A clinical and economic evaluation of Control of Hyperglycaemia in Paediatric intensive care (CHiP): a randomised controlled trial

Article history

The research reported in this issue of the journal was funded by the HTA programme as project number 05/506/03. The contractual start date was in February 2007. The draft report began editorial review in May 2012 and was accepted for publication in April 2013. The authors have been wholly responsible for all data collection, analysis and interpretation, and for writing up their work. The HTA editors and publisher have tried to ensure the accuracy of the authors’ report and would like to thank the reviewers for their constructive comments on the draft document. However, they do not accept liability for damages or losses arising from material published in this report.

Declared competing interests of authors

All authors have completed the unified competing interest form at www.icmje.org/coi_disclosure.pdf (available on request from the corresponding author) and declare Medicines for Children Research Network funding (MD); involvement in the Health Technology Assessment programme SLEEPS (Safety profiLe Efficacy and Equivalence in Paediatric intensive care Sedation) trial and royalties for acting as an editor of handbook of PIC (KM); payment from Baxter for a single advisory meeting and from GlaxoSmithKline for pip contributions (MP); and a grant awarded from the neonatal and paediatric pharmacists group for an in vitro study of drug compatibility (JP).

Copyright statement

© Queen’s Printer and Controller of HMSO 2014. This work was produced by Macrae et al. under the terms of a commissioning contract issued by the Secretary of State for Health. This issue may be freely reproduced for the purposes of private research and study and extracts (or indeed, the full report) may be included in professional journals provided that suitable acknowledgement is made and the reproduction is not associated with any form of advertising. Applications for commercial reproduction should be addressed to: NIHR Journals Library, National Institute for Health Research, Evaluation, Trials and Studies Coordinating Centre, Alpha House, University of Southampton Science Park, Southampton SO16 7NS, UK.

Myocardial infarction

In a meta-analysis,6 patients with acute myocardial infarction, and without diabetes mellitus, who had glucose concentrations in the range 6.1–8.0 mmol/l or higher had a 3.9-fold [95% confidence interval (CI) 2.9- to 5.4-fold] higher risk of death than patients who had lower glucose concentrations. Glucose concentrations higher than values in the range of 8.0–10.0 mmol/l on admission were associated with increased risk of congestive heart failure or cardiogenic shock.

Stroke

Capes et al. 7 conducted a systematic review and meta-analysis of the literature relating glucose levels in the interval immediately post stroke to the subsequent course. A comprehensive literature search was carried out to identify cohort studies reporting mortality and/or functional recovery after stroke in relation to admission glucose level. In total, 32 studies were identified, and predefined outcomes could be analysed for 26 of these. After stroke, the unadjusted relative risk (RR) of in-hospital or 30-day mortality associated with an admission glucose level above the range of 6–8 mmol/l was 3.07 (95% CI 2.50 to 3.79) in non-diabetic patients and 1.30 (95% CI 0.49 to 3.43) in diabetic patients. Non-diabetic stroke survivors whose admission glucose level was above the range of 6.7–8 mmol/l also had a greater risk of poor functional recovery (RR 1.41; 95% CI 1.16 to 1.73).

Head injury and multisystem trauma

Hyperglycaemia has been shown to be an independent predictor of poor outcomes in adults with head injury8 and in cases of multiple trauma. 9

Pulmonary function

Hyperglycaemia has been shown to be associated with diminished pulmonary function in adults, even in the absence of diabetes mellitus,10 and a range of risk factors for lung injury. 11

Gastrointestinal effects

Hyperglycaemia has been shown to be associated with delayed gastric emptying,12 decreased small bowel motility and increased sensation and cerebral-evoked potentials in response to a range of gastrointestinal stimuli in adult volunteers. 13–16

Infections

The in vitro responsiveness of leucocytes stimulated by inflammatory mediators is inversely correlated with glycaemic control. 17 This reduction in polymorphonuclear leucocyte responsiveness may contribute to the compromised host defence associated with sustained hyperglycaemia,17 and, indeed, hyperglycaemia has been shown to be associated with an increased rate of serious infections after adult cardiac18 and vascular surgery. 19

These studies, which associate poorer outcomes with patients with the highest levels of stress glycaemia, raise the question of whether high blood glucose levels simply identify the more severely ill patients, in whom worse outcomes are inevitable, or whether specific homeostatic or allostatic glycaemic dysfunction influences outcomes independently. If the latter were true, then perhaps measures to prevent or limit stress-induced hyperglycaemia would improve clinical outcomes.

Does hyperglycaemia matter for adults in the critically ill setting?

Although the importance of good glycaemic control has long been established in minimising complications of chronic hyperglycaemia in patients with diabetes mellitus,20,21 and a number of mechanisms for glucotoxicity identified,22 in the era up to the year 2000, a permissive approach was typically adopted when managing non-diabetic patients in intensive care settings. A very reasonable question, however, is Could shorter-term hyperglycaemia in non-diabetic populations be associated with clinically important adverse outcomes? Early reports from adult populations started to explore the possible association between acute stress-induced hyperglycaemia and outcome in both diabetic and non-diabetic patients.

Furnary et al. 18 noted that hyperglycaemia is associated with higher sternal wound infection rates following adult cardiac surgery and questioned whether more aggressive control of glycaemia might lead to lower infection rates. In a prospective study of 2467 consecutive diabetic patients who underwent open-heart surgical procedures, patients were classified into two sequential groups. The control group included 968 patients treated with sliding-scale-guided intermittent subcutaneous insulin injections. The study group included 1499 patients treated with a continuous intravenous insulin infusion in an attempt to maintain a blood glucose level of < 11.1 mmol/l. Compared with subcutaneous insulin injections, continuous intravenous insulin infusion induced a significant reduction in perioperative blood glucose levels, which was associated with a significant reduction in the incidence of deep-sternal wound infection in the continuous intravenous insulin infusion group [0.8% (12 of 1499) vs. 2.0% (19 of 968) in the intermittent subcutaneous insulin injection group; p = 0.01]. The use of perioperative, continuous intravenous insulin infusion in diabetic patients undergoing open-heart surgical procedures appeared to significantly reduce the incidence of major infections.

Malmberg et al. 23 randomly allocated patients with diabetes mellitus and acute myocardial infarction to intensive insulin therapy (n = 306) or standard treatment (controls, n = 314). The mean (range) follow-up was 3.4 (1.6–5.6) years. There were 102 (33%) deaths in the treatment group compared with 138 (44%) deaths in the control group (RR 0.72; 95% CI 0.55 to 0.92; p = 0.011). The effect was most pronounced among a predefined group that included 272 patients who had not received insulin treatment previously and who were at a low cardiovascular risk (0.49; 0.30 to 0.80; p = 0.004). Intensive insulin therapy improved survival in diabetic patients with acute myocardial infarction. The effect seen at 1 year continued for at least 3.5 years, with an absolute reduction in mortality of 11%.

In 2001, Van den Berghe and colleagues from Leuven, Belgium,24 extended this approach to non-diabetic hyperglycaemic populations. They performed a single-centre randomised trial in adults undergoing intensive care following surgical procedures which showed that the use of insulin to tightly control blood glucose led to a reduction in mortality from 10.9% to 7.2%, and a significantly lower incidence of a range of important complications of critical illness including renal failure, infection, inflammation, anaemia and polyneuropathy and need for prolonged ventilatory support.

The same group undertook a similar trial in non-surgical, adult, critically ill patients25 and again found benefits from the control of blood glucose with intensive insulin therapy. Patients were randomly assigned to a regimen of strict normalisation of blood glucose (4.4–6.1 mmol/l) with use of insulin or conventional therapy whereby insulin was administered only when blood glucose levels exceeded 12 mmol/l, with the infusion tapered when the level fell below 10 mmol/l. In the intention-to-treat analysis of the 1200 patients, intensive care unit (ICU) and in-hospital mortality were not significantly altered by intensive insulin therapy; however, for those patients who stayed > 3 days in intensive care (an a priori subgroup), mortality was significantly reduced from 52.5% to 43% (p = 0.009). Morbidity was significantly reduced by intensive insulin therapy, with a lower incidence of renal injury and shorter length of mechanical ventilation and duration of hospital stay noted. For patients who stayed > 5 days in intensive care after trial entry, all morbidity end points were significantly improved in the intensive insulin therapy group.

Although the precise mechanisms by which different glucose control strategies might influence clinical outcomes had not been fully elucidated, the clinical effects of ‘tight glycaemic control’ (TGC) for adults in critical care appeared promising. As a result, TGC was widely adopted in adult critical care standards in the years following the publication of Van den Berghe et al. 2001 paper. 24

Stress hyperglycaemia in the critically ill child

Over 12,000 children are admitted to ICUs in England and Wales each year. 26 Hyperglycaemia occurs frequently during critical illness or after major surgery in children, with a reported incidence of up to 86%,3 but children in critical care may not respond to interventions in the same way as adults.

References to hyperglycaemia and its management in critically ill children were identified through searches in MEDLINE27 from 1990 to December 2006. Articles were also identified through searches of the authors’ own files. Only papers published in English were reviewed. The final reference list was generated on the basis of originality and relevance to the genesis of this research proposal. The search terms used were ‘glycaemia’, ‘control’, ‘insulin’, ‘critical illness’ and ‘intensive care’; the limits applied were ‘clinical trials’, ‘meta-analysis’, ‘randomised controlled trial’ and ‘humans’ and ‘age 0–18 years’. No randomised trials or meta-analyses of glycaemic control in childhood critical illness were identified.

