Evaluation of efficacy, outcomes and safety of infant haemodialysis and ultrafiltration in clinical use: I-KID a stepped wedge cluster RCT

Types of ‘dialysis’ treatment

The word ‘dialysis’ is frequently used as a lay term to encompass all of the processes involved in replacing the function of the kidneys, that is cleaning the blood of waste chemicals and removing fluid from the body in a controlled way. However, medically it has a specific and more limited meaning and describes waste products being removed from the blood into fluid by a process of diffusion down their concentration gradients, and this may lead to confusion. In this report, we only use the term in a specific sense, as defined below.

Peritoneal dialysis is the process by which fluid is instilled cyclically into the abdomen and allowed to dwell, and during this time waste chemicals move across the lining peritoneal membrane between the blood that supplies the abdominal organs and the fluid by diffusion, prior to drainage. In PD, fluid removal or ultrafiltration (UF) is generated by the osmotic gradient between the blood and the dialysis fluid which causes water to cross into the peritoneal space.

Haemotherapy; both haemodialysis (HD) and continuous veno-venous haemofiltration (CVVH) are types of treatment in which blood is drawn from the patient into a disposable plastic circuit on a machine, processed through a filter which has a membrane that separates the blood from treatment fluids, and then returns it to the patient. We will call both of these treatments haemotherapies, and they share the need for vascular access into a central vein, an extracorporeal blood circuit which has a tendency to clot and typically requires anticoagulation treatment to stop that, but the mechanisms by which they remove chemical wastes differ.

Continuous veno-venous haemofiltration removes waste chemicals by filtering large volumes of plasma water and replenishing it with a chemically balanced replacement fluid. It generates UF by replacing slightly less fluid than it filters. CVVH machines are the devices conventionally used to provide haemotherapy to critically ill babies in a PICU setting, of which the Prismaflex^® (Baxter Healthcare, www.baxterhealthcare.co.uk) and Aquarius^® (Nikkiso, www.Nikkosomedical.com) are the most commonly used in the UK.

Haemodialysis removes waste chemicals by dialysis as they diffuse across the filter membrane from the plasma into dialysate fluid being pumped across the other side. It generates UF by increasing the pressure gradient between the plasma and the dialysate fluid. Conventional HD devices are widely used to provide therapy to children with chronic renal failure. NIDUS^® (named for the Newcastle Infant Dialysis and Ultrafiltration System) has been developed specifically to treat small infants, either in an acute PICU setting or for chronic use as required. Conventional HD devices generate the pressure gradient necessary for UF by a computer-controlled regulation of the blood and fluid pump speeds. The NIDUS generates the necessary UF pressures by volumetrically controlling the blood flow using syringes.

Haemotherapies can be provided to babies who are on extracorporeal membrane oxygenation (ECMO) support using a simple technique in which the ECMO blood circuit can be used to pump blood through a dialysis filter. Using ECMO plus haemodialysis (ECMO + HD), the flow of dialysis fluid is controlled by adapting intravenous infusion pumps, and UF is controlled by adjusting the difference between the speeds of the inflow and outflow pumps.

The problems

Providing RRT to young babies may be severely challenging because of their small size and immaturity, both for PD, and for haemotherapy solutions which require the use of devices that have disposable extracorporeal blood circuits. Publications indicate similar problems faced by clinicians worldwide who use adult devices because of a lack of alternatives, and the need for new solutions including improved device technology. 10,11

Acute peritoneal dialysis

Acute PD is the technically easiest method for providing RRT, and is carried out manually using simple circuits in small infants, with fresh dialysis fluid being run in and out of the patient’s abdomen through a catheter under gravity, as no suitable automated cycling devices exist. There is no lower size limit, and it is used frequently to support infants after open-heart surgery. 2,12 Complications are common in the smallest participants including leakage of dialysate from the access entry point, drainage difficulties and the risk of developing peritonitis or hyperglycaemia, and it cannot be used in babies who have had abdominal surgery such as for necrotising enterocolitis, which is common in small, unwell babies, or in infants with congenital abdominal wall defects. 1 UF is unpredictable, but can be monitored easily as the dialysate is drained and collected using calibrated burettes. The clearance of waste chemicals is relatively slow. Both the UF and biochemical clearances may fall or fail altogether, especially in unstable babies who develop splanchnic vasoconstriction which limits the supply of blood to the peritoneum.

Continuous veno-venous haemofiltration

Most small infants on RRT haemotherapy are treated with CVVH. These devices were initially designed and built to treat adults, then Gambro developed the Prisma^® with smaller volume circuits which was approved for use in all sizes of children by the Food and Drug Administration (FDA) in the USA, and obtained Conformité Européenne (CE) marking in Europe. However, it was subsequently recognised that its control and reporting of UF was insufficiently precise to guarantee delivering safe treatment to smaller children, even in those being kept in neutral fluid balance. It should be remembered that the typical circulating blood volume of an infant is approximately 80 ml/kg of body weight. For this reason, the FDA withdrew its approval for using the Prisma in children of <20 kg, and Europe limited its CE mark to children <8 kg. 13 Gambro report that their latest iteration, the Prismaflex, has a fluid removal accuracy of ±30 ml/hour or 300 ml/day. 14 Unfortunately, poor fluid control in conventional haemotherapy is a consequence of its inherent technology which relies upon the device computing pressure gradient measurements rather than volumetric monitoring, and erratic and unrecognised variations in the UF have been shown in vitro in the two CVVH devices commonly used in the UK, the Prismaflex and the Aquarius. 15 These volume changes have the potential to cause dehydration or fluid overload in small babies, but such machines are used extensively worldwide outside practice recommendations due to a lack of more suitable devices.

Other problems with using CVVH in babies also result from their small size compared with the blood access, flow and volume requirements of conventional haemotherapy circuits. Specifically, existing CVVH devices require double-lumen central venous access lines with recommended minimum 7-French, size, and continuous 40 ml/minute blood flows, both of which may be difficult to achieve in the smallest babies. Also, they have extracorporeal circuit volumes of 50–70 ml, which exceeds the safe limit of removal of 10–15% of blood volume in babies under 5 kg. This necessitates priming their circuits with blood products, or by using saline and precipitating a sudden haemodilution as the therapy is started. Exposure to foreign blood carries a risk of causing tissue sensitisation, with potential consequences for children who may later require organ transplantation. The sudden exposure of a baby to a relatively large transfusion also has the inherent risk or causing abrupt pH and other aberrant chemical changes, which may be reduced by pre-dialysing the circuit. 16

Because of these device limitations, common clinical consequences of these limitations of CVVH devices include cardiovascular instability with hypotensive episodes on connecting and commencing therapy and at any time due to variations in UF control, difficulties providing vascular access and sufficient blood flows, which may result in the clotting of circuits and multiple blood transfusions. 3–6,17,18

