Multiple frequency bioimpedance devices to guide fluid management in people with chronic kidney disease having dialysis

4.5 Three RCTs reported data on mortality (Onofriescu et al. 2014; Ponce et al. 2014; Huan-Sheng et al. 2016). Use of the BCM – Body Composition Monitor device had no significant effect on mortality rates (pooled hazard ratio 0.69; 95% confidence interval [CI] 0.23 to 2.08; p=0.51) and there was moderate statistical heterogeneity between trials.

4.6 Three non-randomised studies had results for the effects of hydration status on mortality in subgroups of participants monitored with the BCM – Body Composition Monitor device. Kim et al. (2015) reported a higher incidence of mortality in overhydrated participants (defined by relative hydration state; odds ratio 2.57; 95% CI 1.08 to 6.13; p=0.033). In O'Lone et al. (2014), absolute overhydration had a significant effect on the risk of mortality (hazard ratio 1.10; 95% CI 1.01 to 1.20; p=0.025) and Wizemann et al. (2009) reported that hydration state was an important predictor of mortality in patients having haemodialysis (adjusted hazard ratio 2.10; 90% CI 1.39 to 3.18; p=0.003).

4.7 Huan-Sheng et al. (2016) reported significant differences in intradialytic complications between people monitored with and without the BCM – Body Composition Monitor device. But incidences were not consistently higher in 1 group. For people monitored using BCM – Body Composition Monitor, significantly higher incidences of cramping, chest tightness and headaches were reported. However, significantly lower incidences of complications caused by hypotension during dialysis sessions and skin itching were reported. The difference in the number of patient-reported events of intradialytic fatigue in participants monitored with and without the BCM – Body Composition Monitor was not statistically significant (p=0.7).

4.8 Hur et al. (2013) reported no significant difference in the frequency of intradialytic events between groups monitored with and without the BCM – Body Composition Monitor device at 12 months (66.6 and 63.9 events per 1,000 dialysis sessions respectively; p=0.4). Similarly, Onofriescu et al. (2014) found no significant difference in the incidence of hypotension or cramps (p=0.6). Ponce et al. (2014) reported no significant difference in the incidence of hypotensive events (defined as a drop in systolic blood pressure during dialysis by at least 30 mm of mercury [Hg] or to below 90 mm Hg) at 12 months.

4.9 One RCT (Huan-Sheng et al. 2016) reported the incidence of cardiovascular-related events, although this was in combination with the incidence of acute fluid overload events. The incidence rate in people monitored with the BCM – Body Composition Monitor device was significantly lower than the control group (incidence rate ratio 0.50 per patient-year; 95% CI 0.26 to 0.94; p=0.03).

4.10 Three non-randomised studies gave the incidence of cardiovascular events among subgroups of people monitored using the BCM – Body Composition Monitor device. Kim et al. (2015) reported no statistically significant difference in the number of cardiovascular events per year between overhydrated and non-overhydrated subgroups as determined by the level of relative overhydration (p=0.13). Onofriescu et al. (2015) also found no statistically significant difference in the incidence of coronary heart disease, peripheral vascular disease, heart failure or stroke between subgroups with lower relative fluid overload (less than 17.4%) and higher relative fluid overload (over 17.4%). Hoppe et al. (2015) reported a non-significant difference in the incidence of acute myocardial infarction and stroke between people who had been having dialysis for a shorter length of time (short dialysis vintage) and people who had been having dialysis for a longer length of time (long dialysis vintage).

4.11 No RCTs gave data on residual renal function, although 2 reported urinary output which could be used as a surrogate measure. Hur et al. (2013) found a significant increase in the proportion of patients with anuria, that is when the kidneys no longer produce urine, and a significant decrease in urine output in patients without anuria in a group monitored using the BCM – Body Composition Monitor device. In the corresponding control group, there was no change in the proportion of patients with anuria and the decrease in urine output seen in patients without anuria was not significant. Luo et al. (2011) reported non-significant decreases in urine volume in groups monitored with and without the BCM – Body Composition Monitor device.

4.12 All 6 included RCTs reported systolic blood pressure measurements. Use of the BCM – Body Composition Monitor device was associated with a significantly lower systolic blood pressure in a meta-analysis (pooled mean difference −3.48 mm Hg; 95% CI −5.96 to −1.00; p=0.006). When data from Onofriescu et al. (2012) was removed from the meta-analysis, the effect size of BCM – Body Composition Monitor-guided monitoring was reduced and was no longer significant (pooled mean difference −2.46 mm Hg; 95% CI −5.07 to 0.15; p=0.06).

