considered that if more than two doses of an additional drug were required in any 12-hour period then it would indicate that the two-dose regimen was insufficient. The data from the SLEEPS study indicated that neither midazolam plus morphine or clonidine plus morphine was able to provide the efficacy of sedation control alone with conventional doses, and that there remains a requirement for a re-evaluation of drug infusions or the need for a regular third drug.

Recruitment and retention

Recruitment into the trial was slower than expected, in part due to the number of eligible patients being lower than expected; however, retention of randomised participants was high.

The projected numbers for the study were 1000 patients, with 500 in each group. Initially, with 10 participating centres, this appeared feasible, requiring an average of one patient per week to be enrolled. Despite extension of the study, and a screening that totalled 10,023 children, only 129 children were randomised of whom 120 (93.0%) contributed data for the primary outcome of the study. The low recruitment rate, despite every effort – including several protocol amendments to try to improve the figures – has implications for other PICU studies of this type in the future. The specific issues identified are:

1. Conflict with other studies and elective cardiac cases:

   1. Recruitment in a PICU is challenging. The potential competition with other ongoing studies [Control of Hyperglycaemia in Paediatric Intensive Care (CHiP) study; Steroids in Paediatric Sepsis (StePS) study] was recognised early on and discussed during planning of the SLEEPS study. The need for co-enrolment in PICU settings to support the research demands is important and has been discussed elsewhere. 62,63 The main trial that overlapped with recruitment of potential participants with SLEEPS was the CHiP trial, and coenrolment was clinically contraindicated for the two trials. The CHiP trial was more suitable for elective cardiac patients, and, in discussion with the PIs, it was decided that CHiP would concentrate on this, whereas SLEEPS would look to concentrate on non-cardiac patients until the CHiP trial ended.

   2. In the last 5 years it has been recognised that early extubation of many of the postsurgical cardiac cases is beneficial to recovery. With this understanding, anaesthetic techniques have been developed to aid extubation and accelerated recovery. As a result, many infants and children have become ineligible for the SLEEPS trial and the initial entry criteria of expecting to be ventilated for 48 hours was changed to 12 hours. In addition, those children not undergoing early extubation are usually sufficiently unwell that they are paralysed, delaying entry into the study. Again, the protocol was amended to try to extend the time from initial ventilation to study entry from the initial value of ‘within 48 hours’ to ‘within 120 hours’. This allowed extra time for the sickest patient to become well enough to be sedated without additional paralysis. However, despite this the results show that recruitment of patients after cardiac surgery was poor, even after the CHiP trial ended.

2. Parental issues:

   1. Parents of children admitted to PICU in an emergency are highly stressed. Consenting for research studies in children have reduced considerably in the last 10 years, even for simple elective observational studies. There was considerable refusal rate from parents [194/827 (23.5%) eligible patients]. Reasons were multifactorial but a common reason was that if the child was settled on the ventilator in intensive care at the time that consent was asked for parents could see no gain for their child in participating. In addition, clinicians may feel anxious64 about approaching parents for consent, and this may be exacerbated if they begin to expect a negative response.

3. Timing of consent:

   1. Once parents had arrived and settled with their children in the PICU, and the seriousness of their condition had become apparent, parents were more reluctant to give consent to any procedure other than one that was life saving. One of the possible solutions to the above problem would be to have taken consent at the earliest clinical point of contact but still allowing parents to make an informed choice. An additional complication was that many of the patients come from referral hospitals before retrieval to the regional centres. This further delays the ability to access parents for consent. Discussions were held about removing the criteria for children to be adequately sedated before entry into the trial, which would have necessitated a deferred consent approach. However, owing to the retrieval nature of many of the cases and the potential concerns around cardiovascular instability with clonidine this was not considered appropriate. Future studies would be improved by achieving deferred consent, allowing children to be initiated into the study at the outset of critical care, even if muscle relaxants are being used. This would increase patient recruitment considerably, and generalisability of results to evaluate immediate sedative requirements.

