The impact of continuous haemofiltration with high-volume fluid exchange during cardiopulmonary bypass surgery on the recovery of patients with impaired renal function: a pilot randomised trial

Inclusion/exclusion criteria

Consenting men and women with impaired kidney function were included if they were ≥ 18 years old and were scheduled to have elective or urgent on-pump isolated CABG surgery. Patients were excluded if they were undergoing redo surgery, surgery on the great vessels (aortic surgery) or valve surgery; had significant impaired liver function [serum bilirubin > 60 µmol/l or international normalised ratio (INR) > 2 without anticoagulation]; had severe/end-stage renal failure (i.e. eGFR < 15 ml/minute); or were on dialysis. In addition, they were excluded if they could not give informed consent, had a malignancy or were known to be pregnant.

Randomisation

Patients who fulfilled the inclusion and exclusion criteria and gave informed consent to participate in the study were randomised into either of the two study groups on the day prior to surgery as follows:

1. On-pump CABG surgery patients with eGFR < 60 ml/minute to receive haemofiltration during cardiopulmonary bypass (experimental group).

2. On-pump coronary artery bypass graft (CABG) surgery patients with eGFR < 60 ml/minute not to receive haemofiltration during cardiopulmonary bypass (control group).

Treatment assignment was done online and was based on the block randomisation method using randomly varying block sizes of 2, 4 and 6 to ensure numerical balance between the groups. An independent statistician provided the randomisation tables. Each participant had an equal chance of being randomised to an experimental group of Z-BUF or a control group without intraoperative haemofiltration. Patients were stratified at the design stage on the basis of diabetes mellitus and the level of eGFR (eGFR between 15 and 40 ml/minute versus eGFR between 40 and 60 ml/minute). Only trial staff with a unique user identification and password could log on to the bespoke, encrypted database. The allocation was revealed only after unique patient data were entered. Access to any list of previously randomised patients or to case record forms was not permitted; only the research nurse had such access.

Primary outcome measures

The primary outcome measure was the frequency of duration of ICU stay > 3 days for patients with renal impairment. Total ICU length of stay as a continuous variable was also determined.

Secondary clinical outcome measures

Secondary clinical outcome measures were:

• composite of perioperative incidences: bleeding (clinically defined significant loss of blood that needed transfusion of blood products), sepsis, death, arrhythmias, stroke and myocardial infarction

• 30-day mortality

• need for postoperative continuous venovenous haemofiltration (CVVH) in the ICU: indications for requirement of postoperative CVVH adhered to standard surgical guidelines (i.e. onset of hyperkalaemia > 6.0 mmol/l; metabolic acidosis of renal origin; and anuria or oliguria) defined above

• mechanical ventilation time

• postoperative hospital stay; and

• eGFR change from baseline at 6-week follow-up.

Secondary economic outcomes

Resource utilisation and key cost indicators associated with each of the two pilot arms, specifically ICU stay, hospital stay, postoperative renal replacement therapy, mechanical ventilation and medications, were estimated up until hospital discharge or up to 30 consecutive days' in-hospital stay. A health-related quality-of-life questionnaire, European Quality of Life-5 Dimensions (EQ-5D),37 was administered at hospital admission before surgery and at the 6-week follow-up hospital visit.

Adverse events

Adverse events occurring during the perioperative and follow-up periods were documented and reported. Typically, adverse events reported included complications such as anaemia, bleeding, pneumothorax, respiratory/chest infections, atelectasis, in-hospital deaths, gastrointestinal complications, pleural effusion, pulmonary oedema, reintubation, sepsis in ICU, cerebrovascular accidents, myocardial infarctions, cardiac arrests and heart blocks. Other complications that required the return of patients to theatre for reoperation, such as bleeding, tamponade and rewiring, were also included. Patients who developed acute renal failure (defined as eGFR decline of > 50% of the baseline value13) were managed by postoperative haemofiltration or other renal replacement therapies. Some patients also developed wound complications such as sternal dehiscence or infection (including sternal infections). In addition, arrhythmias such as atrial fibrillation were frequently observed and commonly treated with amiodarone infusion or in some cases by cardioversion shock.

Feasibility outcome measures

Barriers to recruitment to a larger full study were documented as observations made during the recruitment period of the trial. Reliability of data collection methods was monitored using a 13-point CRF validation check (see Appendix 1). Three audit clerks who were independent of the trial team performed this check. They randomly chose 12 CRFs and checked the data entry against the patient's clinical notes and the generic database.

Summary of the main reasons for not recruiting eligible patients

Eligible patients were not recruited largely for the following reasons:

1. Patients had an eGFR of > 60 ml/minute, or were undergoing off-pump surgery (n = 103) or combined surgeries/complex surgeries (n = 34): that is, combined CABG and valve surgery, valve replacement surgery, thoracic aneurysms, redo CABG surgery, transcatheter aortic valve implantation, thoracic surgery/bronchoscopy, pacemaker change, or sternal wire removal.

2. Recruitment was restricted to the research nurses' working week of Monday to Friday, 0900–1700 hours. Consequently, 36 patients who were potentially eligible for the trial could not be recruited because they were available for approach outside these hours.

3. Patients were screened but not included because of surgical decisions (n = 44).

4. eGFR was unknown at the time of a patient's hospital admission (n = 30).