The non-randomised studies identified included a number of reports of critically ill children receiving care in general,3,28,29 cardiac surgical,30,31 trauma9,32,33 and burns34 ICUs, all showing that high blood glucose levels occur frequently and that levels are significantly higher in children who die than in children who survive. As in adults, the occurrence of hyperglycaemia was associated with poorer outcomes including death, sepsis and longer length of intensive care stay for critically ill children.

Srinivasan et al. 3 studied the association of timing, duration and intensity of hyperglycaemia with mortality in critically ill children. The study had a retrospective, cohort design and included 152 critically ill children receiving vasoactive infusions or mechanical ventilation. A peak blood glucose of > 7 mmol/l occurred in 86% of patients. Non-survivors had a higher peak blood glucose [mean ± standard deviation (SD)] than survivors (17.3 ± 6.4 mmol/l vs. 11.4 ± 4.4 mmol/l, p < 0.001). Non-survivors had more intense hyperglycaemia during the first 48 hours in the paediatric intensive care unit (PICU) (7 ± 2.1 mmol/l) than survivors (6.4 ± 1.9 mmol/l, p < 0.05). Median blood glucose levels > 8.3 mmol/l were associated with a threefold increased risk of mortality compared with median levels of < 8.3 mmol/l. Univariate logistic regression analysis showed that peak blood glucose and the duration and intensity of hyperglycaemia were each associated with PICU mortality (p < 0.05). Multivariate modelling controlling for age and paediatric risk of mortality scores showed an independent association of peak blood glucose and duration of hyperglycaemia with PICU mortality (p < 0.05). This study demonstrated that hyperglycaemia is common among critically ill children. Peak blood glucose and duration of hyperglycaemia appear to be independently associated with mortality. The study was limited by its retrospective design, its single-centre location and the absence of cardiac surgical cases, a group which make up approximately 40% of paediatric intensive care (PIC) admissions in the UK.

Yates et al. 30 conducted a retrospective review of data from 184 children < 1 year of age who underwent major cardiac surgery over a 22-month period ending in August 2004. Factors analysed included peak glucose levels and duration of hyperglycaemia. The duration of hyperglycaemia was significantly longer in children who developed renal insufficiency, liver insufficiency and infection and those who required mechanical circulatory support or who died, and was associated with longer PICU and hospital lengths of stay (LOS).

Hall et al. 35 investigated the incidence of hyperglycaemia in infants with necrotising enterocolitis (NEC) and the relationship between glucose levels and outcome in these infants. Glucose measurements (n = 6508) in 95 neonates with confirmed NEC admitted to the surgical ICU were reviewed. Glucose levels ranged from 0.5 to 35.0 mmol/l; 69% of infants became hyperglycaemic (> 8 mmol/l) during their admission; and 32 infants died. The mortality rate tended to be higher in infants whose peak glucose concentration exceeded 11.9 mmol/l than in those with peak glucose concentrations of < 11.9 mmol/l, and the late (> 10 days after admission) mortality rate was significantly higher in the former infants (29% vs. 2%; p = 0.0009). Linear regression analysis indicated that peak glucose concentration was significantly related to LOS (p < 0.0001).

Branco et al. 29 showed an association between hyperglycaemia and increased mortality in children with septic shock. They prospectively studied children admitted to a regional PICU with septic shock refractory to fluid therapy over a period of 32 months. The peak glucose level in those with septic shock was 11.9 ± 5.4 mmol/l (mean ± SD), and the mortality rate was 49.1% (28/57). In non-survivors, the peak glucose level was 14.5 ± 6.1 mmol/l, which was higher (p < 0.01) than that found in survivors (9.3 ± 3.0 mmol/l). The RR of death in patients with peak glucose levels of ≥ 9.9 mmol/l was 2.59 (p = 0.012).

Faustino and Apkon28 demonstrated that hyperglycaemia occurs frequently among critically ill non-diabetic children and is associated with higher mortality and longer LOSs in PICUs. They performed a retrospective cohort study of 942 non-diabetic patients admitted to a PICU over a 3-year period. The prevalence of hyperglycaemia was based on initial PICU glucose measurement, peak value within 24 hours and peak value measured during PICU stay up to 10 days after the first measurement. Using three cut-off values (6.7, 8.3 and 11.1 mmol/l), the prevalence of hyperglycaemia was 16.7–75.0%. The RR for death increased for peak glucose within 24 hours of > 8.3 mmol/l (RR, 2.50; 95% CI 1.26 to 4.93) and peak glucose within 10 days of > 6.7 mmol/l (RR, 5.68; 95% CI 1.38 to 23.47).

Pham et al. 34 reviewed the records of children with ≥ 30% total body surface area burn injury admitted to a regional paediatric burn centre during two consecutive periods, during the first of which patients received ‘conventional insulin therapy’ (n = 31), and during the second of which they were managed with TGC (n = 33). Intensive insulin therapy was positively associated with survival and a reduced incidence of infections. The authors concluded that intensive insulin therapy to maintain normoglycaemia in severely burned children could be safely and effectively implemented in a paediatric burns unit and that this therapy seemed to lower infection rates and improve survival.

There was, therefore, mounting evidence to suggest that stress hyperglycaemia occurred in both neonates and children (as in adults). From adult studies, TGC appeared to offer the possibility of clinical benefits, particularly following surgery, but there was no convincing randomised controlled trial (RCT) evidence for children, whether or not admitted to PICUs following surgery. This was of particular importance as approximately one-third of admissions of children to UK PICUs are associated with surgery, in particular cardiac surgery.

Evidence on the cost-effectiveness of tight glycaemic control

The existing evidence on the clinical effectiveness of TGC is derived from studies in both critically ill adults and critically ill children. However, to inform whether or not the NHS should provide TGC rather than conventional management (CM) for critically ill children, it is important to consider whether or not the additional costs associated with implementing a TGC protocol are offset by subsequent reductions in resource use and improved health outcomes. Limited evidence suggests that any additional costs associated with implementing a TGC protocol may be relatively small. 36 A post-hoc analysis of the Van den Berge 2001 RCT24,37 for critically ill adults admitted for surgery reported that TGC can reduce ICU LOS, and hence hospital costs. 37 However, this study had several limitations. The study was not designed to measure costs; resource use after the initial hospital episode was not recorded; the study was undertaken in a single centre and lacked generalisability; and it is unclear whether the results apply to other patient groups (e.g. critically ill children, patients not admitted for surgery).

For critically ill children, any assessment of the effect of a TGC protocol compared with CM on resource use and costs is hindered by the lack of evidence from RCTs. The costs of each PICU bed-day are substantial (ranging from £1000 to £5000 per bed-day),38 so if TGC reduces PICU LOS then it would be anticipated to also reduce short-term costs (i.e. those incurred within 30 days of admission to the PICU). It is also plausible that TGC may have an effect on longer-term costs. A previous study reported that around 10% of PICU survivors had residual long-term disability (median follow-up of 3.5 years from initial admission). 39 Therefore, the long-term costs following PICU survival may be substantial, and may be increased if TGC increases PICU survival, or reduced if improved blood glucose control reduces morbidity. There is little available evidence on the net effect of TGC compared with CM on longer-term morbidity and hence costs, either in general or specifically for critically ill children.

The previous evidence, therefore, raises the hypotheses that TGC may have an impact on costs, both in the short term (e.g. 30 days post PICU admission) and in the longer term (e.g. 12 months post PICU admission). It would, therefore, seem important to consider the net effect of TGC on costs alongside any change in clinical outcomes. No previous study has considered the effect of TGC on health service costs for paediatric patients.

The Control of Hyperglycaemia in Paediatric intensive care (CHiP) trial, therefore, sought to address the question of whether or not a policy of strictly controlling blood glucose using insulin in children admitted to PIC reduces mortality and morbidity and is cost-effective.

Study design

The study was an individually randomised controlled open trial with two parallel arms. The allocation ratio was 1 : 1.

The planned flow of patients through the trial is summarised in Figure 1.

FIGURE 1.

Flow chart summarising the planned flow of patients through the trial. CR, Central Register; GP, general practitioner; IC, Information Centre.

Economic evaluation

Measurement of resource use up to 30 days post randomisation

The trial CRFs recorded the number of inpatient days for the index hospital episode following randomisation, up to day 30. Within this index hospital episode, the CRFs recorded the number of PICU days spent on the unit where the patient was randomised, and any subsequent PICU bed-days following transfers to other hospitals. The CRFs also recorded the LOS on general medical (GM) wards, both within the acute hospital where the patient was randomised and following transfer to other hospitals. The number of day-case admissions was also noted. The total LOS for the initial hospital episode was calculated as the sum of the LOS at PICUs and GM wards up to a maximum of 30 days following randomisation. Any readmissions within 30 days to the PICU where the child was randomised were also recorded. The LOS following these readmissions was added to the total number of days for the initial episode, to give the total hospital LOS up to 30 days post randomisation.

Data on the level of care for PICU bed-days were available through routine collection of the Paediatric Critical Care Minimum Data Set (PCCMDS)54 in 11 of the participating centres via the PICANet database. The PCCMDS consists of 32 items recorded for each PICU bed-day that can be used to define the level of care, that is the paediatric critical care health-care resource group (HRG). 55 The PCCMDS data items were extracted for each PICU bed-day after randomisation. Each PICU bed-day was then assigned to the appropriate HRG (HRG, version 4), using the HRG grouper. 55 [The HRG classification includes items for primary and secondary diagnosis, OPCS (Office of Population, Censuses and Surveys’ Classification of Surgical Operations and Procedures) codes for high-cost drugs, and fields for critical care activity.] Table 1 lists the HRG classifications for PIC with examples of procedures for each category. Figure 2 reports the distribution of PICU bed-days across HRG categories for the 8954 PICU bed-days that were grouped during the first 30 days, both overall and for each site. The most common HRG category was ‘intensive care basic enhanced’ [HRG level 4 (HRG4)]. For 1862 bed-days in the 11 sites that provided information, PCCMDS data were missing or incomplete, and, for three centres (254 bed-days), no PCCMDS information was available. All these bed-days were assigned to the modal HRG category (HRG4) (see Sensitivity analysis). For a total of 326 bed-days, the activity reported was categorised by the HRG grouper as ‘not critical care activity’, and designated as bed-days on GM wards.