4.13 The external assessment group (EAG) also did a subgroup analysis of systolic blood pressure according to the type of dialysis: peritoneal dialysis (1 RCT) or haemodialysis (5 RCTs). In the haemodialysis subgroup, use of the BCM – Body Composition Monitor device was associated with a significant decrease in systolic blood pressure (pooled mean difference −3.09 mm Hg; 95% CI −5.88 to −0.31; p=0.03). For patients having peritoneal dialysis, Luo et al. (2011) reported a mean decrease in systolic blood pressure of −6.08 mm Hg (95% CI −12.57 to 0.41) associated with use of the BCM – Body Composition Monitor device.

4.14 Four non-randomised studies gave data on blood pressure among subgroups of people monitored using the BCM – Body Composition Monitor device. No statistically significant differences in blood pressure were seen in the following subgroup comparisons: patients in whom the average overhydration decreased within 6 months compared with those in whom it did not decrease (Castellano et al. 2014), patients who had been having dialysis for a short length of time compared with those who had been having it for longer (Hoppe et al. 2015), and patients with a high relative fluid overload (more than 17.4%) compared with those in whom it was low (less than 17.4%; Onofriescu et al. 2015). Kim et al. (2012) reported that systolic blood pressure was higher in hyperhydrated patients when compared with dehydrated patients (significance not stated).

4.15 Three RCTs gave data on carotid-femoral pulse wave velocity as a surrogate for arterial stiffness (Hur et al. 2013; Onofriescu et al. 2012; Onofriescu et al. 2014) and were included in a meta-analysis. Arterial stiffness is thought to be associated with an increased risk of cardiovascular events in the longer term. Pulse wave velocity was significantly reduced in patients who were monitored using the BCM – Body Composition Monitor device and standard clinical assessment compared with standard clinical assessment alone (mean difference −1.53 meters per second [m/s]; 95% CI −3.00 to −0.07; p=0.04). There was high statistical heterogeneity between the studies. If data from Onofriescu et al. (2012) were removed from the meta-analysis, the pooled effect was no longer significant (mean difference −1.18 m/s; 95% CI −3.14 to 0.78; p=0.24).

4.16 Five RCTs (Huan-Sheng et al. 2016; Hur et al. 2013; Luo et al. 2011; Onofriescu et al. 2012; Ponce et al. 2014) assessed absolute overhydration; that is, the volume of fluid by which the participants were above their target volume (as assessed by the BCM – Body Composition Monitor device). No data on underhydration were available. A meta-analysis of the mean difference in absolute overhydration volumes showed that absolute overhydration was significantly lower in groups monitored with the BCM – Body Composition Monitor device (mean difference = −0.39 litres, 95% CI −0.62 to −0.15, p=0.001).

4.17 The EAG did a subgroup analysis for absolute overhydration, as assessed by the BCM – Body Composition Monitor device, according to type of dialysis. They compared the pooled effect of using the BCM – Body Composition Monitor device on absolute overhydration in the overall group (all 5 studies) and a subgroup of studies on people having haemodialysis (4 of these studies). A difference in effect between the overall and haemodialysis subgroup was seen, but the EAG stated that this was not large enough to suggest a significant dialysis effect.

4.18 Four RCTs had results for relative overhydration (Huan-Sheng et al. 2016; Onofriescu et al. 2012; Onofriescu et al. 2014; Ponce et al. 2014); that is, a person's absolute overhydration volume normalised against their total extracellular body water volume (both volumes assessed by the BCM – Body Composition Monitor device). A meta-analysis of the reported mean differences in the relative overhydration between groups monitored with and without the BCM – Body Composition Monitor device showed that relative overhydration was significantly lower when the BCM – Body Composition Monitor device was used (mean difference = −1.54; 95%CI −3.01 to −0.07; p=0.04).

4.19 Three RCTs gave data on hospitalisations. Huan-Sheng et al. (2016) reported that the difference in all-cause hospitalisation in patient groups monitored with and without the BCM – Body Composition Monitor device was not significant (hazard ratio 1.19; 95% CI 0.79 to 1.80). Hur et al. (2013) found that the difference in rates of hospitalisation caused by new cardiovascular events in the control and BCM – Body Composition Monitor monitored groups was not statistically significant. In Ponce et al. (2014), 39.6% of participants in the BCM – Body Composition Monitor monitored group and 31.8% of the standard clinical assessment group were hospitalised at least once.

4.20 Two non-randomised studies gave data on hospitalisation. Kim et al. (2015) found no significant differences in the number of hospital days per event between overhydrated and non-overhydrated groups (as determined by the BCM – Body Composition Monitor device). Onofriescu et al. (2015) found a significantly higher all-cause hospitalisation rate for patients classified as overhydrated when a relative overhydration value of 17.4% was used as a cut-off to define people as overhydrated, but not when a value of 15% was used.