5. Patients declined consent for religious or personal reasons (n = 26).

6. Patients were awaiting surgery at the close of trial recruitment (n = 14).

Barriers to recruitment

One of the key objectives of the pilot trial was to obtain an understanding of the barriers to the recruitment of trial patients with an initial plan to recruit 60 patients in 6 months. However, only 37 participants were recruited over a period of 15 months. The recruitment of a relatively small number of patients during the trial period highlights some significant but not insurmountable barriers to recruitment in this setting.

The main barriers were as follows:

1. In our centre, up to 50% of coronary surgery is performed off-pump, a figure that is one of the highest in the UK, and this trial was recruiting on-pump CABG surgery patients only. Our figure demonstrates that 103 patients with pre-operative renal impairment underwent off-pump CABG surgery during the time frame of the trial.

2. Recruitment was restricted to research nurses' working week hours (Monday to Friday, 0900–1700 hours). Consequently, 36 patients who were potentially eligible for the trial could not be recruited outside of those hours because they were available for approach outside these hours.

3. Issues were encountered through the screening process for identifying prospective eligible patients. The primary inclusion criterion for the trial is the presence of mild to moderate impairment in kidney function as measured by eGFRs (< 60 ml/minute and > 15 ml/minute). In the cardiac surgery setting patients may be scheduled for CABG surgery through the elective route or be admitted into the hospital as an urgent case. It is particularly common in patients admitted as urgent cases that their eGFR values are often not documented in the case notes/clinical database or that blood samples are not taken until later on the day before surgery. Consequently, 30 patients who were eligible for this trial were not recruited.

4. Seasonal outbreaks of pandemic influenza and other infectious diseases occurred. In the winter of 2010–11, an outbreak of pandemic influenza led to the closure of all elective cardiac surgery from December 2010 to February 2011. In addition, an outbreak of norovirus within one surgical ward in January 2012 significantly reduced planned cardiac surgical activities for 2 weeks and only urgent cases were considered for operations during that time.

5. Protocol deviations and treatment crossovers arose. There were two protocol deviations and four crossovers. In three cases this was because of a necessary change in clinical strategy intraoperatively. Of the remaining cases, in two there were communication errors and in the last case eGFR had recovered to normal following the initial screening. Treatment fidelity for intraoperative haemofiltration was followed in all cases where the intervention was received in accordance with standard protocol Z-BUF, regardless of whether or not the patients were crossovers.

We have also observed that recruitment could be markedly increased if enrolment criteria were widened to include patients with impaired renal function undergoing other types of cardiac procedures, such as valves or combined valves and CABG surgery. This is based on observations from another ongoing trial (the Haemotracker trial, ISRCTN 48429978) in our centre, which is investigating the impact of intraoperative haemofiltration on oxidative stress for patients with impaired renal function undergoing all cardiac procedures. As shown in Figure 2, within a similar time frame 91 patients were consented and randomised into the Haemotracker trial compared with 37 patients in this trial [the filtration on bypass surgery (FOBS) trial]. If the criteria for the FOBS trial were extended to include all cardiac surgery and recruitment was optimised to all hours we estimate that we could meet a target of 35–50%.

FIGURE 2.

Comparison of recruitment rates for the Haemotracker and FOBS trials.

Reliability of data collection methods

Data were collected by one research nurse, starting from screening and continuing to follow-up. This research nurse also entered the data into the trial database. The accuracy and reliability of this method of data collection was assessed using a 13-point validation check form performed by three audit clerks independent of the trial. These individuals randomly chose four CRFs each and checked the recorded data against the patient clinical notes and the generic database. Twelve CRFs with a total of 9509 database fields were checked for completion and accuracy. This check revealed that 323 database fields (3.4%) were missing and also some information was outstanding from patients awaiting follow-up visits. In addition, no data entry errors were found. A future study could easily follow a similar kind of approach for data collection. However, we plan to work with a registered clinical trials unit in a future multicenter trial, and it is likely that they would have a more robust data collection and entry validation tool.

Urine outputs and fluid balances

We assessed the pattern of urine output and fluid balances during the postoperative period of up to 96 hours after surgery. Table 6 shows similar patterns of urine output between the groups during the perioperative period of up to 48 hours after surgery. The fluid balances showed different patterns between the patient groups and also for diabetic and non-diabetic patients during the perioperative period of up to 48 hours after surgery. After the first 48 hours a significant number of data were missing and therefore are not shown in Table 6. The reason for the incompleteness was the fact that most patients were well enough to be discharged from ICU into wards after 72 hours. In the wards most patients had their urine catheters taken out and were free to use toilets and take in fluids as they wished. Consequently, for some patients, urine outputs and fluid balances were not adequately monitored or documented by staff.

