Adjunctive rifampicin to reduce early mortality from Staphylococcus aureus bacteraemia: the ARREST RCT

Article history

The research reported in this issue of the journal was funded by the HTA programme as project number 10/104/25. The contractual start date was in July 2012. The draft report began editorial review in November 2017 and was accepted for publication in July 2018. The authors have been wholly responsible for all data collection, analysis and interpretation, and for writing up their work. The HTA editors and publisher have tried to ensure the accuracy of the authors’ report and would like to thank the reviewers for their constructive comments on the draft document. However, they do not accept liability for damages or losses arising from material published in this report.

Declared competing interests of authors

Alexander Szubert reports grants from National Institute for Health Research (NIHR) and the Medical Research Council (MRC) during the conduct of the study. Robert Tilley reports personal fees from NIHR Clinical Research Network outside the submitted work. Peter Wilson reports personal fees from 3M Advisory Panel, Roche Drug Safety Monitoring Board and Merck Sharp & Dohme Corp. (MSD; Hoddesdon, UK) outside the submitted work. Achyut Guleri reports receiving fees from Novartis as a member of advisory boards and speaker panels, and consultancy fees from Astellas (Chertsey, UK), AstraZeneca (Cambridge, UK), MSD and Schering-Plough (Hoddesdon, UK); he also received support to attend scientific conferences, including accommodation and travel payments from Becton Dickinson (Winnersh Triangle, UK), Carefusion UK (Winnersh Triangle, UK), Janssen-Cilag (High Wycombe, UK) and MSD. Paul R Chadwick reports non-financial support from Novartis (Frimley, UK), and grants and personal fees from NIHR outside the submitted work. Bernadette Young reports grants from the Wellcome Trust outside the submitted work. M Estee Török reports grants from Academy of Medical Sciences/The Health Foundation, grants from Medical Research Council (MRC), grants from NIHR Cambridge Biomedical Research Centre, grants from MRC/Department of Biotechnology Partnership Grant, and personal fees from Oxford University Press outside the submitted work. Martin J Llewelyn reports personal fees from Pfizer (Walton Oaks, UK) outside the submitted work and is a member of the panel that develops the European Society of Clinical Microbiology and Infectious Diseases/Infectious Diseases Society of America clinical practice guideline on Staphylococcus aureus bacteraemia.

Copyright statement

© Queen’s Printer and Controller of HMSO 2018. This work was produced by Thwaites et al. under the terms of a commissioning contract issued by the Secretary of State for Health and Social Care. This issue may be freely reproduced for the purposes of private research and study and extracts (or indeed, the full report) may be included in professional journals provided that suitable acknowledgement is made and the reproduction is not associated with any form of advertising. Applications for commercial reproduction should be addressed to: NIHR Journals Library, National Institute for Health Research, Evaluation, Trials and Studies Coordinating Centre, Alpha House, University of Southampton Science Park, Southampton SO16 7NS, UK.

2018 Queen’s Printer and Controller of HMSO

Background

Staphylococcus aureus bacteraemia is one of the most common and serious bacterial infections worldwide. There were over 12,000 cases of S. aureus bacteraemia in the UK in 2016/17, and around 25% of patients died. 2,3 Current treatment guidelines recommend that S. aureus bacteraemia should be treated with ≥ 14 days of an intravenous (i.v.) beta-lactam antibiotic, or a glycopeptide if the bacterium is meticillin resistant. Combination antimicrobial therapy is generally not recommended, except in severe meticillin-resistant Staphylococcus aureus (MRSA) infections (e.g. endocarditis) or in the presence of prosthetic joint infections. 4–7 Most of the recommendations are based on uncontrolled observational studies and clinical experience, and views of how to manage S. aureus bacteraemia differ widely. 8,9

Rationale

The results of the meta-analysis, together with data from observational studies indicate, that adjunctive rifampicin may have a surprising and substantial impact on survival from S. aureus bacteraemia. They do not, however, constitute evidence of sufficient rigour to influence current treatment guidelines, clinical practice, or indeed the equipoise of clinicians recruiting patients into the proposed trial – even clinicians in centres using rifampicin in a greater proportion of patients have indicated their willingness to randomise, as they recognise the lack of evidence supporting their practice. In particular, although statistically significant, the results from the trial meta-analysis are not convincing, as they are based on a small number of patients in a small number of trials over a wide period of time. In addition, the potential negative impact of rifampicin toxicity, interactions and resistance cannot reliably be assessed in these studies. Current guidelines recommend adjunctive rifampicin only for the treatment of severe MRSA infections, specifically endocarditis, bone and joint infections, and infections involving prostheses (category II evidence). 5,7 But with weak support for these recommendations it is unsurprising that few physicians follow them in practice. The ARREST (Adjunctive Rifampicin to Reduce Early mortality from STaphylococcus aureus bacteraemia) trial was designed to provide a definitive answer to the role of adjuvant rifampicin therapy in the treatment of S. aureus.

Objectives

The hypothesis addressed by the ARREST trial is that adjunctive rifampicin will enhance the killing of S. aureus early in the course of antibiotic treatment, sterilise infected foci and blood faster, and thereby reduce the risk of dissemination, metastatic infection and death. Therefore, the primary objective of the trial was to investigate the impact of adjunctive rifampicin on bacteriologically confirmed failure/recurrence or death through 12 weeks from randomisation. Secondary objectives included evaluating the impact of rifampicin on all-cause mortality up to 14 days from randomisation, or clinically defined failure/recurrence or death, toxicity [serious and grade 3/4 adverse events (AEs), any modification of treatment as a result of drug interactions], emergence of resistance and duration of bacteraemia, and assessing the cost-effectiveness of adjunctive rifampicin for S. aureus bacteraemia in the NHS.

Substudies

There were three ancillary studies to the main trial. First, with assistance from the trial public and patient representative, Jennifer Bostock, the process of obtaining consent to enter the trial was examined. Patients/legal representatives who did not consent to participation in the trial were offered the opportunity to complete a questionnaire exploring reasons for this; participants/legal representatives at one trial centre who did consent were offered the opportunity to be interviewed by the ARREST trial patient and public representative to explore their experiences of trial participation.

Samples were collected for two further ancillary studies for which funding will be sought separately. Participants enrolled at Guy’s and St Thomas’ NHS Foundation Trust, Cambridge University Hospitals NHS Trust, Oxford University Hospitals NHS Trust, The Royal Liverpool and Broadgreen University Hospitals NHS Trust, and Brighton and Sussex University Hospitals NHS Trust were approached for additional consent for a pharmacokinetic/pharmacodynamic (PK/PD) substudy – a population PK/PD study of rifampicin, flucloxacillin and vancomycin for the treatment of S. aureus. The aim of the substudy is to determine the pharmacological parameters of rifampicin that best predict treatment success and provide a rational basis from which optimal dose, frequency and route of administration can be modelled statistically and/or explored in future studies.

All participants were also approached for additional consent for the host DNA/RNA substudy to investigate the influence of host and bacterial genetics on disease severity and outcome from S. aureus. The aim is to identify host and bacterial genetic factors that influence disease severity (e.g. the development of metastatic complications) and poor outcomes from S. aureus bacteraemia.

The samples for the PK/PD and DNA/RNA substudies have been archived at the King’s College London Biobank until funding has been secured.

Trial setting

Patients were recruited from 29 large UK NHS Hospital Trusts:

• Guy’s and St Thomas’ NHS Foundation Trust

• Oxford University Hospitals NHS Trust

• University College London (UCL) Hospitals NHS Foundation Trust

• Royal Free London NHS Foundation Trust

• King’s College Hospital NHS Foundation Trust

• Brighton and Sussex University Hospitals NHS Trust

• The Royal Liverpool and Broadgreen University Hospitals NHS Trust

• Sheffield Teaching Hospitals NHS Foundation Trust

• Cambridge University Hospitals NHS Foundation Trust

• Royal United Hospital Bath NHS Trust

• Royal Devon and Exeter NHS Foundation Trust

• Plymouth Hospitals NHS Trust

• Hull and East Yorkshire Hospitals NHS Trust

• South Tees Hospitals NHS Foundation Trust

• Heart of England NHS Foundation Trust

• St George’s Healthcare NHS Trust

• Portsmouth Hospitals NHS Trust

• University Hospital Southampton NHS Foundation Trust

• Blackpool Teaching Hospitals NHS Foundation Trust

• The Leeds Teaching Hospital NHS Trust

• Aintree University Hospital NHS Foundation Trust

• Bradford Teaching Hospitals NHS Foundation Trust

• County Durham and Darlington NHS Foundation Trust

• Dartford and Gravesham NHS Trust

• North Cumbria University Hospitals

• University Hospitals of Leicester NHS Trust

• Wirral University Teaching Hospital NHS Foundation Trust

• The Newcastle Upon Tyne Hospitals NHS Foundation Trust

• Salford Royal NHS Foundation Trust.

The main criteria for selecting participating hospitals were that they had an existing S. aureus bacteraemia ward consultation service and sufficient numbers of S. aureus bacteraemias to be able to recruit patients (potential to recruit a minimum of one patient per month), as well as the necessary research infrastructure to conduct the trial.

The overall trial design is summarised in Figure 1.

FIGURE 1.

Trial schema. a, Incapacitated adults would be eligible, provided that they had an appropriate legal representative.

Patient selection

As S. aureus bacteraemia is a serious infection for which the standard treatment requires i.v. antibiotics, all eligible patients were hospital inpatients at the time of recruitment. Patients were identified via the clinical microbiology laboratory and the infectious diseases/microbiology consult service at each centre. When possible, patients were screened for eligibility on the day that their blood cultures were flagged positive with S. aureus. Written informed consent was obtained from patients. Incapacitated adults were eligible provided that they had an appropriate legal representative to provide consent. The principal investigator (PI) or another experienced independent physician was required to follow the Mental Capacity Act 200536 to formally assess the capacity of the individual to make an informed decision to participate in the trial. If incapacity was confirmed, then written informed consent was sought from either a personal (e.g. a relative) or a nominated (e.g. consultant intensivist caring for the patient, but not involved in the trial) legal representative.

Randomisation

Eligibility was confirmed by the ARREST trial site investigators (PI, co-PI or research nurse) via the online ARREST trial database, and patients were randomised into two parallel groups in a 1 : 1 ratio, to standard i.v. antibiotic therapy plus 14 days of placebo, or standard i.v. antibiotic therapy plus 14 days of rifampicin. The choice and duration of the standard antibiotic therapy was left to the attending physician. Randomisation was stratified by clinical centre, as blinded drug (in fully made-up and labelled treatment packs) was pre-shipped to local pharmacies. A computer-generated sequential randomisation list using variably sized permuted blocks was prepared by the trial statistician and incorporated securely into the online trial database. The list was concealed until allocation, after eligibility was confirmed by researchers at the local hospitals, who then performed the randomisation. A 24-hour web-based randomisation service was provided via the online ARREST trial database.

Trial intervention

Rifampicin/placebo was given by oral or i.v. route, depending on the attending physician’s preference and the patient’s status. Providing that a patient could swallow safely, the preference was to use rifampicin orally. Intravenous administration was permitted for patients who were not able to swallow or absorb tablets. Rifampicin is a well-established, widely used drug, and was not used outside its licensed indication during the course of the trial.

The oral investigational medicinal product (IMP) was prepared by a Clinical Trials Supplier (Sharp Clinical Services, Crickhowell, UK). It was supplied as 300 mg of rifampicin capsules (Sanofi-Aventis, Guildford, UK) [summary of product characteristics (SPC); www.medicines.org.uk/emc/medicine/21223/SPC/Rifadin±300mg±Capsules/ (accessed 4 June 2018)] or placebo oral 300-mg capsules containing cellulose. The rifampicin capsules were over-encapsulated so that they were identical in appearance to the placebo capsules. The capsules were supplied to trial centres as individual participant-blinded treatment packs so that they were dosed and dispensed in the same way.

The i.v. IMP was provided via standard hospital stock and consisted of either rifampicin for i.v. infusion [600 mg of rifampicin for i.v. injection (Sanofi-Aventis) SPC www.medicines.org.uk/emc/medicine/6435 (accessed 4 June 2018)] or standard saline as the placebo. Participants receiving i.v. infusions in the intensive care unit (ICU) could have their infusion volume altered in accordance with standard local practice and the SPC. The trial pharmacist at each hospital had access to a copy of the randomised allocations for each ARREST trial number for their centre in order to prescribe i.v. rifampicin if required.

Assessments and follow-up

Procedures for assessing efficacy

The primary outcome was:

• time to death or bacteriologically confirmed failure/disease recurrence up to 12 weeks from randomisation.

This outcome measure was assessed by visiting the patient on days 3, 7, 10 and 14, and weekly thereafter until discharge from hospital, and the final clinical assessment 12 weeks after recruitment [either by a ward visit (if the patient was still admitted to hospital) or by a clinic visit or telephone call]. Consent to contact the patient’s general practitioner (GP) was also obtained.

The definition of bacteriologically confirmed failure was:

1. symptoms and signs of infection ongoing for > 14 days from randomisation

2. the isolation of same strain of S. aureus (confirmed by genotyping) from either blood or another sterile site (e.g. joint fluid, pus from tissue) indicating blood-borne dissemination of the bacteria.

The definition of bacteriologically confirmed disease recurrence was:

1. the isolation of the same strain of S. aureus from a sterile site after > 7 days of apparent clinical improvement.

As defined, failure reflected both the speed of killing of S. aureus and sterilisation of infected foci/blood, and both failure and recurrence reflected the risk of dissemination and metastatic infection. Outcome measures included S. aureus infection of sterile sites other than just blood, because such disseminated infection can be the consequence of failure to treat initial infections adequately. Asymptomatic bacteraemia without any sign or symptom of infection was not considered failure. Additional blood cultures were requested as soon as the PI/study physician suspected failure or recurrence. All bacterial isolates (initial and all subsequent) from patients randomised in the trial were originally intended to be genotyped by multilocus sequence and spa-typing and tested for susceptibility to rifampicin.

A substantial proportion of bacteriological failure/recurrences did not have both baseline and failure/recurrence isolates stored [17 (61%) of 28 failures/recurrences where S. aureus was isolated from a sterile site]. In order to avoid excluding a substantial proportion of potential primary end points, the statistical analysis plan specified that the primary analysis would include all bacteriologically confirmed failures and recurrences (i.e. without restricting to the same strain).

In the 11 pairs of baseline and failure/recurrence isolates that were stored, same strain was defined by whole-genome sequencing using Illumina technology (San Diego, CA, USA) on the basis of 40 single nucleotide variants between baseline and failure/recurrence isolates. All failure/recurrence isolates were within 12 single nucleotide variants of the baseline isolate {median 1 [interquartile range (IQR) 1–6; range 0–12] single nucleotide variant}.