NIDUS technology

The NIDUS began development in 1995 specifically to provide RRT to very small infants, and has a novel circuit that operates using different physical principles from conventional systems. 19 It uses syringes rather than peristaltic pumps to drive the blood flow, which provide precise volumetric control of UF, and uncouple the baby’s blood flow capacity from the requirements of the dialysis filter, allowing it to sample more slowly. Its minimum circuit volume of <10 ml does not require blood priming and only requires a relatively small single-lumen central venous access line. By 2005 an early automated version had been used to treat four babies of 0.8 to 3.4 kg, and this was subsequently re-engineered to produce the I-KID study intervention device, the NIDUS. 20 During development it was tested on piglets of 1 to 8 kg, and was used to dialyse 10 babies of 1.8 to 5.9 kg in a PICU and a paediatric nephrology setting, including 354 dialysis sessions totalling 2475 hours, where it was found to be safe. 21 Its UF precision and biochemical clearances were consistently superior to PD in both the animal and clinical studies.

Other novel infant RRT devices

There are two other haemotherapy devices being developed to provide RRT in babies as small as 2.5 kg, but neither was available for us to include in the I-KID study. A group in the USA have used the Aquadex^® (Nuwellis, www.nuwellis.com) adult haemofiltration device in parallel with intravenous pump controllers to regulate the flow of replacement fluid and to generate UF. 22 An Italian group has produced and CE marked an infant CVVH device, the CARPEDIEM^®, using a miniaturised conventional circuit. 23,24 Although this has been used in some European centres, and has recently gained FDA approval, it was not available in the UK to enable comparisons when the I-KID study started.

Main study objectives

To compare the novel non-CE-marked infant HD machine, the NIDUS, to conventional standard RRT in children under 8 kg in PICU. The study aimed to evaluate the clinical efficacy of the NIDUS in improving accuracy of UF fluid removal and to monitor safety and patient outcomes using a cluster-randomised stepped wedge (SW) study design. The study also compared NIDUS separately with each of CVVH and PD.

In addition, the study was designed to look at the incidence and severity of the adverse effects of renal replacement, and to generate a safety profile in the application of NIDUS in the clinical environment.

Inclusion criteria

• Participants with a body weight of 0.8 kg–7.99 kg who require continuous RRT for acute renal insufficiency or fluid overload as part of their standard clinical care.

• Person with legal parental responsibility (PR) for the patient able to provide written informed consent for the patient to take part in the study.

Exclusion criteria

• Patient with known chronic renal failure already on established adequate RRT.

• Patient already established on adequate RRT for whom entry into the study would require additional central venous access, if that access is not clinically indicated.

• Patient with an underlying metabolic diagnosis, including hyperammonaemia.

• A clinical decision is made that the patient should not receive RRT using NIDUS.

• Unable to receive written informed consent for data collection from a person with legal PR for the patient.

Two main changes were made to the eligibility criteria during the course of the study as shown in Appendix 1, Table 33. These were the introduction of deferred consent for the study, reflecting common practice in emergency situations in PICU so as not to delay treatment, and the acceptance of use of estimated body weight.

Control therapy (usual treatment)

Participants were treated with current RRT options available at the participating site, either PD or CVVH, when in the control phase of the design. Staff in PICU were already trained in the clinical use of these RRT methods. Additional training was given regarding the procedures needed to obtain measurements of UF and biochemical clearance.

All sites used PD. Each site was also able to perform CVVH either using the Gambro Prismaflex, or the Baxter Aquarius. In the absence of suitable and safe alternatives, these machines are used off licence during standard care. In one site, RRT was also provided for infants on ECMO by connecting a dialysis filter between the arterial and venous ends of the ECMO circuit and controlling the rate of dialysate flow using pumps designed to regulate intravenous infusion lines (ECMO + HD). The NIDUS machine was not available for use during the control period.

Control therapy was used in the control period according to usual clinical practice until changeover to NIDUS according to the SW design. Eligible participants who declined consent to the I-KID study received standard therapy.

Tests of RRT efficacy

Children recruited to this study had two types of tests of the efficacy of their RRT, measurement of the (UF precision) and measurement of the rate of clearance of chemicals from their blood. The details of these are given below, and separate detailed information and bedside data recording sheets for PD, the Prismaflex, the Aquarius, for ECMO + HD, and the NIDUS are available in the project documents.

Measurement of UF precision

The quantity of fluid removed by RRT was measured over an accurately timed period to calculate the actual UF rate that the therapy had achieved, and for each modality this was compared with the UF rate that the clinical team had documented they required. For the CVVH and NIDUS therapies, the measured UF was also compared with the volume of UF that the device displayed it had achieved.

For PD, the total volumes of dialysate fluid infused and drained were measured volumetrically using the calibrated burettes that are integral to the clinical circuits, as is undertaken for every complete dialysis cycle in normal clinical practice. To minimise errors that may occur due to variations in the completeness of emptying of the peritoneal space during some PD cycles, the collection periods were between 5 and 7 hours long, and were timed to ensure they only included completed cycles. The achieved hourly UF rates were calculated by subtracting the mean volume infused from the mean volume drained.

For CVVH and the NIDUS, the UF rates which were set by the operators and recorded by the devices were volumetric (ml/hour), and these were compared with the actual UF rates that were measured gravimetrically by assuming that all of the fluids had a density of 1 g/ml. This method was employed because both types of infant haemotherapy use closed fluid circuits, which means that the total combined weight of the fresh and waste fluid bags remains constant in the absence of any UF, and it can be assumed that any increase in their net weight will represent fluid added by removal from the baby’s circulation. In the case of CVVH machines, the saline infusion used to deliver heparin into the circuit also enters the device’s closed circuit and adds to the increase in net weight of the bags, so it was deducted from the measured weight gain, but this was not so for the NIDUS because in that circuit it is removed without entering the closed fluid system.

Thus, the weights of all of the dialysate and/or replacement fluid bags, plus the waste-fluid drainage bags were recorded at the start and end of each study period by suspending them over a stable weighing balance capable of weighing up to 16 kg, and sensitive to changes of 0.1 g, and by recording the volume of heparin solution infused by the CVVH devices. The calibration of the weigh-scales was checked to be accurate to within ±1 g using a 5 kg weight, and ±0.1 g using a 10 g weight before each study, and the fluid bags were suspended carefully in precisely the same manner each time, avoiding any stretching or contact of their connection lines with the device. For both types of device, the minimum collection study period was set at one hour. If the clinical requirement for UF changed during any of the study periods, the timing of the altered settings was recorded, and this was accounted for in the rate calculations.