Postoperative acute renal failure

As summarised in Table 7, there were three patients who developed postoperative acute renal failure out of the 37 randomised patients. These three patients were given postoperative renal support [haemofiltration, dopexamine (Dopacard^®, Cephalon) and furosemide infusions]. One non-diabetic patient with pre-operative eGFR of 29 ml/minute who was randomised to the no-intraoperative haemofiltration arm developed postoperative metabolic acidosis and anuria/oliguria. The patient was given postoperative haemofiltration for 67 hours and intravenous dopexamine and furosemide infusions for 107 hours. The second patient was diabetic and had a pre-operative eGFR of 50 ml/minute, was randomised and given intraoperative haemofiltration and then developed anuria/oliguria postoperatively. The patient was given postoperative haemofiltration for 19 hours together with intravenous furosemide infusions for < 1 hour. The third patient was non-diabetic with an eGFR of 25 ml/minute, was randomised to the intraoperative haemofiltration group and then developed postoperative anuria/oliguria and was given haemofiltration for 22 hours together with intravenous dopexamine for 7 hours in total. The pattern of the duration of postoperative haemofiltration appears to favour the use of intraoperative haemofiltration, although owing to the small numbers no definitive conclusion can be reached based on this observation.

Changes in estimated glomerular filtration rate

The difference in changes in eGFR between 6 weeks' follow-up and baseline values indicated a trend towards higher values for the haemofiltration group than the group without haemofiltration (Figure 4). Again, the sample size is too small to draw any conclusions from this.

FIGURE 4.

Mean eGFR change (95% CI) between baseline and at 6 weeks' follow-up.

Framework

One of the key issues that needs to be considered is the type of framework to adopt for the economic analysis. Acute kidney injury is reported in the literature to occur in up to 30% of postcardiac surgery patients,27 leaving a large scope to provide benefit in terms of both clinical outcomes and costs. However, whether or not this will be seen in the feasibility analysis and for this particular patient group is uncertain.

The gold standard in economic analysis is cost–utility analysis, in which results are expressed cost per quality-adjusted life-year (QALY). However, one aspect that needs to be explored within this pilot study is the feasibility of quality-of-life measurement tools in sensitively differentiating between the two groups of patients.

Methodology

The economic model aims to assess the quality of life and cost variation between the two groups and thus assess the incremental cost-effectiveness of using intraoperative haemofiltration during surgery. The perspective of the cost analysis is that of the NHS. The benefits included are quality of life in the form of the EQ-5D. No adverse events are considered although their occurrence, if the effects remain at follow-up, should be encompassed within the quality-of-life measurement within the EQ-5D.

Quality of life

A small sample of EQ-5D questionnaires was used with one measurement pre surgery and one post surgery as a late addition to the protocol whose follow-up is not yet complete for all patients at the time of producing this report. The sample with pre-admission and postdischarge values is 20 patients. The validity of any differences between the two groups depends on precision of the matching of the two treatment groups so that theoretically all patients are the same at the start of the trial. If this is the case then any variation in quality-of-life outcomes can be attributed to the use of haemofiltration within theatre.

The quality-of-life measure used in this study was the EQ-5D, which classifies 243 different health states with five different dimensions of health, each with its own level of severity. The EQ-5D is a generic measure and is the preferred method of the National Institute of Health and Care Excellence (NICE). NICE recommends the use of the weighted values from the EQ-5D questionnaire to be taken into consideration when evaluating health-care technologies.

The EQ-5D produces two estimates of quality of life. The VAS records the estimated quality of life of the individual who marks the scale. The answers to the EQ-5D questionnaire can be weighted according to the values generated by the UK population's preferences for each of the health states. In this pilot study we have considered both elements of the EQ-5D.

Costs

In order to assess the cost implications of the use of haemofiltration within theatre it is necessary to appropriately identify the resource use that differs between the two groups as a direct result of the change to treatment.

The key elements of resource use data that are utilised within the economic analysis concern the duration of hospital stay and what combination of ICU and ward days this stay consisted of, the use of haemofiltration, whether it was used within the theatre or in the ICU environment, and drugs prescribed specifically to aid renal function.

• The costs of nursing time are not included for within theatres because no additional nursing support is required in order to facilitate the intraoperative haemofiltration, as this is performed by the perfusionists running the pump. Postoperatively, within the ICU, the nurse to patient ratio is already 1 : 1 and such nurses are trained to be able to use the haemofiltration machine.

• The cost that has been allocated for 1 day on a ward is £264, which is the cost for the ward in which the patients are most likely to end up.

• The cost of 1 day's stay in ICU has been divided by 24 and used alongside the number of hours that the patient stays within the ICU, in order to allow for more precision in the estimates. This does not, however, take into consideration the fact that staying in ICU for 1 hour during the night may in effect cost less as, for example, meals would not be necessary, or the fact that charges may have to be made for an entire day, even if the patient occupies the bed for only a short period of time.

• The postoperative haemofiltration machine is estimated to last for 10 years. The number of patient hours over a 1-year period between November 2010 and October 2011 has been recorded as 1375 and used to calculate the number of hours of use of the haemofiltration machine over a 10-year period (13,750). The fixed cost of the purchase of the machine (£24,713) has, therefore, for the purpose of this analysis, been calculated as a cost per hour.

• The cost of haemofiltration consumables within the theatre has been calculated at £63 per patient on average. This estimate is derived from the Liverpool Heart and Chest Hospital patients who receive intraoperative haemofiltration in theatre. The duration of the use of the intraoperative haemofiltration and the cost of consumables used are, however, unlikely to differ in this patient group. This does mean, however, that the cost of the use of haemofiltration within theatre does not currently take into consideration the duration of the filtration.