TABLE 1 - Health-care resource groups for paediatric critical care: definitions and examples

BSA, burns surface area; CVP, central venous pressure; ECG, electrocardiogram; ECLS, extracorporeal life support; ECMO, extracorporeal membrane oxygenation; HDU, high-dependency unit; VAD, ventricular assist device.

HRG Level	Critical care unit	Examples of procedures
HRG1	HDU basic	ECG or CVP monitoring, oxygen therapy plus pulse oximetry
HRG2	HDU advanced	Non-invasive ventilation, acute haemodialysis, vasoactive infusion (inotrope, vasodilator)
HRG3	ICU basic	Invasive mechanical ventilation, or non-invasive ventilation + vasoactive infusion + haemofiltration
HRG4	ICU basic enhanced	Invasive mechanical ventilation + vasoactive infusion, or advanced respiratory support
HRG5	ICU advanced	Invasive mechanical ventilation or advanced respiratory support + haemofiltration
HRG6	ICU advanced enhanced	Invasive mechanical ventilation or advanced respiratory support + burns > 79% BSA
HRG7	ICU ECMO/ECLS	ECMO or ECLS, including VAD or aortic balloon pump

FIGURE 2.

Distribution of paediatric critical care bed-days within 30 days post randomisation across HRG categorises for CHiP patients. Proportion of paediatric critical care bed-days within each HRG category, overall and by centre.

Ancillary studies

In addition to the main study, the grant holders welcomed more detailed or complementary studies, provided that proposals were discussed in advance with the TSC and appropriate additional research ethics approval was sought. These will not be discussed further in this monograph.

Publication policy

To safeguard the integrity of the trial, data from this study were not presented in public or submitted for publication without requesting comments and receiving agreement from the TSC. The primary results of the trial will be published by the group as a whole in collaboration with local investigators and local investigators will be acknowledged. The success of the trial was dependent on the collaboration of many people. The results were, therefore, presented first to the trial local investigators. A summary of the results of the trial will be sent to parents of participating children on request and also made available on the trial website.

Organisation

A TSC (see Appendix 9) and a DMEC were established (see Appendix 10). Day-to-day management of the trial was overseen by a Trial Management Group (TMG) (see Appendix 11). Each participating centre identified a paediatric intensivist as a PI (see Appendix 12). Each participating centre was allocated funding (from the core trial grant, from the MCRN and/or from local Comprehensive Local Research Networks) for research nursing time, and employed or reallocated a research nurse to support all aspects of the trial at the local centre.

Confidentiality

Patients were identified by their trial number to ensure confidentiality. However, as the patients in the trial were contacted about the study results (and patients recruited until November 2010 were followed up for 12 months following randomisation), it was essential that the team at the DCC had the names and addresses of the trial participants recorded on the data collection forms in addition to the allocated trial number. Stringent precautions were taken to ensure confidentiality of names and addresses at the DCC.

The chief investigator and local investigators ensured conservation of records in areas to which access is restricted.

Audit

To ensure that the trial was conducted according to the International Conference on Harmonisation of Technical Requirements for Registration of Pharmaceuticals for Human Use (ICH) and Good Clinical Practice (GCP) guidelines, site audits were carried out on a random basis. The local investigator was required to demonstrate knowledge of the trial protocol and procedures and ICH and GCP. The accessibility of the site file to trial staff and its contents were checked to ensure all trial records were being properly maintained. Adherence to local requirements for consent was examined.

If the site had full compliance, the site visit form was signed by the lead research nurse. In the event of non-compliance, the DCC and/or the lead research nurse addressed the specific issues to ensure that relevant training and instruction were given.

The CHiP trial also passed an inspection by the Medicines and Healthcare products Regulatory Agency (MHRA) (August 2009).

Termination of the study

At the termination of planned recruitment, the DCC contacted all sites by telephone, email or fax in order to terminate all patient recruitment as quickly as possible. After all recruited patients had been followed until 30 days post randomisation (or hospital discharge if stay > 30 days), a declaration of the end of trial form was sent to EudraCT and the Multicentre Research Ethics Committee (MREC). The following documents will be archived in each site file and kept for at least 5 years: original consent forms, data forms, trial-related documents and correspondence. At the end of the analysis and reporting phase, the trial master files at the clinical co-ordinating centre and DCC will be archived for 15 years.

Funding

The costs for the study itself were covered by a grant from the HTA programme. Clinical costs were met by the NHS under existing contracts.

Indemnity

If there is negligent harm during the clinical trial, when the NHS body owes a duty of care to the person harmed, NHS indemnity covers NHS staff, medical academic staff with honorary contracts and those conducting the trial. NHS indemnity does not offer no-fault compensation.

Recruitment

Trial recruitment began on 4 May 2008. As indicated in Chapter 2, recruitment was slower than expected. This was mainly a result of delays in trial initiation at some sites, clinical constraints and a ‘research learning curve’ in many of the participating units which had no previous experience of recruiting critically ill children to clinical trials. These delays necessitated an application to the HTA programme for an extension to the trial. The HTA programme granted funding to allow recruitment to be extended to allow the trial to achieve sufficient power (1500 children) to identify whether or not there was a differential effect for the primary end point (VFD-30) in the two strata (cardiac and non-cardiac).

The DMEC confidentially reviewed unblinded interim analyses on two occasions. In addition, they met to discuss SAEs and recruitment rates on three further occasions.

Recruitment closed on 31 August 2011, as agreed in the HTA funding. A total of 19,924 children were screened from 13 sites. Of these, 1384 were recruited and randomised (701 to TGC and 683 to CM). The reasons for non-recruitment are shown in Table 4. Of the 1384, 15 were subsequently found to be ineligible (Table 5), leaving 1369 eligible children (694 to TGC and 675 to CM) randomised into the trial – 91% of the original target of 1500. The flow of patients is shown in Figure 3 and cumulative recruitment in Figure 4. Recruitment by site is shown in Table 6.

FIGURE 3.

A flow chart showing the flow of patients. Note that the information on vital status up to 12 months was available from the Office for National Statistics, aside from for 17 non-UK nationals.

FIGURE 4.

Cumulative recruitment – actual accrual vs. revised expected.

Comparability at baseline

The characteristics of the children at baseline are shown in Table 7. The randomised groups were broadly comparable at trial entry. Sixty-two per cent were randomised within 1 day of admission to PICU. In terms of the prespecified stratifying factors, two-thirds were aged < 1 year, and 60% of the children were in the cardiac surgery stratum. Seven per cent of children in the cardiac surgery stratum were considered to be undergoing surgical procedures associated with a high risk of mortality (RACHS1 score 5 or 6), and 19% of children in the non-cardiac group had a PIM2 score indicative of a ≥ 15% risk of PICU mortality.

Actual management

Table 8 describes the observed management of blood glucose after randomisation, and shows a clear difference between the two arms of the study. In the TGC arm, 461 of the children (66%) received insulin compared with 109 of 675 (16%) in the CM arm. Children in the TGC arm received more insulin, and continued on insulin for longer. Figure 5 shows the mean daily blood glucose level by arm. There was a clear separation between the two randomised arms, with children in the TGC arm having a significantly lower blood glucose profile than those in the CM arm.

FIGURE 5.

Mean glucose level.

Thirty-day clinical outcomes

Secondary outcomes

The secondary outcomes up to 30 days are shown in Table 10, and the duration of ventilation in Figure 6 and of vasoactive drug use in Figure 7. In general, the secondary outcomes are similar between the arms over the 30-day period, although less RRT was undertaken in the TGC arm (odds ratio 0.63; 95% CI 0.45 to 0.89). Additionally, mean caloric intake (Figure 8) was similar between the two groups.

TABLE 10 - Secondary outcomes 30 days post randomisation

PELOD, paediatric organ dysfunction; SE, standard error.

Outcome missing for two patients (one in TGC arm and one in CM arm).

Patients who experienced moderate episodes may also have experienced a severe episode and vice versa.