The secondary efficacy outcome measures were:

• time to all-cause mortality up to 14 days

• time to clinically defined failure/recurrence/death by 12 weeks

• duration of bacteraemia

• AEs (grade 3/4 AEs, SAEs, AEs of any grade leading to modification of rifampicin/placebo dose or interruption/early discontinuation) (all AEs reported, primary comparisons based on time to first event)

• the proportion modifying any treatment (including concomitant medications) as a result of drug interactions

• the proportion developing rifampicin-resistant S. aureus

• cost-effectiveness of rifampicin.

Mortality was reported on the ARREST trial database on a SAE electronic case report form (eCRF). Clinically defined failure/recurrence was assessed clinically in the same manner as bacteriologically confirmed failure or recurrence; however, microbiological confirmation was not required (e.g. patients who failed clinically but for whom blood cultures were not taken). Clinically defined failure/recurrence was primarily determined by radiological evidence for an ongoing or new active infection focus by 12 weeks and the requirement for ongoing or new antibiotic therapy.

The PIs were required to report all potential failures/recurrences and these were adjudicated as trial end points by an independent end-point committee. The blinded independent review committee consisted of two infectious disease physicians with experience in acute/general medicine (Professor Tim Peto, Oxford University Hospitals NHS Foundation Trust, Oxford, UK; Dr Graham Cooke, Imperial College Healthcare NHS Trust, London, UK; see Acknowledgements). Potential failures/recurrences were also identified through questions regarding signs and symptoms of ongoing or new S. aureus infections on routine case report forms, and S. aureus isolated from any microbiological specimen. For all such potential failures/recurrences a structured clinical narrative was completed by the site physician and approved by the site PI. All reported failures, recurrences and deaths were then adjudicated using standardised proformas by the committee without knowledge of randomised allocation.

Blood cultures were taken on days 3 and 7 following randomisation to assess duration of bacteraemia. Sensitivity to rifampicin was repeated on the day 3 and 7 blood cultures and in all subsequent S. aureus isolates grown at scheduled time points or at failure/recurrence in order to assess the secondary end point: development of rifampicin-resistant S. aureus.

The levels of CRP were measured longitudinally as a continuous measure of response to infection.

Procedures for assessing safety

Hepatitis is the most important side effect of rifampicin. Liver function tests were performed twice while on rifampicin/placebo (days 3 and 10) to assess laboratory safety parameters. Additional safety blood tests or investigations were performed to investigate symptoms or monitor emergent laboratory test abnormalities as clinically indicated.

Grade 3 and 4 AEs and SAEs were elicited at the regular clinical assessments, through consultation with the patient, their medical team or their medical records. All such AEs were reported on eCRFs, together with AEs of any grade leading to modification of rifampicin/placebo dose or its interruption/early discontinuation. All AEs (clinical and laboratory) were graded using the Common Toxicity Criteria grading scale v3.0. The SAEs were defined following the International Conference on Harmonisation of Technical Requirements for Registration of Pharmaceuticals for Human Use37 as events that led to death, were life-threatening, caused or prolonged hospitalisation (excluding elective procedures), caused permanent disability, or were other important medical conditions or with a real, not hypothetical risk of one of the previous categories. The SAEs were reported to the Medical Research Council (MRC) Clinical Trial Unit (CTU) at UCL according to standard timelines. All SAEs were reported on study eCRFs, unless they were specifically related to the S. aureus bacteraemia episode for which the patient was originally admitted (in which case they were reported as infection-related events). The protocol specifically exempted events related to S. aureus bacteraemia from AE reporting, unless the event was fatal, to avoid double-counting. The severity and likely relationship of any AEs to rifampicin/placebo were documented by a physician. All reported AEs were coded centrally at the MRC CTU at UCL using the Medical Dictionary for Regulatory Activities (MedDRA).

All modifications to rifampicin/placebo dose or administration were recorded, as were all significant drug interactions requiring modification of study and non-study medication.

Procedures for assessing health-related costs of S. aureus and quality of life

Health-care-related costs of S. aureus bacteraemia in the NHS and the evaluation of health-related quality of life (HRQoL) were evaluated using the EQ-5D. These assessments were used further to inform the cost-effectiveness of adjunctive rifampicin and relevant antibiotic regimens for S. aureus bacteraemia (see Chapter 5). Information on health-care-related costs of patients in the trial was collected, starting from when the first positive blood culture was taken and continuing for the duration of follow-up. Information on hospitalisation costs (including procedures, laboratory tests and concomitant medications) was collected at the regular clinical assessments, and data on other health-care resource utilisation (post-discharge outpatient visits, medications and procedures) were collected at the 12-week visit.

Within-trial assessments of HRQoL (using the EQ-5D) were also used in the economic analysis. The EQ-5D scores were used to weight lifetime lived by its quality; the EQ-5D tariff developed for the UK was used to derive the scores from the participants’ responses to the EQ-5D descriptive system. The cost-effectiveness analysis thus used quality-adjusted life-years (QALYs) as the outcome measure.

Sample size

The trial was originally designed with two coprimary outcomes: all-cause mortality by 14 days and bacteriological failure/recurrence or death by 12 weeks. Assuming 80% power, a two-sided alpha of 0.025 (to adjust for multiple testing given two coprimary outcomes) and a 10% loss to follow-up by 12 weeks, 920 participants were needed to detect a 30% relative reduction in bacteriological failure/death from 35% to 25%, an absolute difference of 10% corresponding to an number needed to treat (NNT) of 10 participants. Assuming 80% power, a two-sided alpha of 0.025 and a lower 4% loss to follow-up by 14 days (as most participants remained in hospital over this time scale), 940 participants were needed to detect a 45% relative reduction in mortality from 16% to 9%, an absolute 7% difference and a NNT of 14 participants. The total sample size was originally therefore 940 participants.

Recruitment to the trial was slower than anticipated. To facilitate successful completion of the trial and at the request of the trial funder, after 3 years of recruitment 14-day mortality was moved from a coprimary to a secondary outcome. Therefore, 12-week bacteriological failure/recurrence or death became the sole primary outcome with consequent decrease in sample size [as a result of the increase in the two-sided alpha (type I error) from 0.025 (two coprimary outcomes) to 0.05 (one primary outcome)]. With 12-week bacteriological failure/recurrence or death as the sole primary outcome, the total sample size became 770 participants (alpha = 0.05, other assumptions as above).

The protocol and statistical analysis plan specified that the primary outcome (bacteriologically confirmed failure/recurrence or death) would be analysed using time-to-event methods as described in Statistical methods. The sample size calculation treated this outcome as binary, in order to produce a conservative estimate of sample size given uncertainties in the underlying assumptions, and because all patients were to be followed for a fixed 12-week period (i.e. no additional power was gained from longer follow-up in some patients).

Statistical methods

Randomised groups were compared following the principle of intention to treat, including all follow-up regardless of changes to treatment. The statistical analysis plan prespecified that any patient who was randomised in error (defined as realising that the patient should not have been randomised before taking blinded study drug and not ever taking study drug and, hence, not followed up) would be excluded. The blinding means that there was no possibility that knowledge of randomised allocation affected this judgement about what was an error. Any participants who were randomised in good faith (i.e. not by mistake) but never took study drug were included in all analyses.

The time-to-event analyses measured time from randomisation. Analyses of clinical outcomes censored at the earliest of 12 weeks from randomisation and the last clinical information. Analyses of mortality censored at the earliest of the time scale being considered (2 weeks, 12 weeks) or last vital status information (including that ascertained at trial closure through the NHS records).

The primary analyses were unstratified because the randomisation stratification factor (centre) was expected to have some small strata and participants in these strata might then not contribute to comparisons. Results from secondary stratified analyses (stratified log-rank test and stratified Cox regression) were very similar (data not shown). Lost to follow-up was defined as not having been assessed in person or by telephone at the 12-week final visit (within a window of 1 week before to 8 weeks after week 12) by a trial clinician and not having information on whether or not signs/symptoms of S. aureus were present (e.g. from the patient’s GP).

Primary analysis of the primary end point included all randomised participants other than those considered randomised in error (following the statistical analysis plan): secondary analysis of the primary end point was to exclude those (expected < 1%) who were subsequently identified as having had a rifampicin-resistant S. aureus bacteraemia on susceptibility testing. As no patients were identified after randomisation as having had a rifampicin-resistant S. aureus bacteraemia at enrolment, this analysis was identical to the primary analysis. In the statistical analysis plan (but not the protocol), a per-protocol analysis was also specified for the primary end point, including all participants in the primary intention-to-treat analysis who received active/placebo for ≥ 80% of days from the start of trial drug to earliest of 14 days subsequently/death/discontinuation of active antibiotics (not including trial drug).

Safety analyses included all data between randomisation and 12 weeks post randomisation (inclusive). Non-fatal events related to S. aureus bacteraemia were not considered AEs/SAEs in the protocol.

When composite outcomes did not include all-cause mortality as part of the composite, competing risks analysis methods were used. Analogous to a Kaplan–Meier estimate, competing risk methods use cumulative incidence functions to estimate the probability of the event. The effect of randomised group on the subdistribution hazard that corresponds to this cumulative incidence function was estimated. Stratification is not possible with the estimating equation approach used to estimate these subdistribution hazards and so these analyses were conducted unstratified.

The CRP and liver function test results were compared between randomised groups over time using generalised estimating equations (normal distribution, independent correlation structure) with randomised group, adjusting for the stratification factor, baseline values and scheduled visit week as categorical independent variables and interaction between baseline values and scheduled visit week. The closest measurement to each scheduled visit date within equally spaced windows was used as the measurement at each scheduled visit. The midpoint between two scheduled assessment days was taken as belonging to the latter window (e.g. a measurement taken on day 12, midpoint between day 10 and day 14, would be considered as day 14 in the analysis). When there were two values within one of these equally spaced windows, but both equidistant from the nominal assessment day, the later value was used. Analyses were based on observed data. To account for CRP values above the limit of quantification in one centre (i.e. CRP levels only reported as > 156 mg/l if above this threshold), mean CRP level was estimated using normal interval regression. For analyses of change from baseline, these values were assumed equal to the limit of quantification.

For blood cultures, baseline (used to define baseline resistance/susceptibility) was defined as the closest up to and including day 0, and up to 1 day post randomisation providing this was on or before date of start of the trial drug. Cultures prior to randomisation were used in preference to cultures the same number of days after randomisation, but on or before the date of start of trial drug. As eligibility was based on the screening of positive blood cultures, and because the intention was to characterise persisting bacteraemia, baseline bacteraemia included cultures on day 1 when a culture on the day of or on the day prior to randomisation was not available. For duration of bacteraemia, baseline was defined as the closest up to and including day 0 within the preceding day, and up to 1 day post randomisation.

For laboratory measurements (e.g. CRP levels), baseline was defined as the closest up to and including day 0 within the preceding 4 days, and up to 1 day post randomisation providing this was on or before date of start of trial drug. Measurements prior to randomisation were used in preference to measurements the same number of days after randomisation, but on or before the date of start of trial drug.

A deep infection focus was defined as infection of implanted vascular device, native/prosthetic heart value, native/prosthetic bone/joint, or deep tissue infection/abscess (including vertebral bone/disc or other bone infection, epidural or intraspinal empyema, infected intravascular thrombus, brain infection).

Information on all antibiotics received through 12 weeks was collected, but not according to specific indication. Primary antibiotic treatment, and its duration, was therefore defined by complete cessation of all antibiotics for 2 days, with the exception of vancomycin when intermittent dosing up to 1 week was allowed. The cessation of vancomycin was defined by adding the number of days between the last two doses to the date of the final dose.

Data collection and handling

Data were entered by staff at each NHS trust hospital on to eCRFs on the online ARREST trial database. Staff with data entry responsibilities were required to complete database training before they were granted access to the database. Data were exported into Stata^® (v15.1) (StataCorp LLC, College Station, TX, USA) for analysis.

Interim analyses

The trial was reviewed by the ARREST trial’s Data Monitoring Committee (DMC). It met four times in strict confidence over the course of the trial: 14 November 2013, 31 October 2014, 26 May 2015 and 24 February 2016. DMC recommendations were communicated through a letter to the Trial Steering Committee (TSC) following each meeting.

Clinical site monitoring

Trial monitoring was carried out in accordance with the protocol. Trial centres agreed to provide access to source data and consent was gained from patients for direct access to patient notes. All centres that had a minimum of four patients who had completed follow-up (week 12 visit or death) were monitored on site at least once during the trial. The following data were validated from source documents:

• eligibility and signed consent

• trial drug and antibiotic management

• safety events

• any data concerns raised by central monitoring.

Patient and public involvement

The ARREST trial was developed with the Healthcare Associated Infection Service Users Research Forum (SURF) [www.hcaisurf.org (accessed 4 June 2018)], in particular, Jennifer Bostock, who was the patient and public involvement (PPI) representative on the ARREST Trial Steering Committee. Jennifer Bostock advised on the inclusion of incapacitated adults, the application of the Mental Capacity Act 200536 and the information provided to patients. SURF is no longer active, but Ms Bostock is helping to disseminate the trial results beyond the academic and health-care professional community to other patient groups that she works with, including MRSA Action UK.

In particular, given recruitment challenges, Ms Bostock developed and led the substudy investigating patients’ and carers’ reasons for, and for not, participating in the trial. This is reported in full in Chapter 4.

Protocol changes

The trial was approved by the London (Westminster) Research Ethics Committee (reference number 12/LO/0637). See Appendix 1 for changes to the protocol.

Participant flow diagram

Between 10 December 2012 and 25 October 2016, 770 participants from 29 UK hospital groups were randomised to add placebo (n = 396) or rifampicin (n = 374) to their ‘backbone’ antibiotic treatment (Figure 2). A total of 2896 participants were screened for entry to the trial. The most common reason for not randomising a potentially eligible participant was that they had already received > 96 hours of antibiotics (n = 664). In 364 cases, the participant was not willing. Rifampicin was considered mandatory in 232 cases. Known rifampicin resistance occurred in only 19 cases; however, 139 cases were not eligible because of pre-existing liver disease raising concerns about rifampicin treatment and 167 cases because of predicted drug interactions.

FIGURE 2.