• The costs of consumables used for postoperative haemofiltration within the ICU environment were estimated from the resource use of three patients similar to those in the trial. In order to estimate the consumables and other resources used within a full 24-hour period, costs were estimated and an average generated. In each case, although the full day of usage may have fallen part way through the treatment period, the cost of the initial set-up material (dialysis line, which is used for the duration of the treatment unless infection is suspected, and an aqua-set, which is changed every 72 hours) is included to most accurately represent what the costs of patients within the trial would be. As all the patients within the pilot study who required postoperative haemofiltration within the ICU were on the machine for less than 24 hours, the costs of consumables for the second day, when the initial set-up materials were not needed, did not need to be included.

• The drug use that has been considered within the economic analysis regards only those drugs that specifically target renal function. Within the ICU, dopexamine is used intravenously. It is assumed that this drug is used for the duration of the period that the patient remains in the ICU and therefore drug costs (£19.80 per 24-hour period), along with the cost of saline solution (£4.27 per 24 hours), are allocated according to hours spent in the ICU.

• In terms of drugs prescribed on discharge, furosemide, assumed at a dose of 20 mg, has been included in the costs (at £0.80 per pack of 28 × 20-mg tablets) when the patient was not already being prescribed the drug on admission. The furosemide prescription at follow-up has not been included in the cost analysis as it may be influenced by the variation in time until follow-up.

Short-term modelling

The economic analysis is largely focused on the data collected within the pilot study. The outcome of interest for the economic evaluation is HRQL as measured by the EQ-5D. This information can be used to calculate an incremental benefit of the use of intraoperative haemofiltration within cardiac surgery. The benefit of measuring quality of life post surgery but close to discharge from hospital is that it means that any effects of the surgery that last beyond the hospital stay and impact on quality of life will be picked up. While quality of life throughout a patient's stay may have varied somewhat, this may not be important over the longer term. On the other hand, what a postsurgery measurement close to discharge does not capture is the time that the individual has spent in hospital. This will be assessed to some extent within the costing element of the economic analysis; however, it means that those patients who have had to reside within the hospital for a longer period will not have their overall assessment of quality of life adjusted accordingly.

The EQ-5D follow-up scores could be adjusted to take into consideration the variation in follow-up time, which could vary between 6 and 12 weeks. Currently, the follow-up time is assumed to be similar in the two groups on average and therefore the quality-of-life change is compared unadjusted. Such adjustment would enable greater accuracy in the estimation of QALYs and would need to be considered in a further trial.

A study assessing the HRQL of chronic kidney disease patients was conducted in Japan. 38 The authors in that case also used the EQ-5D questionnaire along with eGFR and took into consideration some comorbidities including hypertension, diabetes and cardiovascular disease. In this case, patients were categorised as having cardiovascular disease if they had ischaemic or congestive heart disease or stroke in their record. EQ-5D values were produced for each of the five chronic kidney disease states with the subgroups mentioned above and analysed separately. These can be compared with the results obtained from the sample of patients in the pilot study, although it must be noted that the preference weights attached to the EQ-5D values are those of a Japanese population and therefore may differ from those within the UK population.

The clinical outcome of this pilot study is related to the number of days' ICU care that is required by each patient. This will be captured within the economic analysis as one of the costs to be compared between groups. The cost of the use of the filtration machine both within theatre and subsequently within the ICU will also be compared along with the costs of stay within a general ward and any drug use or additional nursing care that is necessary as a result of the renal impairment. The resource use at the individual level will have a cost attached to it and then the two groups can be compared.

Quality of life

The EQ-5D consists of two elements: the VAS and a questionnaire. The VAS asks the patients to rate their health on a scale from 0 to 100. The results of this section are detailed below in Table 10.

The mean on admission differs by almost 10 points between the two groups; however, by follow-up this difference has narrowed to around 3 points. The mean difference between the two time points is also recorded in the table showing that the quality-of-life improvement is greater in patients who received haemofiltration.

The answers from the EQ-5D questionnaire can be converted into quality-of-life estimates based on the preferences of the UK population. These weights have been derived and used to generate the means for each group as shown in Table 11. The quality-of-life estimated preoperatively is slightly different between the two groups and this gap then widens at follow-up. The incremental benefit of using intraoperative haemofiltration, ceteris paribus, is 0.11 within the 0–1 range of the quality-of-life scale.

The mean estimated quality of life of patients with chronic kidney disease in Japan who have cardiovascular disease is 0.826 (stage 3) and 0.843 (stage 4). 38 The improvement in quality of life from stage 3 to stage 4 is unexpected and may represent abnormalities in the sample used. It could, however, be a true reflection of quality of life within this group of patients. The quality of life of the patients within our pilot study is lower than that of the patients in the Japanese study. This could be simply due to the demographic differences between the people of Japan and the UK but may also be because the patients within this pilot study all are undergoing CABG surgery, and therefore the severity of their cardiovascular disease may be greater than that of the patients within the Japanese study sample.

Costs

Table 12 depicts the mean costs for each of the groups for those elements of resource use assumed to be directly influenced by the use of haemofiltration in theatre. The averages are for all patients for whom data were recorded.

The largest average cost difference came from the duration of ICU stay, followed by the duration of stay on the ward. However, only three patients within the pilot study were prescribed dopexamine, and therefore this element may play a more significant role within a larger data set if the proportion of patients prescribed the drug was to increase. The total average cost per patient is 24% lower in those patients who received haemofiltration within theatre.