Outcome	TGC (N = 694)	CM (N = 675)	Mean difference/odds ratio (95% CI)
Death within 30 days of trial entry [n (%)]	35 (5.04)a	34 (5.04)a	1.00 (0.62 to 1.63)
Number of days in PICU [mean (SE)]	6.50 (0.21)	6.96 (0.24)	–0.47 (–1.12 to 0.15)
Number of days in hospital [mean (SE)]	16.40 (0.34)	16.73 (0.36)	–0.33 (–1.24 to 0.62)
Duration of mechanical ventilation (days) [mean (SE)]	5.30 (0.19)	5.61 (0.21)	–0.31 (–0.87 to 0.25)
PELOD score [mean (SE)]	9.79 (0.19)	9.79 (0.21)	–0.31 (–0.87 to 0.25)
Duration of vasoactive drug use (days) [median (IQR)]	3 (2 to 6)	4 (2 to 6)	–0.20 (–0.64 to 0.25)
RRT [n (%)]	62 (8.93)	91 (13.48)	0.63 (0.45 to 0.89)
Bloodstream infection [n (%)]	38 (5.48)	43 (6.37)	0.85 (0.54 to 1.34)
Use of antibiotics > 10 days [n (%)]	62 (8.93)	74 (10.96)	0.80 (0.56 to 1.14)
Number of red blood cell transfusions [mean (SE)]	1.00 (0.11)	1.12 (0.11)	–0.11 (–0.43 to 0.18)
Number of patients who experienced at least one hypoglycaemic episode (moderate or severe) [n (%)]	110 (15.85)	25 (3.70)	4.90 (3.13 to 7.67)
Number of moderate hypoglycaemic episodes (< 2.0–2.5 mmol/l)
Number of episodes	127	30
Number of patients who experienced at least one episodeb [n (%)]	87 (12.54)	21 (3.11)	4.46 (2.73 to 7.28)
Mean number of episodes per patient (SE)	0.18 (0.03)	0.04 (0.01)
Number of severe hypoglycaemic episodes (< 2.0 mmol/l)
Number of episodes	70	11
Number of patients who experienced at least one episodeb [n (%)]	51 (7.35)	10 (1.48)	5.27 (2.65 to 10.48)
Mean number of episodes per patient (SE)	0.1 (0.02)	0.02 (0.005)
Seizures given pharmacological treatment [n (%)]	23 (3.31)	15 (2.22)	1.15 (0.77 to 2.98)

FIGURE 6.

Proportion of patients on mechanical ventilation.

FIGURE 7.

Proportion of patients receiving vasoactive drugs.

FIGURE 8.

Mean caloric intake.

In terms of adverse effects, there were 135 patients whose blood glucose level was below the threshold that defined moderate hypoglycaemia; 61 of these had one or more episodes that were considered severe. Hypoglycaemia occurred in 33 (4.1%) patients not given insulin, but was more commonly observed in patients who received insulin [102 (17.9%)].

Hypoglycaemia occurred in a greater proportion of patients in the TGC arm than in the CM arm of the study (moderate, 12.5% vs. 3.1%, p < 0.001; severe, 7.3% vs. 1.5%, p < 0.001). Of the patients who experienced any hypoglycaemic episode, 11.1% died as opposed to 4.4% of those who did not experience any hypoglycaemic episode (p = 0.001).

Thirty-day economic outcomes

For the index hospital episode, the mean PICU bed-days, LOS on GM wards and total LOS for the index hospital episode were similar between arms (Table 12). The mean total number of hospital days up to day 30, including both the initial episode and readmissions to the initial PICU before day 30, were similar between arms (see Table 12). For the stratum admitted for cardiac surgery, the mean total LOS was again comparable between arms (Table 13). As regards the non-cardiac stratum, for the initial hospital episode, the mean numbers of PICU days, LOS on GM wards and total LOS were lower for the TGC than the CM arm (Table 14).

Tables 15–17 report that the mean numbers of PICU bed-days, by HRG level, were similar between arms.

Overall, the mean total costs at 30 days post randomisation were similar between arms (Table 18). For the cardiac subgroup, the mean total costs per patient were £16,228 (TGC) and £17,005 (CM) (Table 19). For the non-cardiac subgroup, the TGC arm had lower mean costs than the CM group, with an incremental cost of –£2319 (95% CI –£4702 to £124) (Table 20). Including the treatment by cardiac interaction term led to a statistically significant improvement in model fit (p < 0.001).

Twelve-month results

Index hospital episode and readmissions to paediatric intensive care unit within 30 days post randomisation

Table 21 reports the mean total number of hospital days up to 12 months post randomisation, including the initial hospital episode and any readmissions to PICU within 30 days. A lower proportion of patients in the TGC than in CM arm had an index hospital admission or relevant readmission that continued beyond day 30. Between 30 days and 12 months post randomisation, the mean number of days in PICU, on GM wards and in total, was lower for the TGC than the CM arm (see Table 21).

TABLE 21 - Lengths of stay (days) within 12 months post randomisation: whole study cohort

Includes hospital days between randomisation and 12 months for initial hospital episode and readmissions to initial PICU that were within 30 days.

Source: CRFs.

	TGC (N = 694)	CM (N = 675)
Mean (SD) total days up to 30 days post randomisation	16.62 (8.81)	16.98 (9.25)
30 days to 1 year
n (%) continuing admission	130 (18.73)	152 (22.52)
Mean (SD) PICU days	1.98 (11.86)	2.79 (15.19)
Mean (SD) days on general wards	5.67 (25.89)	9.75 (36.22)
Mean (SD) total days	7.64 (31.18)	12.54 (41.47)
Mean (SD) total hospital daysa	24.26 (35.40)	29.52 (46.18)

Four patients were still in hospital at the date of administrative censoring, with LOS ranging from 119 to 359 days. Each of these patients was assumed to have the mean total LOS taken across the whole sample still in hospital at the respective time point. (For example, for the patient censored at a LOS of 119 days, the assumed LOS was 228 days, according to the mean across the 46 patients still in hospital 118 days post randomisation.) One patient withdrew consent for participation in the study, after 8 days in hospital, and was assumed to have a total hospital LOS of 60 days, the mean across the whole sample of patients who were still in hospital after day 8.

For the cardiac stratum, the mean total LOS at 12 months was similar between arms (Table 22). For the non-cardiac subgroup, the TGC arm had a lower proportion of patients who had a hospital admission that continued beyond 30 days post randomisation, and on average reported fewer days on PICUs, and on GM wards (Table 23), than the CM arm. For the non-cardiac stratum, the mean total LOS for the initial episode and readmissions to PICU within 30 days was 31.0 days for the TGC arm compared with 44.5 days for the CM arm (see Table 23).

TABLE 22 - Lengths of stay (days) within 12 months post randomisation: cardiac surgery subgroup

Includes hospital days between randomisation and 12 months for initial hospital episode and readmissions to initial PICU that were within 30 days.

Source: CRFs.

	TGC (N = 421)	CM (N = 416)
Mean (SD) days at 30 days post randomisation	15.27 (8.29)	14.54 (8.25)
30 days to 1 year
n (%) continuing admission	59 (14.01)	52 (12.50)
Mean (SD) PICU days	1.39 (8.81)	1.26 (6.70)
Mean (SD) days on general wards	3.25 (18.54)	4.40 (26.48)
Mean (SD) total days	4.64 (22.17)	5.65 (29.30)
Mean (SD) total hospital daysa	19.90 (26.40)	20.19 (33.19)

TABLE 23 - Lengths of stay (days) within 12 months post randomisation: non-cardiac surgery subgroup

Includes hospital days between randomisation and 12 months for initial hospital episode and readmissions to initial PICU that were within 30 days.

Source: CRFs.

	TGC (N = 273)	CM (N = 259)
Mean (SD) days at 30 days post randomisation	18.70 (9.20)	20.91 (9.43)
30 days to 1 year
n (%) continuing admission	71 (26.01)	100 (38.61)
Mean (SD) PICU days	2.88 (15.40)	5.25 (22.82)
Mean (SD) days on general wards	9.40 (33.97)	18.35 (46.66)
Mean (SD) total days	12.27 (41.03)	23.60 (53.98)
Mean (SD) total hospital daysa	30.98 (45.18)	44.51 (58.59)

Figure 9 plots the proportion of patients over time who were still in hospital following the index admission. For the overall sample, and the cardiac patients, the proportion still in hospital was similar between arms at each time point. For the non-cardiac patients, a higher proportion of the CM than the TGC arm were still in hospital 60 and 90 days post randomisation.

FIGURE 9.

Proportion of patients remaining in hospital (index admission) up to 180 days post randomisation. Proportion of patients remaining in hospital: (a) whole study cohort; (b) cardiac surgery subgroup; and (c) non-cardiac surgery group.

Other hospital and community service use (after discharge from index hospital episode but excluding any readmissions to the initial paediatric intensive care unit within 30 days)

Figure 10 shows the flow of patients from randomisation to response to the service-use questionnaire. In the overall sample, a total of 397 patients (203 in the TGC arm, 194 in the CM arm) were randomised after 30 October 2010 and could not be followed up for 1 year; that is, for the purposes of collecting information on service use, these patients were administratively censored. Patients were also ineligible for the service-use questionnaire if their GP did not confirm that it was appropriate to contact them to administer the questionnaire. Of the eligible patients, the response rate to the service-use questionnaire was 63% in the TGC arm and 61% in the CM arm. For those who responded to the questionnaire, the mean LOS following hospital readmissions after 30 days post randomisation, and the mean number of contacts with hospital and personal social services, was similar between the randomised arms (Tables 26–28). The mean total costs of hospital and community health services were also similar between the randomised arms (Tables 29–31).

FIGURE 10.

Flow chart for 12-month follow-up for service-use questionnaire.

TABLE 26 - Levels of service use for questionnaire responders within 12 months post randomisation: whole study cohort

Readmissions to GM wards, to PICUs other than the one the patient was originally randomised to or readmissions to the same PICU that were after day 30.

Source: service-use questionnaire.

Service	TGC (n = 207) [mean (SD)]	CM (n = 201) [mean (SD)]
Inpatient hospital days following readmissiona	6.12 (1.01)	6.63 (1.29)
Hospital outpatient visits	6.81 (0.46)	6.93 (0.45)
GP contacts	1.23 (0.12)	1.08 (0.99)
Practice nurse contacts	1.76 (0.21)	2.14 (0.29)
Health visitor contacts	5.85 (0.74)	6.87 (0.69)
Social worker contacts	0.10 (0.02)	0.08 (0.02)
Speech and language therapist contacts	2.18 (0.38)	1.88 (0.32)
Occupational therapist contacts	7.83 (1.80)	11.94 (2.30)
Other health service contacts	9.52 (1.13)	10.04 (1.11)

TABLE 27 - Levels of service use for questionnaire responders within 12 months post randomisation: cardiac surgery subgroup

Readmissions to GM wards, to PICUs other than the one the patient was originally randomised to or readmissions to the same PICU that were after day 30.