Participant flow diagram. a, Reasons are not mutually exclusive and, therefore, total is more than the number of participants not randomised. b, Seven participants with predicted drug interaction, two misdiagnosed (S. aureus not grown from blood but only other samples), rifampicin considered mandatory in one, a clinician considered one participant should not have been randomised because of acute kidney injury, a clinician considered one participant should not have been randomised as they were in another study (not of an IMP). c, Final 12-week visit could occur any time from 11 weeks onwards in accordance with the protocol. Consent withdrawals not included in these numbers. d, Time-to-event analyses included all time at risk from randomisation to the earliest of the event or last clinical follow-up if the event had not occurred. © The Author(s). Published by Elsevier Ltd. This is an open access article published under the CC BY 4.0 licence.

A total of 12 participants (placebo, n = 8; rifampicin, n = 4) were randomised in error (the participant should not have been randomised and never received trial drug) and were excluded following the statistical analysis plan. Of these 12 participants, seven participants had predicted drug interactions, two were misdiagnosed (S. aureus was not grown from blood), rifampicin was considered mandatory in one patient, a clinician considered that one participant should not have been randomised owing to acute kidney injury, and a clinician considered that one participant should not have been randomised as they were in another study (not of an IMP, allowed according to the protocol).

Thus, 758 (placebo, n = 388; rifampicin, n = 370) participants were included in the analyses. The median number of patients recruited per centre was 11 (IQR 4–30 patients, range 1–163 patients). A total of 415 (54.7%) participants were recruited from five centres (Oxford, n = 163; Guy’s and St Thomas’, n = 99; Liverpool, n = 62; Plymouth, n = 48; and Sheffield, n = 43). The large number of centres recruiting small numbers of participants, together with the relatively large block size (6–8), led to a small imbalance in the numbers included randomised to the placebo (n = 388) and rifampicin groups (n = 370).

Baseline characteristics

Baseline characteristics were well-balanced between randomised groups (Tables 1 and 2).

A total of 495 (65.3%) participants were men (see Table 1). The median age was 65 years (IQR 50–76 years), median weight was 76.0 kg (IQR 64.0–90.0 kg) and the median Charlson Comorbidity Index score was 2 (IQR 0–3). Diabetes (30.1%), renal disease (18.2%), cancer (16.6%) and chronic lung disease (11.9%) were all common comorbidities. A total of 83 (10.9%) participants were active injecting drug users, 70 (9.2%) participants were in an ICU, 90 (11.9%) participants had undergone surgery in the past 30 days, and 127 (16.8%) participants had consent provided by a legal representative because of incapacity. Reflecting disease severity, mean CRP level was 164 mg/l [standard error (SE) 3.7 mg/l] and a median Sequential Organ Failure Assessment (SOFA) score was 2 points (IQR 1–4 points).

At randomisation, participants had already received a median of 62 hours (IQR 42–75 hours) of active antibiotics, with their first blood culture taken a median of 3 days (IQR 2–3 days) previously and their first symptoms occurring a median of 4 days (IQR 3–6 days) previously. A total of 157 out of 642 (24.5%) participants still had a positive blood culture on the day of randomisation.

A total of 485 (64.0%) infections were community acquired, with only 132 (17.4%) infections being nosocomial; 47 (6.2%) were caused by MRSA. No patients were known to have rifampicin-resistant S. aureus bacteraemia at randomisation.

The initial focus was deep in 301 (39.7%) patients, including 33 (4.4%) with endocarditis and 14 (1.8%) with infected prostheses. A total of 130 (17.2%) infections were due to infected central/peripheral lines, and 138 (18.2%) were associated with skin/soft tissue infections. Another type of focus was identified in 49 (6.5%) participants and not established in 139 (18.3%) participants.

In 255 (33.6%) participants, the most likely portal of entry of S. aureus into the bloodstream was a clinically apparent skin or soft tissue infection unrelated to a surgical intervention. Central or peripheral lines were the most likely portal of entry in 141 (18.6%) participants, although 191 (25.7%) had a vascular catheter in situ at randomisation. For 218 (18.6%) participants, the portal of entry was unknown.

Follow-up and treatment received

Overall, completeness of scheduled visits was high up to 14 days. Excluding visits after death or discharge, day 3 visits were missed in 10 out of 372 (2.7%) participants in the placebo group versus 12 out of 350 (3.4%) participants in the rifampicin group, day 7 visits were missed in 15 out of 337 (4.5%) participants in the placebo group versus 16 out of 311 (5.1%) participants in the rifampicin group, day 10 visits were missed in 22 out of 293 (7.5%) participants in the placebo group versus 26 out of 262 (9.9%) participants in the rifampicin group, and day 14 visits were missed in 9 out of 230 (3.9%) participants in the placebo group versus 13 out of 204 (6.4%) participants in the rifampicin group. Completeness dropped after 14 days when patients started to be discharged, for example visits were missed in 21 out of 149 (14.1%) participants in the placebo group versus 19 out of 134 (14.2%) participants in the rifampicin group at day 21, 23 out of 115 (20.0%) participants in the placebo group versus 23 out of 93 (24.7%) participants in the rifampicin group at day 28, and 25 out of 89 (28.1%) participants in the placebo group versus 19 out of 58 (32.8%) participants in the rifampicin group at day 35.

A total of 22 (2.9%) participants withdrew consent. At the 12-week visit, only 39 (5.1%) participants had unknown vital status and 65 (8.6%) were not assessed for signs/symptoms of S. aureus infection (including consent withdrawals).

A total of 23 (3.0%) participants were still in hospital at 12 weeks [15 (3.9%) participants in the placebo group vs. 8 (2.2%) participants in the rifampicin group; p = 0.17). The median initial hospitalisation duration was 21 days (IQR 14–50 days) versus 22 days (IQR 13–43 days) in placebo and rifampicin groups, respectively (p = 0.80) (Figure 3). A total of 132 (39.8%) participants in the placebo group versus 138 (44.8%) participants in the rifampicin group were discharged on outpatient parental antibiotic therapy (p = 0.35). A total of 94 (24.2%) participants in the placebo group versus 83 (22.4%) participants in the rifampicin group were readmitted post discharge and before 12 weeks (p = 0.56), spending a median of 9 nights (IQR 4–20 nights) and 10 nights (IQR 3–20 nights) in hospital post original discharge, respectively. Any admission was considered relating to S. aureus bacteraemia in 16 (4.1%) participants in the placebo group and 9 (2.4%) participants in the rifampicin group (p = 0.19).

FIGURE 3.

Days from admission to current hospital to original post-randomisation discharge. p = 0.80 subhazard regression. Note: death treated as competing risk.

A total of 744 (98.2%) participants initiated blinded trial drug a median of 4.3 hours (IQR 2.3–7.8 hours) after randomisation. Reasons for not initiating blinded trial drug were patient decision (n = 7), increasing liver enzyme levels (n = 2), starting on open-label rifampicin (n = 2), withdrawn for palliation (n = 1), incorrectly believing that the bacteraemia was rifampicin resistant (n = 1) and unable to access i.v. trial drug from trials pharmacy at weekend (n = 1).

A total of 96 (12.7%) participants initiated i.v. trial drug rather than oral trial drug (Table 3). A total of 595 (78.5%) participants initiated 900 mg daily rather than 600 mg daily and 362 (52.2%) participants twice daily rather than once daily. The median dose was 11.1 mg/kg (IQR 10.0–12.9 mg/kg). The trial drug was initiated a median of 68 hours (IQR 48–85 hours) after starting active antibiotics for the current infection. The trial drug was continued for a median of 12.6 days (IQR 6.0–13.2 days) for participants in the rifampicin group versus 13.0 days (IQR 11.3–13.5 days) in participants in the placebo group (p < 0.0001; discontinuations primarily because of antibiotic-modifying AEs and drug–drug interactions, see Safety). A total of 60 (15.5%) participants in the placebo group versus 51 (15.6%) participants in the rifampicin group received the i.v. trial drug. Percentages reporting missing any doses of trial drug ranged from 9.5% to 16.2% but did not differ between randomised groups (Figure 4; global p = 0.71).

FIGURE 4.

Percentage reporting missing one or more doses of trial drugs since the previous scheduled visit.

A substantial variety of ‘backbone’ active antibiotics were used (Table 4; details in Table 28, Appendix 2). Flucloxacillin was given in 619 (81.7%) participants, and vancomycin or teicoplanin in 380 (50.1%) participants at some point in the primary treatment course, with no evidence of difference between randomised groups (p = 0.44 and p = 0.34, respectively). Stat (one-off) doses of gentamicin or amikacin were used in 199 (26.3%) participants (p = 0.89). There was no evidence that the number of antibiotics used [median of 3 (IQR 2–4)] or the total duration of active anti-staphylococcal treatment (including therapy received before randomisation) (median 29 days (IQR 18–45 days)] differed between groups (p = 0.98 and 0.64, respectively) (see Table 4). Post-randomisation active anti-staphylococcal treatment was taken for a median of 27 days (IQR 15–41 days) in the placebo group versus 26 days (IQR 15–43 days) in the rifampicin group.

A total of 32 (8.6%) participants in the rifampicin group versus 52 (13.4%) participants in the placebo group used open-label rifampicin at some point after randomisation (p = 0.04). Median time from randomisation to initiation of open-label rifampicin was 14 days (IQR 7–18 days) (see Table 4). There was a trend to slightly fewer participants initiating open-label rifampicin from 14 days onwards [i.e. after stopping trial drug; 14 (3.8%) participants in the rifampicin group vs. 27 (7.0%) participants in the placebo group; p = 0.053]. Open-label rifampicin was used in participants with a range of original infection foci (Table 5). The median duration of open-label rifampicin was 25 days (IQR 13–45 days) in participants in the placebo group versus 32 days (IQR 26–48 days) in participants in the rifampicin group. A total of 60 (15.5%) participants in the placebo group received antibiotics after the primary course versus 34 (9.2%) participants in the rifampicin group (p = 0.01).

A total of 159 participants in the placebo group versus 142 participants in the rifampicin group had a deep focus that was drained/removed in 35 (22.0%) versus 29 (20.4%) participants, and a median (IQR) of 5 (2–12) and 3 (1–6) days from randomisation respectively (Table 6). A total of 88 participants in the placebo group versus 76 participants in the rifampicin group had an intravascular device that was removed in 62 (70.5%) versus 60 (78.9%) participants, and a median (IQR) of 2 (0–3) and 1 (0–2) days prior to randomisation, respectively.

Primary end point

By 12 weeks, bacteriological failure/recurrence or death occurred in 62 (16.8%) participants in the rifampicin group versus 71 (18.3%) participants in the placebo group [absolute risk difference (RD) –1.4%, 95% CI –7.0% to 4.3%; HR 0.96, 95% CI 0.68 to 1.35; p = 0.81] (Figure 5a). In exploratory post hoc analyses, comparing rifampicin with placebo there were 4 (1.1%) versus 5 (1.3%) failures (competing risks p = 0.82), 3 (0.8%) versus 16 (4.1%) recurrences (competing risks p = 0.01) and 55 (14.9%) versus 50 (12.9%) deaths without bacteriological failure/recurrence respectively (competing risks p = 0.30) (Table 7). The NNT to prevent one bacteriologically confirmed recurrence was 29 participants.

FIGURE 5.

Bacteriological failure/recurrence or death (a) overall (HR 0.96, 95% CI 0.68 to 1.35; p = 0.81) and (b) according to three priority subgroups. Note: see Figures 6 and 7 for other subgroup analyses. © The Author(s). Published by Elsevier Ltd. This is an open access article published under the CC BY 4.0 licence.

A total of 242 (65.4%) participants in the rifampicin group versus 290 (74.7%) participants in the placebo group were included in the per-protocol population [received active rifampicin/placebo for ≥ 80% of days from start of trial drug to 14 days subsequently, death or discontinuation of active antibiotics (not including trial drug), whichever came earliest]. By 12 weeks, 39 (16.1%) participants in the rifampicin group versus 49 (16.9%) participants in the placebo group experienced bacteriological failure/recurrence or died (absolute RD –0.8%, 95% CI –7.3 to 5.6; HR 1.00, 95% CI 0.65 to 1.52; p = 0.99). An exploratory post hoc analysis was also done, excluding participants in either group who started open-label rifampicin at any time during follow-up. A total of 225 (60.1%) participants in the rifampicin group versus 262 (67.5%) participants in the placebo group were included in this post hoc per-protocol population. By 12 weeks, 37 (16.4%) participants in the rifampicin group versus 37 (14.1%) participants in the placebo group experienced bacteriological failure/recurrence or died (absolute RD 2.3%, 95% CI –4.3 to 8.8; HR 1.23, 95% CI 0.78 to 1.93; p = 0.38).

Of the 28 failures/recurrences in which S. aureus was isolated from a sterile site, paired baseline and failure/recurrence isolates were stored for 11 (39%) participants. All failure/recurrence isolates were whole genome sequenced and within 12 single nucleotide variants of the baseline isolate [median 1 (IQR 1–6) (range 0–12) single nucleotide variants].

Subgroup analyses according to the three most important characteristics [time between starting active antibiotics and trial drug, meticillin resistance, and foci of infection (deep versus not deep)] suggested no heterogeneity in lack of effect of rifampicin (p_{heterogeneity} 0.42, 0.07, 0.10, respectively, see Figure 5b. The rifampicin effect varied significantly according to the initial antibiotic given at randomisation, with some suggestion of benefit in those with meticillin-sensitive infection treated with flucloxacillin alone (p_{heterogeneity} = 0.01, Figure 6), but across none of 16 other subgroup analyses (p_{heterogeneity} > 0.05, Figure 7). At the suggestion of a reviewer, subgroup analyses were also considered by diabetes (p_{heterogeneity} = 0.37), weight (p_{heterogeneity} = 0.13), body mass index (BMI) (p_{heterogeneity} = 0.58) and dose in mg/kg (p_{heterogeneity} = 0.42).

FIGURE 6.

Five other priority subgroup analyses for bacteriological failure/recurrence or death through 12 weeks (primary end point). Note: presenting class-level and antibiotic-level categorisation of initial active antibiotics (as per the statistical analysis plan). See Figure 5b for the three other priority subgroup analyses defined in the statistical analysis plan [time between starting active antibiotics and trial drug, meticillin resistance and foci of infection (deep versus not deep)]. All eight priority subgroup analyses were prespecified in the protocol and the statistical analysis plan.

FIGURE 7.

Twelve other subgroup analyses for bacteriological failure/recurrence or death through 12 weeks (primary end point). Bc, positive blood culture; first/gent, first administered/gentamicin; first+, first positive blood culture. a, Subgroup analysis prespecified in the statistical analysis plan but not the protocol. b, p = 0.07 using continuous interactions (splines); p = 0.01 using continuous interaction (linear). c, Additional subgroup analysis not in protocol or statistical analysis plan.