In such a small sample, the costs of exceptions to the protocol, such as the patient who was randomised to the no-haemofiltration arm but received haemofiltration in theatre, can make a difference to the overall cost-effectiveness analysis.

However, we had EQ-5D data for only 20 patients in total and thus, to be consistent, we used the costs from only these patients. This does not greatly affect costs; the incremental cost for haemofiltration now becomes –£1744 (Table 13).

Subgroup analysis

One of the aims of the pilot trial was to stratify the patients into subgroups according to the presence or absence of diabetes, and kidney functioning levels. It would have been interesting to consider the patients by both their diabetes status and their eGFR ranges; however, owing to the small sample sizes, this was not possible. It is worth mentioning that in a larger trial it would be important to explore this aspect further as it may be possible to make predictions or acquire costs from the patient ranges.

Table 14 categorises the number of patients in each trial arm. As shown in Table 14 the numbers in each arm are relatively small, which means that the results may not be representative of the patient population of interest, hence a larger group of patients is required in a future trial.

The impact of diabetic/non-diabetic status on the cost analysis

Part of the study's aim was to stratify the patients where possible and then perform a cost analysis for the patient subgroups and in particular for those patients with or without diabetes. From this we added costs for each of the patients in their groups (as shown in Table 14) and then calculated as an average cost for each of these subgroups. However, one caveat to note at this point is that at such a small scale we can only discuss the patterns of trends shown from the average costs. Table 15 depicts the averages from the subgroup costings.

Table 15 has itemised the costs for diabetic and non-diabetic patients in both groups (non-haemofiltration and haemofiltration) in order to establish whether or not there are any trends. As shown in Table 15, in both groups the costs are greater for the diabetic patients than for the non-diabetic patients. The sample is too small to definitively say if there is a relationship between the two, although it is feasible that additional care is required for diabetic patients and this could be the reason for the increase in overall costs. Additionally, it is interesting to note that the length of time spent in ICU (and therefore the costs associated) for the diabetic groups, regardless of whether or not they had haemofiltration, was approximately twofold higher than for the non-diabetic patients. This too may relate to costs associated with diabetes as a disease although with such a small sample the evidence cannot be conclusive.

As we can expect, the cost of longer ward-days is a large driver of the overall average cost. For the non-haemofiltration patients this was higher; however, when this was broken down between diabetic and non-diabetic patients what was found, which we may not have expected, is that costs accrued are greater for the non-diabetic patients. Despite this we cannot categorically claim that this would apply to all non-diabetic patients. On further analysis we see that one patient in particular within the non-diabetic group is driving up the costs through an unusually longer ward stay (than average for the group). Therefore, owing to the smaller sample size of the groups and any unusual characteristics that the patient might have, this can impact dramatically on the average costs. Consequently we cannot draw any definitive conclusions from this finding and so must look at the costs as provisional estimates owing to the small sample size of the groups.

It is important to note here that some patients allocated to the no-haemofiltration group have some costs arising from haemofiltration given in theatre because of treatment crossover. Although the costs are only small, they are attributed to one patient in the group who was given intraoperative haemofiltration during surgery. This is, therefore, a study treatment crossover and although this should not have occurred, it did, and is included in the costs for the purpose of completeness.

European Quality of Life-5 Dimensions in diabetic/non-diabetic patients

The EQ-5D was conducted in 20 of the patients from the whole sample; from this it was possible to compare costs between the haemofiltration and non-haemofiltration patients. It was not feasible to analyse the data any further in terms of whether patients were diabetic or non-diabetic. The comparisons would have been only one versus seven patients in the group with no haemofiltration for diabetics and non-diabetics respectively, and two versus nine patients in the group with haemofiltration for diabetics and non-diabetics respectively. It would be valuable in a larger study to assess this to see if there are any significant cost and quality-of-life differences.

Incremental cost-effectiveness ratio

The incremental cost-effectiveness ratio (ICER) is calculated by dividing the difference in costs by the difference in benefit. In order to enable like-for-like comparisons to be made, only the patients who have complete EQ-5D data have been included in the costing element of the ICER. In the case of the current population the incremental benefit of the use of haemofiltration within theatre is an improvement of 0.11. This comes at an incremental cost of –£1744, so haemofiltration saves money. The ICER is therefore –£15,851. Thus, haemofiltration in theatre dominates the non-haemofiltration strategy, as it is associated with lower costs and increased quality of life.

This figure should be viewed with caution as the sample size of 20 patients in total (group A = 9, group B = 11) with complete EQ-5D pre-operatively and post discharge is far too small to have any statistical significance. It is also hugely dependent on the costs that have been included being influenced directly by the use of the haemofiltration machine within theatre, which cannot be proven without a larger data set, and so their inclusion is based on clinical assumption of the impact of the use of haemofiltration within theatre.

Owing to the small sample any single large fluctuations in costs will have large implications for the resulting ICER estimate. For this reason, the patients who had to be readmitted either to ICU or to hospital following discharge have been excluded from the main analysis above. As none of the patients was recorded as being readmitted as a result of their renal functioning, this should not alter the comparability of the two groups. In a further trial readmission should, however, be considered for its relationship between haemofiltration in surgery and, if found to be an important factor, included within the costing.