4. Clinicians’ issues:

   1. All of the PIs and the PICU teams were committed to the study but, despite this, the clinicians in charge felt, in some cases, that the child was too unstable to enter into the study. This was one of the first studies to institute a blinded randomised controlled medication trial in the PICU on sedation. Although the study itself has led to more confidence with the use of clonidine in the PICU, at the time there were concerns about cardiovascular instability or ineffectiveness. In situations when there was clinical concern this led to abandoning of consent for the study and, although frustrating, was understandable. The severity of illness in the children under study and the effects on clinician decision-making cannot be overestimated. When faced with a critically ill child, clinicians will naturally tend towards conservatism. For example, although many intensivists will use morphine liberally in the child with asthma, a few will avoid the drug because of the potential concern of exacerbation of bronchospasm. Similarly, although there is evidence that muscle relaxation increases complications, including the risk of nosocomial infection, a common failure to recruit was due to the use of these agents. These drugs were often used to simplify a complex situation so that the physician could concentrate on facilitating treatment of the primary disease while effectively removing the need to consider sedation in a generic fashion. The results of the SLEEPS study would raise some concerns about this practice, in that at the standard doses of midazolam–morphine or clonidine–morphine reliable sedation cannot be guaranteed.

5. Research nurse time:

   1. The amount of adequate research nurse support in complex interventional trials should not be underestimated. It is required in terms of education of staff and thus avoidance of protocol violation. Paucity of research nurses cannot be underestimated as an impediment to recruitment.

6. Delay in study start:

   1. From the time of obtaining funding for the study there were significant delays in opening to recruitment. The key delay was the production of the blinded drugs, which required feasibility work, bioburden and postfiltration validation and testing, analytical method development and validation, stability protocol development, and obtaining of stability data. The manufacturers experienced delays with receiving supplies of the midazolam active ingredient, which delayed the feasibility work and consequently all time points with regards to the Investigational Medicinal Product (IMP). We were also required to carry out a systematic review and to produce a Simplified Investigational Medicinal Product Dossier as part of the clinical trial application (CTA) approval process. This meant that the process took a year from the signing of the contract with the IMP manufacturers until the CTA was granted by the MHRA. Furthermore, there were significant delays of several months in opening sites to recruitment once the regulatory approvals were in place. This was due to the training needs associated with the study and the volume of staff on each PICU. With research nurse time equivalent to 1 day per week, it was very difficult for the research nurse to train the large number of staff required to run the trial, especially as it was necessary to find time within their clinical roles for this to take place. During the time between first applying for funding and opening to recruitment, shifts in practice became apparent, with moves towards oral sedation and reductions in the length of sedation.

7. Compliance with the protocol:

   1. This study highlights the difficulties of adhering to a tight sedation protocol and attempting to apply this to a wide age group in children with different disease processes. This approach led to difficulty in recruitment and in those who were recruited, and difficulties in maintaining the patients within the tight sedation regimen. Future studies in sedation will need to approach this by having more relaxed entry criteria, by allowing the use of additional drugs and to accept periods when data acquisition is not possible. Such a study will require large patient numbers and careful stratification by age and disease in order to further our understanding. It will also require considerable resources and funding, with research staff independent of the clinical care so that there is tighter matching of drug delivery in response evaluations, and more rapid response to behavioural change that would include allowance for incremental dosing with study drugs or other alternatives. Alternatively, there should be a reversion to small tightly controlled single-centre studies focusing on one age group and one disease group.

Interpretation

This is the first study to document and compare the applicability of two commonly used sedation and analgesia regimens in critically ill ventilated children in a prospective blinded randomised fashion. The results have indicated that although both clonidine or midazolam can provide effective sedation some of the time, neither are able to achieve a target of 80% time in the targeted sedation zone. Additional supplementary medication can be used to maintain sedation but, as the trial did not allow more than two doses per 12-hour period, treatment failure occurred (clonidine in 18.8%, midazolam 11.5%). Although the drugs were not shown to be equivalent, this is not surprising, given the low statistical power to detect equivalence. However, these drugs do have very different pharmacology, with separate target sites, and so different profiles may be expected. Although their ability to provide controlled dose-dependent sedation is broadly similar, their characteristics and side effect profile are different. The study demonstrated the need to be aware of cardiovascular side effects, with clonidine in particular within the first 12 hours, and that patients who have been sedated with midazolam may require additional treatment for withdrawal phenomena afterwards.