Secondary end points

Clinically defined failure/recurrence or death occurred in 76 (20.5%) participants in the rifampicin group versus 86 (22.2%) participants in the placebo group (RD –1.4%, 95% CI –7.4% to 4.7%; HR 0.97, 95% CI 0.71 to 1.32; p = 0.84 (Figure 8). In exploratory post hoc analyses comparing rifampicin and placebo there were 23 (6.2%) versus 25 (6.4%) failures (competing risks p = 0.97), 8 (2.2%) versus 23 (5.9%) recurrences (competing risks p = 0.01) and 45 (12.2%) versus 38 (9.8%) deaths without clinically defined failure/recurrence, respectively (competing risks p = 0.22) (see Table 7). The NNT to prevent one clinically confirmed recurrence was 26 participants.

FIGURE 8.

Clinically defined failure/recurrence or death. HR 0.97, 95% CI 0.71 to 1.32; p = 0.84. © The Author(s). Published by Elsevier Ltd. This is an open access article published under the CC BY 4.0 licence.

The end-point review committee adjudicated that failure of infection focus management was implicated in 38 out of 48 (79%) on placebo versus 24 out of 31 (77%) failures/recurrences on rifampicin (see Table 7). Of these failures of infection focus management, there were five participants in the placebo group versus 12 participants in the rifampicin group for whom the focus was not recognised, 16 participants in the placebo group versus 14 participants in the rifampicin group in which the focus was recognised but not actively managed (e.g. because it was in an inaccessible site, or other patient characteristics made intervention impossible) and 10 participants in the placebo group versus five participants in the rifampicin group for whom the focus was recognised and actively managed, but despite this failure/recurrence still occurred. Failure of antibiotic therapy was implicated in the failure/recurrence in only 3 (6%) placebo versus 1 (3%) rifampicin failures/recurrences, with the cause being impossible to distinguish in the remaining 7 (15%) versus 6 (19%), respectively.

By 12 weeks, 56 (15.1%) participants in the rifampicin group versus 56 (14.4%) participants in the placebo group had died (RD 1.0%, 95% CI –4.3% to 6.2%; HR 1.10, 95% CI 0.76 to 1.60; p = 0.60) (Figure 9). A total of 25 (6.8%) participants in the rifampicin group versus 17 (4.4%) participants in the placebo group died before 2 weeks (HR 1.60, 95% CI 0.86 to 2.95; p = 0.13). Fourteen deaths in the rifampicin group versus 16 deaths in the placebo group were adjudicated definitely S. aureus related, 14 deaths in the rifampicin group versus 12 deaths in the placebo group were probably S. aureus related, and 8 deaths in the rifampicin group versus 4 deaths in the placebo group were possibly S. aureus related (see Appendix 2, Table 29). A total of 18 deaths in the rifampicin group versus 23 deaths in the placebo group were not attributed to S. aureus (remainder unattributable) (overall p = 0.64).

FIGURE 9.

Mortality through 12 weeks. (Week 2: HR 1.60, 95% CI 0.86 to 2.95; p = 0.13. Week 12: HR 1.10, 95% CI 0.76 to 1.60; p = 0.60.) © The Author(s). Published by Elsevier Ltd. This is an open access article published under the CC BY 4.0 licence.

As for clinically defined and bacteriologically confirmed failures/recurrences, the end-point review committee adjudicated that failure of infection focus management was implicated in most S. aureus-related deaths, 21 out of 32 (66%) on placebo versus 18 out of 36 (50%) on rifampicin (see Table 7). Of these failures of infection focus management, there were three participants in the placebo group and four participants in the rifampicin group for whom the focus was not recognised, eight participants in the placebo group and eight participants in the rifampicin group for whom the focus was recognised but not actively managed, and 10 participants in the placebo group and six participants in the rifampicin group for whom the focus was recognised and actively managed, but despite this the participant still died from S. aureus. Failure of antibiotic therapy was implicated in only one (3%) S. aureus-related death in the placebo group versus three (8%) S. aureus-related deaths in the rifampicin group, with the relationship to antibiotics/focus management being impossible to distinguish in the remaining 10 (31%) and 15 (42%) participants, respectively. Three (9%) S. aureus-related deaths in the placebo group versus 11 (31%) S. aureus-related deaths in the rifampicin group were considered to have occurred as a consequence of late presentation to health care (i.e. were not preventable).

There was no difference in longer-term (post week 12) survival between the groups, based on consented updates of vital status from routine electronic health records (p = 0.69) (Figure 10).

FIGURE 10.

Mortality over the longer term. (HR 0.95, 95% CI 0.72 to 1.24; p = 0.69.)

Two (0.5%) participants in the rifampicin group developed new rifampicin-resistant S. aureus bacteraemia at 7 and 42 days after randomisation (p = 0.24). One occurred on day 7 (followed by rifampicin discontinuation on day 11 and bacteriological failure on day 14), and the other on day 42 (prescribed 14 days of rifampicin and bacteriological recurrence on day 42). One additional participant had rifampicin-resistant S. aureus isolated from a permanent pacemaker wire removed on day 1 (within 4 hours of the first dose of trial drug). The screening blood culture had isolated a rifampicin-sensitive S. aureus. Further blood cultures were sterile for the remainder of follow-up. Following whole-genome sequencing, the rifampicin-resistant pacemaker isolate was 11 single nucleotide polymorphisms from the screening isolate and another isolate taken from the pacemaker on day 1, whereas these last two isolates did not differ genetically, suggesting a diversity between isolates of > 3 days in origin, and thus suggesting that the patient had a mixed infection with both rifampicin-resistant and rifampicin-susceptible strains that were not detected at screening.

There was no evidence that duration of bacteraemia was significantly shorter in those randomised to the rifampicin group (Figure 11; global p = 0.66). Eighty-eight patients in the rifampicin group had positive blood cultures at enrolment. Of these 88, only one failed bacteriologically, none had bacteriological recurrence and none developed rifampicin-resistant infection. Eight failed clinically (including the one who failed bacteriologically) and two had clinical recurrence.

FIGURE 11.

Persistence of bacteraemia. Global test of difference between groups across days 3 and 7: p = 0.66. Test of difference between groups at day 3 in those bacteraemic at day 0: p = 0.36.

The CRP level declined significantly in both the rifampicin and the placebo groups, but decreases were smaller in participants in the rifampicin group (global p = 0.001, Figure 12).

FIGURE 12.

C-reactive protein levels over 2 weeks from randomisation. Note that p-values compare change from baseline across randomised groups.

Safety

By 12 weeks, 101 (27.3%) participants in the rifampicin group versus 94 (24.2%) participants in the placebo group experienced 112 versus 116 SAEs, respectively (HR 1.21, 95% CI 0.92 to 1.61; p = 0.17) (Figure 13 and Table 8; see also Appendix 2, Table 30). The most common type of SAE was related to infections and infestations, and the vast majority were because of fatal events caused by S. aureus bacteraemia (non-fatal S. aureus-related AEs were exempted from AE reporting in the protocol to avoid double-counting disease failure/recurrence events).

FIGURE 13.

Time to first SAE. (p = 0.17 log-rank test.)

Two participants in the rifampicin group with pre-existing liver disease experienced non-fatal hepatic failure.

One 47-year-old female required prolongation of hospitalisation for acute hepatic failure (grade 3) with, raised international normalised ratio (INR) (grade 2), ascites (grade 3) and acute renal failure (grade 3), which developed on ICU following 5 days of rifampicin (at a dose of 900 mg daily) with flucloxacillin. The participant had pre-existing hepatitis C and chronic liver disease. Acute hepatic and renal failure were considered to have been triggered by sepsis. The participant recovered.

One 51-year-old female required prolongation of hospitalisation for decompensated liver disease (grade 3) with ascites (grade 3) following 14 days’ rifampicin (initially on a dose of 900 mg daily) with flucloxacillin. The participant did not mention liver disease at screening/enrolment and there was nothing in her medical notes regarding any past history of liver problems. When she developed decompensated liver disease with ascites, it was discovered that she had a previous diagnosis of non-alcoholic steatosis at another hospital several years previously, but was no longer being followed up. The participant recovered.

A total of 129 (34.9%) participants in the rifampicin group versus 131 (33.8%) participants in the placebo group experienced 209 versus 193 grade 3/4 AEs, respectively, (HR 1.12, 95% CI 0.88 to 1.43; p = 0.36) (Figure 14 and Table 9; see also Appendix 2, Table 31). Most notable was a trend towards more renal grade 3/4 AEs with rifampicin, which occurred in 19 (5.1%) participants in the rifampicin group and 9 (2.3%) participants in the placebo group (p = 0.053); 17 and 6 of which, respectively, were acute kidney injury.

FIGURE 14.

Time to first grade 3 or 4 AE. (p = 0.36 log-rank test.)

A total of 63 (17.0%) participants in the rifampicin group versus 39 (10.1%) participants in the placebo group experienced 89 versus 52 antibiotic-modifying AEs, respectively, (subdistribution HR 1.78, 95% CI 1.20 to 2.65; p = 0.004) (Figure 15 and Table 10; see also Appendix 2, Table 32). Gastrointestinal disorders (24 vs. 8 participants, respectively, p = 0.003) and renal/urinary disorders (8 vs. 1 participants, respectively, p = 0.02) were more common in the rifampicin group, as were events classified as general disorders and administration site conditions (13 vs. 4 participants, respectively), which included some drug interactions (see below).

FIGURE 15.

Time to first antibiotic-modifying AE. (p = 0.004 subhazard regression.)

A total of 24 (6.5%) participants in the rifampicin group versus 6 (1.5%) participants in the placebo group experienced drug interactions with antibiotics or other drugs (p = 0.0005). This led to the discontinuation of trial drug in 13 participants in the rifampicin group versus 4 participants in the placebo group (p = 0.03), grade 1/2 AEs in 14 versus 3 participants, respectively (p = 0.006), and grade 3/4 AEs in five versus two participants, respectively (p = 0.27).

There was no evidence of differences between groups in changes in ALT (global p = 0.18, Figure 16) or alkaline phosphatase (global p = 0.11, Figure 17). Bilirubin increased significantly in the rifampicin group at day 3 (p < 0.0001; global p < 0.0001; Figure 18). Very few participants experienced grade 3 or 4 elevations in these laboratory parameters (Table 11).

FIGURE 16.

Alanine transaminase levels over 2 weeks from randomisation. Note that p-values compare change from baseline across randomised groups.

FIGURE 17.

Alkaline phosphatase levels over 2 weeks from randomisation. Note that p-values compare change from baseline across randomised groups.

FIGURE 18.

Bilirubin levels over 2 weeks from randomisation. Note that p-values compare change from baseline across randomised groups.

Experiences of being approached for trial participation, the consenting process and trial participation

The overall objective of this substudy was to identify patient and personal legal representative barriers to recruitment. The substudy was led by Jennifer Bostock, the trial PPI representative. The substudy had two components: the first involved patients/legal representatives who did not consent to trial recruitment, and the second involved patients/legal representatives who did consent to trial recruitment.

Findings

The study revealed two findings: first, there was a disappointing uptake of both questionnaire completion and interview. A total of 20 questionnaires were sent and only seven people responded and three patients/legal representatives were interviewed. Although it was expected that the study would be challenging and unorthodox in such a trial (especially seeking to explore views of those who did not consent), it was not anticipated that uptake would be as low as it was.

For patients who did not consent, the reasons given were ‘everything else going on was too much’ (patients A, B, C, D and E), ‘could not make a decision either way’ (patients A, B and E), ‘best to play safe’ (patients A, C, D and E), ‘felt too ill/tired’ (patients A, B, C, D and E), and ‘did not have enough time to decide’ (patients A, D and E). Another added, that they ‘would have liked to take part but the side effects were too risky and I didn’t want to take any risks. Everything was explained really well, sorry I couldn’t help’ (patient B). The same patient said that ‘more time to decide’ was very important and would have improved the likelihood of him participating. Another patient who gave similar reasons said that they would have been more likely to have participated if the ‘information sheet was shorter’ (patient C).

One questionnaire was sent back without any questions being answered but with a narrative arguing that it was not appropriate for patients who were ‘very unwell in A&E [accident and emergency] to be hassled by a research nurse about studies that are going on’. The patient went on to state:

I fully understand and appreciate trials take place and have taken part in clinical trials. Timing is the key and explaining when people feel a bit more human and can think straight about partaking when they have had time to read and digest it.

Patient F

Although this was only one patient, it is important that all studies seeking to recruit those with serious illness do so in a manner that is sensitive to the needs of patients. The fact that only one patient used the questionnaire or interview as an opportunity to complain in this way is evidence of the careful and considerate method of recruitment displayed by the recruiting staff.

Reasons given by personal legal representatives as to why they did not want their relative to participate were: ‘felt too worried’, ‘too much responsibility’, ‘worried that my relative might GET the study drug’ and ‘worried about side effects and liver problems’ (legal representative Z).

For patients who did consent, the reasons given were: ‘I didn’t really need time to think about it, I was ill and so it’s all rather difficult’, and ‘I don’t remember what the information sheet was like but XXXX explained it all to me and I don’t think I had any questions, but I’m sure she would have answered them if I had some’ (patient G).

Another said:

It’s not about the information, I just thought well I’ve got nothing to lose, but I did ask them to come back the next day and I thought about it, asked some people and still came to the same conclusion that I had nothing to lose so the next day I just signed up.

Patient H

The reasons given by personal legal representatives as to why they consented to their relative participating were ‘to help my mum and perhaps other people, it’s a 50/50 chance of her getting the medicine or the placebo and I just thought she might be helped’ (legal representative Y). On the information given they said:

But to be honest I don’t think I even read that sheet. Well I suppose the stuff about safety they told me about and I read it, it wasn’t difficult to understand. I just signed it and they were helpful the people who told me about it.

Legal representative Y

Conclusion

It is unusual for a trial such as this to explore these issues and it is ethically challenging to gain approval to conduct a study approaching patients/representatives who did not consent to participate in the primary study. However, it was deemed important both by the trial team for this and future research, and by the PPI advisor for the benefit of patients and their representatives. Having gained ethics approval for this study and having learned lessons on how to improve in future, the authors are confident that other research will benefit from the lessons, methods and findings of this small study. A number of practical suggestions were made based on the findings and were presented at an investigator meeting by the PPI advisor. It is hoped that these suggestions and the model for this substudy will be used by those at the meeting and their wider research networks.

Introduction

The ARREST trial was designed to evaluate the efficacy of adjunctive rifampicin in reducing bacteriologically confirmed failure/recurrence of S. aureus bacteraemia or death in 12 weeks. The trial did not provide evidence that rifampicin improves the composite primary end point. However, analyses of the components of this composite primary end point suggested that adjunctive rifampicin reduced the risk of disease recurrence. Nevertheless, the trial did not find any impact of rifampicin on short- or long-term mortality (secondary outcomes). Rifampicin also significantly complicated other drug treatment. Hence, clinically, adjunctive rifampicin was not considered to provide overall benefit over standard antibiotic therapy in adults with S. aureus bacteraemia.