Source: service-use questionnaire.

Service	TGC (n = 127) [mean (SD)]	CM (n = 121) [mean (SD)]
Inpatient hospital days following readmissiona	7.10 (1.50)	8.25 (2.03)
Hospital outpatient visits	6.81 (0.59)	6.64 (0.55)
GP contacts	1.37 (0.17)	1.19 (0.13)
Practice nurse contacts	1.88 (0.25)	2.11 (0.05)
Health visitor contacts	6.24 (0.78)	8.24 (0.98)
Social worker contacts	0.05 (0.02)	0.03 (0.01)
Speech and language therapist contacts	2.32 (0.71)	1.77 (0.50)
Occupational therapist contacts	5.84 (1.99)	7.20 (2.26)
Other health service contacts	8.86 (1.31)	9.06 (1.44)

TABLE 28 - Levels of service use for questionnaire responders within 12 months post randomisation: non-cardiac surgery subgroup

Readmissions to GM wards, to PICUs other than the one the patient was originally randomised to or readmissions to the same PICU that were after day 30.

Source: service-use questionnaire.

Service	TGC (n = 79) [mean (SD)]	CM (n = 80) [mean (SD)]
Inpatient hospital daysa	4.54 (1.00)	4.15 (1.00)
Hospital outpatient visits	6.82 (0.75)	7.36 (0.75)
GP contacts	1.02 (0.19)	0.91 (0.13)
Practice nurse contacts	1.58 (0.39)	2.18 (0.53)
Health visitor contacts	5.22 (1.47)	4.78 (0.86)
Social worker contacts	0.17 (0.04)	0.17 (0.04)
Speech and language therapist contacts	2.32 (0.71)	1.77 (0.50)
Occupational therapist contacts	11.04 (3.45)	11.17 (4.58)
Other health service contacts	10.60 (2.08)	10.63 (1.97)

TABLE 29 - Costs (£) of health and personal social services for questionnaire responders, between discharge from the index admission and 12 months post randomisation: whole study cohort

Refers to readmissions to GM wards, to PICUs other than the one the patient was originally randomised to or readmissions to the same PICU that were after day 30.

Source: service-use questionnaire.

Service	TGC (n = 207) [mean (SD)]	CM (n = 201) [mean (SD)]
Inpatient hospital daysa	1367 (243)	1402 (264)
Hospital outpatient visits	1019 (66)	1002 (68)
GP contacts	51 (5)	58 (6)
Practice nurse contacts	13 (3)	10 (4)
Health visitor contacts	67 (8)	78 (8)
Social worker contacts	3 (1)	3 (1)
Speech and language therapist contacts	18 (3)	15 (3)
Other health service contacts	115 (15)	119 (15)

TABLE 30 - Costs (£) of health and personal social services for questionnaire responders, between discharge from the index admission and 12 months post randomisation: cardiac surgery subgroup

Readmissions to GM wards, to PICUs other than the one the patient was originally randomised to or readmissions to the same PICU that were after day 30.

Source: service-use questionnaire.

Service	TGC (n = 127) [mean (SD)]	CM (n = 121) [mean (SD)]
Inpatient hospital daysa	1576 (358)	1732 (414)
Hospital outpatient visits	976 (82)	1001 (87)
GP contacts	56 (6)	65 (8)
Practice nurse contacts	12 (5)	5 (3)
Health visitor contacts	71 (9)	94 (11)
Social worker contacts	2 (1)	1 (1)
Speech and language therapist contacts	17 (16)	16 (3)
Other health service contacts	107 (19)	104 (19)

TABLE 31 - Costs (£) of health and personal social services for questionnaire responders, between discharge from the index admission and 12 months post randomisation: non-cardiac subgroup

Readmissions to GM wards, to PICUs other than the one the patient was originally randomised to or readmissions to the same PICU that were after day 30.

Source: service-use questionnaire.

Service	TGC (n = 79) [mean (SD)]	CM (n = 80) [mean (SD)]
Hospital inpatient daysa	1029 (256)	900 (211)
Hospital outpatient visits	1082 (111)	1004 (111)
GP contacts	48 (7)	48 (9)
Practice nurse contacts	16 (7)	17 (8)
Health visitor contacts	59 (17)	54 (10)
Social worker contacts	4 (11)	5 (11)
Speech and language therapist contacts	19 (6)	14 (4)
Other health service contacts	127 (25)	141 (26)

Twelve-month total costs

Tables 32–34 report the total costs at 12 months across all the resource-use items recorded. The values presented are the results after using MI to handle missing values for health and community service costs at 12 months. Table 32 reports that, overall, the mean total costs were lower in the TGC than in the CM group, but with 95% CIs that encompass zero. For the cardiac surgery stratum, the mean total costs were similar between the groups (see Table 33), but, for non-cardiac patients, the mean costs were lower in the TGC than in the CM group, with an incremental cost of –£9865 (95% CI –£18,558 to –£1172) (see Table 34).

TABLE 32 - Total costs (£) at 12 months post randomisation: whole study cohort

Includes index hospital admissions and readmissions to initial PICU that were within 30 days but that continued beyond day 30. Source: CRF.

Includes other hospital readmissions, and other hospital and community service use. Source: service-use questionnaire.

	TGC (n = 694) [mean (SD)]	CM (n = 675) [mean (SD)]	Incremental [mean (95% CI)]
Hospital costs at 30 days	16,228 (13,504)	17,005 (12,913)
Hospital costs between 30 days and 12 monthsa	5683 (27,978)	8463 (35,366)
Other hospital and community health service costs at 12 monthsb	2388 (3659)	2452 (4010)
Grand total costs up to 1 year	24,300 (34,503)	27,920 (42,775)	–3620 (–7743 to 502)

TABLE 33 - Total costs (£) at 12 months post randomisation: cardiac surgery subgroup

Includes index hospital admissions and readmissions to initial PICU that were within 30 days but that continued beyond day 30. Source: CRF.

Includes other hospital readmissions, and other hospital and community service use. Source: service-use questionnaire.

	TGC (n = 421) [mean (SD)]	CM (n = 416) [mean (SD)]	Incremental [mean (95% CI)]
Hospital costs at 30 days	14,465 (12,437)	14,350 (11,285)
Hospital costs between 30 days and 12 monthsa	3811 (20,497)	3815 (17,720)
Other hospital and community health service costs at 12 monthsb	2652 (3890)	2630 (4319)
Grand total costs up to 1 year	20,929 (27,385)	20,796 (26,520)	133 (–3568 to 3833)

TABLE 34 - Total costs (£) at 12 months post randomisation: non-cardiac surgery subgroup

Includes index hospital admissions and readmissions to initial PICU that were within 30 days but that continued beyond day 30. Source: CRF.

Includes other hospital readmissions, and other hospital and community service use. Source: service-use questionnaire.

	TGC (n = 273) [mean (SD)]	CM (n = 259) [mean (SD)]	Incremental [mean (95% CI)]
Hospital costs at 30 days	18,949 (14,614)	21,268 (14,183)
Hospital costs between 30 days and 12 monthsa	8569 (36,495)	15,927 (51,688)
Other hospital and community health service costs at 12 monthsb	1979 (3056)	2167 (3437)
Grand total costs up to 1 year	29,498 (4267)	39,363 (58,551)	–9865 (–18,558 to –1172)

Figures 11–13 report SAs that investigate whether or not the base-case results are robust to alternative assumptions. The results show that the incremental costs under these alternative scenarios are similar to the base case. For example, in the SAs that include additional costs of staff time and tests associated with monitoring TGC, and further costs for managing hypoglycaemic episodes, the mean incremental costs of TGC overall and for the non-cardiac subgroup are similar to the base case (see Figures 11–13). Moreover, when alternative approaches were taken to unit costing, this had little impact on the results.

FIGURE 11.

Sensitivity analysis reporting mean total costs at 12 months post randomisation according to alternative assumptions: whole study cohort.

FIGURE 12.

Sensitivity analysis reporting mean total costs at 12 months post randomisation according to alternative assumptions: cardiac surgery subgroup.

FIGURE 13.

Sensitivity analysis reporting mean total costs at 12 months post randomisation according to alternative assumptions: non-cardiac surgery subgroup.

Lifetime cost-effectiveness results

The Kaplan–Meier survival curves show that when the time horizon was extended beyond 1 year, for those for whom survival data were available, the probability of survival remained similar between arms (Figure 14).

FIGURE 14.

Kaplan–Meier survival curves. Overall cohort, TGC vs. CM.

Figure 15 considers alternative parametric extrapolations for both treatment arms combined, using the observed survival data after day 30. Of the alternative survival functions, the Gompertz function appears to fit the observed data best in that it reports the lowest Akaike and Bayesian information criterion (Table 35). The Gompertz function also offers the most plausible projections of future survival (Table 36), in that the levels of excess death compared with those for the age- and gender-matched general population remain constant over time from 2 years post randomisation onwards.

FIGURE 15.

Comparison of alternative parametric extrapolations for survival from 12 months to 5 years post randomisation, across both randomised arms.

Tables 37–39 present the resultant life-years, QALYs, lifetime costs and INBs according to the base-case assumptions. Overall, at a threshold of £20,000 per QALY, the INBs are positive, but with wide 95% CIs that include zero. For cardiac patients, the INBs are close to zero with wide CIs. For non-cardiac patients, the INBs are positive but with 95% CIs that include zero.