For ECMO + HD, the change in the weight of the fresh dialysate fluid bag was recorded, and the volume of waste dialysate fluid was measured in a calibrated drainage container. The rates of the dialysate inflow and outflow infusion pumps were recorded, and the clinically set UF rate was taken as the difference between these rates.

Measurement of biochemical clearance

We calculated the clearance of creatinine, urea and phosphate by measuring the rate of accumulation of each chemical in the dialysate and/or replacement fluid and comparing that to its plasma concentration. This requires the measurement of the plasma and waste fluid concentrations of these three metabolites, and a knowledge of the waste fluid flow rate. The urea and phosphate were measured using standard methods in the clinical laboratories of the participating centres, and creatinine was measured by an enzymatic assay to avoid measuring non-creatinine chromogens in the plasma by the previously widely used alkaline picrate reaction. Clearances were expressed as ml of plasma totally cleared of that chemical per minute of therapy.

For PD, the biochemical clearance test was performed at the same time as the UF test, that is over a timed period of five to seven hours. The whole volume of waste fluid collected during that study was mixed, and a sample of this was used for the chemical assays.

For the CVVH and NIDUS the biochemical clearances were measured either just before or just after doing the UF test because each study would have interfered with the precision of the other one. When the standard extracorporeal circuit was mounted on the device, an extra extension tube and three-way tap were inserted into the waste fluid drain line close to the main drainage bag, with a small collection bag attached to the side connection. During the collection period of at least 20 minutes, the tap was turned to allow waste fluid to be collected as it was produced, and a sample of this was assayed for the metabolites. The rate of fluid drainage was calculated as the total of dialysis and/or replacement fluid flow rates set on the machine plus the volume of ultrafiltrate set.

For EMCO + HD, as for PD, the whole volume of waste fluid collected during the study period was mixed, and an aliquot was chemically assayed.

In all clearance studies, the test was performed within one hour of a blood sample being taken for creatinine, urea and phosphate measurement for clinical reasons. No extra blood samples were taken for the tests; instead, the timing of the biochemical clearance tests was adjusted to coincide with routine blood sampling, which is typically twice daily in babies on RRT in PICUs.

Details of the calculations used for each modality for both UF and clearances are available in Report Supplementary Material 1.

Intervention therapy (NIDUS)

The NIDUS was only available for use by trained staff during the intervention period for that site.

The main instrument is used in conjunction with a device-specific blood tubing set and a NeoFlux1^® (Allmed, www.allmedgroup.com) HD filter, and withdraws small volumes of blood (5–10 ml) from the patient, passes it through the filter and then returns it to the patient, via a single lumen vascular access catheter. Blood movement is controlled using driven syringes and pinch valves. Its extracorporeal circuit volume with the syringes empty is 4.9 ml, which is small enough to prime safely with saline, without the need for blood products. UF is controlled by the differential movement of the two operating syringes, with the difference in volume between the two syringes is removed as UF. Dialysate is pumped around the outside of the filter via a peristaltic pump, allowing dialysis to occur by diffusion. The NIDUS makes a constant recording of all activity data, including volumes, flows, pressures, alarms and response to alarms, downloadable for safety purposes.

Allmed are the manufacturers of NIDUS device, circuit and filters. Devices were loaned, three to each site (except Newcastle that has two), in case of breakdown or multiple recruits.

Sites were supported in their use of the investigational device, both clinically and technically. A 24-hour helpline was available to contact on-call renal nurses from Newcastle upon Tyne Hospitals (NuTH), experienced in the use of NIDUS, in order to provide immediate support to study nurses at the bedside. Clinical telephone or videolink support was also available from Dr Heather Lambert (Chief Investigator) and Dr Malcolm Coulthard at all times as backup to the nursing advice. Allmed and NuTH Medical Physics provided continued support in response to immediate technical queries.

Support was both reactive and proactive. A monthly teleconference with representatives from each site and the I-KID Trial Management Group (TMG) enabled users from different sites, with different experiences, a platform to feedback to one another and Trial Steering Committee (TSC) members. Regular meetings between the clinical and technical teams were essential to manage and review the ongoing support offering and to respond to user requests regarding interventional aspects of the study.

During the intervention period, standard therapy continued to be used for those participants who did not meet the criteria for the NIDUS machine (see Inclusion criteria and Exclusion criteria in the Introduction).

Clinicians caring for participants patients under 8 kg who started on conventional dialysis methods, which had failed and where the patients did not meet the inclusion criteria for the study, had the option to use/switch to the NIDUS machine for compassionate use. These cases were initially discussed with the Chief Investigator. Local trust process and the process set out by the Medicines and Healthcare products Regulatory Agency (MHRA) processes for compassionate use were followed by sites when appropriate. No data from these patients were used for the study.

Study induction and control phase

During the control phase, sites followed routine clinical practice with the addition of conducting a number of research activities; and so, training involved the detailing and demonstration of these activities, namely, bag-weighing and fluid sampling methods. Initiation visits were conducted at each site in the months prior to the start of the study. The site PI and senior members of the clinical and/or research team received the training in person. These trained individuals then cascade-trained other team members at their own site. In addition to the introduction of research activities, information was disseminated with regard to the study rationale and protocol, the principles of the NIDUS (how it differs from conventional RRT) and the content and location of specific documentation required when carrying out research activities. This included an overview of safety reporting procedures and document version control.

Intervention phase

As per study design, intervention phase training followed a stepped approach, where sites were trained approximately two months before they were due to crossover to use the NIDUS. Each site had a minimum of four training days.

These sessions aimed to ensure that key members were competent to use the device and its components. This involved detailing and demonstrating how to set up and run the NIDUS, and how to troubleshoot potential issues that may occur in practice. The device’s operating principles, and key differences compared with conventional RRT devices were discussed. Largely, the emphasis of this face-to-face training was to allow hands-on time with the device. As with the control phase, members were also shown, and then asked to demonstrate, how to accurately perform device-related research activities, i.e. bag-weighing and fluid sampling methods.

Trainees were considered, and signed off as, competent to use the device if they showed that they could correctly follow guidance documentation to set up the device, perform procedures required when running the device, and if they showed a comprehensive understanding of the NIDUS and how they might troubleshoot issues via answering a set of trainer-led questions and discussion. A number of key individuals including senior nurses and the PI at each site were signed off as competent to both use the device and cascade train others. Sites were responsible for ensuring their skills were maintained during the study and this was supported by the I-KID training team, for example by conducting additional training sessions.

Parents/guardians were asked about their experiences and staff were asked about acceptability and usability of the device, using questionnaires.