The pragmatic design of this trial means that the population included is clinically relevant, and non-comparative findings can be considered generalisable. The clinical results highlight the severity of S. aureus bacteraemia and also show the high degree of heterogeneity in the patient population. In this component of the analyses, the trial evidence on the HRQoL and economic consequences of a S. aureus bacteraemia episode in this patient population are described, which can inform the burden to patients (in terms of HRQoL) and to health systems (in terms of health system costs). Heterogeneity was also explored by evaluating determinants of costs and HRQoL. Quantifying the burden of S. aureus bacteraemia allows better informed future evaluations of alternative treatment and prevention strategies, a research area that has been highlighted. 38 Although the literature on the economic impact of S. aureus bacteraemia is substantial, particularly for meticillin-resistant strains (such as MRSA), the evidence it is based on is poor and often does not rely on any empirical data. 39

Although evaluation of the costs and HRQoL impact of S. aureus bacteraemia, and potential determinants of these, are the main focus of the analyses, the effect of adjunctive rifampicin on HRQoL and cost outcomes was also investigated, given the primary aim of the trial. From the results of the clinical analyses it can be hypothesised that rifampicin adjunctive treatment may be associated with cost savings and improvements in HRQoL via the small but significant reduction in bacteriologically and clinically defined disease recurrences (hypothesised to arise from the sterilisation of deep infection foci). The trial data will be used to determine the potential cost-effectiveness of adjunctive rifampicin treatment, and given the likely high degree of uncertainty, the value of further research will be determined.

Methods

Cost and health outcomes for patients with S. aureus bacteraemia were evaluated using data from the ARREST trial. Health outcomes were measured as QALYs. The QALY combines survival and HRQoL into a single metric, in which time spent with poorer HRQoL is downweighted. Costs considered in analysis were those incurred by the NHS and Personal Social Services, as recommended by the National Institute for Health and Care Excellence (NICE). 40 Costs and QALYs were only measured for 84 days (i.e. 12 weeks) from the date of randomisation, which was also the maximum duration of active follow-up (longer follow-up through electronic health records was done only for mortality). When considering determinants of costs and QALYs, the effect of adjunctive rifampicin was evaluated, which allowed for a cost-effectiveness analysis to be conducted. Given the short time horizon, neither costs nor health benefits were discounted. The analysis was conducted using the statistical software R version 3.4.1 (The R Foundation for Statistical Computing, Vienna, Austria).

Details of each component of analyses (analysis of health benefits, analysis of costs and analysis of cost-effectiveness and value of information) are presented in the following sections. These are followed by a description of the statistical methods used.

Statistical methods of analyses

Results

A total of 758 participants were recruited: 388 were randomly allocated to receive standard antibiotic therapy (placebo) and 370 to receive adjunctive rifampicin. Baseline characteristics of participants by treatment group can be found in Table 12. Note that one rifampicin participant withdrew shortly after randomisation without an enrolment form having been completed. This patient has been excluded from all tables after baseline as they had no post-baseline data, leaving the number in the rifampicin group as 369 rather than 370 in Chapter 3.

Health benefits

Utility and quality-adjusted life-years (unadjusted and not using multiple imputation)

At baseline, there were approximately 10% (n = 80) comatose patients and 3% (n = 20) of patients unable or unwilling to provide answers to the EQ-5D questionnaire owing to their poor health. Descriptive statistics on HRQoL at different assessment times can be found in Table 21. Observed EQ-5D scores by domain/level and by time period can be found in Appendix 2, Table 37.

TABLE 21 - Unadjusted EQ-5D index scores over time

Deceased patients received an EQ-5D index score of 0; comatose patients received an EQ-5D index score of –0.402; patients reported to be unable/unwilling to provide EQ-5D answers received an EQ-5D index score of –0.261, corresponding to the bottom decile of the EQ-5D index score distribution of all trial patients for which a EQ-5D index score was available.

Unadjusted EQ-5D index scorea	Treatment group		Total (N = 757)
	Placebo (N = 388)	Rifampicin (N = 369)
Baseline
Number of patients, n (%)	381 (98.2)	365 (98.9)	746 (98.5)
Number responded, n (%)	329 (84.8)	317 (85.7)	646 (85.2)
Number in coma, n (%)	43 (11.1)	37 (10.0)	80 (10.5)
Number unwilling/unable, n (%)	9 (2.3)	11 (3.0)	20 (2.6)
Mean of unadjusted EQ-5D index score (SD)a	0.09 (0.35)	0.12 (0.34)	0.10 (0.34)
Day 7
Number of patients, n (%)	314 (80.9)	293 (79.4)	608 (80.3)
Number responded, n (%)	283 (72.9)	258 (69.9)	542 (71.6)
Number in coma, n (%)	24 (6.2)	24 (6.5)	48 (6.3)
Number unwilling/unable, n (%)	7 (1.8)	11 (3.0)	18 (2.4)
Number died, n (%)	7 (1.8)	13 (3.5)	20 (2.6)
Mean of unadjusted EQ-5D index score (SD)a	0.19 (0.34)	0.19 (0.35)	0.19 (0.34)
Day 14
Number of patients, n (%)	240 (61.9)	213 (57.7)	453 (59.8)
Number responded, n (%)	215 (55.4)	188 (50.9)	403 (53.2)
Number in coma, n (%)	20 (5.2)	18 (4.9)	38 (5.0)
Number unwilling/unable, n (%)	5 (1.3)	7 (1.9)	12 (1.6)
Number died, n (%)	17 (4.4)	25 (6.8)	42 (5.5)
Mean of unadjusted EQ-5D index score (SD)a	0.20 (0.34)	0.17 (0.32)	0.19 (0.33)
Day 84
Number of patients, n (%)	280 (72.2)	251 (68.0)	531 (70.1)
Number responded, n (%)	273 (70.4)	244 (66.1)	516 (68.2)
Number in coma, n (%)	2 (0.5)	2 (0.5)	4 (0.5)
Number unwilling/unable, n (%)	5 (1.5)	5 (1.4)	10 (1.5)
Number died, n (%)	56 (14.4)	56 (15.2)	112 (14.8)
Mean of unadjusted EQ-5D index score (SD)a	0.29 (0.31)	0.32 (0.28)	0.30 (0.29)

Descriptive statistics of the EQ-5D index scores (unadjusted) show that the mean score is fairly balanced across treatment groups, irrespective of time point of assessment. The baseline unadjusted mean EQ-5D index score was 0.10 (SD 0.34, n = 746), reflecting the very poor quality of life of patients affected with S. aureus bacteraemia. At 7 days, the unadjusted mean EQ-5D index score was 0.19 (SD 0.34, n = 608) and at 14 days this was also 0.19 (SD 0.33, n = 453). At this assessment point, 42 (5.5%) patients had died and hence were allocated an EQ-5D index score of 0. In interpreting these figures, care is needed, as 40% fewer patients completed the EQ-5D questionnaire at 14 days. At the end of the active follow-up (84 days), the mean unadjusted EQ-5D index score was 0.30 (SD 0.29, n = 531). Again, at this point 112 (14.8%) patients were deceased and received an EQ-5D index score of 0. Although only about 70% (n = 531, including values allocated for deceased/comatose/unable to answer patients as per Statistical methods of analyses, Missingness, denoted ‘hard’ imputations subsequently) of the total number of patients who were recruited into the trial completed an EQ-5D at 84 days, it shows that the selective group of patients for whom a EQ-5D index score at 84 days was obtained had a better (higher) mean EQ-5D score than the remaining individuals, at baseline. Distributions of EQ-5D index score at baseline, 7, 14 and 84 days (not using multiple imputation, but including hard imputations for coma/unwilling/unable to complete and death) are shown in Appendix 4, Figure 22.

The unadjusted total QALYs (over the whole of the follow-up period, including hard imputed values as per Statistical methods of analyses, Missingness) are presented in Table 22. It is highlighted again that results correspond to only about 70% of the sample as information for remaining patients was missing. Given that the period of assessment is 84 days (i.e. 3 months = one-quarter of 1 year), the maximum total QALYs that were observed is 0.25. Thus, the distribution of total QALYs will be inherently be both left and right truncated. Mean unadjusted total QALYs were similar between treatment groups, with a total mean QALY of 0.06 (SE 0.06). The distribution of total unadjusted total QALYs, including hard imputed values as per Statistical methods of analyses, Missingness, can be seen in Figure 19.

TABLE 22 - Unadjusted total QALYs (not using multiple imputation, but including hard imputations for coma/unwilling/unable to complete and death)

Unadjusted total QALYs	Treatment group		Total (n = 757)
	Placebo (n = 388)	Rifampicin (n = 369)
Mean (SD)	0.054 (0.063)	0.059 (0.059)	0.057 (0.061)
Number of patients (%)	275 (70.9)	249 (67.3)	524 (69.1)

FIGURE 19.

Distribution of (a) total QALYs; and (b) imputed total QALYs from one randomly selected imputed data set using multiple imputation techniques.

Quality-adjusted life-years (using multiple imputation)

A multiple imputation procedure was used to impute missing total QALYs at 84 days, which occurred in approximately 31% of the sample. Following a suggestion from the literature, in which the number of imputations should be similar to the percentage of cases that are incomplete,62,63 30 imputations of 20 iterations each were performed. Mode of acquisition of infection, Charlson Comorbidity Index score, BMI, deep infection foci, endocarditis, neutrophil count, coma status and EQ-5D index score were the baseline patient characteristics used as predictors by the imputation model. This process generated 30 different data sets with a calculated imputed outcome variable (i.e. total QALYs at 84 days). The distribution of total QALYs at 84 days for one of the imputed data sets, randomly chosen, can be seen in Figure 19. On these multiple imputed data sets, alternative GLM models for the total QALYs at 84 days were considered. As for the cost data, the null model, a model with only randomised treatment (model TQ) and a model with all covariates were implemented (model 1Q). A regression model assuming a normally distributed outcome with identity link (i.e. ordinary least squares model) was chosen (AIC statistic in model 1Q: –2109). Other modelling distributional assumptions tested either did not run or did not converge.

Base-case model (model 1Q)

The results of the base-case model (model 1Q) are shown in Table 23. These results are complemented with results of the model considering randomised treatment only (model TQ, first column). Regression models presented are linear and, therefore, additive, so coefficients can be interpreted directly to assess their impact on the outcome variable. In the model 1Q, being randomised to rifampicin was associated with slightly higher total QALYs (mean 0.004, SE 0.004) than being randomised to placebo, although this association was not statistically significant (p = 0.40, similar for model TQ). As expected, the EQ-5D index score at baseline was one of the main predictors of total QALYs accrued over 84 days, with one unit higher baseline EQ-5D estimated to be associated with higher total QALYs (model 1Q: mean difference of 0.06, 95% CI 0.05 to 0.08). Conversely, those aged ≥ 72 years (model 1Q: –0.043, 95% CI –0.072 to –0.014), those with any comorbidities as indicated in the Charlson Comorbidity Index score (gradient from –0.015, 95% CI –0.027 to –0.003 for index scores of 1–2, and up to –0.024, 95% CI –0.041 to –0.006, for higher index scores) and those in a coma (–0.020, 95% CI –0.037 to –0.004) were associated with significantly lower total QALYs. These covariates, together with meticillin resistance and neutrophil count, were retained in model 1Qp, showing that the latter are also relevant to explain variation in total QALYs.

TABLE 23 - Modelling total QALYs at end of active follow-up period (84 days) using multiple imputation: base-case and parsimonious model results

***p < 0.001; **p < 0.01; *p < 0.05; ^†p < 0.1.

TQ, total QALYs.

Thirty parsimonious models were obtained, one for each imputed data set. The covariate set retained in the parsimonious models was slightly different across models. Thus, results shown use a majority rule (i.e. when the covariate was retained three or more times).

Model specification	Model
	TQ	1Q	1Qpa
Type of regression model	OLS	OLS	OLS
Equation	Total QALYs (imputed)	Total QALYs (imputed)	Total QALYs (imputed)
Covariates (baseline)	Coefficient (SE)	Coefficient (SE)	Coefficient (SE)
EQ-5D index baseline score	–	0.064 (0.008)***	0.064 (0.008)***
Randomised treatment (1 – rifampicin; 0 – placebo)	0.007 (0.005)	0.004 (0.004)	–
Age (years)
54–71	–	–0.026 (0.020)	–0.027 (0.020)***
≥ 72	–	–0.043 (0.014)**	–0.044 (0.014)**
Sex (1 – male; 0 – female)	–	0.004 (0.005)	–
Acquisition
Nosocomial infection	–	–0.005 (0.006)	–
Health care associated	–	–0.001 (0.006)	–
Charlson Comorbidity Index score
1 or 2	–	–0.015 (0.006)**	–0.015 (0.006)*
3 or 4	–	–0.019 (0.008)**	–0.020 (0.007)**
≥ 5	–	–0.024 (0.009)**	–0.024 (0.009)**
BMI (kg/m2)
18.5–24.9	–	0.005 (0.010)	–
25.0–29.9	–	0.005 (0.010)	–
30.0–39.9	–	0.010 (0.011)	–
≥ 40	–	0.004 (0.013)	–
Deep focus (1 – yes; 0 – no)	–	0.0004 (0.005)	–
Endocarditis (1 – yes; 0 – no)	–	0.003 (0.011)	–
Meticillin resistance	–	–0.027 (0.020)	–0.024 (0.021)
Neutrophils
6–9 109/l	–	0.004 (0.006)	0.005 (0.006)
> 9 109/l	–	–0.010 (0.006)†	–0.011 (0.006)†
Comatose (1 – yes; 0 – no)	–	–0.020 (0.008)*	–0.020 (0.008)*
Intercept	0.076 (0.009)***	0.104 (0.012)***	0.115 (0.006)***
Observations	757	724	724

For the mean patient in the trial across all covariates used in the regression, the weighted mean predicted total QALYs for the placebo group were similar in the rifampicin and the placebo groups (Table 24, first set of results). As expected, and because of model linearity, the difference in predicted total QALYs between treatment groups was estimated to be mean 0.004 (SE 0.004). Results of the sensitivity analysis on patients unable/unwilling to provide EQ-5D answers can be found in Appendix 2, Tables 38–40.