The ICER estimation is influenced by each of its constituent elements. Sensitivity analysis can be undertaken in order to quantify the impact of each of the structural elements of the economic model (Table 16). A countless number of scenarios can be built to test the influence of the parameters within the model.

As standard care is currently dominated by the use of haemofiltration within theatre due to providing a benefit coupled with a cost saving, sensitivity analyses have been conducted to test the influence of some of the key parameters and how much they would need to change in order for the costs of non-haemofiltration to outweigh the costs of haemofiltration.

• Sensitivity analysis (SA) 1: if, for example, it was proven that the length of time in a general ward has no relationship to the renal function of the patient, then the costs of the ward stay would need to be excluded from the analysis. This reduces the cost saving but, as long as there is an incremental quality-of-life benefit, haemofiltration would still dominate standard care.

• SA2: the cost of consumables used in the use of haemofiltration within theatre was given as a set cost of £63 per patient in this analysis. Within the current model structure, including the cost of stay in a general ward, the cost per patient within theatre of haemofiltration consumables would have to increase to £2600 in order for group A to cost more on average than group B. With the quality-of-life improvement of 0.11, this would result in an ICER of £456.

• SA3: hours spent in ICU is the primary outcome measure within this pilot study and therefore its impact on the cost-effectiveness estimates has been explored. The average number of hours spent in ICU in the pilot result for group A was 79 and for group B was 65. The average ICU hours for group B needed to increase to 100 (keeping the average hours of group A the same) in order for group B to cost more on average per patient. With the quality-of-life improvement of 0.11, this would result in an ICER of £4035.

• SA4: the final scenario in the sensitivity analysis excluded the costs of ward-days as in SA1 but also looked at an increase in ICU hours from group B. In this case the average stay had to be 82 hours in comparison with 65 in group A in order for group B to be more costly on average per patient. With the quality-of-life improvement of 0.11, this would result in an ICER of £2335.

An additional one-way scenario analysis was undertaken to see how much costs would have to increase to reach the NICE threshold of £30,000. Assuming that the benefit remains at 0.11 for haemofiltration, the incremental cost for haemofiltration would need to increase from –£1744 to £3300 per patient for the ICER to reach £30,000 per QALY. This would obviously change if the benefit was also reduced.

Potential cost savings

There were 1276 patients who were operated on during the same time period as the trial. Of these patients, 107 were identified as meeting the eligibility criteria for the trial. If we assume that all of these patients could have received haemofiltration within theatre, using the average cost saving estimated from the trial, the cost saving to the Liverpool Heart and Chest Hospital could be in excess of £150,000 annually. This example aims to demonstrate the potential for cost saving within a single unit; however, it is important to recognise that the cost savings estimates are tentative only and therefore their implications in terms of overall cost savings should be viewed with caution. In such a small sample the costs of exceptions to the protocol, such as the patient who was randomised to the non-haemofiltration group but received intraoperative haemofiltration in theatre, can make a difference to the overall cost-effectiveness analysis. Likewise, any changes to the variables included within the costing analysis as a result of clinical findings from a more robust sample, could have a considerable impact upon the final cost difference.

Quality of life

In this feasibility analysis only EQ-5D was used to value patient quality of life. Many studies, such as that of Fructuoso et al. ,39 who measured quality of life in chronic kidney disease patients, have used the Short Form questionnaire-36 items (SF-36). 40 Like the EQ-5D, this is a generic measure and well validated, although it is not as widely used in the UK. Along with the EQ-5D score, a more disease-specific measure such as the Kidney Disease Quality of Life (KDQoL)41 or KDQoL Short Form (KDQoL-SF)39 may be useful in capturing any sensitive changes that a generic measure would not. The KDQoL-SF includes SF-36 scales as well as dimensions specific to kidney disease; the dimensions contain 43 items that can be summarised in 12 scales. 39 Fructuoso et al. 39 also used the KDQoL in their study and found that it is sensitive enough to detect important changes in the effects and burden of kidney disease, overall health and patient satisfaction; in addition to this, validation studies have been conducted on the KDQoL-3642 and found it to be informative with high validity, which is one main argument in favour of including it in a future trial.

Resource use and costs

Given the small sample size and the exploratory nature of the economic evaluation, it is unclear if all of the relevant costs and resources have been adequately captured. For example, only three patients received haemofiltration postoperatively, two of whom were in the non-haemofiltration group. This difference is only a one-patient variance but in such a small data set the effect is magnified. In a future trial, all of these differences would need to be carefully explored to ensure that they are real differences and not spurious results.

One cost that might be relevant to a future economic analysis is the cost of treating anaemic adverse events.

Longer-term economic model

The long-term model could follow a structure as below. The data collected from the larger trial would end at follow-up; however, using estimations from the published literature and from clinical estimates the future implications of any prevention in chronic renal impairment can be analysed. Based on the data generated at follow-up the patients would be categorised as having chronic, acute or no presence of renal impairment. If the follow-up period was not sufficient to estimate whether or not the impairment is chronic then the likelihood of the current functioning worsening or improving would be modelled to aid with the categorisation of patients. Patients assumed to have acute renal impairment would have their quality of life and costs of treatment estimated and patients would move between subgroups depending on their probability of worsening or improving in that period.