Updates to guidance and training documents

User documentation, including documents used for training, device-use guidance documents and research activity guidance and recording documents (to be used at the bedside), were continuously reviewed and amended throughout the trial. On-going user feedback from sites largely facilitated this process. Aspects of accessibility and clarity of content were reviewed. Documentation, such as the training packages, began as paper copies that were also available online via study specific shared drives (Microsoft Teams and Google Drive); then, were adapted to become direct online material (accessible via a QR code or web link), in an app-like format, created using the Google Forms forum. This ‘app’, in its final form, provided a step-by-step, image-guided walkthrough of device set up and running procedures; it was used as a training tool and for guidance during clinical use of the NIDUS. Additional training was provided where appropriate, including prior to resumption of the study post the interruption caused by the COVID-19 pandemic, and included virtual video-linked training.

Primary outcome

The first observation of precision of fluid removal (UF) from an episode lasting at least an hour within 48 hours of the start of RRT.

Secondary outcomes: related to the primary outcome

• average of all precision values observed on the patient

• biochemical clearance rates for creatinine, urea and phosphate

• precision of observed versus reported fluid removal.

Secondary outcomes: mortality data

• death within 30 days of the start of RRT (collected by the I-KID team)

• death before discharge from PICU [collected via PICANet (Paediatric Intensive Care Audit Network) – see below].

Secondary outcomes: collected through PICANet

Paediatric Intensive Care Audit Network is an audit database recording details of the treatment of all critically ill children in PICU. Data are routinely collected on admission to PICU and thereafter daily returns are made, with detailed renal data. This custom renal data set uses precise definitions and terms for the information collected. PICANet publishes an annual report and makes regular download of data to the site of origin for audit and quality improvement purposes. Some of the descriptive and secondary outcome data in I-KID was collected via PICANet, and with consent from parents/guardians as part of the I-KID study-specific data were shared with the I-KID study by the sites. The reason for using this process for some data collection was to try to ensure as near as possible complete data collection as these data were being collected routinely by sites who were used to doing this in a regular way. The I-KID study also aimed to not cause additional workload by duplicating similar data sets.

• haemodynamic status (drop in blood pressure after connection to a RRT device, requiring intervention of fluid bolus or administration of inotropes)

• completion of intended RRT course

• need for additional vascular or dialysis access

• unplanned change in circuits

• exposure to blood transfusion

• bleeding events

• anticoagulant use

• number of ventilator-free days (calculated as number of days on RRT minus number of days on RRT and on ventilator).

Secondary outcomes: questionnaire results

• parent/guardian experience measured using questionnaires

• staff acceptability and usability of device measured using questionnaires.

Changes to primary outcome variable

The protocol initially specified that the primary outcome variable should be measured during data phase 1 (0–7 hours), provided the data collection episode exceeded one hour. However, because of the time constraints for clinical teams managing critically ill babies and the complexity of the study testing, the protocol was formally updated so that; if the site team were unable to collect the primary outcome during the first phase of data collection, then the next available collection period was used to measure the primary outcome, provided this was within the first 48 hours of RRT.

Primary analysis

The primary analysis of the primary outcome used a linear model with a categorical period effect, to allow for time effect during the study, and a binary treatment effect comparing NIDUS and controls. Differences between sites were accommodated by a fixed categorical effect. The log of the duration over which X and A were observed was included as a covariate. The fit of the model was assessed using standard diagnostic methods. The treatment effect estimated log σ_control/σ_NIDUS and results are presented in terms of the estimated ratio and 95% confidence interval (CI). It should be noted that the outcome measure is based on the assumption that X-A has zero mean: if this assumption was not borne out by the data, alternative methods would be used.

Secondary analyses

Subgroup analyses (all pre-specified)

Trial management

The TMG was responsible for overseeing management of the study. The TMG met approximately every four to eight weeks during the course of the study. TMG meetings involved the Chief Investigator, trial statisticians, local co-applicants, members from the Northern Medical Physics and Clinical Engineering Directorate for technical support and development of the NIDUS device, a sponsor representative and trial management team members from Newcastle Clinical Trials Unit (NCTU).

Data management

A study-specific MACRO database was designed and built by the database manager with input from the TMG. Data for participants were entered into the electronic case report forms by local site staff.

Sites were asked to send all serious adverse device event (SADE) reporting forms, serious adverse event (SAE) logs and device deficiency (DD) logs were sent from site to the central study team in Newcastle using secure email.

The occurrence of events such as blood transfusions and access line changes were recorded by local site staff via the PICANet enhanced renal audit reporting system. Staff at site downloaded the PICANet data for their participants who had been consented to the study. The downloaded dataset was sent to the Database Manager at NCTU by secure email.

Data were handled, digitalised and stored in accordance with the Data Protection Act 2018.

Study oversight

Study oversight was provided by the study sponsor (the NuTH NHS Foundation Trust), the TSC and the Independent Data Monitoring Committee. A Clinical Safety Sub-Group consisting of clinicians from the TMG, representatives from Medical Physics and a sponsor representative reviewed all SADE and SAE safety reports. Device deficiencies that led to a SADE or had the potential to become a SADE underwent expedited reporting by Sponsor to the MHRA, in line with the requirements for a clinical investigation (see Appendix 7, Figure 28). SAEs that were not consistent with the usual clinical pattern for participants requiring RRT in PICU were recorded by site on the study SAE log. SAEs which were excluded from reporting in the protocol were recorded in the study database. All device deficiencies in the intervention arm were recorded by site on a DD log.

Protocol deviations and violations

In total there were 23 deviations and violations. Full details are in Appendix 2, Table 34.

Losses to follow up, withdrawals and death

There were no withdrawals and no losses to follow up. Twenty-two participants died by the 30-day follow-up or by the time they were discharged from PICU, whichever came earlier (10 control, 12 NIDUS). Full details are in Appendix 2, Table 35 and see also Key findings in Chapter 4.

Availability of primary outcome

The actual fluid removal rate, X, and the prescribed removal rate, A, were observed during several episodes throughout the period that a patient was on RRT. The primary outcome variable was defined as log|X-A| from the first episode at which X and A were available, provided that the episode lasted at least an hour and started within 48 hours of the inception of RRT.

For the 62 participants who received the control treatment, the primary outcome was available for 61 participants. For one patient, who was receiving PD, the first episode to provide X-A started 67 hours after the inception of RRT. However, as the analysis of the primary outcome was by ITT, this observation was included in the analysis of the primary outcome.

For the 35 participants who received NIDUS, a primary outcome was available for 21 participants. One of these 21 participants was a transition baby and was excluded from the primary analysis. For 14 participants receiving NIDUS no value of X-A was available. The reasons for the 14 missing values are summarised in Table 4, with more detail provided in Appendix 2, Table 37. In most cases the reason why no value of the primary outcome was available was that technical difficulties were experienced in establishing or sustaining RRT using NIDUS. In some of these 14 cases, the patient spontaneously started to pass urine so RRT was no longer needed; others were so sick that they died very quickly. In five cases it was recorded that another method of RRT was attempted but in none of these cases was a value of X-A available, which probably reflects the clinical circumstances at the time. Three of the 14 participants without a primary outcome were transition babies, so would not have been included in the primary analysis even if they had had outcomes recorded.