TABLE 24 - Predicted total QALYs at the end of the active follow-up period, by treatment group (using multiple imputation)

HRQoL predictions (QALYs)	Treatment group
	Placebo	Rifampicin
Model 1Q
Mean predicted total QALYs (SE)	0.077 (0.008)	0.080 (0.009)
Median predicted total QALYs (IQR)	0.077 (0.071–0.082)	0.080 (0.074–0.086)
Mean predicted total QALYs difference (SE)	0.004 (0.004)
Model 2Q
Mean predicted total QALYs (SE)	0.076 (0.010)	0.080 (0.013)
Median predicted total QALYs (IQR)	0.076 (0.070–0.083)	0.080 (0.071–0.088)
Mean predicted total QALYs difference (SE)	0.004 (0.003)

Scenario analysis: consideration of treatment effect modifiers (model 2Q)

As for total costs, a scenario analysis was implemented for total QALYs (model 2Q) considering treatment interactions. The results of this scenario analysis can be found in Table 25. In general, similar results to model 1Q were obtained. Following model 1Q, model 2Q did not find treatment to be significantly associated with total QALYs. Model 2Qp results (also in Table 25) show that the most parsimonious model retained the following covariates as important to explain the outcome variable: EQ-5D baseline score, age, Charlson Comorbidity Index score, meticillin resistance, neutrophil count and coma status. Thus, randomised treatment and randomised treatment interactions were not selected for the most parsimonious model based on AIC statistics. As for model 1Q, and considering the mean trial patient, as no statistically significant difference was found between treatment groups, both groups had similar mean total QALYs (see Table 24, second set of results). Total QALY predictions for each patient subgroup and randomised treatment are presented in Cost-effectiveness and decision uncertainty.

TABLE 25 - Modelling total QALYs at the end of the follow-up period (multiple imputation)a

***p < 0.001; **p < 0.01; *p < 0.05; ^†p < 0.1.

OLS, ordinary least squares.

Thirty parsimonious models were obtained, one for each imputed data set. The covariate set retained in the parsimonious models was slightly different across models. Thus, results shown use a majority rule (i.e. when the covariate was retained three or more times).

Model specification	Model
	2Q	2Qp
Type of regression model	OLS	OLS
Equation	Total QALYs (imputed)	Total QALYs (imputed)
Covariates (baseline)	Coefficient (SE)	Coefficient (SE)
EQ-5D index score	0.064 (0.011)***	0.064 (0.008)***
Randomised treatment (1 – rifampicin; 0 – placebo)	0.016 (0.022)	–
Age (years)
54–71	–0.028 (0.020)	–0.027 (0.020)
≥ 72	–0.041 (0.014)**	–0.044 (0.014)**
Sex (1 – male; 0 – female)	0.005 (0.007)	–
Acquisition
Nosocomial infection	–0.002 (0.008)	–
Health care associated	–0.005 (0.009)	–
Charlson Comorbidity Index score
1 or 2	–0.011 (0.008)	–0.015 (0.006)*
3 or 4	–0.017 (0.010)†	–0.020 (0.007)**
≥ 5	–0.012 (0.010)	–0.024 (0.009)**
BMI (kg/m2)
18.5–24.9	–0.0003 (0.014)	–
25.0–29.9	0.004 (0.014)	–
30.0–39.9	0.015 (0.014)	–
≥ 40	0.015 (0.018)	–
Deep focus (1 – yes; 0 – no)	0.008 (0.007)	–
Endocarditis (1 – yes; 0 – no)	–0.005 (0.016)	–
Meticillin resistance	–0.020 (0.028)	–0.024 (0.021)
Neutrophils
6–9 × 109/l	0.010 (0.009)	0.005 (0.006)
> 9 × 109/l	–0.018 (0.008)*	–0.011 (0.006)†
Comatose (1 – yes; 0 – no)	–0.020 (0.012)†	–0.020 (0.008)*
Treatment × EQ-5D index score	0.003 (0.016)	–
Treatment × age, 54–71 years	0.005 (0.011)	–
Treatment × age, ≥ 72 years	–0.007 (0.011)	–
Treatment × sex (1 – male; 0 – female)	–0.003 (0.009)	–
Treatment × acquisition, nosocomial infection	–0.010 (0.012)	–
Treatment × acquisition, health care associated	0.008 (0.013)	–
Treatment × Charlson Comorbidity Index score of 1–2	–0.007 (0.011)	–
Treatment × Charlson Comorbidity Index score of 3–4	–0.003 (0.016)	–
Treatment × Charlson Comorbidity Index score of ≥ 5	–0.024 (0.016)	–
Treatment × BMI of 18.5–24.9 kg/m2	0.009 (0.020)	–
Treatment × BMI of 25.0–29.9 kg/m2	0.003 (0.020)	–
Treatment × BMI of 30.0–39.9 kg/m2	–0.009 (0.021)	–
Treatment × BMI of ≥ 40 kg/m2	–0.020 (0.026)	–
Treatment × deep focus (1 – yes; 0 – no)	–0.018 (0.011)†	–
Treatment × endocarditis (1 – yes; 0 – no)	0.019 (0.021)	–
Treatment × meticillin resistance	–0.013 (0.022)	–
Treatment × neutrophils, 6–9 × 109/l	–0.007 (0.011)	–
Treatment × neutrophils, > 9 × 109/l	0.015 (0.011)	–
Treatment × comatose (1 – yes; 0 – no)	–0.001 (0.017)	–
Intercept	0.098 (0.015)***	0.115 (0.006)***
Observations	724	724

Cost-effectiveness and decision uncertainty

The results presented on the analysis of costs and health outcomes (QALYs) showed that participants randomised to receive rifampicin for the treatment of S. aureus bacteraemia were expected to attain higher QALYs over the duration of the trial, and were expected to incur lower costs than those participants allocated to receive the placebo. These results were not, however, statistically significant.

Considering the mean total costs and mean total QALYs at face value, adjunctive rifampicin could promote relevant cost savings over an 84-day time horizon without compromising health outcomes, and that actually there may be even positive, although small, implications for total QALYs (Table 26). This means that rifampicin dominates placebo, that is, it costs less but provides additional health benefits compared with placebo. If releasing £20,000 to the NHS is assumed to result in 1 additional QALY (the cost-effectiveness threshold), the mean incremental net health benefit (INHB) of adjunctive rifampicin is approximately 0.06 QALYs (SE 0.04 QALYs).

TABLE 26 - Cost-effectiveness: base-case and scenario analysis results

At cost-effectiveness thresholds of £13,000, £20,000 and £30,000 per QALY gained, respectively.

Cost-effectiveness outcomes	Treatment group, mean (SE)
	Placebo	Rifampicin
Base-case results (using results from regression models 1C and 1Q)
Predicted total costs (£)	12,142 (546.0)	11,050 (509.7)
Predicted total QALYs	0.077 (0.008)	0.080 (0.009)
Incremental predicted total costs (£)	–1092 (749.8)
Incremental predicted total QALYs	0.004 (0.004)
ICER (£/QALY gained)	Rifampicin dominates (i.e. costs less and has positive health benefits in relation to placebo)
Incremental net health benefita
£13,000/QALY	0.088 (0.058)
£20,000/QALY	0.058 (0.038)
£30,000/QALY	0.040 (0.025)
Probability of being cost-effectivea
£13,000/QALY	0.07	0.93
£20,000/QALY	0.06	0.94
£30,000/QALY	0.06	0.94
Scenario analysis results (using results from regression models 2C and 2Q)
Predicted total costs (£)	11,969 (535.8)	10,900 (500.2)
Predicted total QALYs	0.076 (0.010)	0.080 (0.013)
Incremental predicted total costs (£)	–1068 (726.6)
Incremental predicted total QALYs	0.004 (0.003)
ICER (£/QALY gained)	Rifampicin dominates (i.e. costs less and has positive health benefits in relation to placebo)
Incremental net health benefita
£13,000/QALY	0.086 (0.056)
£20,000/QALY	0.057 (0.037)
£30,000/QALY	0.039 (0.024)
Probability of being cost-effectivea
£13,000/QALY	0.06	0.94
£20,000/QALY	0.06	0.94
£30,000/QALY	0.06	0.94

Figure 20 shows the cost-effectiveness plane using results from the Monte Carlo simulation to represent uncertainty over incremental mean costs and QALYs (joint density plot). Each green dot represents a simulated incremental cost and QALY pair; the cloud of green points is representative of the uncertainty around the cost-effectiveness outcomes. The blue dot displays the mean incremental costs and effects. The majority of green points lie in the fourth quadrant of negative incremental costs and positive incremental benefits. The ARREST trial data show that rifampicin is likely to be less costly and is associated with small QALY gains (although uncertain) in the 84 days of follow-up (post randomisation). This suggests that rifampicin is cost-effective with very little associated uncertainty, [i.e. a very high probability of being cost-effective (93–94%)]. Similar results to the base case were found for the scenario analysis (see Table 26).

FIGURE 20.

Cost-effectiveness plane for the base-case results.

Considering a technology time horizon of 10 years and an annual effective population of 12,500 patients per year in the UK,2 the EVPI for the MRSA/MSSA bacteraemia population is estimated to be approximately £2M at the commonly used cost-effectiveness thresholds. This estimate represents the maximum amount that the health-care system should be willing to pay for further information and resolve identified uncertainties. At the individual level and for the same cost-effectiveness threshold values, the EVPI is estimated to be approximately £20 per MRSA/MSSA patient.

Discussion

The ARREST trial aimed to determine whether or not adjunctive rifampicin improved outcomes following S. aureus bacteraemia, but found no evidence of an effect either on resolution of bacteraemia or on mortality (design and effectiveness results of the trial are reported in detail in Chapters 3 and 4, respectively). In this chapter, evaluating the cost and HRQoL implications of S. aureus bacteraemia using the trial data were focused on first.

This first set of analyses found that an episode of S. aureus bacteraemia costs, on average, £12,197 over 12 weeks (unadjusted results). The cost categories that contribute the most to costs (descriptive analyses) are length of stay (primary hospital admission and readmissions) and procedures undertaken in hospital. Determinants of higher episode costs (variables evaluated at baseline), evident from the trial population, were whether or not the primary infection was nosocomial (episode costs 41% higher), whether or not there was a deep focus primary infection (episode costs 43% higher), whether or not the patient had endocarditis (episode costs 65% higher), whether or not the patient had a high neutrophil count (> 9 × 10⁹/l, episode costs 34% higher), and whether or not the patient was comatose (episode costs 33% higher). For example, for an infection classified as having a deep focus, the mean costs of the episode were estimated at £12,514, whereas for infections without a deep focus the mean costs were £8752. In the ARREST trial population, neither age, sex, BMI, Charlson Comorbidity Index score nor meticillin resistance were found to determine costs at standard levels of standard statistical significance.

Analyses indicate that adjunctive rifampicin may save 10% of episode costs, although this result was not statistically significant at the standard 95% level (p = 0.14). Descriptive, unadjusted, analyses suggest that these savings start in the first 14 days of treatment (unadjusted difference in the first 14 days was £413), but that are perhaps most relevant after 14 days (unadjusted difference of £787). Because the trial was not powered on this outcome, the relevance of this finding (had a larger sample size been recruited) is unclear. However, this result is consistent with the reduction in recurrences, which occurred in a small proportion of participants, but significantly fewer in the rifampicin group (1%) than in the placebo group (4%).

As expected in this population of acutely ill patients, very low values of the EQ-5D score were observed at baseline (mean EQ-5D score of 0.10). A high proportion of patients were comatose, and a high proportion of individuals had health states that the valuation algorithm ascribed as worse than death (i.e. returning negative EQ-5D score values). Unadjusted figures show, however, that mean HRQoL score was significantly higher at 84 days (mean 0.30). The measure of benefit in the adjusted analysis considered QALYs over 84 days. QALYs are often the recommended measure of benefit for societal decision-makers, as they are generic and thus allow comparisons to be made across different treatments, conditions and patient populations. Given the high mortality and low HRQoL that this population is subjected to, total QALYs over the 84 days were on average 0.077 per patient, only 33% of the maximum innings for this period (0.23 QALYs or 84 out of 365 days). Determinants of QALYs in the sample were baseline EQ-5D score (0.0064 QALYs lost for every 0.1-point decrease in baseline EQ-5D), older age (up to 0.044 QALY loss), higher Charlson Comorbidity Index score (up to 0.024 QALY loss) and being comatose (mean QALY loss of 0.020). As opposed to total costs, deep foci infection did not affect total QALYs. After adjustment, the effect of rifampicin on total QALYs was positive (0.004 QALYs) but not statistically significant (SE 0.004 QALYs). Given the lack of statistical significance, the relevance of the finding that rifampicin has a positive (but small) effect on total QALYs is unclear; however, it is in accordance with the reduction in recurrences in the rifampicin group.

Public Health England has conducted mandatory enhanced surveillance of MRSA bacteraemia since October 2005 and of MSSA bacteraemia since January 2011. From April 2017 to March 2017, 823 cases of MRSA and 11,486 cases of MSSA were reported in England. 2 At the episode cost determined in the ARREST trial, these incidence figures imply a £150M burden to the NHS.

Based on the analyses from ARREST, adjunctive rifampicin could result in cost savings and negligibly small gains in mean QALYs. The cost savings possibly arise from reductions in hospital stay and readmissions in the short term. In cost-effectiveness terms, adjunctive rifampicin could be said to dominate placebo. The within-trial economic analysis, however, excluded potentially important outcomes, such as resistance arising from increased use of rifampicin and the clinical consequences of its drug–drug interactions (which may not have been captured fully in the analyses as costs of non-antibiotic drugs were not included, nor were costs of monitoring tests, e.g. for toxicity). This was a pragmatic decision because patients enrolled in this trial had wide range of underlying conditions and will have required a very large number of other drugs. A decision was therefore made not to try to record all of these other drugs on case report forms (CRFs), which would make them impossible to cost. Similarly, it was difficult to know what quantitative data to record to assess drug interactions. Rather than collect a large amount of free text to try to code and risk missing different items for different episodes, a pragmatic decision was made to not include these on CRFs either. Moreover, the ARREST trial was conducted under experimental conditions, and, despite providing unbiased estimates of treatment effects, practice may not be as homogeneous, and, hence, further research could confirm whether or not any predicted cost savings would be effectively realised in practice.

Summary and future research

In summary, the ARREST trial provides high-quality data when there are almost none. The clinical management of infectious diseases and, in particular, the treatment of many common severe bacterial infections, lacks high-quality clinical trial evidence. The situation is especially stark for S. aureus bacteraemia, probably the commonest life-threatening community- and hospital-acquired infection worldwide. But although the ARREST trial addresses some of these inadequacies, it also leaves many questions unanswered as to how to improve outcomes from S. aureus bacteraemia.