The cost-effectiveness acceptability curves consider alternative thresholds of willingness to pay for a QALY gain, and show that, overall and for the cardiac surgery stratum, it is highly uncertain whether or not TGC is cost-effective (Figures 16 and 17). For the non-cardiac stratum, the probability that TGC is cost-effective is relatively high. For example, at ceiling ratios of £10,000 to £30,000 per QALY, the probability that TGC is cost-effective ranges from 90% to 70% (Figure 18).

FIGURE 16.

Probability that TGC vs. CM is cost-effective at alternative levels of willingness to pay for a life-year gained: whole study cohort.

FIGURE 17.

Probability that TGC vs. CM is cost-effective at alternative levels of willingness to pay for a life-year gained: cardiac surgery subgroup.

FIGURE 18.

Probability that TGC vs. CM is cost-effective at alternative levels of willingness to pay for a life-year gained: non-cardiac surgery subgroup.

The SA on the lifetime results suggests that these findings are robust to alternative assumptions about the extrapolation of long-term survival, QoL for PICU survivors or long-term costs (Figures 19–21).

FIGURE 19.

Sensitivity analysis reporting lifetime INBs (£) according to alternative assumptions: whole study cohort.

FIGURE 20.

Sensitivity analysis reporting lifetime INBs (£) according to alternative assumptions: cardiac surgery subgroup.

FIGURE 21.

Sensitivity analysis reporting lifetime INBs (£) according to alternative assumptions: non-cardiac surgery subgroup.

Limitations

The primary end point in the study was selected based on the best evidence at the time and took into account usual practice in reporting short-term outcomes in studies conducted on critically ill patients. Our results, however, indicate that for non-cardiac surgical cases, it may be better to assess VFDs or QoL at a later time point, for example day 60, in future studies.

The cost-effectiveness analysis has some limitations. First, the level of activity was measured only within PICUs and only for up to day 30, after which general PICU costs were assumed. But analysis using different assumptions made little difference to the results. Second, there were some missing data on follow-up costs and at 12 months from administrative censoring owing to study funding, but also non-response to questionnaire. This was handled using MI, which assumes that data are missing conditional on baseline factors and other end point and process measures that are observed; therefore, if missingness is driven by unobserved prognostic factors, this could have led to biased estimates. However, for questionnaire responders, the follow-up costs were similar between the groups. Third, like previous RCTs in PIC, this study did not measure QoL. Instead, the cost-effectiveness analysis used QoL data from a previous study that included children admitted to UK PICUs who were at least 6-months-old. 71 It is unclear whether or not the QoL values from this sample of older children (median age of approximately 5-years-old) apply directly to the population represented by CHiP patients. However, the results of the SA suggest that the lifetime cost-effectiveness results were robust to the QoL value assumed for PICU survivors. Four, the cost-effectiveness analysis required that survival data from the CHiP trial were extrapolated over the lifetime. The analysis took a standard approach and applied the parametric survival function that was judged most plausible. There was no evidence from the survival data available from CHiP of any differences in survival up to 24 months, and the SA again shows results were robust to alternative assumptions.

Strengths

The CHiP trial addressed both clinical and economic questions, with follow-up at 12 months. The trial was the largest RCT in PIC, and was rigorously designed and conducted. Randomisation reduced the potential for selection bias at trial entry, and there was no loss to follow-up for the primary outcome. Although the local clinicians could not be blinded to allocation post randomisation, the use of a hard outcome (mortality) as an integral component of the 30-day primary outcome ensured the risk of biased outcome assessment was low, and it is not plausible that LOS would be influenced by allocation across the range of PICUs involved. The reasonably small number of secondary outcomes reduced the risks of problems associated with multiple testing.

There was careful preparation and training for the introduction of the clinical management protocol with a run-in period and minor adjustments to address any concerns about risks of hypoglycaemia. The delivery of TGC may be further improved by the use of continuous glucose monitoring systems,72 including those with paediatric decision support.

The biggest drivers of cost differences between the arms were PICU and total hospital LOS. For the vast majority of patients (99%), these resource-use data were measured until they were discharge from their index hospital admission or following readmission to the PICU within 30 days, and up to a maximum of 12 months from randomisation. The study employed careful resource-use and cost measurement, estimated from large sample patient-level data drawing on the PCCMDS, which provided reasonable inference on costs. State-of-the-art approaches to handling missing 12-month cost data were used. The study also recorded use of other hospital and community health services and found that reductions in hospital LOS for the TGC arm were not offset by increased use of other services.

The CHiP trial was conducted in a large proportion of the PICUs in England, increasing the generalisability of the clinical and cost-effectiveness results to other UK PICUs, and other settings with similar populations and systems of care.

As the largest PICU trial to date, CHiP had sufficient power to reliably assess whether the balance of costs and benefits differed depending on if the admission was following cardiac surgery or not and to provide reasonable inference on costs.

The results of the economic evaluation were subject to extensive SAs and were robust to alternative assumptions, including, for example, the approaches taken to unit costing or the projection of life expectancy from the trial data.

The CHiP trial results in context

In considering how the CHiP trial results compare with other research, we have drawn on the trials of TGC with adults in critical care, with preterm neonates in neonatal ICUs (NICU) and with children in PICUs, using studies published prior to CHiP’s initiation and reviewed in Chapter 1, and later relevant studies in different populations of critically ill patients. The particular emphases here are the size and type of effect of the intervention, the rate of hypoglycaemia and the differential effect in medical and surgical populations. There are some distinctions in these classifications between the CHiP trial and the adult studies. About a third of the patients in the Leuven surgical trial24 underwent surgery that was not cardiovascular, and, in the CHiP trial, general surgery was incorporated in the non-cardiac surgery group. The CHiP trial stratification of cardiac surgical and non-cardiac surgical does, however, allow direct comparison with the Leuven paediatric study73 and the ongoing US paediatric cardiac surgical study. 74

Implications for health care

For the cardiac surgery subgroup, average costs at 12 months post randomisation were similar between arms and TGC was unlikely to be cost-effective. For children admitted to PICUs following cardiac surgery, our study suggests that PICUs should not adopt TGC and should continue with CM.

For children admitted to PICUs for other reasons, although TGC did not improve short-term clinical outcomes compared with CM, children in this subgroup were discharged earlier from hospital. This clinical benefit (earlier discharge) was associated with substantially reduced NHS costs. These average reductions in NHS costs were not offset by increased use of community health services in the TGC versus CM arm.

This is the first study to find that TGC may be cost-effective in children in PIC who were not admitted for cardiac surgery. Before a policy of TGC can be recommended for this important subgroup of NHS patients (around 1200 patients per year), careful consideration should be given to the balance of risk and benefit of this intervention: between the small increased risk of hypoglycaemia and the potential reduction in length of hospital stay and associated cost savings.

Recommendations for further research

The findings of the CHiP trial raise the following important questions:

• Does the excess rate of moderate and severe hypoglycaemia during TGC for children admitted to PICUs for reasons other than cardiac surgery have an impact on long-term neurodevelopmental outcomes?

  1. Before a policy of TGC can be recommended for children admitted to PICUs for reasons other than cardiac surgery, further information is needed about the long-term implications of the small increased risk of hypoglycaemia. Over 100 children in the CHiP trial experienced at least one episode of moderate or severe hypoglycaemia. In the first instance, we will undertake a post-hoc analysis of the data along the lines of the recent analysis of hypoglycaemia complicating TGC in adults that was published in the New England Journal of Medicine. One of the CHiP investigators has been in contact with the lead investigator of NICE SUGAR and plans are being made for the post-hoc data evaluation. Subsequently, a neurodevelopmental follow-up study of these children would inform clinicians of the long-term risk of hypoglycaemia in this population.

• Can we improve the delivery of TGC to minimise the risk of hypoglycaemia?

  1. If a policy of TGC is to be recommended for children admitted to PICUs for reasons other than cardiac surgery, research is needed to further refine clinical algorithms for the delivery of TGC, and to assess whether or not delivery of TGC may be further improved by the use of continuous glucose monitoring systems,72 including those with paediatric decision support.

• Does TGC in critically ill children protect the kidneys from injury?

  1. One hypothesis raised by the CHiP trial is that TGC can reduce acute kidney injury (AKI). AKI is a common complication of critical illness, which in its most severe form requires RRT. Children requiring RRT have a prolonged stay in PICU and increased mortality (≈ 40%). Children in the TGC arm in CHiP had a lower incidence of RRT than those in the CM arm. However, RRT represents the most severe end of the spectrum of AKI and is a rare outcome compared with lesser degrees of AKI. An add-on study in CHiP patients could provide more precise measures of AKI, through estimation of daily glomerular filtration rate from sequential plasma creatinine values obtained as part of routine clinical care. Creatinine values would be retrieved from laboratory databases and statistical analysis would be undertaken to investigate this research question.

• Do the findings from CHiP apply to routine clinical practice?

  1. The CHiP RCT had a pragmatic design, included the majority of English PICUs and had broad patient eligibility criteria. However, as with any RCT, the findings might not apply directly to routine clinical practice. Inevitably, the CHiP centres did not screen all potential patients, and eligible patients were excluded for various reasons, for example because of refused consent. Further, the CM delivered in the CHiP study may differ from that provided routinely. These potential concerns illustrate the general challenge that RCT findings might not apply directly to routine clinical practice. To address these issues, further research is required that develops an approach for maximising external validity. One possible approach to extending the CHiP findings would be to carefully assess the external validity of the study findings using PICANet data. This further research could examine whether or not the patients’ baseline characteristics, resource use, costs and outcomes following CM in CHiP differed from those observed in routine clinical practice. This setting, with a large RCT nested within an observational data set, would provide an opportunity for considering whether or not results from pragmatic trials such as CHiP have external validity.