Description of primary outcome

Summaries of the baseline characteristics of the 20 participants receiving NIDUS who were included in the primary analysis, are given in Appendix 3, Tables 40 and 41. Also, in these tables are the summaries of the 11 participants allocated to NIDUS that would have been included in the primary analysis had values for X-A been available. Summaries of the same variables comparing the 62 control participants and the 20 Intervention participants included in the primary analysis are in Appendix 3, Tables 42 and 43.

Descriptive statistics for the data used for the primary analysis are in Table 5. The use of log|X-A| for the primary outcome was based on the assumption that the expectation of X-A is zero. The means and SDs for X-A shown in Table 5 lend support to this assumption. A histogram and boxplot of X-A are shown in Figures 5 and 6 to illustrate that this variable appears to be symmetrically distributed about zero: it should also be noted that the spread is considerably less in the intervention than the control group. Further descriptive statistics of X-A by period and sequence are presented in Report Supplementary Material 1 Table SPO1.

FIGURE 7.

Boxplot and histogram of first computable precision (X-A) by arm. Top: Boxplot of precision by arm for n = 83 participants with a computable precision measurement. Mid-blue points denote the participants in the control arm and the Orange points denote participants in the intervention arm. Point in dark blue indicates the NIDUS participant recruited during a transition period (precision = −13.12). Bottom: Histogram of precision by arm. Colour scheme is in accordance with the boxplot.

A linear model with categorical covariates for centre, period and treatment arm was fitted to X-A to investigate further. Only the effect for period 2 was significant out of the 11 fitted parameters and this seemed to be due to a marked outlier in period 2 and in the control group (see Appendix 4, Figure 14). There was no compelling evidence against the assumption that X-A has zero mean, so the primary analysis was based on log|X-A|.

Average value of log|X-A|

The average value of all the available log|X-A| collected within 48 hours of inception of RRT was computed (see Table 7). This analysis excluded the patient on PD whose only value of log|X-A| was collected more than 48 hours from the start of RRT. There were up to four observations on each patient: in the control group the number of participants providing 1, 2, 3 or 4 observations was 6, 18, 13, 24, respectively, whereas in the NIDUS group the corresponding numbers were 9, 3, 5, 4.

The same linear model used for the primary outcome was fitted. The estimated treatment difference (NIDUS – Control) (95% CI) was −2.05 (−3.22 to −0.88), with P = 0.001. This corresponds to a ratio of SDs NIDUS: control of 0.13 (0.04 to 0.41), which is in line with the primary analysis but with a narrower CI. Full details of the fitting of the model are in Appendix 5.

Rates of biochemical clearance

The rates of clearance of creatinine, urea and phosphate (in ml/min) were computed from all episodes of RRT where the necessary data were available. Data from transition babies were excluded (three for each metabolite). Two analyses were performed for each metabolite, one on the first observed clearance and the second on the mean of all clearances available on a patient.

The SAP prescribed comparing two groups, namely NIDUS and non-NIDUS groups. The results of this analysis are shown in Tables 8 and 9 where we see that for each metabolite there is a difference between NIDUS and control for the residual SD but no evidence of any difference in the mean clearances.

However, this analysis is misleading because, for each metabolite, the mean clearance rate was generally larger for CVVH than for NIDUS and for NIDUS was larger than PD (see Figures 8–10). To combine these two groups, one with generally larger values and one with smaller values than for NIDUS would seriously misrepresent the data. The analysis comparing three interventions, CVVH, PD and NIDUS (with the single Manual HD and ECMO excluded) was specified as a subgroup analysis in the SAP, and this is presented in Tables 10 and 11. It is also clear from Figures 8–10 that the residual SD differs between PD, NIDUS and CVVH, so generalised least squares was used to fit a model with categorical covariates for centre, period, treatment at three levels and separate residual variances in each treatment group.

FIGURE 8.

Boxplot of first computable biochemical clearance by mode of RRT – creatinine.

FIGURE 9.

Boxplot of first computable biochemical clearance by mode of RRT – urea.

FIGURE 10.

Boxplot of first computable biochemical clearance by mode of RRT – phosphate.

Descriptive statistics of the first computable clearances are given in Table 10, with the results from fitting the model given in Table 11.

Further details of the fitting of the model and its assessment are in Appendix 5 and Appendix 6 (Tables 46 and 47).

Table 11 shows that the biochemical clearance rate for PD was smaller than that for NIDUS, which in turn was smaller than that for CVVH for each metabolite. The SD of the clearance rate was also smaller for PD than NIDUS, and smaller for NIDUS than CVVH, again for each metabolite.

The results for the mean biochemical clearances were very similar to those shown here for the first clearance (details not shown).

Analysis of mortality and of variables collected through PICANet

Descriptive statistics for mortality, using measures collected through PICANet and by the NIDUS team, are shown in Table 12. Other variables collected through the routine returns made to PICANet are presented in Table 13.

The mortality outcomes are binary as are the other PICANet variables apart from ventilator-free days. The intervention and control groups were compared for binary outcomes using standard χ² tests, with differences described by both absolute differences and odds ratios, each with associated 95% CIs.

Comparisons were based on the full data, including transition babies, as these variables should not be affected by unfamiliarity with a new device and most are routinely collected through PICANet.

Pre-specified subgroup analyses

The SAP specified the following subgroup analyses:

1. Repeat the primary analysis but with the control group split into CVVH and PD, so that NIDUS is compared with CVVH and PD separately.

2. Descriptive statistics of mortality and of the variables collected through PICANet presented separately for those allocated to PD, CVVH and NIDUS.

3. Comparison of the actual filtration rate (X) and the filtration rate reported by the dialysis device (A₂). This quantity is only relevant to participants allocated to NIDUS or CVVH.

The patient receiving Manual HD with ECMO was omitted from these analyses.

Primary outcome compared between NIDUS, CVVH and PD

Descriptive statistics of the primary outcome between NIDUS, PD and CVVH are presented in Table 14, with the results of analysis in Table 15. The analysis used the same model used for the primary analysis, but with the treatment factor now having three levels: full details of the fitted model are in Appendix 6. The results were similar to the results when CVVH and PD are combined, with the SD of log|X-A| on NIDUS being 14% of that for PD and 11% of that on CVVH.

Descriptive statistics of mortality by NIDUS, CVVH and PD

Mortality statistics by PD, CVVH and NIDUS are shown in Table 16.