The ARREST trial has exposed two possible windows in which to intervene. The first is in the acute phase, when S. aureus can be cultured from the bloodstream and a severe inflammatory response (or ‘sepsis’) can have rapidly fatal consequences. The interventions in this early phase are those targeted at more rapid recognition or diagnosis of S. aureus bacteraemia, and those that might enhance bacterial killing and control the detrimental effects of the inflammatory response. Future research therefore might investigate novel molecular techniques, perhaps based on rapid next-generation sequencing, to identify S. aureus from the blood and predict drug susceptibility such that effective antibiotic treatment can be given quickly. The question of whether or not intensified antibiotic therapy – be it different drugs, doses, or drug combinations – might speed bloodstream sterilisation very early in the infection and thereby improve outcomes, has not been resolved by the ARREST trial, although it has delineated the considerable challenges of conducting a trial to address this question. Likewise, early control of the inflammatory response, using corticosteroids or newer drugs targeted at specific molecules in the inflammatory cascade, might reduce early mortality and would be amenable to testing by clinical trials.

However, given the challenges experienced conducting this trial, this is probably not the most important priority for future studies. Rather, the second interventional window in S. aureus bacteraemia is more easily accessible to triallists than the first window and, as the health economic analysis suggests, may also bring substantial cost savings. It opens after around 72–96 hours of active antibiotic treatment, once the acute phase is over, and concerns interventions to prevent, detect and manage the longer-term complications of the S. aureus bacteraemia, including disease recurrence. These complications include endocarditis, vertebral osteomyelitis, and other deep and potentially occult infection foci. As recent PET scan studies have shown,72 perhaps the most promising strategies that should be prioritised for future research are those that aim to speed the detection of these complications and improve their antibiotic and surgical management; these could have major impacts on outcome and costs. A clinical trial investigating these strategies against current standards of care would be both feasible and likely to have a major impact on clinical practice.

The ARREST co-applicants

Professor Tim Peto, Dr Gerraint Davies, Dr Martin Llewelyn, Professor Peter Wilson, Dr Duncan Wyncoll and Marta Soares.

The ARREST Trial Steering Committee

Dr Adrian Martineau (chairperson), Dr Geoff Scott, Dr Achim Kaasch, Ms Jennifer Bostock, Professor Guy Thwaites, Professor A Sarah Walker, Dr Gavin Barlow and Dr Susan Hopkins.

The ARREST Trial Management Group

Professor Guy Thwaites, Professor A Sarah Walker, Fleur Hudson, Janet Cairns, Denise Ward, Alex Szubert, Helen Webb, Charlotte Russell, Chiara Borg, Brooke Jackson, Damilola Otiko, Lindsey Masters, Zaheer Islam, Carlos Diaz and Debbie Johnson.

Data Monitoring Committee

Professor David Lalloo (chairperson), Professor Mark Wilcox and Professor Doug Altman.

End-point review committee

Professor Tim Peto and Dr Graham Cooke.

Recruiting sites

Oxford University Hospitals NHS Foundation Trust: M Scarborough, M Kamfose, A de Veciana, NC Gordon, L Peto, G Pill, T Clarke, L Watson, B Young, D Griffiths, A Vaughn, L Anson, E Liu, S Perera, L Rylance-Knight, C Cantell and R Moroney.

Guy’s and St Thomas’ NHS Foundation Trust: JD Edgeworth, G Thwaites, K Bisnauthsing, A Querol-Rubiera, C Gibbs, A Patel, C Hemsley, AL Goodman, D Wyncoll, J Biswas, J Fitzpatrick, L Roberts, J Millard, N Stone, A Cape, L Hurley and C Kai Tam.

The Royal Liverpool and Broadgreen University Hospitals NHS Trust: E Nsutebu, M Hoyle, K Maitland, L Trainor, H Reynolds, J Harrison, J Anson, J Lewis, J Folb, L Goodwin, N Beeching, S Dyas, H Winslow, E Foote, P Roberts, P Natarajan, A Chrdle, M Fenech and H Allsop.

Plymouth Hospitals NHS Trust: R Tilley, R Austin-Hutchison, L Barrett, K Brookes, L Carwithen, A Conbeer, R Cunningham, C Eglinton, R Fok, H Gott, S Hughes, L Jones, M Kalita, A King, L March, M Marner, T Mynes, A Plant, S Price, J Sercombe, A Stolton, M Wallis, M West, J Westcott, C Williams, R Wosley and L Yabsley.

Sheffield Teaching Hospitals NHS Foundation Trust: J Greig, L Butland, J Sorrell, T Mitchell, A Alli, J Meiring, B Masake, C Rowson, L Smart, L Makey, S Moll, J Cunningham, K Ryalls, K Burchall, J Middle, Y Jackson, D Swift, J Cole, B Subramanian, F Okhuoya, M Edwards, C Bailey, R Warren, G Islam, M Ankcorn, S Birchall, P Jones, J Humphries, S Booth and C Evans.

Portsmouth Hospitals NHS Trust: S Wyllie, A Flatt, L Strakova, M Hayes, S Valentine, C James, M Wands, N Cortes, N Khan, R Porter, Z Martin, K Yip, H Preedy, H Chesterfield, T Dobson and C Walker.

Brighton and Sussex University Hospitals NHS Trust: M Llewelyn, A Dunne, L Latter, A Porges, J Price, J Paul, L Behar, L Robinson, A Murray, J Fitzpatrick, T Sargent, C Ridley, L Ortiz-Ruiz de Gordoa, D Gilliam, C McPherson, S Matthews, E Foreman, R Jarghese, A Beddoe, S Martin, S Shaw, D Wlazly, M Cole, A Gihawi and K Cole.

Cambridge University Hospitals NHS Foundation Trust: ME Török, T Gouliouris, L Bedford, RB Saunderson, I Mariolis, R Bousfield, I Ramsay, D Greaves, S Aliyu, K Cox, L Mlemba, L Whitehead, N Vyse and M Bolton.

South Tees Hospitals NHS Foundation Trust: J Williams, P Lambert, D Chadwick, K Baillie, M Cain, R Bellamy, J Wong, J Thompson, H Vassallo, A Skotnicka, A Boyce and A Donnelly.

University College London Hospitals NHS Foundation Trust: P Wilson, G FitzGerald, V Dean, K Warnes, A Reyes, S Rahman, L Tsang, J Williams and S Morris-Jones.

Royal Free London NHS Foundation Trust: S Hopkins, E Witness, O Brady, E Woodford, T Pettifer, A McCadden, B Marks, S Collier, D Mack, S Warren, C Brown, A Lyons, S Taiyari, S Mepham, A Sweeney and L Brown.

Royal Devon and Exeter NHS Foundation Trust: C Auckland, A Potter, J Mandiza, M Hough, S Williams, C Renton, F Walters, M Nadolski, M Hough, A Evans, P Tarrant, S Williams, K Curley, S Whiteley, J Halpin, M Hutchings, S Todd, C Lohan, T Chapter, E Folland, A Colville, K Marden, M Morgan, R Fok, R Porter and M Baxter.

The Leeds Teaching Hospital NHS Trust: J Minton, S Rippon, M Cevik, J Chapman, T Kemp, R Vincent, D Osborne, T Platt, J Calderwood, B Cook, C Bedford, L Galloway-Browne, N Abberley, K Attack and J Allen.

Aintree University Hospital NHS Foundation Trust: P Lal, M Harrison, S Stevenson, C Brooks, P Harlow, J Ewing, S Cooper, R Balancio-Tolentino, L O’Neil, R Tagney and D Shackcloth.

St George’s Healthcare NHS Trust: T Planche, J Fellows, R Millett, J Studham, C de Souza, G Howell, H Greaves, E Foncel, R Kurup, J Briggs, M Smith, C Suarez, G Sorrentino, A Scobie, A Houston, F Ahmad, A Breathnach, R Chahuan and K Wilkins.

Blackpool Teaching Hospitals NHS Foundation Trust: A Guleri, N Waddington, R Sharma, P Flegg, V Kollipara, M Alam, A Potter, S Donaldson, C Armer and J Frudd.

King’s College Hospital NHS Foundation Trust: D Jeyaratnam, M Joy, A Mathews, SK Glass, A Ajayi, A Fife, S Qaiser, S Sheehan, S Muñoz Villaverde, NO Yogo, I De Abreu, G Notcheva, J Flanagan, C Watson, E Sais, A Adedayo, V Chu, G Shaw, MA Graver, R Palmer, D Palmer, S Haile, J Gordon, C Kai Tam, Mandar K and W Szypura.

Heart of England NHS Foundation Trust: N Jenkins, J Marange, V Shabangu, K Moore, J Lyons, M Munang, M Sangombe, E Moran and A Hussain.

University Hospitals of Leicester NHS Trust: M Wiselka, A Lewszuk, S Batham, K Ellis, L Bahadur, H White, M Pareek, A Sahota, S Coleman, H Pateman, A Kotecha, C Sim and A Rosser.

County Durham and Darlington NHS Foundation Trust: D Nayar, J Deane, R Nendick, C Aldridge, A Clarke, M Wood, A Marshall, L Stephenson, T Matheson-Smith, J Sloss, K Potts, J Malkin, L Ftika and V Raviprakash.

University Hospital Southampton NHS Foundation Trust: J Sutton, A Malachira, M Kean, K Criste, K Gladas, C Andrews, C Hutchison, E Adams, J Andrews, B Romans, N Ridley, M Ekani, J Mitchell, N Smith, T Clark, S Glover, R Reed, T Yam, H Burton and R Said.

Wirral University Teaching Hospital NHS Foundation Trust: D Harvey, A Janvier, R Jacob, C Smalley and A Fair.

Dartford and Gravesham NHS Trust: A Gonzalez-Ruiz, S Lord, K Ripalda, H Wooldridge, L Cotter, G Cardoso, E Strachan, G Kaler, A Mohamoodally, E Lawrence, Z Prime and R Abrahams.

Newcastle Upon Tyne Hospitals NHS Foundation Trust: DA Price, L Rigden, L Shewan, K Cullen, I Emmerson, K Martin, H Wilson, C Higham, K L Taylor, E Ong, B Patel, H Bond, J Gradwell and J Widdrington.

North Cumbria University Hospitals: C Graham, S Prentice, S Thornthwaite, U Poultney, H Crowther, H Fairlamb, E Hetherington, C Brewer, S Banerjee, C Hamson and A McSkeane.

Bradford Teaching Hospitals NHS Foundation Trust: P McWhinney, P Sharratt, J Thorpe, S Kimachia, H Wilson, B Jeffs, L Masters, J Wilson, J Platt and L Burgess.

Salford Royal NHS Foundation Trust: P Chadwick, A Jeans, C Keatley, A Moran, Z Swann, K Pagett, A Peel and J Howard.

Royal United Hospital Bath NHS Trust: S Meisner, K Maloney, A Masdin and L Wright.

Hull and East Yorkshire Hospitals NHS Trust: G Barlow, S Crossman, V Lowthorpe, E Moore, P Moss, A Parkin, A Wolstencroft, B Warner, C Tarbotton, A Eyre, A Anderson, T Burdett and A Driffill.

Contributions of authors

Guy E Thwaites (Professor of Infectious Diseases) drafted the report.

Matthew Scarborough (Consultant in Infectious Diseases and Microbiology) helped draft the report.

Alexander Szubert (Medical Statistician) performed the analysis.

Pedro Saramago Goncalves (Health Economist) designed and implemented the cost-effectiveness analysis and drafted Chapter 5.

Marta Soares (Health Economist) designed the cost-effectiveness analysis and drafted Chapter 5.

Jennifer Bostock (Public and Patient Representative) drafted the PPI sections.

Emmanuel Nsutebu (Consultant in Infectious Diseases) helped draft the report.

Robert Tilley (Consultant Microbiologist) helped draft the report.

Richard Cunningham (Consultant Microbiologist) helped draft the report.

Julia Greig (Consultant in Infectious Diseases) helped draft the report.

Sarah A Wyllie (Consultant Microbiologist) helped draft the report.

Peter Wilson (Consultant Microbiologist) helped draft the report.

Cressida Auckland (Consultant Microbiologist) helped draft the report.

Janet Cairns (Clinical Trials Manager) helped compile data and draft the report.

Denise Ward (Clinical Trials Manager) helped compile data and draft the report.

Pankaj Lal (Consultant Microbiologist) helped draft the report.

Achyut Guleri (Consultant Microbiologist) helped draft the report.

Neil Jenkins (Consultant in Infectious Diseases and Microbiology) helped draft the report.

Julian Sutton (Consultant in Infectious Diseases and Microbiology) helped draft the report.

Martin Wiselka (Consultant in Infectious Diseases and General Medicine) helped draft the report.

Gonzalez-Ruiz Armando (Consultant Microbiologist) helped draft the report.

Clive Graham (Consultant Microbiologist) helped draft the report.

Paul R Chadwick (Consultant Microbiologist) helped draft the report.

Gavin Barlow (Consultant Infectious Diseases) helped draft the report.

N Claire Gordon (Infectious Diseases/Microbiology Trainee) helped draft the report.

Bernadette Young (Infectious Diseases/Microbiology Trainee) sequenced and analysed bacterial isolates.

Sarah Meisner (Consultant Microbiologist) helped draft the report.

Paul McWhinney (Consultant Microbiologist) helped draft the report.

David A Price (Consultant Microbiologist) helped draft the report.

David Harvey (Consultant Microbiologist) helped draft the report.

Deepa Nayar (Consultant Microbiologist) helped draft the report.

Dakshika Jeyaratnam (Consultant Microbiologist) helped draft the report.

Timothy Planche (Consultant in Infectious Diseases) helped draft the report.

Jane Minton (Consultant in Infectious Diseases) helped draft the report.

Fleur Hudson (Clinical Trials Manager) helped compile data and draft the report.

Susan Hopkins (Consultant in Infectious Diseases and Microbiology) helped draft the report.

John Williams (Consultant infectious diseases) helped draft the report.

M Estee Török (Consultant in Infectious Diseases and Microbiology) helped draft the report.

Martin J Llewelyn (Professor of Infectious Diseases) helped draft the report.

Jonathan D Edgeworth (Consultant Microbiologist) helped draft the report.

A Sarah Walker (Professor of Medical Statistics) analysed data and drafted the report.

Publications

Thwaites GE, Scarborough M, Szubert A, Nsutebu E, Tilley R, Greig J, et al. Adjunctive rifampicin for Staphylococcus aureus bacteraemia (ARREST): a multicentre, randomised, double-blind, placebo-controlled trial. Lancet 2018;391:668–78.

