Cost-effectiveness of digoxin versus beta blockers in permanent atrial fibrillation: the Rate Control Therapy Evaluation in Permanent Atrial Fibrillation (RATE-AF) randomised trial

The RATE-AF trial

The clinical trial design, protocol and outcomes have previously been published.4 8 In brief, RATE-AF was a pragmatic, NHS-embedded, open-label, blinded, endpoint RCT in patients with permanent AF and symptoms of heart failure in need of heart rate control. All participants provided written informed consent after review of participant information material created by the PPI team. The trial was publicly funded by the UK National Institute for Health and Care Research (CDF-2015-08-074); registration ClinicalTrials.gov (NCT02391337) and clinicaltrialsregister.eu (EudraCT 2015-005043-13).

Participants and randomisation

Recruitment occurred from primary care practices and three hospitals in the West Midlands region of England (2016–2018). Patients had to be diagnosed with permanent AF, aged 60 years or older and with symptoms of possible heart failure with breathlessness equivalent to New York Heart Association class II or above. Exclusion criteria were limited to ensure the population enrolled reflected routine clinical practice (see online supplemental table 1 for full eligibility criteria). Participants were randomised using a telephone or web-based portal to either digoxin (62.5–250 µg daily) or a beta-blocker (bisoprolol 1.25–10 mg daily) in a 1:1 ratio. Alternative beta- blockers were acceptable for those with intolerance to bisoprolol. Note that both randomised groups received an intervention, as necessitated by their clinical status. The vast majority of patients in usual standard care in the UK would receive beta-blockers, with digoxin typically reserved as a second-line agent for patients intolerant or not responding to beta-blocker therapy. Randomisation used a minimisation algorithm to ensure balance between the treatment groups for gender and modified European Heart Rhythm Association (mEHRA) functional class score.

Economic analysis overview

The within-trial economic analysis estimates the cost-utility of digoxin versus beta-blockers for heart rate control from a UK NHS perspective. Costs are expressed in British pound sterling (2019/2020 price year) and health outcomes in quality-adjusted life years (QALYs), as recommended by the National Institute for Health and Care Excellence (NICE).9 The time horizon covered the period from randomisation to the end of follow-up at 12 months, and all data included in the analysis were collected via study data collection forms and self-reported questionnaires. Differences in mean costs and QALYs between treatment arms were used to calculate the incremental cost-effectiveness ratio, or to determine dominance if a comparator was found to be more effective and less costly,10 with NICE thresholds of £20 000 and £30 000 per QALY gained.9

Costs and outcomes were not discounted as the trial was limited to 12 months of follow-up. NICE recommendations were used to guide methods9 and findings reported in accordance with the Consolidated Health Economic Evaluation Reporting Standards. The health economists performing this analysis were independent of the RATE-AF team that deployed the trial.

Resource use and cost

Resource use data were collected prospectively for individual patients, covering primary healthcare, secondary healthcare and medication use. Primary and secondary healthcare resources were considered in the analysis if they were related to AF or other cardiovascular events, in consultation with clinical experts. Data on primary and secondary healthcare service utilisation were collected at the 6 and 12-month follow-up visits using specific case report forms by a study research nurse not involved in the usual clinical care of each patient. All adverse events (as listed in the Summary of Product Characteristics for each drug) were compiled by directly asking participants at each trial visit and review of the electronic medical record. The number and the type of primary care contacts, including general practitioner (GP), nurse or allied health professionals, were involved in discussion with the patient and updated with information from the primary care team. Cardiovascular events and any serious adverse events (SAE) underwent a process of independent clinical adjudication.

SAEs were costed according to the Healthcare Resource Group (HRG) code that included the costs of medical management incurred. An average of the HRG unit costs for each clinical inpatient event was calculated from NHS reference costs.11 To reflect complicated cases with prolonged hospitalisation, the high-end HRG cost that corresponded to the cardiovascular event was used. Hospital stay was assumed to be prolonged if the patients spent more than 10 days in the hospital for a specific cardiovascular event. It was assumed that patients who died of cardiovascular events did not consume any resources after the date of death, therefore missing cost values were replaced with zero. Costs of trial and additional medications were estimated using the average cost of the daily dose multiplied by the average duration of receiving the treatment. Unit costs for all types of health service resources were obtained from Personal Social Services Research Unit, NHS reference costs and the British National Formulary (online supplemental table 2). Unit costs were assigned to health service resources to generate total healthcare costs per patient over 12 months of follow-up.10

Health outcomes

Patient-reported QoL using the EuroQol 5-Domain 5-Level (EQ-5D-5L) questionnaire was the primary health outcome measure for the economic analysis. Patients completed the questionnaire at baseline and 6 and 12 months of follow-up. Individual patient responses were used to calculate health index scores applying the UK crosswalk value set. Score values range from 1 (state of perfect health) to 0 (state of death), with negative scores representing a health state that is worse than death. The area under the curve approach was employed to calculate QALYs based on patient-specific utility scores at each time point over the 12-month period.10 For patients who died due to a cardiovascular event, a utility value of zero was assigned based on the date of death.

Statistical and economic evaluation analyses

The base case analysis was conducted using complete data on both resource use and health outcomes at baseline, 6 months and 12 months. For health outcomes, patient responses to all EQ-5D-5L items at all follow-up points were required for the analysis to generate QALYs, otherwise the data were defined as missing. Similarly, data on resource use were deemed to be missing if any of the follow-up forms were not completed. QALYs, resource use and cost per patient were estimated and presented as means and SD by treatment arm. As cost and health outcome data are likely to be skewed, mean differences between treatment arms with the 95% CIs were calculated from 1000 resamples using bias-corrected and accelerated bootstrapping. The difference in QALYs was adjusted for baseline EQ-5D-5L score, mEHRA and left ventricular ejection fraction (LVEF; analysed from blinded cardiac ultrasound), in addition to age and gender.12 The mean difference in total costs was adjusted for baseline mEHRA and LVEF scores, age and gender.

To account for uncertainty due to sampling, a bootstrapping approach was used to generate 5000 paired estimates of differences in mean costs and health outcomes (QALYs). Estimates were presented as a scatterplot on a cost-effectiveness plane.13 The net monetary benefits were generated from these estimates to calculate the probability of digoxin being cost-effective at different threshold values of willingness to pay per additional QALY, with results plotted as a cost-effectiveness acceptability curve.14 A multiple imputation approach was employed to account for the impact of missing data on the results.15 Missing values were predicted from the observed values using chained equations with predictive mean matching.15 16 The imputation model included gender, age, baseline mEHRA class and LVEF for cost data, and the same variables plus baseline EQ-5D-5L score for QALYs. Treatment allocation was also included as a variable in all multiple imputation models. 10 imputed datasets were used, reflecting the percentage of incomplete cases17 and Rubin’s rule used to combine the imputed datasets into one final imputed variable.18 An additional analysis was conducted according to the intention-to-treat principle including all randomised patients. The imputed data of costs and QALYs were used to determine cost-effectiveness.

The health economic analysis was prespecified in the RATE-AF trial protocol prior to the inclusion of any participants. Extrapolation from the RATE-AF trial to an annual UK NHS cost for AF was calculated post hoc using published sources of data on AF prevalence, medication usage and the cost of managing AF to the NHS (online supplemental table 4).3 17 19–21 A two-tailed p value of <0.05 was considered statistically significant, with data analysis carried out using Stata (V.17.0; StataCorp, Texas).