Descriptive statistics of variables collected through PICANet, by PD, CVVH and NIDUS

Descriptive statistics of these variables are in Table 17.

Difference between actual fluid removal and value reported by the procedure (X-A₂)

For NIDUS and CVVH, an important measure was to compare how close the actual fluid removal (X) was to that reported by the device (A₂). As this was not relevant for PD nor for HD and ECMO, the analysis used data from the 48 NIDUS and CVVH participants in the study. At least one measurement of X-A₂ was available for 34 participants (13/13 CVVH and 21/35 NIDUS). Descriptive statistics of X-A₂ are presented in Table 18.

The analysis anticipated in the SAP assumed that X-A₂ would have zero mean but stipulated that this should be confirmed before proceeding to the proposed analysis. A general linear model fitted to the first computable X-A₂ on treatment arm, site, period and duration was used to assess this assumption (see Appendix 6, Table 48). This suggested that X-A₂ does not have zero mean and also that the residual SDs varied between arms. Consequently, the analysis fitted the usual model of centre, period, treatment and duration to the difference in X-A₂ using generalised least squares, with different residual variances in the two treatment groups. Results are in Table 19. Further details on the fitting and assessment of the model are presented in Appendix 6, Table 49.

Table 19 shows that X-A₂ is substantially smaller on NIDUS than on CVVH, with a SD on NIDUS of about 17% that on CVVH and the adjusted mean of X-A₂ is 28.2 ml/hour (95% CI 6.8 to 49.5) smaller on NIDUS than CVVH. However, some caution should be exercised in interpreting this analysis, which considers only 33 participants. The adjusted mean difference is substantially larger than the difference in unadjusted means (mean 11.6 ml/hour on CVVH vs. −0.4 on NIDUS), so the conclusions could be highly dependent on the effect of a small number of influential observations on the fitted model.

The adjusted difference in means is very different to the difference in unadjusted means. The results should be treated with caution as this analysis is based on small samples and the adjusted estimates could be affected by a few points that have a marked effect on the fitted model.

Composition of control group

Of the 62 babies in the control group 48 received PD, 13 CVVH and one baby on ECMO had HD via a filter in the blood circuit. From baseline PICANet data from 2011 to 2013 when this study was first proposed, we had expected in I-KID that the modality of RRT in the control group to be more evenly split between PD and CVVH. In the study development we had sought data on dialysis modality by weight but as PICANet collected age and not weight data at that time, we therefore made some assumptions using weight for age centiles to estimate numbers of potential recruits.

There are no clinical guidelines for choice of dialysis modality in babies under 8 kg and the modality used is influenced by individual clinicians own experience. Moreover, it is well recognised that simply performing a study in a clinical environment can subtly change clinical practice and fine details of routine data collection. 36 Although it was not intended to influence current practice for the control group it is possible that starting I-KID actually influenced choice of RRT modality. At study launch, site initiation and study training visits the issues of reported problems and the current regulatory status of CVVH devices being used off-licence and against manufacturers advice was highlighted. In addition, on reflection, the study tasks for collection of data for the primary outcome and related secondary outcomes (filtration and clearances) were probably easier to perform for the babies on PD than babies on CVVH or NIDUS because of wider nursing familiarity with PD and how these tasks fitted with normal clinical practice. For the volumetric study for babies on PD the main bedside study tasks were accurate timing and collecting and sending of a fluid sample, whereas for the CVVH devices and intervention NIDUS device there were additional bag-weighing tasks and the collection of fluid for clearance calculation was more complex for bedside staff.

Age and weight of participants

There was a wide age range of participants: the median [interquartile range (IQR)] age in controls 10.5 (7–38) days was similar to that in the intervention group 11 (7–61) days; the range of age of participants was between 1 and 477 days (c. 16 months). It is recognised from previous PICANet reports that use of RRT is skewed towards the younger end of the age range of babies expected to fall in the under 8 kg weight range. 7,8 The 50th centile for weight of a 9-month-old baby is around 8 kg, whereas, at the upper end of the age range of 15 months, 8 kg is on the 2nd centile for weight. However, it should be noted that weight and age do not closely correlate in sick babies. 37

Admission to PICU and use of RRT commonly occur for babies following surgery for congenital abnormalities especially cardiac, which tends to be required in the early postnatal period. This is also reflected in the median (IQR) weights 3.2 (2.9–3.9) kg and 3.7 (3.1–5.6) kg which were similar between control and intervention and not dissimilar to the average birth weight of a term baby (37–40 weeks gestation) in the UK which is around 2.5–4 kg.

One participant was 10 kg with estimated dry weight of 7.7 kg; estimated dry weight was permitted in the protocol, as some babies can become severely oedematous by several litres of fluid. The median gestational age was 38 weeks in both the control and intervention groups with a range of 26–41 weeks which is not unexpected.

Descriptive data were similar in both control and intervention groups; around half the participants had unplanned admissions to PICU and approximately a third were transferred from outside hospitals. RRT was required post surgery in 52% of control and 40% of intervention participants. For those requiring RRT post surgery this involved cardiac bypass surgery in 97% of controls and 84% of intervention participants. Systolic blood pressure median (IQR) was 68 (59–78) mmHg for control, and 68 (60–86) mmHg for intervention, and the need for mechanical ventilation (>80%) was similar in the two groups.

PIM3 score

The Paediatric Index of Mortality 3 (PIM3) is a score to quantify a patient’s probability of death was developed as mortality risk prediction model (range 0–1) for all admissions to PICUs, a heterogeneous patient population of which those receiving RRT from a very small subset. It is based upon information from the first hour of admission. The median (IQR) PIM score was similar in control 0.023 (0.013–0.065) and in intervention 0.027 (0.014–0.131) groups, although the corresponding values for PD of 0.02 (0.01–0.05) and CVVH 0.06 (0.01–0.20) suggest a lack of homogeneity in the control group.

Laboratory data pre-initiation of RRT

Laboratory data collected pre-initiation of RRT were similar for control and intervention groups for plasma sodium, median (IQR) control 146 (141–149) mmol/l, intervention 143 (136–146) mmol/l, and for plasma potassium control 4.45 (4.0–5.2) mmol/l, intervention 4.8 (4.2–5.5) mmol/l. Plasma creatinine and urea appeared slightly higher in the intervention group than in the control group: creatinine 74 (56–110) µmol/l versus 51.5 (40–68) µmol/l and urea 6.15 (3.5–10.6) µmol/l versus intervention 9.5 (4.7–17.2) µmol/l, respectively. Within the control group plasma creatinine was higher in the CVVH group 94 (42–218) µmol/l than the PD group 51 (39.5–62) µmol/l. This raises the possibility that the severity of renal failure was different in these subgroups or that the threshold for starting this modality of RRT may have been different.