Thwaites G, Auckland C, Barlow G, Cunningham R, Davies G, Edgeworth J, et al. Adjunctive rifampicin to reduce early mortality from Staphylococcus aureus bacteraemia (ARREST): study protocol for a randomised controlled trial. Trials 2012;13:241.

Data-sharing statement

All data requests should be submitted to the corresponding author for consideration. Access to available anonymised data may be granted following review.

Patient data

This work uses data provided by patients and collected by the NHS as part of their care and support. Using patient data is vital to improve health and care for everyone. There is huge potential to make better use of information from people’s patient records, to understand more about disease, develop new treatments, monitor safety, and plan NHS services. Patient data should be kept safe and secure, to protect everyone’s privacy, and it’s important that there are safeguards to make sure that it is stored and used responsibly. Everyone should be able to find out about how patient data are used. #datasaveslives You can find out more about the background to this citation here: https://understandingpatientdata.org.uk/data-citation.

Disclaimers

This report presents independent research funded by the National Institute for Health Research (NIHR). The views and opinions expressed by authors in this publication are those of the authors and do not necessarily reflect those of the NHS, the NIHR, NETSCC, the HTA programme or the Department of Health and Social Care. If there are verbatim quotations included in this publication the views and opinions expressed by the interviewees are those of the interviewees and do not necessarily reflect those of the authors, those of the NHS, the NIHR, NETSCC, the HTA programme or the Department of Health and Social Care.

Version 1.01

Original version approved by the London-Westminster Research Ethics Committee (REC).

Version 1.02

Amended safety reporting text, as requested by the Medicines and Healthcare Products Regulatory Agency:

• page 3, SAE reporting: ‘within 1 working day’ changed to ‘within 24 hours’

• page 44, 7.3 investigator responsibilities: ‘within 1 working day’ changed to ‘within 24 hours’

• page 46, 7.3.1.E notification: ‘within 1 working day’ changed to ‘within 24 hours’.

Version 2.0

Amended TSC chairperson, as requested by the Health Technology Assessment:

• page i, authorisation: Professor Jeremy Farrar changed to Dr Adrian Martineau

• page 72, appendix I: Dr Adrian Martineau named as chairperson.

Major edits:

• page ix, 55, 56: addition of UCL Hospitals to sites enrolling patients to intensive PK/PD study

• page ix, 38, 57: DNA and RNA stored for later genetic testing by storing 2.5 ml of blood in PAXgene Blood RNA tube at day 0 (no change in total volume of blood taken at that time point)

• page ix, 38, 56, 57: clarification of sample storage and subsequent archiving

• page 25, 32: clarification that rifampicin susceptibility testing by disc test or Vitek is acceptable

• page 76, 77, Patient information sheets: text changed to explain that DNA and RNA will be stored for later genetic testing.

Minor edits:

• pages i and v: added Clinical Trials Authorisation and Medical Research Ethics Committee (MREC) reference numbers

• page iii, MRC CTU staff: added clinical project manager

• page iv, site PIs: amended typographical errors in site names

• page vi, duration: changed to patient ‘recruited’ rather than ‘randomised’ over 3 years

• page vii, trial schema: ‘consenting’ changed to ‘consented’

• page 14, abbreviations: King’s College London (KCL) and King’s Health Partners (KHP) Clinical Trials Office (CTO) added

• page 23, 2.3 trial centres: amended typographical errors in site names

• page 27, 4.1 randomisation practicalities: amended CRF names. Amended text to clarify sample storage.

• page 30, treatment schedule: clarified that those ≥ 60 kg will receive 900 mg of rifampicin.

• page 31, 5.3.1 dispensing: amended text, ‘will be prescribed by the local pharmacist’ changed to ‘will be dispensed by the local pharmacist’

• page 44, 7.2.1 pregnancy: amended typographical error, ‘rifampicin in safe’ changed to ‘rifampicin is safe’

• page 48, 8.1 risk assessment: Quality Management Committee changed to Research Governance Committee

• page 58 and 63, site compliance and Finance: agreement between site, KCL and MRC

• page 81, 83, appendix III: version and date of patient information sheets and consent forms changed to version 2.0, 23 August 2012. Reformatted optional consent boxes

• page 85, appendix IV: added MREC reference number.

Version 3.0

Reasons for substantial amendment: (1) remove KCL as co-sponsor and (2) add four new trial sites:

• page i: remove KCL logo

• pages ii and vii, sponsor: remove KCL

• pages iv and v, site PIs: Drs Sutton, Guleri, Minton and Munthali

• page 24, 2.3 trial centres: Southampton, Blackpool, Leeds and Coventry

• page 63, indemnity: remove King’s College London

• page 66, 15.6: remove paragraph ‘Role of Study Sponsor’.

Major edits to intensive and sparse PK/PD substudy information:

• Pages ix–xi, trial assessment schedule (h), (i), (k): changed time point of PK/PD lithium heparin blood collections from day 0 to day 1. Clarified total volume of blood to be collected. Added reference to PK/PD laboratory manual.

• Page x, (j): clarified procedure for intensive PK/PD component studies.

  • Pages 56–58, 11.1–11.1.2: updated sampling time points and procedures.

  • Page 81, appendix II, Intensive substudy: updated sampling time points.

  • Page 82, appendix II, Sparse substudy: updated sampling time points.

Major edits to bacterial genetic substudy information:

• Pages ix and x, trial assessment schedule (g): single nasal bacterial swab added at baseline.

• Page 58, 11.2: nasal swab added to study details. Clarification that bacteria will be archived in Oxford and Brighton.

• Page 77, appendix II, What will happen if I take part?, point 1: updated to include baseline nasal swab.

Minor edits:

• Page iii: updated randomisation summary and SAE reporting summary.

• Page iii, MRC CTU Staff: added Trial Statistician Alex Szubert.

• Page iv, chief investigator: updated CI address and e-mail address.

• Page v, site PIs: updated Sheffield PI to Dr Julia Greig.

• Page x, (h): amended typographical errors ‘University of Liverpool’ and ‘Davies’.

• Page 23, 2.2: amended pack name and document name.

• Pages 26, 3.4: ‘Trial register’ changed to ‘Screening & Randomisation Register’.

• Page 28, 4 randomisation: added text to clarify randomisation via the ARREST trial database.

• Pages 30, 32, 35 and 51, 5.1, 5.2.2, 5.2.3, 5.3.1, 5.7, 9.1: updated name of Clinical Trials Supplier to Sharp Clinical Services (formerly Bilcare Global Clinical Supplies).

• Page 31, 5.2.3 blinding issues, second paragraph, last sentence: added text to clarify that infusion volumes used in the ICU may be altered in accordance with local standard practices and the product’s SPC.

• Page 31, 5.3 treatment schedule, third paragraph: changed ‘intended’ to ‘initial’.

• Page 32, 5.3.1 dispensing, second paragraph, last sentence: added text to clarify that temperature monitoring of treatment packs is not required after dispensing.

• Page 32, 5.3.1 dispensing, fourth paragraph: changed ‘Dispensing Log’ to ‘Accountability Log’, and added ‘temperature monitoring’.

• Page 34, 5.5.2 unblinding by the MRC CTU: updated Trial Manager contact number.

• Page 38, 6 assessments and follow-up, second paragraph: added text to clarify that the final follow-up visit may take place over the telephone.

• Page 41, 6.6 early stopping of follow-up, first paragraph: amended CRF name.

• Page 47, 7.3.2 notification procedure, points 1, 2 and 3: updated text to clarify notification practicalities using the ARREST trial database.

• Page 49, 8.2 central monitoring: amended typographical error ‘Case Report Forms’.

• Page 51, 9.1 method of randomisation: changed ‘website randomisation service’ to ‘the ARREST trial database’. Updated name of Delegation Log.

• Page 52, 9.2 outcome measures, last paragraph: added ‘clinical’.

• Page 53, 9.4 interim monitoring and analyses, fifth sentence: update text to clarify Haybittle-Peto rules in the context of this trial.

• Page 53, 9.5 analysis plan (brief), second paragraph: added more detailed text regarding analysis of end points.

• Page 54, 9.5 analysis plan (brief), third paragraph: changed ‘intended’ to ‘initial’.

• Page 65, 15.1–15.2: added ‘as required’ to clarify Trial Management Team and Trial Management Group membership.

• Page 77, appendix II, What will happen if I take part?, point 3, second paragraph, second sentence: amended error in text by removing day 7 time point.

• Page 79, appendix II, storing samples: added more detailed name of storage facility.

• Pages 84 and 86, appendix III: updated protocol version number and date. Corrected typographical errors in legal representative consent form.

Version 4.0

Reasons for substantial amendment: addition of substudy – experiences of being approached for trial participation, the consenting process and trial participation.

Major edits: addition of new substudy –

• page 56: section 11 text amended to include participation and consenting substudy and number of ancillary studies changed from two to three

• page 59–61: section 11.3 new section and text added to give details of the participation and consenting substudy

• page 95: appendix VII questionnare for non-consenting patients and legal representatives who are not health-care professionals added

• page 98: appendix VIII questionnaire for non-consenting health-care professionals added

• page 100: appendix IX information sheet for consenting patients or legal representatives added.

Minor edits:

Throughout document: change of name of co-ordinating site from MRC CTU to MRC CTU at UCL.

Throughout document: change of e-mail address for co-ordinating centre from arrest@ctu.mrc.ac.uk to mrcctu.arrest@ucl.ac.uk.

• Page ii: infections theme corrected to ARREST trial team.

• Page iii: co-ordinating site generic telephone number removed.

• Page iii and iv: MRC CTU at UCL staff titles, names, e-mail addresses and telephone numbers updated.

• Page iv: site PIs table removed.

• Page v: protocol version and date updated.

• Page vi: trial manager name updated; Project Lead title amended.

• Page vii: amended trials schema error – day 10 added to follow up visits.

• Page viii: added text to Trial assessment schedule footnote to clarify that if a patient is discharged before day 14 but is attending outpatient appointments then blood samples should be collected if possible.

• Page 16: MRC CTU amended to MRC CTU at UCL; percutaneous endoscopic gastrostomy (PEG) and UCL added to abbreviations table.

• Page 24: section 2.3 trial centres removed.

• Page 25: section 3.1 added text to clarify that stat doses should be excluded from the 96 hour active antibiotic therapy in inclusion criteria.

• Page 31: section 5.3 added text to clarify that it is not permissible for IMP capsules to be opened and administered via PEG.

• Page 36: section 5.8 added text to clarify that if a patient receives additional doses of IMP for > 24 hours (i.e. 15 days of trial treatment) then this must be reported as a protocol deviation.

• Page 39: section 6.1 added text to clarify that if a patient is discharged before day 14 but is attending outpatient appointments then blood samples should be collected, if possible.

• Page 44: section 7.1.2 moved bullet points describing AEs that should be reported to page 44, section 7.1 in order to prevent confusion. Added text to clarify that disease-related events that are not fatal are exempt from being reported as SAEs.

• Page 49: amended text to clarify that site initiation visits may occur either as a site visit or by WebEx (Cisco, San Jose, CA, USA) teleconference.

• Page 66: section 13 MRC indemnity text removed, UCL insurance text added.

• Page 79: appendix I Professor Jeremy Farrar removed from TSC membership, Dr Achim Kaasch added; trial statistician title amended.

• Page 80–86: appendix II friend/relative amended to friend/relative/patient to clarify that a doctor primarily responsible for a patient’s medical treatment may also act as a legal representative.

• Page 87: appendix II what is a legal representative information sheet. Text added to clarify that a doctor primarily responsible for a patient’s medical treatment may also act as a Legal Representative.

• Page 88: appendix III trial participant consent form. Updated protocol and version number. ‘Doctor’s signature’ changed to ‘Signature of person delegated to take consent’ since some trusts allow nurses to take consent.

• Page 90: appendix III Legal Representative consent form. Updated protocol and version number. Friend/relative amended to friend/relative/patient to clarify that a doctor primarily responsible for a patient’s medical treament may also act as a legal representative. ‘Doctor’s signature’ changed to ‘Signature of person delegated to take consent’ since some Trusts allow nurses to take consent. Box for participant’s name added.

Version 4.1

Minor edits requested by REC.

• Page 92: appendix IV, GP letter amended to clarify that consent may have been granted by a legal representative. Version number and date of GP letter updated.

Protocol version number and date updated throughout protocol.

Version 5.0

Reason for substantial amendment: sample size reduced and co-primary end point (all-cause mortality up to 14 days) reassigned as a secondary end point at the request of the funder.

Major edits: sample size reduced and co-primary end point (all-cause mortality up to 14 days) reassigned as a secondary end point.

• Page v, summary of trial: ‘all-cause mortality up to 14 days from randomisation’ removed as a primary end point, added as a secondary outcome measure.

• Page vi, summary of trial: number of patients to be studied changed from ‘940’ to ‘770’.

• Page viii, trial schema updated to reflect change in sample size and end points.

• Section 1.5, hypothesis and objectives: ‘all-cause mortality up to 14 days from randomisation removed as a primary objective, ‘evaluating the impact of rifampicin on all-cause mortality up to 14 days from randomisation’ added as a secondary objective.

• Section 3.3, number of patients: number of patients to be studied changed from ‘940 patients (470 in each treatment arm)’ to ‘770 patients (385 in each treatment arm)’; ‘enrolled over a target of 3 years’ changed to ‘enrolled over a target of 3.5 years’.

• Section 6.2, procedures for assessing efficacy: ‘all-cause mortality up to 14 days from randomisation’ removed as a primary end point, added as a secondary outcome measure.

• Section 9.2, outcome measures: ‘all-cause mortality up to 14 days from randomisation’ removed as a primary end point, added as a secondary outcome measure. Text modified to clarify.

• Section 9.3, sample size: updated to reflect sample size change from ‘940’ to ‘770’.

• Section 9.5, analysis plan (brief): ‘all-cause mortality up to 14 days from randomisation’ removed as a primary end point.

Minor edits:

• Page vi, summary of trial: ‘experiences of being approached for trial participation, the consenting process and trial participation’ added to Ancillary/substudies section (omitted in error in version 4.1 of protocol).

• Section 5.3.1, ‘the ward nurse’ changed to ‘the ward or the ARREST trial research nurse’ to clarify that the ARREST trial nurse may take the prescription form to the pharmacy.

• Section 5.3.2, unblinding by MRC CTU at UCL: trial manager telephone number corrected.

• Section 9.4, interim monitoring analyses: changed from ‘the DMC will in general meet’ to ‘the DMC will meet at least once ber year’ for clarification.

• Appendix VII, substudy questionnare: ‘relative/spouse’, ‘relative’ ‘friend/relative’ have been corrected to ‘relative/friend’ for consistency.