Cost-effectiveness of a two-layer compression bandage versus standard bandage following total knee arthroplasty

Population and trial details

KreBS was a pragmatic two-arm, open label, parallel-group, randomized controlled trial conducted in 26 UK NHS hospitals, with the primary outcome of the OKS at 12-month follow-up, to represent change in knee function. Patients scheduled for primary TKA aged 18 years and over were included in the study and randomized on a 1:1 basis to receive either a two-layer compression bandage (Coban 2; 3M, UK) or a standard wool and crepe bandage (i.e. the control group). Full details of the trial design and clinical effectiveness results are available elsewhere.⁷

Ethical approval

The study (ISRCTN 87127065) had ethical approval provided by Newcastle & North Tyneside 2 Research Ethics Committee (16/NE/0400).

Data collection

Data were collected by means of participant self-completed questionnaires at baseline and 12-month follow-up, and also via NHS Digital’s patient-reported outcome measures (PROMs) project and local patient administration system (PAS) data from study sites (provided by 66% of sites). Specifically, for the economic analysis, HRQoL, resource use, and cost data were obtained using participant questionnaires; PAS data obtained from study sites were used for initial length of hospital stay, complications, comorbidities, and demographics; adverse event forms were used to capture complications; and additional EuroQol five-dimension (EQ-5D) data were obtained via PROMs.

Analysis overview

The economic evaluation took the form of a cost-utility analysis, which evaluated the compression bandage versus standard bandage in terms of the incremental cost per quality-adjusted life year (QALY). The base case analysis adopted the perspective of the UK NHS and Personal Social Services (PSS), including healthcare services used by participants over the 12-month follow-up period in relation to their knee arthroplasty and the costs of the bandaging systems. A secondary analysis explored the broader societal perspective by also considering indirect costs of lost productivity. Analyses used multiple imputation, with a complete case analysis also undertaken. Intention-to-treat principles were followed and analyses were conducted over a 12-month period, therefore discounting of costs and health outcomes was not necessary. Costs are presented in pounds sterling (£) for 2018 to 2019, with the NHS cost inflation index used where required to inflate costs to this year of pricing.⁸ Sensitivity analyses explored the impact of altering key assumptions in terms of the overall findings. All analyses were conducted using Stata version 16 (StataCorp, USA).

Resource use and costs

Healthcare resource use data were collected for each participant over a 12-month period using self-completed participant questionnaires at 12 months post-randomization. Consultations occurring within primary care and the community (general practitioner (GP), nurse, physiotherapist, occupational therapist visits) and the hospital setting (outpatient attendances, day case visits, inpatient nights, and emergency department (ED) attendances) were recorded. Participants were asked to record resource use relating to their knee arthroplasty and also resource use relating to any ‘other reason’. Only knee arthroplasty-related resource use was included in the base case analysis. As no statistically significant difference was found between the groups in terms of resource use relating to any other reason, no further exploration via sensitivity analysis was undertaken, as set out in the health economics analysis plan. The base case assumed that any resource use relating to complications was captured through participants’ responses to resource use questions in the participant questionnaires. The total cost per participant was calculated by multiplying the resources used by the corresponding unit costs. Unit costs of healthcare resource use items (Table I) were sourced from established costing sources comprising NHS Reference Costs⁹ and PSSRU Unit Costs of Health and Social Care.⁸ In addition to healthcare use, the base case analysis included costs associated with the bandage used (see ‘bandage costs’ for details). A secondary analysis undertaken from a societal perspective considered costs regarding lost productivity, specifically days missed from work. Productivity loss over the 12-month period was estimated by multiplying the total number of missed workdays self-reported by participants by a daily wage of £114 (based on weekly earnings of £569).¹⁰

Bandage costs

Costs associated with the bandage system comprised staff time for bandage application and removal and the cost of the bandage itself. In addition, training was provided for the surgical staff involved in applying the compression bandage; therefore, a training cost was applied for participants who received this type of bandage. The different bandage cost components for both groups are shown in Table II. For training on compression bandage application, information provided by study personnel was used to assume that: each trainee received ten minutes’ training (watching a training video, reading an instruction leaflet, and discussion with other team members); the principal investigator at each site received ten minutes’ support from the trainer; and there were six trainees per site for each of the 26 sites. Costs of the relevant healthcare professionals’ time (Table II), for both trainees and trainers, were attached to the time spent on training, to generate a total training cost. This was divided by the number of participants who received the compression bandage to generate a training cost per participant, and applied to each participant who received a compression bandage.