The median platelet count was lower in the intervention, 124 (95–209), than control 207 (113–257) groups.

Indication for starting RRT

The clinical decision that the participant required RRT was made by the attending clinical team irrespective of the I-KID study. Decisions about clearances and UF aims and prescriptions were made based on individual patient clinical parameters by the clinical team. For the NIDUS device, advice was given in training sessions and as written information about the need for anticoagulation with heparin, and how to monitor this. The need for very frequent (hourly) monitoring of activated clotting time (ACT) assays was reinforced after we received early reports of filter clotting.

The primary indication for starting RRT in the control and intervention groups, was similar in that in about half of the cases it was for fluid volume control alone, the rest having RRT started for biochemical control or both.

Ultrafiltration (fluid removal)

The results show that the UF obtained with NIDUS was closer to that prescribed than with control. If the precision denotes the UF obtained minus the UF prescribed, X-A, then when calculated for the primary outcome, that is the first observation on X-A from each patient, Figure 7 shows this is much less dispersed about 0 with NIDUS than with control. Results in Tables 5 and 6 indicate that SD of X-A using NIDUS is about 13% of that using control 95% CI (3 to 71). If the information from all determinations of X-A in the first 48 hours post RRT is considered, then the SD using NIDUS is still estimated to be 13% of that using control, but with a narrower 95% CI (4 to 41).

When the results for the primary outcome in participants receiving CVVH and PD are each compared with NIDUS the picture is very similar. The SD for NIDUS is 14% of that for PD, 95% CI (3 to 70) and 11% of that for CVVH 95% CI (2 to 64).

A concern is that while UF values were obtained for all participants allocated to control, values were obtained from only 21 of the 35 participants receiving NIDUS. Of the 14 participants with no primary outcome, three were transition babies in the transition period so would have been excluded from the primary analysis. The reasons for this are given in Table 4 with more details in Appendix 2, Table 37 and are almost all related to technical issues with NIDUS, in particular its filter. Nevertheless, the failure to obtain the primary outcome in such a high proportion of one treatment group in a conventional therapeutic study would raise serious concerns about bias. While these missing data are a definite weakness, it should be borne in mind that in this study the comparison is between properties of methods of RRT – we are concerned with how closely the method of RRT delivers the prescribed UF – and this is likely to be less affected by patient characteristics.

We note that prescribed UF rates were substantially lower for NIDUS median 12 (8.6–21) range 0–40 ml/hour, and for PD median 10 (4–13.5) range 0–60 ml/hour than for CVVH median 30 (20–39) range 6–58 ml/hour. The difference in settings may be related to the understanding that CVVH achieves biochemical clearance by convection (and is increased by increasing UF and replacement/dilution fluid). Whereas PD and NIDUS primarily achieve biochemical clearance by diffusion.

Precision of RRT device

For NIDUS and CVVH devices, an important measure is to compare the difference between the actual fluid removal and that reported by the device: The results in Table 18 show this to have a mean closer to zero for NIDUS than CVVH (means −0.44 vs. 11.6 ml/hour, respectively), with less variation in NIDUS than CVVH (SDs 3.2 vs. 28.4 ml/hour). The formal analysis reported in Table 19 shows a similar ratio of SDs (17%, 95% CI 9% to 31%), but with a markedly different adjusted mean difference. This, and the difference in SDs, may be heavily influenced by large discrepancies on CVVH in period 2.

It is clinically very important to be able to rely on the information given by the device is accurate; then the clinician is able to make adjustments to the participants overall fluid input to counter discrepancies. Conversely if the device gives inaccurate information to the clinical team it contributes to uncertainty and difficulty in overall fluid management. Manufacturers are aware of the inherent imprecision of their devices and give warning in their technical documentation (e.g. Prismaflex ±30 ml/h, ±300 ml per 24 hours). In vitro studies of the fluid removal precision of RRT devices when set to ‘treat’ bags of saline using infant settings showed that the variances between the displayed and actual UF volumes for Prismaflex and Aquarius were wide at 14.0 and 30.3 ml by 15 minutes, respectively, while that for NIDUS was close to zero at 0.17 ml by 15 minutes. 15 These variances were seen whether the devices were set at relatively high UF rates (40 ml/hour) or to maintain neutral fluid balance.

Clearances

Secondary outcomes (mortality)

Of the 62 participants receiving control 54 survived to 30 days (87%) and 52 (84%) survived until discharge. For the 35 participants in the NIDUS group, 25 survived to 30 days (71%) and 23 (66%) survived to discharge. These comparisons gave p = 0.057 and 0.040, respectively (see Table 12).

When the groups receiving PD and CVVH are compared with NIDUS there is a suggestion that survival was highest in PD: out of 48 participants 47 survived to 30 days (98%) and 46 (96%) to discharge, whereas for the 13 participants on CVVH the corresponding values were 7 (54%) and 6 (46%) (see Table 16).

No formal statistical modelling of the effect of variables on death rates in the groups has been carried out. However, it is noticeable that the means of a variables measuring the risk of death (PIM3) and extent of renal disease (baseline creatinine) are in the same order as the death rates in the treatment groups: in the order PD, NIDUS, CVVH the means of PIM3 are 0.02, 0.027, 0.06 and for creatinine before RRT 52, 106 and 172 mmol/l.

While I-KID has provided evidence that NIDUS delivers more precise UF, there is no indication that this has translated into lower mortality. I-KID was not designed to detect differences in mortality. While it is to be hoped that improved methods for RRT will lead to better patient outcomes, mortality will be principally dependent on the diagnosis, so it is likely that the effect of the performance of the method of RRT on mortality will only be detected in larger studies.

Causes of death are presented in Table 31. Most deaths were assessed to be unrelated to the RRT device with the exception of two which were possibly related to the ECMO + HD and NIDUS devices, respectively.

Secondary outcomes (collected through PICANet)

Study design

The study design chosen was a result of various factors. We knew from PICANet data that most RRT in small babies in the UK is concentrated at 11 sites (this is largely related to case mix; those sites with a paediatric cardiothoracic surgery unit doing more RRT. 7 As we only had 18 NIDUS devices the maximum number of sites, we could use was six to enable sites to have three intervention devices each – potentially one in use, one for a second baby recruited and one as backup in case of device problem. By the time the study started there were only 17 devices available, one having been lost to regulatory testing, thus it was decided that Newcastle as the original development site with small recruitment potential and with onsite engineers and scientists should have two devices.