The patients with chronic renal impairment enter the staging model in accordance with their eGFR. A Markov model with Monte Carlo patient-level simulation will then move them between stages depending on the transition probabilities of disease progression and death within each period. Owing to the progressive nature of the disease patients cannot retreat through stages, but jumping over progressive stages is possible. The period length used within the model will be guided by clinical experts and the evidence in the literature as to which is the most appropriate period for measuring change within chronic renal disease. Orlando et al. 43 produced an economic model of chronic kidney disease modelling natural history and incorporating seven treatment interventions. Their Markov model uses, for example, monthly cycle lengths and includes adverse events such as strokes and myocardial infarction.

The probability of adverse events occurring within each period will also be modelled owing to the close relationship between kidney function and coronary-related events and disease. 44 The adverse event node in Figure 5 represents modelling of key adverse events such as heart attacks, strokes and the development of diabetes.

FIGURE 5.

Model schema.

Quality of life and costs of treatment for each of the diseases and stages of chronic renal disease will be estimated to allow QALYs and ICERs to be calculated. For example, when a patient reaches stage 5, kidney failure is assumed and the costs of dialysis will be included in the analysis, as will the probability of a transplant and its likelihood of success. Discounting of longer-term costs and benefits would also need to be included within the model structure. The treatment and control groups will be analysed separately to estimate whether or not the use of haemofiltration within theatre is a cost-effective use of scarce resources over the long term.

It is important, however, to be aware that extrapolations of short-term data far into the future is done with decreasing confidence in the precision of the estimates over time, and therefore bootstrapping could be conducted to quantify the uncertainty in the estimates. In order to justify the comparison of the patient groups in such a long-term model, the data must first prove a significant difference in renal impairment at follow-up, which is likely to be as a result of the use of haemofiltration within theatre, as the patient groups have as far as possible been matched to be equivalent in every other sense.

Uncertainties and issues

• The type of economic analysis most suitable for this condition is uncertain. If the full trial shows that there are no clinical benefits of haemofiltration, then the assessment becomes a cost-minimisation analysis. If there are important benefits with haemofiltration, such as quality-of-life gains, then a cost-effectiveness analysis will be most appropriate.

• The time frame of the analysis will also affect the economic model developed. If a long-term perspective is taken, then the long-term costs and outcomes of chronic kidney disease would likely need to be estimated, rather than just the short-term differences in EQ-5D.

• Whichever economic framework is developed for the full trial, a more detailed microcosting of the consumables used within haemofiltration treatment at an individual level would enable the differences in terms of haemofiltration consumables within theatre and ICU to be incorporated. However, if these elements are unlikely to change between groups then the total costs as used here may be adequate.

• The number of times that the machine clogs and has to be reset has implications for the length of time it is used and the consumables needed. Whether the likelihood of coagulation is random or related to patient-specific characteristics needs to be considered along with the cost implications of the use of more consumables and potentially more nursing time.

• Does the length of time on the haemofiltration machine play a role in the final outcome? If the assumption is that the haemofiltration machine is only needed as the CPB machine is causing the reduction in functional ability, then the use of the haemofiltration machine for the duration of the CPB usage should suffice. What happens, however, if the patient is on the CPB machine longer than on the haemofiltration machine? Should the ratio of time be considered and banded into subgroups for analysis in a further trial?

• In a further trial, it would also be useful to compare the use of off-pump surgery. Over the period between November 2012 and March 2012, 19 patients with on-pump surgery and 17 off-pump patients required haemofiltration postoperatively. This would create a second comparator group within the trial, but may help to clarify if the use of the CPB machine is in fact the cause of the renal impairment within CABG surgery, and also help to determine which is the most clinically effective and cost-effective strategy.

• Although only three patients received haemofiltration postoperatively, two of these patients were in the non-haemofiltration group; therefore, the cost for this element is greater than for those patients within standard care. This highlights an important consideration that needs to be made in any further trial. If the use of haemofiltration within theatre does not reduce the need for its use within the ICU environment, then it is imperative that we can be confident that any variation in other resource use, such as ICU stay, is not simply due to other external factors.

• Another important consideration for the economic analysis, which is also essential in the clinical assessment, is the matching of the treatment and control groups. Without confidence that the groups being compared are similar prior to the initiation of the randomised controlled trial (RCT), we cannot read too much into the results. An initial assessment of the similarity between the groups in terms of quality of life would need to be conducted to ensure comparability between the groups. If the quality of life varies between the two groups initially then adjustments would need to be made to ensure that any resulting difference is not as a result of baseline variation. In the case of this pilot study the groups are unlikely to be precisely comparable, as owing to the recruitment volume being smaller than expected the randomisation was not completed.

• It is conceivable that the use of haemofiltration within surgery may not be cost-effective for all patients. Renal injury during cardiac surgery appears to be related to a number of issues, such as pre-existing renal dysfunction, diabetes mellitus, ventricular dysfunction, older age and hypertension. 27 The data are already sectioned into eight subgroups and, although the sample size is not sufficient in this pilot study to enable analysis at this level, the subgroups could be explored separately in the economic analysis of any future trial.