Estimates of time spent applying and removing bandages, provided by taking an average of responses given by study healthcare staff, were multiplied by the unit cost for the relevant healthcare professional’s time, as detailed in Table II. Bandages were removed between 24 and 48 hours after application. Where bandages were removed less than 24 hours after application and there was relevant information recorded, an additional cost of application and removal was included if bandages were re-applied; study notes were used to attach a cost for the particular bandage that was used for the second application.

Outcome data

Participants’ HRQoL represented by utilities were derived from their EuroQol five dimension five-level (EQ-5D-5L)¹² questionnaire responses, collected at baseline and 12 months using self-completed questionnaires. Responses provided for the five domains of the EQ-5D-5L generate a specific health state, where full health is represented by a utility score of one, death is represented by a score of zero, and negative scores indicate states worse than death. Following UK National Institute for Health and Care Excellence (NICE) guidance regarding use of EQ-5D-5L data for analyses,¹³ utilities have been estimated using the mapping function developed by van Hout et al.¹⁴ Regression methods explored the difference in EQ-5D-5L index scores between the two groups, including covariates consistent with those included in the primary statistical analysis model. Utilities were converted into QALYs using the area under the curve approach,^15,16 and the difference in QALYs between groups over the 12 months was adjusted for baseline utility,¹⁷ hence allowing for any baseline differences between the groups. In addition to the EQ-5D-5L data, further EQ-5D data were collected to enable a comparison of responses obtained from the EQ-5D-5L and EQ-5D-three level questionnaire (3L).

Missing data

Multiple imputation (MI) was used to deal with missing data; MI generates unbiased, plausible estimates for the missing data, dependant on the assumption of the data being missing at random. Logistic regression was used to investigate the plausibility of the missing at random assumption, by exploring the association between missingness (of costs and QALYs) and baseline covariates and observed outcomes. With the exception of the complete case analysis, all analyses used multiple imputation with chained equations (MICE) with predictive mean matching on QALY and cost estimates, thereby ensuring plausible imputation values. The imputation model included age, sex, baseline OKS, study site, utilities (at baseline and 12 months), and total costs at the resource use level. A total of 40 imputations were undertaken for the base case, following guidance that the number of imputed datasets should be at least greater than,¹⁸ or similar to,¹⁹ the proportion of missing data, which was 38% in the base case. This guidance was followed similarly for all sensitivity analyses. Imputed datasets were combined following Rubin’s rules to generate mean cost and QALY estimates.²⁰ Usual imputation methods were followed for participants who died during the study, for any missing data arising before their death. For data that would have been gathered after their death, zero QALYs and resource use were assumed. A complete case analysis was undertaken for comparison, which relies on the data being not missing at random, with available case analysis used for initial investigation of the data.

Data analysis

Mean incremental costs and QALYs for the compression versus standard bandage groups were estimated using seemingly unrelated regression for 10,000 replications, while adjusting for baseline covariates (age, sex, study site, OKS score, and baseline utility), with 95% CIs estimated using bias corrected and accelerated bootstrap methods. Incremental cost-effectiveness ratios (ICERs) were used to express cost-effectiveness, where appropriate, by dividing the incremental costs by the incremental QALYs. A treatment option is found to be dominated if it is more costly and generates fewer QALYs than the alternative. The current threshold of £20,000 to £30,000 per QALY recommended by NICE has been used for the analyses.²¹ Results were also presented in terms of incremental net monetary benefit (NMB),²² using the cost-effectiveness threshold of £20,000 to translate the health benefits into monetary terms. An intervention is deemed cost-effective if the mean incremental NMB is positive. Cost-effectiveness acceptability curves (CEACs) represent uncertainty around the findings.

Analysis of uncertainty

Sensitivity analyses were undertaken to explore the uncertainty around the cost-effectiveness results, while controlling for covariates. Sensitivity analysis (SA)1 comprised a complete case analysis where only participants with complete economic data profiles were included, thereby exploring the impact of excluding participants with missing data. Costs for complications based on PAS data and adverse event forms were included in SA2; specifically study sites provided PAS information about whether participants had encountered myocardial infarction, urinary tract infection, stroke, transient ischaemic attack, thrombocytopenia, pneumonia, or readmission to hospital (all within 30 days of TKA), or a pulmonary embolism or deep vein thrombosis (within 60 days of TKA). SA3 used alternative costs for inpatient nights, outpatient attendances and day case attendances, using the average for all Health Resource Group (HRG) codes, rather than only those that related to knee arthroplasty, as used in the base case (Table III). The cost of initial inpatient stay for TKA surgery using PAS length of stay data was included in SA4. For both SA2 and SA4, the relevant costs for each of the HRG codes reported by PAS data were sourced from NHS Reference Costs and applied accordingly.^9,23

Resource use and costs

The healthcare services used by participants in relation to their knee arthroplasty are presented in Table IV for all available cases. The proportion of responses missing for each of the resource use items was found to be between 19% and 29%. Use of healthcare services was similar for both groups, with participants most commonly accessing healthcare services for physiotherapist visits, hospital outpatient appointments, and inpatient stays in hospital. The largest mean cost differences between the study groups (compression minus standard bandage), based on the multiply imputed dataset, were seen for hospital day case attendances (£45.24), inpatient nights spent in hospital (£11.19), and outpatient appointments (-£13.17) (Table V).