4.21 Measures of left ventricular hypertrophy, and surrogates of this, such as left ventricular mass index, may be associated with longer-term cardiac morbidity. Hur et al. (2013) reported that left ventricular hypertrophy was present at 12 months in 44% of participants monitored using the BCM – Body Composition Monitor device and in 50% of participants monitored using standard clinical assessment alone. This was a non-significant reduction from baseline in both groups (67% and 53% respectively). But there was a statistically significant reduction in left ventricular mass index from baseline in the group monitored using the BCM – Body Composition Monitor device (p<0.001), although not in the group monitored using standard clinical assessment (p=0.9).

4.22 Two non-randomised studies gave data on the use of antihypertensive medication in subgroups of people monitored using the BCM – Body Composition Monitor device. In Castellano et al. (2014), consumption of antihypertensive medication was significantly higher in a subgroup of patients who did not have reduced relative overhydration after 6 months of monitoring. Kim et al. (2012) found no significant difference in medication use between people who were dehydrated or hyperhydrated.

4.27 A Markov model was developed to simulate the effects of monitoring the fluid status of cohorts of people on dialysis, using a multiple frequency bioimpedance device with standard assessment or by standard assessment alone. The model was run as a cohort simulation over a 30‑year time horizon for a 66 year old mixed dialysis population in the base-case analyses. All future costs and benefits included in modelling were discounted at a rate of 3.5% per annum.

4.28 In the model, people started in a stable state on either haemodialysis or peritoneal dialysis and over time could either stay in that state or move to others when events (such as a kidney transplant, cardiovascular event or death) happened. These events could occur in each cycle of the model, which was set as 3 months. The characteristics of the cohort of patients modelled (for example, their age, the proportion of people on haemodialysis or peritoneal dialysis, gender, and incidence of comorbidities) were based on the UK Renal Registry Report (2015). Mortality rates and hospitalisation rates were informed by a combination of European (ERA-EDTA Registry Annual Report 2013) and UK Renal Registry data, and other published sources.

4.29 The model also had an option to allow people in the 'stable' and 'post-CV event' dialysis states to be further classified as either severely overhydrated or normohydrated (based on their relative overhydration). This allowed scenarios to be run in the model in which mortality and hospitalisation rates were increased for dialysis patients who were overhydrated. No 'underhydrated' state was included because of a lack of evidence on the prevalence of underhydration in UK dialysis cohorts, the effect of underhydration on the risk of adverse events and quality of life, and the effectiveness of the BCM – Body Composition Monitor device in reducing the incidence of underhydration.

4.30 Parameter values used in the model were taken from focused reviews of the literature to identify baseline risks for mortality and hospitalisation, and also sources for cost and utility data, and the clinical-effectiveness review. Several possible outcomes that may be affected by using the BCM – Body Composition Monitor device were not included in base-case modelling because of a lack of evidence. These included changes in quality of life (independent of effects of hospitalisation and cardiovascular events), maintenance of residual renal function and effects on dialysis requirements (number and duration of sessions).

4.31 The clinical-effectiveness review only found data on using the BCM – Body Composition Monitor, therefore only this device has been assessed in base-case analyses. Several scenarios were used to model the effect of BCM – Body Composition Monitor-guided fluid management on baseline model parameters. Direct evidence was only available for the effect of BCM – Body Composition Monitor-guided monitoring on all-cause mortality. Several identified trials also reported the effects of BCM – Body Composition Monitor-guided monitoring on surrogate endpoints, such as pulse wave velocity as a measure of arterial stiffness. The EAG did a further literature search to identify evidence that could be used to link changes in these surrogate endpoints to final health outcomes. Using this linked approach, estimated effects of BCM – Body Composition Monitor-guided monitoring on mortality and non-fatal cardiovascular events were calculated. The EAG also modelled an effect of assuming that BCM – Body Composition Monitor-guided monitoring reduced the proportion of people who were seriously overhydrated (with relative overhydration over 15%). This was applied by classifying people in dialysis states in the model as either severely overhydrated or normohydrated, which allowed mortality and hospitalisation rates to be adjusted upwards for proportions of people in the dialysis cohorts who were estimated to be severely overhydrated. Table 1 gives a summary of the relative effects applied to different parameters in the base-case scenario analyses.

Abbreviations: CI, confidence interval; CV, cardiovascular; HR, hazard ratio.

4.32 The model incorporates health service costs associated with maintenance dialysis, blood pressure medication, erythropoietin stimulating agents, all-cause inpatient hospitalisation, renal transplantation (including work-up, surgery and follow-up), post-transplantation immunosuppression and outpatient visits. Dialysis costs, per session (haemodialysis) or per day (peritoneal dialysis), were taken from NHS reference costs (2014 to 2015). For haemodialysis, the average cost of £154 per haemodialysis session was calculated based on the cost per type of session, at home or at a unit, weighted by relative incidence. For peritoneal dialysis, the average cost per day of £69 was taken from the NHS reference costs.