The SLEEPS Study Group

Alertness

Rates the patient’s response to ambient stimulation in the environment including responses to sound (noises from monitors, intercoms, people talking, pagers, etc.), movement, light, etc. To rate this category, no stimulus is introduced by the observer.

1. Deeply asleep The state of least responsiveness to the environment. The patient’s eyes are closed, breathing is deep and regular, and the patient shows minimal responses to changes in the environment.

2. Lightly asleep The patient has their eyes closed throughout most of the observation period, but still responds somewhat to the environment as evidenced by slight movements, facial movements, unsuccessful attempts at eye openings, etc.

3. Drowsy The patient closes their eyes frequently or makes laboured attempts to open eyes and is less responsive to the environment.

4. Alert and awake The patient is responsive and interactive with the environment, but without an exaggerated response to the environment. The patient’s eyes remain open most of the time or open readily in response to ambient stimuli.

5. Hyper-alert The patient is hyper-vigilant, may be wide-eyed, attends rapidly to subtle changes in the environmental stimuli and has exaggerated responses to environmental stimuli.

Guidelines: If two or more of the following items achieve a score of ≥ 2 then the child is classified as lightly asleep – respiratory response, physical movement, muscle tone.

Calmness/agitation

Rates the patient’s level of emotional arousal and anxiety.

1. Calm The patient appears serene and tranquil. There is no evidence of apprehension or emotional distress.

2. Slightly anxious The patient is not completely calm. The patient shows slight apprehension and emotional distress.

3. Anxious The patient appears somewhat apprehensive and emotionally distressed, but remains in control.

4. Very anxious The patient appears very apprehensive. Emotional distress is apparent but the patient remains somewhat in control.

5. Panicky The patient’s total demeanour conveys immediate and severe emotional distress with loss of behavioural control.

Respiratory response

Rates the patient’s oral and respiratory responses to an endotracheal tube and intermittent ventilation.

1. No coughing or no spontaneous respiration Only ventilator-generated breaths are apparent. No respiratory movement is apparent between ventilator breaths. No oral movement or chest wall movement occurs except as created by the ventilator.

2. Spontaneous respiration The patient breathes at a regular, normal respiratory rate in synchrony with the ventilator. No oral movement or chest wall movement occurs which is contrary to the ventilator movement.

3. Occasional cough/resists ventilator The patient has occasional oral or chest wall movement contrary to the ventilator pattern. The patient may occasionally breathe out of synchrony with the ventilator.

4. Actively breathes against ventilator The patient has frequent oral or chest wall movement contrary to the ventilator pattern, coughs regularly or frequently breathes out of synchrony with the ventilator.

5. Fights ventilator – coughs/chokes/gags The patient actively makes oral or chest wall movement contrary to the ventilator pattern, and coughs and/or gags in a manner which may interfere with ventilation.

Physical movement

Rates frequency and intensity of physical movement.

1. None The patient shows complete absence of independent movement.

2. Occasional, slight movements The patient shows three or fewer small amplitude movements of the fingers or feet, or very small head movement.

3. Frequent, slight movement The patient shows more than three small amplitude movements of the fingers or feet, or very small head movements.

4. Vigorous movements of extremities only The patient shows movements of greater amplitude, speed or vigour of hands, arms or legs. The head may move slightly. Movement is vigorous enough to potentially disrupt cannulas.

5. Vigorous movements of extremities, torso and head The patient shows movements of greater amplitude, speed or vigour of the head and torso, such as head-thrashing, back-arching or neck-arching. Extremities may also move. Movement is vigorous enough to potentially disrupt placement of an endotracheal tube.

Guidelines Occasional movement is defined as less than once per minute. Frequent movement is defined as more than once per minute.

Blood pressure

Mean arterial blood pressure (MAP) rates the frequency of elevations above (or below) a normal baseline. The baseline may need to be reset on a daily basis or occasionally more frequently, depending on changes in clinical conditions (e.g. change in temperature or the addition of inotropes, etc.). Each re-evaluation will set the cardiovascular baselines for each ‘rating period’.