Bandage costs

The mean costs associated with bandages were £15.46 in the compression bandage group and £5.15 in the standard bandage group. These costs incorporate crossovers (e.g. where a participant randomized to receive a compression bandage actually received a standard bandage); costs for a standard bandage were applied. A breakdown of the bandage cost components can be seen in Table V.

Health-related quality of life

Baseline utility levels were very similar across groups, with mean baseline scores of 0.514 (SD 0.215) and 0.511 (SD 0.220) for the compression and standard bandage groups, respectively. At 12 months, mean utilities were 0.705 (SD 0.263) for the compression bandage group and 0.709 (SD 0.255) for the standard bandage groups; hence, utilities remained similar. There was no statistically significant difference in QALYs over the 12 months for all available cases; a mean difference (for compression minus standard) of -0.0027 (95% CI -0.0137 to 0.0082) when controlling for baseline utility, and 0.0003 (95% CI -0.0112 to 0.0118) when controlling for all covariates.

Base case analysis

Participants in the compression bandage group incurred additional costs; a total mean cost of £1,720 (95% CI £1,678 to £1,763) was incurred per participant in the compression bandage group versus £1,669 (95% CI £1,626 to £1,712) for the standard bandage group, based on the multiply imputed dataset (Table VI). In terms of the EQ-5D-5L findings, QALYs were found to be marginally lower in the compression bandage group (0.602 mean QALYs (95% CI 0.599 to 0.606)) compared to participants in the standard bandage group (0.607 mean QALYs (95% CI 0.603 to 0.610)) (Table VI). The incremental analysis found a mean incremental cost of £52.68 (95% CI £50.56 to £54.80) and a mean reduction of -0.0050 (95% CI -0.0051 to -0.0049) QALYs, based on a bivariate model that used seemingly unrelated regression and adjusting for covariates. The incremental net monetary benefit was -£152.01 (95% CI -£155.30 to -£148.73) at the £20,000 per QALY threshold, and -£201.68 (95% CI -£205.85 to -£197.51) at the £30,000 per QALY threshold, thereby indicating that compression bandages do not represent a cost-effective option when compared to standard bandages.

Although the cost-effectiveness point estimate indicates that compression bandages are dominated by standard bandages (i.e. the compression bandage was found to be more expensive and less beneficial in terms of QALYs gained), there was considerable uncertainty around this finding, as illustrated by the point estimates of the 10,000 bootstrap sample estimates in Figure 2. Estimates appear in all four quadrants of the cost-effectiveness plane, although more heavily populated in the north-west and south-west quadrants. Figure 3 illustrates the probability of compression bandages being cost-effective for different willingness-to-pay thresholds. The probability of compression bandages being cost-effective for a threshold of £20,000 per QALY is 18.3%, and 17.4% at the £30,000 per QALY threshold.

Fig. 2

Scatter plot on the cost-effectiveness plane: incremental costs and incremental quality-adjusted life years (QALYs).

Fig. 3

Cost-effectiveness acceptability curve.

Secondary analysis

The secondary analysis (i.e. inclusion of productivity losses) found that the compression bandage group was associated with an additional mean cost of £416 (95% CI 411 to 422) compared with the standard bandage group. This was driven by participants in the compression bandage group encountering more sick leave days than the standard bandage group (23 days (SD 51) versus 18 days (SD 40), based on all available cases; n = 585 in the compression bandage group and n = 507 in the standard bandage group) over 12 months.

Sensitivity analysis

Table VII summarizes the incremental mean QALYs, incremental mean costs, cost-effectiveness finding (i.e. ICER, or dominated), and probability of compression bandages being cost-effective for the four sensitivity analyses. The cost-effectiveness findings remained robust to the sensitivity analyses undertaken, with the compression bandage dominated by the standard bandage for all four sensitivity analyses. The complete case analysis undertaken for SA1 was the only sensitivity analysis to find that the incremental costs and QALYs were not statistically significant, and the probability of being cost-effective at the £20,000 threshold increased to 29%.