• According to NICE guidelines, there is a causal relationship between chronic kidney disease and cardiovascular risks (in both directions); however, the level of risk is not known, although it has been recorded that patients with chronic kidney disease have a much greater prevalence of dying from cardiovascular disease-related issues than going on to develop kidney failure. 45 This relationship would need to be explored further in order to incorporate it into any longer-term modelling that was required.

• There are a number of reports that have commented on the relationship between cardiac function and renal impairment; many of these studies used serum creatinine as an index in order to derive the eGFR; however, it has been suggested by Levey et al. 45 and Moist et al. 46 that creatinine levels can be affected by factors other than eGFR. If this is proven to be true, a full haemofiltration study should look at the effects of factors other than simply eGFR to determine changes in kidney function, and such factors could be incorporated into the disease staging in the long-term model. Essentially owing to their close relationship, the prevalence rates of cardiac problems in patients with renal impairment are also explored in a number of studies. Sarnak et al. 47 suggest that in higher-risk populations in particular, albuminuria levels and reduced eGFR are the greatest risk factors to cause kidney disease patients to encounter cardiac problems. It is also suggested that individual patient factors such as sex and race play a bigger role in kidney function, and therefore more focus within the study should be given to the different racial groups and the sexes within each. Such evidence could be used to incorporate risk profiles of development of chronic kidney disease and progression within the disease stages in the economic model related to variable demographics and comorbidities.

• Although during this feasibility study longer-term data were not available, it has been suggested that time frame may have a considerable impact on overall cost-effectiveness. In a similar study, Klarenbach et al. 48 found that, during sensitivity analysis, extending the time frame to a patient's lifetime in the haemofiltration group managed to reduce the cost of a QALY, and therefore increased overall cost-effectiveness.

• Cox proportional hazard regression can be used to estimate the hazard ratios for the treatment and control arms. This information coupled with the means would allow estimations of the impact of haemofiltration within theatre on the time in ICU to be analysed. This would enable the length of stay to be taken into account when the current point estimates of quality of life do not allow for such variation.

What is the most appropriate structure of economic analysis for evaluating filtration on bypass surgery?

The appropriateness of any economic methodology depends on the nature and context of the underlying clinical analysis. Evaluations based on inappropriate or poor-quality clinical data will fail to provide a reliable basis for health care decision-making. The primacy of clinical data is particularly evident in the choice of economic methodology that is appropriate in any context. In the case of FOBS, two potential options were available given that the feasibility study indicated that there were likely to be significant improvements in outcome as a consequence of the use of filtration on bypass surgery. The first option would be to undertake the full-scale trial in the context of a cost-effectiveness analysis. Such an analytical structure would identify measure and value variations in both cost and outcome arising between the two strategies to generate a cost per unit of outcome.

If we were unable to conduct a cost-effectiveness analysis, the second-best methodology that we would look to employ would be a cost–utility analysis. The use of a cost–utility analysis, however, would be conditional on the ability to capture the additional health benefits arising from the use of filtration by the preference-based generic measure employed from the EQ-5D. Analysis of the clinical data obtained in the feasibility study appears to indicate significant variations in EQ-5D results between the two arms of the trial. Therefore, this would appear to indicate that the nature of the clinical outcome variations exhibited between both arms of the trial might be appropriately captured in a cost-per-QALY (i.e. cost–utility) framework. Such a framework would greatly facilitate the ability to place the results obtained from this analysis in the context of QALY analyses undertaken in a wide range of alternative therapeutic areas.

The value of having access to the results of this feasibility study arises from the fact that they can be used to effectively inform and structure decisions with regard to any subsequent economic study that is undertaken. By definition, the appropriate structure of economic analysis to be employed in a clinical trial cannot be known with certainty in advance in the absence of a feasibility study. Such a study provides a priori evidence that allows a full trial to be structured in the most appropriate manner49–51 using empirically generated evidence concerning comparative patient outcomes. In the case where evidence has been generated in a feasibility study in which the EQ-5D appears to adequately capture outcomes, a cost–utility analysis can be adopted as an appropriate methodology for subsequent health economic analyses. If this crucial and indispensable element underpinning the decision to use a cost–utility analysis is found to be erroneous in the light of further evidence gained in a full trial then it would be comparatively simple for the analysis to revert to a cost-effectiveness analysis.

What lessons have been learnt concerning the structure of any future trials of filtration on bypass surgery?

Randomised controlled trials can be structured to evaluate superiority therapeutic equivalence or therapeutic non-inferiority. The greatest support for the use of cost–utility analysis occurs when a feasibility study identifies the EQ-5D as being an effective instrument for identifying and measuring significant variations in quality of life between patients on the two arms of the trial. The implications of adopting an inappropriate clinical trial design or misinterpreting the results of a clinical trial are often considerable:

. . . wrongly discounting treatments as ineffective will deprive patients of better care. Wrongly accepting treatments as effective exposes patients to needless risks and wastes. 52,53

However, such certainty in trial outcomes, is rare and in practice there exists a myriad of ‘grey’ areas that may be indicative of various ‘shades’ of applicability and appropriateness of different structures of economic evaluation, which are likely to require more careful analytical consideration and judgement. In large part, the interpretation of clinical evidence will depend on the specific context of the clinical trial, the range of outcomes being measured and the judgement of the analyst. In such circumstances, although a feasibility study will provide helpful guidance, the results obtained should always be viewed as being indicative rather than definitive.