Cost-utility analysis of cast compared to removable brace in the management of adult patients with ankle fractures

Background of trial

Data were available from the Ankle Injury Rehabilitation (AIR) trial; its protocol and trial results are reported in detail elsewhere.^6,7 AIR was a UK-based, pragmatic, multicentre, RCT that recruited patients between October 2017 and September 2019. Patients were eligible for inclusion if they were aged 18 years and over and presented to the hospital with an ankle fracture, whether treated operatively or nonoperatively, for which the clinician would consider cast a reasonable management option. Further details on eligibility can be found in the protocol⁷ and the main clinical results.⁶ The AIR trial was approved by the UK National Research Ethics Committee (17/WM/0239).

Trial recruitment and demographics

There were 334 patients randomized to cast and 335 patients randomized to removable brace. The mean age of participants was 46.3 years (standard deviation (SD) 17), and more than half the participants were female (57%). All participants had a clear ankle fracture on a radiograph, with 624 (93%) showing lateral malleolar involvement, 194 (29%) medial malleolar involvement, and 120 (18%) posterior malleolar involvement. Surgery was performed in 364 of 669 the randomized participants. Further details on trial recruitment and demographic information of participants can be found in the clinical paper.⁶

Resource use data

Data were recorded using case report forms at six, ten, 16, and 24 weeks, and 12 months post-randomization. Use of medication, outpatient visits, community care services, and in-patient care were recorded by participants. Non-medical health care resource use, such as aids and adaptations and personal social services (PSS), were also recorded. Data on lost productivity due to injury were collected at each time point. The resource items collected reflects an NHS and PSS perspective, and a broader societal perspective, additionally including lost productivity costs and private health and non-medical healthcare costs incurred.

Collection and valuation of resource use

We used methods recommended by the National Institute of Health and Care Excellence (NICE) guide to methods of technology appraisal.⁸ Medication used was identified and valued using 2019/2020 NHS Prescription cost analysis data for England,⁹ and the 2019 British National Formulary.¹⁰ Inpatient care procedures were derived from the relevant Health Care Resource Group (HRG) and OPCS Classification of Interventions and Procedures version 4 (OPCS4) codes of reported procedures and/or clinical diagnosis. These procedures were valued using Department of Health and Social Care Reference Costs.¹¹ Costs associated with out-patient services were derived using Department of Health and Social Care Reference Costs. Unit costs for community and PSS were obtained from the Personal and Social Service Research Unit (PSSRU) cost compendium.¹² Costs for aids and adaptations used were obtained from the NHS supply chain catalogue 2019.¹³

Costs were calculated by multiplying resource use by item unit costs, in pounds sterling using the 2019/2020 prices. If necessary, costs were inflated or deflated using the Health and Community Health Prices Pay and Price Inflation Index.¹³ No discounting of costs was applied since cost-effectiveness was determined in the first year following treatment.

For the base case analysis NHS and PSS costs were included. Non-prescribed medications, private physiotherapy services, and services not normally provided by the NHS were considered private expenses. Costs associated with lost productivity were estimated using the human capital approach, where median daily wage was multiplied by number of days off work due to injury.

Health outcomes

The primary health outcome for the economic evaluation was the QALY in accordance with NICE guidelines.⁸ Health-related quality of life of trial participants was assessed at baseline (pre and post injury), six, ten, 16, and 24 weeks, and 12 months post-randomization using the EuroQol five-dimension five-level (EQ-5D-5L) instrument.¹⁴ The EQ-5D-5L defines health in relation to five domains: mobility, self-care, usual activities, pain and discomfort, anxiety, and depression. Each domain is expressed in five levels of increasing severity ranging from no problems to extreme problems.

Responses to the EQ-5D-5L instrument were converted to utilities by mapping responses from the EQ-5D-5L to the EuroQol five-dimension three-level (EQ-5D-3L) valuation set using a mapping function¹⁵ as recommended by NICE.¹⁶ QALYs were generated for each patient using the trapezoidal rule.

Statistical analysis

Analysis followed intention to treat (ITT) principles, providing summaries, and estimates of effects based on allocated (as per randomization) treatment groups. Free text resource-use responses were recoded within cost categories where possible or removed if they were deemed not relevant to the trial treatments by clinical experts. Resource use categories, costs, and utilities were summarized by trial arm and differences were tested for statistical significance by independent-samples t-test. A two-sided significance level of 5% was used.

Cost-utility analysis

Cost-effectiveness is presented as an incremental cost-effectiveness ratio (ICER), calculated as the ratio between of the mean difference in costs and QALYs. The removable brace was selected as the test intervention and cast as usual care (control). For the base case model, 5000 bootstrapped replicates of incremental costs and QALYs were used to populate the ICER plane, reflecting sampling uncertainty. The cost-effectiveness acceptability curve (CEAC) was generated showing the probability that brace is cost-effective compared to cast at different willingness-to-pay thresholds. The net monetary benefit (NMB) of using a removable brace was calculated at the different willingness-to-pay thresholds, where a positive incremental NMB would indicate that the brace is cost-effective compared to the cast. The expected value of perfect information (EVPI) was calculated and represented graphically, indicating the monetary value of removing all uncertainty from the cost-effectiveness estimate at a range of willingness to pay thresholds.