• What can be learnt from triallists, clinicians, parents and older children about their experiences of participating in CHiP, to aid the design and conduct of future PICU trials?

  1. As more PICU RCTs assessing treatments in the NHS are funded, it is important for their success to learn from the experiences of participants in existing trials. A substantial body of research on participants’ experiences of NICU trials aided development of CHiP trial procedures, but there is currently no equivalent literature from PICUs. As the largest UK PICU trial to date, CHiP offers a timely opportunity for such work. Qualitative research exploring clinician and family experiences of implementing the CHiP protocol would inform the conduct of future PICU trials.

Trial Steering Committee

P Barnes, S Edwards, D Field, J Hooper, M Preece (chairperson), T Quick, C Snowdon, L Tume, D Vlasselaers and P Williamson.

Project management group

E Allen, H Betts, D Elbourne, R Grieve, C Guerriero, M Loverage, D Macrae, K Morris, J Pappachan, R Parslow, D Piercy, Z Sadique, L Van Dyck, N Smith-Wilson and R Tasker.

Trial co-ordinating centre

E Allen, L Brooks, K Diallo, D Elbourne, C Frost, A Gadhiya, P Henley, E Morris, D Piercy, P Ramos, J Sturgess and A Truesdale.

Clinical coordination centre (Royal Brompton Hospital)

S Bacon, H Betts and D Macrae.

Economic evaluation

R Grieve, C Guerriero and Z Sadique.

Data Monitoring Committee

D Dunger, D Harrison, D Hatch, G Peek and J Smith.

Design, branding and case report form development

S Moncrieff.

Official scoring

A Kirby (Clinical Psychologist).

Recruiting centres: collaborating doctors and nurses [the principal investigator(s) and lead research nurse(s) are listed first and italicised]

National Institute for Health Research Health Technology Assessment

R Davies, L Dodd and C Gregory.

Medicines for Children Research Network

V Poustie, R Smyth and W Van’t Hoff.

Research and development (Royal Brompton Hospital)

W Butcher, A Cooper, L Henderson and J Fatukasi.

The UK Paediatric Intensive Care Society Study Group

Publication

Macrae D, Grieve R, Allen E, Sadique Z, Morris K, Pappachan J, et al. A randomized trial of hyperglycemic control in pediatric intensive care. N Engl J Med 2014;370:107–18.

Contributions of authors

Duncan Macrae (Consultant in Paediatric Intensive Care) was the chief investigator. He was involved in the design and conduct of research and was a member of the trial development group, the project management group and the TSC. He was also involved in the development and refinement of the trial protocol, the conduct of the research interpretation and the reporting of results, and he revised the manuscript for important intellectual content.

Richard Grieve (Reader in Health Economics) led the health economic evaluation and was involved in the design and conduct of research, the trial development group, the project management group and the TSC, the statistical analysis, and the interpretation and reporting of results. He wrote the economic evaluation sections of the report and revised the manuscript for important intellectual content.

Elizabeth Allen (Senior Lecturer in Medical Statistics) was the lead statistician and was involved in the design and conduct of research, the Data Monitoring Committee, the project management team, the interpretation and reporting of results, and the draft and revision of the manuscript for important intellectual content.

Zia Sadique (Research Fellow in Health Economics) undertook the majority of the health economics analysis and was involved in the design and conduct of research, the project management team, the interpretation and reporting of results, and revision of the manuscript for important intellectual content.

Helen Betts (Lead Research Nurse, London) was involved in the set up, conduct and acquisition of data of this research and in the preparation of the manuscript.

Kevin Morris (Consultant Paediatric Intensivist) was a member of the trial development group, the TMG and the TSC. He was the PI for Birmingham Children’s Hospital. He was involved in the development and refinement of the trial protocol, the conduct of the research, acquisition of data, the interpretation and reporting of results, and revision of the manuscript for important intellectual content. In addition, he was chief investigator for an add-on study to CHiP investigating the mechanism of hyperglycaemia in critically ill children.

Vithayathil John Pappachan (Consultant Paediatric Intensivist) was the PI for University Hospitals Southampton NHS Foundation Trust and was involved in the conduct of research, trial design and management acquisition of data, the interpretation and reporting of result and manuscript preparation.

Roger Parslow (Senior Lecturer at the University of Leeds) was the PI for PICANet and was a member of the TMG involved in the conduct of research facilitating the acquisition of data, the interpretation and reporting of results and revision of the manuscript for important intellectual content.

Robert C Tasker (Professor of Neurology and Anaesthesia) was part of the senior management structure for the trial: he was involved with the conception and design of the trial, its conduct of research, and interpretation and reporting of results. He helped draft and revise the manuscript for important intellectual content.

Paul Baines (Consultant Paediatric Intensivist) was the PI at Alder Hey Hospital and was involved in the conduct of research, acquisition of data, and interpretation and reporting of results.

Michael Broadhead (Consultant Cardiac Intensivist) was the PI at Great Ormond Street Hospital for Children and was involved in the conduct of research, patient consent, data review and reporting.

Mark L Duthie (Consultant Paediatric Intensivist) was the PI for University Hospitals of Leicester and was involved in conduct of the trial, recruitment of subjects, reporting and interpretation of acquired data.

Peter-Marc Fortune (Consultant Paediatric Intensivist) was the PI for Royal Manchester Children’s Hospital and was involved with the drafting of the trial protocols, conduct of the research and acquisition of data.

David Inwald (Senior Lecturer and Honorary Consultant Paediatric Intensivist) was the PI for Imperial College Healthcare NHS Trust and was involved in the conduct of research, acquisition of data, and interpretation and reporting of results.

Paddy McMaster (Consultant Paediatrician) was the PI for the University Hospital of North Staffordshire and was involved in the conduct of research, acquisition of data and interpretation and reporting of results.

Mark J Peters (Senior Lecturer in Paediatric Intensive Care) was the PI for Great Ormond Street Hospital for Children paediatric and NICUs, and was involved in the design and conduct of research, acquisition of data, and interpretation and reporting of results.

Margrid Schindler (Consultant Paediatric Intensivist) was the PI for Bristol Royal Hospital for Children and was involved in the conduct of the research, acquisition of data and reporting of results.

Carla Guerriero (Research Fellow in Health Economics) helped with the design and analysis of the health economic evaluation and was involved in the project management team and in the interpretation and reporting of results.

Deborah Piercy (Trial Manager) was the trial manager (until 2009) and was involved in the set up, running and management of the trial.

Zdenek Slavik (Consultant Paediatric Cardiologist/Intensivist) originally suggested carrying out a prospective study involving TGC in paediatric patients requiring intensive care treatment, and was involved in the initial discussion of its merits and initial planning stages of the trial.

Claire Snowdon (Qualitative Researcher) was a member of the TSC, was involved in the design of the trial communication strategies with parents, advised on the conduct of research and the acquisition of data in her role as TSC member and contributed to the reporting of results.

Laura Van Dyck (Trial Manager) was involved in the day-to-day running of the trial, in particular the data management, and was a member of the project management team.

Diana Elbourne (Professor of Healthcare Evaluation) was the lead investigator for trial design and management, and was involved in the design and conduct of the research, the trial development group, the project management group and the TSC, the statistical analysis, the interpretation and reporting of results, and revision of the manuscript for important intellectual content.

Disclaimers

This report presents independent research funded by the National Institute for Health Research (NIHR). The views and opinions expressed by authors in this publication are those of the authors and do not necessarily reflect those of the NHS, the NIHR, NETSCC, the HTA programme or the Department of Health.

Appendix 1a

Appendix 1b

Appendix 1c

Appendix 1d

Appendix 1e

Appendix 1f

Amendment 1 (protocol version 2, 20 November 2007)

• Changes to the Patient Information Sheets and Consent Forms for parents to include why their child had been chosen the this study, that blood taken specifically for this study would be painless and taken only from a line that already been inserted, and clarity on where and who was signing the consent form.

• Changes to the protocol to more clearly define how the insulin infusions should be prepared and managed.

Amendment 2 (protocol version 3, 13 February 2008)

• Changes to the protocol to clarify the dosing in the control group.

Amendment 3 (16 October 2008)

• Poster to be displayed in intensive care units.

Amendment 4 (17 March 2009)

• Addition of the following two sites:

  • Leicester Royal Infirmary and Glenfield Hospital (in Leicester).

  • University Hospital of North Staffordshire (in Stoke on Trent).

Amendment 5 (5 February 2009)

• Letter to be sent to parents at discharge.

• Diary to be given to parents at discharge.

• Letter to be sent to parents at 12 months.

• Questionnaire to be sent to parents at 12 months.

Amendment 6 (protocol version 4, 15 May 2009)

• Changes to the protocol to make minor alterations to the insulin control guidelines for the tight group.

Amendment 7 (not applicable)

• Not applicable, due to misnumbering by ethics (incorrectly number amendment 6 as amendment 7, and then unable to correct the numbering in their database).

Amendment 8 (28 June 2009)

• Follow-up questionnaires to be sent at 1 year to TBI patients (Conners’ Parent Rating Scale - Revised, Health Utilities Index-2, and Child Behaviour Checklist).

Amendment 9 (18 November 2009)

• Letter to parents of children who have died.

Amendment 10 (protocol version 5, 5 March 2010)

• Changes in the protocol, patient information sheets and GP and parent letters to reflect changes in the TBI follow-up (all questionnaires now to be sent to them, rather than one to be completed over the telephone).

• Changes in the protocol to add the publication policy.