In most circumstances the best design for a clinical study comparing two interventions is to randomise individual participants to one or other treatment. However, in I-KID this would open the possibility of two participants on the same PICU being treated differently. Discussion with parents and public revealed a strong feeling that asking parents to consent to individualised randomisation to dialysis type was not possible or desirable when most parents would have no prior knowledge of what dialysis was. Moreover, they would already be overwhelmed by the sheer amount of medical information they were being given at the time and thus most parents in that situation would defer to the clinician’s decision. It was also felt that having babies in the same unit on different devices at the same time could add to staff confusion and cause additional distress to parents to see an adjacent baby had the ‘new’ device when theirs had not been offered it – or vice versa.

Consequently, we opted for a form of cluster-randomised design, in which random allocation was applied to PICUs, not participants. As pointed out, I-KID was designed to study RTT in participants under 8 kg, where only a limited number of sites in the UK could participate and where resources limited this further to just six sites. With six participating sites it was always going to be difficult, notwithstanding random allocation, to ensure that the intervention groups in a conventional parallel-group cluster-randomised trial were comparable. This difficulty was heightened because the tertiary centres which participated in I-KID will inevitably have developed slightly different specialisms – for example some are co-located with quaternary cardiothoracic units.

This led to consideration of forms of cluster-randomised trials which can yield within-cluster information on intervention effects. A cluster-randomised crossover design would achieve this, in which three sites received the control intervention at the outset, and the other three NIDUS. Halfway through the study the sites would change to offering the other intervention. This form of design was not adopted because (1) there could be carryover effects, which in this case would largely be to do with the way staff treat participants needing RRT and, more importantly, (2) because there was a strong indication from staff in Newcastle who had used NIDUS during its development and compassionate use, of a reluctance to change from NIDUS to conventional therapy for these very sick children. Taken together, these considerations, where cluster allocation, availability of within-cluster information on intervention effects and reluctance to change back from NIDUS once used, led to the choice of SW design.

We anticipated that training sites to use new technology in a busy clinical environment and supporting them in the early stages of use would be challenging and time consuming. A further advantage of the SW design was that not all sites crossed over to the new device at the same time.

The SW design is a form of crossover designs, one in which there is substantial confounding of the treatment effect with time. Consequently, a valid analysis will almost certainly have to allow for the effect of time through the study, as well as any effect of treatment. In addition, the model needs to accommodate the fact that observations from each cluster, that is site, may be correlated. This is done by allowing a separate term in the model for each site, and usually this term is assumed to be random with a dispersion matrix taking a form specified up to the value(s) of the dispersion parameters in the model.

The model originally posited for I-KID assumed a general period effect, modelled by a categorical variable, and assumed that different observations within a site would have a given correlation, with those from different sites being uncorrelated – an equi-correlation structure. While there were some relatively short interruptions to the study when recruitment to NIDUS alone was paused, there was a much more extended pause due to COVID-19 from March 2020 for eight months for five sites and longer for a sixth site. This interruption was from the start of period 4 in the design and the general form of the period effect meant that this did not need modification. However, the much greater interval between observations taken in the fourth period compared with the intervals in earlier periods, called into question the form of the dispersion assumptions that had been made. Moreover, for a study with fewer than 100 participants, modelling a more elaborate dispersion structure would be inadvisable. A compromise was to assume a fixed site effect, so no dispersion structure needed to be specified. A disadvantage is that this model is potentially less efficient (Matthews and Forbes) but will be less vulnerable to the effect of mis-specification of the dispersion structure.

There were technical problems in establishing RRT in some participants using NIDUS and, when this occurred in the earlier parts of the intervention phase at a site (period 2 in sequence one and period 3 in sequence two), the rate of recruitment to NIDUS seemed slow. In parallel group studies, there is always an option to extend the study, however, with a SW design of three sequences and four periods, by the end of period 3 the only option would be to extend period 4. However, the information on the treatment effect contained in period 4 is limited – in the extreme case where the intra-class correlation is zero, then there is no information on the treatment effect in period 4. While it remains the case that there less information on the treatment effect in the final period than in periods 2 and 3, the reduction in information is lessened if fixed centre effects are fitted, which was a useful consequence of the decision to use fixed centre effects. For a related discussion of the information content of cells in a SWD, see Kasza and Forbes. 38

Questionnaires

Research team and wider involvement

I-KID research team composition reflected the range of professional disciplines, skills, expertise and experience required to deliver the study: paediatric intensive care medicine and nursing; medical physics; biostatistics; clinical trial design and management; data management (including representation from the central PICANet team at Leeds University); project management. The I-KID study co-applicants were all experienced in their respective disciplines, and generally at a relatively senior point in their careers. Within Newcastle University and the NuTH NHS Foundation Trust (NuTH), allocation of staff (e.g. trial and data management staff from the NCTU, project management staff from NUTH) was from within the existing complement of staff (i.e. no new staff were recruited specifically to work on I-KID), and was determined by capacity of the available staff members. NCTU staff allocation followed the unit’s normal model of a senior trials manager, trial manager, clinical trials assistant and database manager, and therefore included members of staff with various levels of seniority and experience. As might be expected in a study carried out over a number of years, there was some staff turnover (especially in the medical physics, NCTU and NUTH project management teams) over the course of the study. Replacement of members of staff who left was again determined by experience and spare capacity within the relevant organisation.

At site level, those delivering the clinical aspects of the study, and collecting and entering study data, were also drawn from the existing staff of the participating PICUs. In keeping with the responsibilities of the role, the Principal Investigators at all six sites were experienced paediatric intensive care consultants. As indicated in Chapter 2, all site staff participated in training for the roles in the I-KID study.

The I-KID study was closely aligned to and fully supportive of the equality, diversity and inclusion policies, including in respect of staff recruitment and training, of the university and NHS organisations employing the research team and site staff. Nonetheless, we note and are critical of these organisations still having a gender pay gap.

We did not collect data on the age, sex, gender, race, ethnicity or other protected characteristics of research team members or of site staff.

Collaboration with manufacturer (Allmed)

The I-KID study was designed, managed and analysed completely independently from Allmed, the device manufacturer, and no funding has been received from them.

However, this study could not have taken place without their support; the 17 NIDUS devices used were loaned by Allmed free of charge to study sites, and delivery and movement of devices was undertaken by them. The consumables used (filters and tubing) were purchased from Allmed by study sites for the NIDUS device and for the alternative forms of RRT (Prismaflex, Aquarius) and PD through the usual NHS purchasing routes. NUTH medical physics/engineering provided practical support in troubleshooting, training, maintenance and minor NIDUS device repair as the on-site NHS dialysis technicians would not have had sufficient knowledge and expertise to provide this specialist support of a new device.

Regulatory

The NIDUS device requires completed approvals by appropriate regulatory authorities (CE, UKCA, FDA, etc.) in order to be introduced more widely, and these should be pursued without delay.

Research