Missing data

Missing costs and health utility data were imputed under the missing at random (MAR) assumption, at each time point using fully conditional multiple imputation by chain equations implemented through the MICE package in Stata 16 (StataCorp, USA). The appropriateness of using MAR assumption was evaluated by investigating the missing data patterns and comparing attributes with and without missing cost and health-related quality of life data.

The multiple imputation model used baseline covariates (age, sex, and ankle fracture treatment, either operatively managed or nonoperatively managed). Unobserved costs and QALYs were imputed at each follow-up time point using observed values. Iterations of the imputation model were assessed for convergence, and the distribution of observed, imputed, and completed data for costs and QALYs were compared graphically.

Imputation was conducted separately for each trial arm in line with best practice,¹⁷ generating 50 imputation draws. The number of draws was initially selected reflecting the extent of missing data and verified by assessing the fraction of missing information reported by the base case model. Bivariate regression using seemingly unrelated regression (Sureg) was used to estimate costs and QALYs, adjusting for baseline covariates (age, sex, and ankle fracture treatment) and baseline utility. The Mi function in Stata provided estimates for the imputed model applying Rubin’s rule,¹⁸ reflecting the variability within and across imputations. Joint distributions of costs and outcomes were generated through non-parametric bootstrapping of the imputed model, and incremental costs and QALYs were estimated.

Sensitivity analysis

Sensitivity analyses were conducted to test the robustness of the cost-effectiveness estimates. These include re-estimating the cost-effectiveness estimates (i) using complete case data (ii) adopting a wider societal perspective, including private costs incurred and productivity losses due to work absences.

Completeness of data

Of 669 participants randomized, data for 228 were missing at the six-week follow-up time and 285 were missing at the 12-month follow-up. The level of missingness of data was similar in each treatment arm. Imputation was done at the aggregate resource use level at each follow-up timepoint. A detailed breakdown of the missingness of health economic data for each resource use category, and at each follow-up time point, can be found in Supplementary table i.

Health and social care costs and resource use

There were no statistically significant differences in resource use categories between the treatment groups. A detailed breakdown of the health and social care resource use by follow-up timepoints in each group can be found in Supplementary Material table ii. At follow-up time periods between baseline and 12 months, the mean NHS and PSS costs for those with complete data were also generally similar with small, non-significant differences. Intervention costs were £108.64 for removable brace and £35.71 for cast.¹⁹ Over the entire follow-up period, mean total NHS and PSS costs were non-significantly higher in the brace group compared to the cast group. With the addition of intervention costs,¹⁹ the mean between group difference for those with complete data was £278.08 (95% CI £-153.80 to £709.95).

Table I reports the mean cost for each resource use category, alongside productivity loses, over the follow-up period for cases with complete data. A detailed breakdown of mean NHS and PSS costs, and productivity loses at each follow-up timepoint can be found in Supplementary table iii.

Health outcomes

Table II summarizes the EQ-5D utility estimates and estimated mean difference between the two trial groups for each study period. There were no statistically significant differences in EQ-5D utility between treatment groups at any time point. The mean QALYs for brace recipients were similar to cast (0.723 vs 0.720; p = 0.890). The mean difference of 0.013 was not statistically significant at a 5% level.

Cost-utility analysis

Cost effectiveness results are presented in Table III with removable brace as the intervention and cast as the control treatment. Point estimates of the incremental costs and QALYs (and 95% CIs) are presented, generated from the bootstrapped, imputed model. The probability of brace being cost-effective is presented at cost-effectiveness thresholds (£15,000, £20,000, and £30,000); the mean net monetary benefit (and 95% CI) is similarly presented. The analytic time horizon covers the period from baseline to 12 months.

Participants in the brace group had a small non-significant increase in cost (£46.73; 95% CI £-9 to £147) a small and non-statistically significant increase in QALYs (0.0141; 95% CI -0.005 to 0.033). Together, these increases result in a base case ICER of £3318 per QALY gained from the NHS and PSS perspective. Assuming thresholds of £15,000/QALY, £20,000/QALY and £30,000/QALY, the probability that the brace is cost-effective was 0.81, 0.84, and 0.88 respectively. The mean net monetary benefit was positive at the reported thresholds and the expected value of perfect information was £17.60 per patient at a threshold of £30,000/QALY. supported the base case findings showing brace was cost-effective at the £30,000/QALY threshold. For the complete case analysis, the ICER increased to £21,499/QALY, with a lower probability that the use of brace was cost-effective (p = 0.578 at £30,000/QALY willingness-to-pay). Figures 1 to 4 visualizes the ICER plane, EVPI, as well as the probability that brace is cost-effective and net monetary benefit of brace at varying willingness-to-pay thresholds (£0 to £100,000/QALY). Use of the brace is likely to be cost-effective at a cost-effectiveness threshold of £30,000/QALY in the population studied. However, the ICER is composed of small non-significant increases in cost and QALYs. Sensitivity analysis conducted (complete case and societal perspective) supported the base case findings showing brace was cost-effective at the £30,000/QALY threshold. For the complete case analysis, the ICER increased to £21,499/QALY, with a lower probability that the use of brace was cost-effective (p = 0.578 at £30,000/QALY willingness-to-pay).

Fig. 1

Incremental cost-effectiveness ratio plane with 95% confidence region.

Fig. 2

Cost-effectiveness acceptability curve.

Fig. 3

Net bonetary benefit with 95% confidence interval.

Fig. 4

Expected value of perfect information.