• Letter for parents of children who are inpatients for over 1 year.

Amendment 11 (protocol version 6, 23 August 2010)

• Changes in the protocol, patient information sheets, consent forms, GP and parent letters due to the 1 year follow-up questionnaires no longer being sent out to parents (due to reasons of cost).

• Changes in patient information sheets and consent forms to reflect the change in name of the NHS information centre.

• Changes in the protocol, patient information sheets and consent forms to make it clearer that their patient data (including personal details) would be sent to the London School of Hygiene and Tropical Medicine.

• Changes in the protocol to update the study background section to include more recent studies and literature.

Amendment 12 (23 August 2010)

• Addition of the following site:

  • St George’s Healthcare NHS Trust (in London).

Amendment 13 (14 December 2010)

• Change in PI at University Hospitals of North Staffordshire NHS Trust.

Definition of traumatic brain injury

Accidental trauma to the head resulting in need for intubation and mechanical ventilation.

Population

Seven hundred and fifty ICU admissions per year in UK. Estimate of 150 recruited to the trial.

Outcomes assessment

This will comprise four components:

Overall health status: measured by the Health Utilities Index (HUI-3) [HUI-3. Copyright 2002 Health Utilities Inc. (HUInc), 88 Sydenham Street, Dundas ON, Canada L9H 2V3. www.healthutilities.com].

Global neurological outcome: measured by the Kings Outcome Scale for Childhood Head Injury (KOSCHI). 46

Attention and behavioural assessment: measured by the CBCL (CBCL. Copyright 2000 Achenbach T and Rescorla L. ASEBA, University of Vermont, 1 South Prospect Street, Burlington, VT 05401-3456. www.ASEBA.org).

The CRS-R:S. 47

The HUI, CBCL and CRS-R:S are written questionnaires that will be posted out to the families. They take approximately 30 minutes to complete.

Health Utilities Index is a multi-attribute health status classification system. Seven attributes (sensation, mobility, emotion, cognition, self-care, pain, fertility) are categorised according to one of 4 or 5 levels. In this population fertility will be excluded. The algorithm (from death to perfect health scale) provides a single numerical value.

KOSCHI is a 5-point categorical scale, ranging from death to normal neurological function, and is similar in structure to the Glasgow Outcome Scale, which is widely used in adult studies. In addition the KOSCHI is further subdivided into two subcategories at points 4 and 5 on the scale (moderate outcome and good outcome). Patient outcomes will be dichotomised between patients in categories 1, 2, 3 and 4A, and those in 4B, 5A and 5B.

Appendix 7a

Appendix 7b

Appendix 7c

Appendix 7d

Appendix 7e

Appendix 7f

Appendix 7g

Appendix 7h

Terms of reference

• The TSC should approve the protocol and trial documentation in a timely manner.

• In particular the TSC should concentrate on progress of the trial, adherence to the protocol, patient safety and consideration of new information of relevance to the research question.

• The safety and well-being of the trial participants are the most important consideration and should prevail over the interests of science and society.

• The TSC should provide advice, through its chair, to the chief investigator, the trial sponsor, the trial funder, on all appropriate aspects of the trial. Specifically, the TSC will:

  • Monitor recruitment rates and encourage the TMG to develop strategies to deal with any recruitment problems.

  • Monitor completion of data sheets and comment on strategies from TMG to encourage satisfactory completion in the future.

  • Monitor follow-up rates and review strategies from TMG to deal with problems including sites that deviate from the protocol.

  • Approve any amendments to the protocol, where appropriate.

  • Approve any proposals by the TMG concerning any change to the design of the trial, including additional sub-studies.

  • Oversee the timely reporting of trial results.

  • Approve and comment on the statistical analysis plan.

  • Approve and comment on the publication policy.

  • Approve and comment on the main trial manuscript.

  • Approve and comment on any abstracts and presentations of any results during the running of the trial.

  • Approve external or early internal requests for release of data or subsets of data or samples including clinical data and stored biological samples.

  • Receive reports from the DMEC.

  • The TSC will make decisions as to the future continuation (or otherwise) of the trial.

• Membership of the TSC should be limited and include an independent chair, at least two other independent members, two collaborators and two members of the public. The investigators and the trial project staff are ex officio.

• Representatives of the trial sponsor and the HTA should be invited to all TSC meetings.

• Responsibility for calling and organising the TSC meetings lies with the chief investigator. The TSC should meet at least annually, although there may be periods when more frequent meetings are necessary.

• There may be occasions when the trial sponsor or the HTA will wish to organise and administer these meetings in exceptional circumstances.

• The TSC will provide evidence to support any requests for extensions, including that all practicable steps have been taken to achieve targets.

• The TSC will maintain confidentiality of all trial information that is not already in the public domain.

Membership

Professor Michael Preece (chairperson) Consultant Paediatrician, Great Ormond Street Children’s Hospital

Mrs Pamela Barnes Lay member

Ms Sian Edwards Paediatric Pharmacist, Royal Brompton Hospital

Professor David Field Neonatologist, Leicester Royal Infirmary and the University of Leicester

Dr James Hooper Consultant Clinical Biochemist, Royal Brompton Hospital

Mrs Tara Quick Lay member, parent

Dr Claire Snowdon Lecturer, Medical Statistics Department, LSHTM, and Centre for Family Research, University of Cambridge (until 2011)

Ms Lyvonne Tume Research Nurse, Royal Liverpool Children’s Hospital

Dr Dirk Vlasselaers Consultant Paediatric Intensivist, Leuven, Belgium

Professor Paula Williamson Professor of Medical Statistics, University of Liverpool

Terms of reference

To safeguard the interests of trial participants, monitor the main outcome measures including safety and efficacy, and monitor the overall conduct of the CHiP study.

The DMEC should receive and review information on the progress and accruing data of CHiP and provide advice on the conduct of the trial to the Trial Steering Committee (TSC).

The DMEC should inform the Chair of the TSC if, in their view the results are likely to convince a broad range of clinicians, including those supporting the trial and the general clinical community, that, on balance, one trial arm is clearly indicated or contraindicated for all participants or a particular category of participants, and there was a reasonable expectation that this new evidence would materially influence patient management.

Interim review of the trial’s progress including updated figures on recruitment, data quality, adherence to protocol, follow-up, and main outcomes and safety data. Specifically, these roles include to:

• monitor evidence for treatment differences in the main efficacy outcome measures

• monitor evidence for treatment harm (e.g. toxicity, SAEs and SARs, treatment related deaths)

• assess the impact and relevance of external evidence

• decide whether to recommend that the trial continues to recruit participants or if recruitment should be terminated either for everyone or for some treatment groups and/or some participant subgroups

• decide whether or not trial follow-up should be stopped earlier

• assess data quality, including completeness (and by so doing encourage collection of high quality data)

• maintain confidentiality of all trial information that is not in the public domain

• monitor recruitment figures and losses to follow-up

• monitor compliance with the protocol by participants and investigators

• consider the ethical implications of any recommendations made by the DMEC

  • monitor planned sample size assumptions, preferably with regards to

    1. (i) a priori assumptions about the control arm outcome and/or

    2. (ii) emerging differences in clinically relevant subgroups, rather than on emerging, unblinded differences between treatment groups, overall

• suggest additional data analyses if necessary

• advise on protocol modifications proposed by investigators or HTA (e.g. to inclusion criteria, trial endpoints, or sample size)

• monitor continuing appropriateness of patient information

• monitor compliance with previous DMEC recommendations.

Membership

Professor David Dunger (chairperson) Department of Paediatrics, University of Cambridge

Dr David Harrison Statistician, Intensive Care Audit and Research Network (ICNARC), Professor David Hatch Emeritus Professor of Paediatric Anaesthesia and Intensive Care, Great Ormond Street Hospital

Mr Giles Peek (till Sept. 2009) Consultant Cardiac Surgeon, Glenfield Hospital, Leicester

Dr Jon Smith (from Sept. 2009) Consultant Cardiothoracic Anaesthetist, Newcastle General Hospital

Data co-ordinating centre responsibilities

• To ensure that all members of the study team are able by knowledge, training and experience to undertake the roles assigned to them and to comply with requirements as specified by the host organisation.

• To provide overall efficient day-to-day management of the trial ensuring compliance with GCP.

• To ensure each centre is put on-line with the randomisation service after Local Research Ethics Committee (LREC), research and development (R&D) approval and the signed local collaborator agreement have been received from the sponsor.

• To provide site folders and relevant documentation to each centre.

• To contribute to the development of the protocol, and all study documentation including data sheets.

• To design, produce and regularly update all trial materials and arrange printing and supply of documentation.

• To monitor recruitment and advise on remedial action if targets are not being met.

• To set up and maintain the website.

• To service the Project management Committee, Steering Committee and any other relevant advisory groups.

• To use all reasonable efforts to ensure that the data collected and reported are accurate, complete and identifiable at source; and that record keeping and data transfer procedures adhere to the Data Protection Act 1998.

• To undertake the interim and final analyses and report regularly to the DMEC in a timely way at their request.

• To supply documentation and reports as deemed necessary by the sponsors to fulfil their obligations.

• To co-ordinate the preparation and publication of data, reports and information, ensuring that these meet legislative, contractual and ethical requirements.

• To co-operate with audits or inspections undertaken by the host institution, the sponsors and regulatory authorities including the MHRA as required.

• To assist investigations into any alleged research misconduct undertaken by or on behalf of the co-sponsors.

• To ensure safe storage of data, including trial site file, data sheets and other records for a period of 15 years after the conclusion of the trial.

• To inform the chief investigator of any changes in the trial protocol that affect the conduct of the trial.