At the beginning of the rating period, ‘baseline’, ‘15% below baseline’ and ‘above baseline’ values are recorded on the rating sheet in an easily observable location. The rater observes the monitor for MAP during the observational period of an hour and records, with a hash mark, each observation above or below the baseline. Ratings are made upon the number of readings above the baseline.

1. BP 15% below baseline

2. BP consistently at baseline

3. infrequent elevations of ≥ 15% (one to three during observation period)

4. frequent elevations of ≥ 15% (more than three during observation period)

5. sustained elevation of ≥ 15%.

Guidelines The baseline to use initially for BP will be an average of the measurements taken hourly over the 4 hours previous to trial entry. Following this, the baseline should be recalculated on a daily basis for each patient (or occasionally more frequently) if this is felt to be clinically appropriate for the individual.

Heart rate

Heart rate score is based on the frequency of elevations above (or below) a normal baseline. The baseline may need to be reset on a daily basis or occasionally more frequently, depending on changes in clinical conditions (e.g. change in temperature or the addition of inotropes, etc.). Each re-evaluation will set the cardiovascular baselines for each ‘rating period’. At the beginning of the rating period, baseline, 15% above baseline and 15% below baseline values are recorded on the rating sheet in an easily observable location. The observer observes the heart rate throughout the hour and records, with a hash mark, each episode of elevation above the baseline or episodes below the baseline. Ratings are made based upon the number of readings above the baseline.

1. heart rate 15% below baseline

2. heart rate consistently at baseline

3. infrequent elevations of ≥ 15% (one to three during observation period)

4. frequent elevations of ≥ 15% (more than three during observation period)

5. sustained elevation of ≥ 15%.

Guidelines The baseline to use initially for BP will be an average of the measurements taken hourly over the 4 hours previous to trial entry. Following this, the baseline should be recalculated on a daily basis for each patient (or occasionally more frequently), if this is felt to be clinically appropriate for the individual.

Muscle tone

Muscle tone is assessed in relation to normal tone in a patient who is awake and alert. The rating is based upon patient response to rapid and slow flexion and extension on a non-instrumented extremity (i.e. elbow or knee without an i.v. line, tape, arterial line or physical restraint). A wrist or ankle may be used if no other joint is available. This rating is the only one that requires active intervention by the rater and is performed at the end of the observation period.

1. Relaxed/none Muscle tone is absent. There is no resistance to movement.

2. Reduced muscle tone The patient shows less resistance to movement than normal but muscle tone is not totally absent.

3. Normal muscle tone Resistance to movement is normal.

4. Increased tone/flexion – fingers/toes The patient shows resistance to movement that is clearly greater than normal but the joint is not rigid.

5. Extreme rigidity/flexion – fingers/toes Muscle rigidity is the patient’s predominant state throughout the observation period. This may be observed even without manipulating an extremity.

Facial tension

Facial tension assesses tone and tension of facial muscles. The standard of comparison is a patient who is awake and alert.

1. Relaxed The patient shows no facial muscle tone, with absence of normal mouth and eye closing. The mouth may look slack and the patient may drool. Brow smooth.

2. Normal tone The patient shows no facial muscle tension with mouth and eyes closing appropriately. Small movements of the lips, mouth or tongue. Brow smooth.

3. Some tension This does not include sustained tension of muscle groups – such as the brow, forehead or mouth – but you may see a frown or eye squeezing.

4. Full facial tension The patient shows notable, sustained tension of facial muscle groups, including the brow, forehead, mouth, chin or cheeks.

5. Hyper-alert The patient demonstrates facial grimacing with an expression that conveys an impression of crying, discomfort and distress. This generally includes extreme furrowing of brow.

Substantial amendment version 4.0 (6 May 2010) to version 5.0 (1 March 2011)

Substantial amendment version 3.0 (5 October 2010) to version 4.0 (6 May 2010)

Substantial amendment version 2.1 (14 September 2009) to version 3.0 (5 October 2009)

Non-substantial amendment version 2.0 (5 May 2009) to version 2.1 (14 September 2009)

Substantial amendment version 1.0 (1 October 2008) to version 2.0 (5 May 2009)