4.33 Costs of bioimpedance monitoring included in modelling were purchase costs for devices (annuitised over 5 years), maintenance costs, staff costs related to using the device, training costs and device consumable costs (such as electrodes). The costs of the bioimpedance devices are shown in table 2.

4.34 Health state utility values for people on dialysis and post-renal transplant were identified through a focused search of the literature. Two systematic reviews were found that published EQ‑5D data for UK patients (Liem et al. 2008; Wyld et al. 2012). Further searches did not identify any other studies reporting EQ‑5D data for UK patients after 2010 (the end date for searches in the most recent systematic review). Short and longer-term utility multipliers associated with cardiovascular events were calculated based on data from the Health Survey for England (2003 and 2006). Decreases in health state utilities resulting from hospitalisations were taken from the NICE guideline on peritoneal dialysis.

4.40 No clinical-effectiveness data were found for the InBody S10 or MultiScan 5000. These devices were therefore not included in base-case cost-effectiveness modelling. But they were included in scenario analyses which assumed that these devices reduced mortality and non-fatal cardiovascular events to the same extent as the BCM – Body Composition Monitor device in scenario 3 (but with different costs). The ICERs produced for these devices were very similar, being between £15,000 and £16,000 per QALY gained.

4.42 One-way sensitivity analyses were done on model parameters for base-case scenario 3 (both with and without dialysis costs). When dialysis costs were included, adjusting the hazard ratio for all-cause mortality to 1.00 resulted in the most favourable ICER for BCM – Body Composition Monitor-guided monitoring. This was because these people have the same survival as those having standard monitoring, and therefore do not have higher dialysis costs, but do have the benefit of a reduced cardiovascular hospitalisation rate. When dialysis costs are included, ICERs produced by varying model parameters within their specified ranges generally stayed above £30,000 per QALY gained.

4.43 When dialysis costs were not included, the ICERs stayed sensitive to varying all-cause mortality. But, in this case, the least favourable ICER occurs when the hazard ratio is equal to 1.00.

4.44 The EAG did probabilistic sensitivity analyses for base-case scenarios 1, 3 and 4 (both with and without dialysis costs included). Results are shown in table 4. The probabilistic ICERs produced for all 3 base-case scenarios were similar to the deterministic ICERs (shown in table 3 above). If dialysis costs were included, the probability of BCM – Body Composition Monitor-guided monitoring being cost effective at a maximum acceptable ICER of £20,000 per QALY gained was 26% in scenario 1 and less than 6% in scenarios 3 and 4. If dialysis costs were excluded, BCM – Body Composition Monitor-guided monitoring was 67% to 75% likely to be cost effective at this maximum acceptable ICER in the 3 scenarios. The EAG warned that the uncertainty in the parameters produced by linking the effects of monitoring with the BCM – Body Composition Monitor device on arterial stiffness to mortality and non-fatal cardiovascular events (as in base-case scenarios 3 and 4) may not be fully captured in the probabilistic modelling.

Abbreviations: BCM, BCM – Body Composition Monitor; ICER, incremental cost-effectiveness ratio; QALY, quality-adjusted life year.

4.45 As noted in the clinical-effectiveness section, removing the Onofriescu et al. (2012) data from meta-analysis reduced the estimated effect of BCM – Body Composition Monitor-guided monitoring on reducing arterial stiffness. Because the pooled estimate of arterial stiffness was used to estimate the relative treatment effects of the BCM – Body Composition Monitor in modelling (in base-case scenarios 2, 3 and 4), a revised cost-effectiveness analyses was done with BCM – Body Composition Monitor-guided modelling assumed to have a smaller and more uncertain effect on hospitalisation for cardiovascular events and mortality. Similar ICERs were produced for revised base-case scenarios 2, 3 and 4 and also for most of the further revised sensitivity, subgroup and scenario analyses. But there was greater uncertainty about the cost-effectiveness results in the revised probabilistic analyses. When dialysis costs were included, the probability of BCM – Body Composition Monitor being cost effective increased from less than 6% to about 13% for scenarios 3 and 4. When dialysis costs were excluded, the probability of BCM – Body Composition Monitor being cost effective decreased for revised scenarios 3 and 4 (from about 72% to about 62%). This reflected the greater uncertainty in the effect of BCM – Body Composition Monitor-guided monitoring on reducing arterial stiffness, and so the linked effect on all-cause mortality and hospitalisation for cardiovascular events.