Early switch from intravenous to oral antibiotic therapy in patients with cancer who have low-risk neutropenic sepsis: the EASI-SWITCH RCT

Article history

The research reported in this issue of the journal was funded by the HTA programme as award number 13/140/05. The contractual start date was in October 2015. The draft report began editorial review in September 2021 and was accepted for publication in October 2022. 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’ manuscript 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 manuscript.

Copyright statement

Copyright © 2024 Coyle et al. This work was produced by Coyle et al. under the terms of a commissioning contract issued by the Secretary of State for Health and Social Care. This is an Open Access publication distributed under the terms of the Creative Commons Attribution CC BY 4.0 licence, which permits unrestricted use, distribution, reproduction and adaptation in any medium and for any purpose provided that it is properly attributed. See: https://creativecommons.org/licenses/by/4.0/. For attribution the title, original author(s), the publication source – NIHR Journals Library, and the DOI of the publication must be cited.

2024 Coyle et al.

Neutropenic sepsis

Neutropenic sepsis (NS) is a long-recognised and common complication of systemic anticancer treatment (SACT). 2 The term broadly refers to a significant inflammatory response to a presumed bacterial infection, in a person with or without fever and a low blood neutrophil count. 3 A low neutrophil count occurs commonly following SACT with a trajectory that varies depending on type and timing of SACT, typically reaching a nadir around 7 days post SACT then recovering over 2–3 weeks. 4 There is significant variation in the definition of neutropenia and sepsis, with a lack of comparative data to guide threshold-setting for neutrophil count or fever in patients with potential NS. 3,5–15 Based on the available data, the National Institute for Health and Care Excellence (NICE) Guideline Development Group (GDG) concluded using a threshold of 0.5 × 10⁹/l for neutrophil count and ≥ 38°C for temperature for diagnosis of NS reflected an acceptable trade-off between over- and under-treatment of what could be a potentially fatal infection. 3 However, in the NHS, NS care pathways commonly use a temperature threshold of ≥ 38°C and an absolute neutrophil count threshold of either ≤ 0.5 × 10⁹/l or < 1.0 × 10⁹/l and falling/expected to fall.

Robust predictors of NS risk are lacking in patients with cancer receiving SACT. It seems that NS is more common soon after treatment is initiated (within the first two cycles)16 and following administration of anthracycline or taxane-containing regimens in treatment of early-stage breast cancer. Other factors associated with risk of developing NS include age, performance status and a diagnosis of blood cancer rather than solid tumour. 17,18

Despite widespread adoption of prophylactic colony-stimulating factors (CSF) and fluoroquinolone antibiotics for patients at high risk of septic complications, NS remains potentially life-threatening, with an in-hospital mortality rate of approximately 9.5%19 and, in the setting of severe sepsis or septic shock, as high as 50%. 20 NS deaths recorded by the Office of National Statistics more than doubled in England and Wales between 2001 and 2010 to approximately two deaths per day (716 deaths in 2010),3 with significant increases in chemotherapy use likely contributing. 21

Significant patient morbidity can also occur through hospitalisation, with a strong desire not to be hospitalised during treatment cited as a common barrier to patients promptly seeking help for symptoms. 22 An episode of NS can also result in dose delays and reductions to patients’ planned SACT, potentially compromising treatment efficacy in certain tumour types and settings. 3,23–25 There are associated financial implications on healthcare systems managing NS episodes, with hospital, antibiotic, diagnostic and additional therapeutic costs involved resulting in an estimated average cost per inpatient admission in the NHS ranging from approximately £257226 to £3163. 27

Empirical management of NS

Neutropenic sepsis continues to be viewed as a time-critical medical emergency with widespread agreement that early recognition and prompt administration of broad-spectrum empirical antibiotics are essential to successful treatment. 3,21,28,29 However, there is much less consensus on optimal patient management thereafter, including when to switch from intravenous (i.v.) to oral antibiotics, and duration of antibiotic treatment and hospital admission. Widely variable practice has been noted among cancer centres in the UK. 30 A review of 51 English and Welsh centres, prior to the introduction of a national UK NS clinical guideline, highlighted the heterogeneity that existed in almost all aspects of NS management, commenting that there was ‘surprisingly little agreement’ and ‘dramatic variations’ in clinical practice. These findings were consistent with previously published audits of both adult and paediatric haemato-oncology practice. 6,31,32

The NICE GDG recommended use of empirical beta-lactam monotherapy (piperacillin/tazobactam) as immediate treatment in patients with suspected NS in the absence of local microbiological indications to use an alternative agent or combination therapy (such as addition of an aminoglycoside) based on local resistance patterns. 3 Evidence supporting a specific duration of treatment was found to be lacking by the Group but the principle of switch to oral antibiotics following risk assessment after 48 hours of i.v. antibiotic therapy could be considered. While current European guidelines suggest that following initial assessment, including prompt institution of empirical broad-spectrum antibiotics, patients identified as low risk may be suitable for inpatient oral antibiotic treatment, the authors noted clinician preference was often to continue i.v. treatment for at least 48 hours and then consider a change to oral antibiotics if fever resolves. 28

Risk stratification

A spectrum of NS severity exists, encompassing a heterogeneous group of patients with variable risk of septic complications such as organ failure, need for critical care support and death. 33

At the low-risk end of the spectrum, there are patients who do not demonstrate clear clinical or microbiological evidence of proven infection, have uncomplicated hospital admissions and are at low risk of developing septic complications. These patients potentially receive overtreatment, with the associated distress of hospitalisation and additional burden to the healthcare system. 34

Risk stratification tools have therefore been developed in an attempt to identify patients predicted to be at low risk of an adverse outcome. The Multinational Association of Supportive Care in Cancer (MASCC) score (Table 1) is the most widely validated risk score for SACT-induced NS. 3,28,29,33

In the original validation set, a risk index score of 21 or greater out of 26 identified low-risk patients with a positive predictive value of 91%, a sensitivity of 71% and a specificity of 68%. A low rate of adverse outcomes (6% serious medical complication, 1% mortality) was observed in patients with a risk index score ≥ 21 compared with 49% in those with a risk score of < 21. 35

Oral antibiotic therapy as treatment

A UK single-centre prospective randomised controlled trial (RCT) investigated the effectiveness of oral antibiotics (ciprofloxacin and co-amoxiclav) with early hospital discharge for patients with low-risk NS in comparison to standard i.v. antibiotics (tazobacam-piperacillin and gentamicin) with hospital admission. One hundred and twenty-six NS episodes from 102 patients were evaluated for the ‘success and safety’ primary end point, which comprised: lysis of fever, resolution of symptoms and signs, absence of modification of antibiotic regimen, absence of recurrence within 7 days and occurrence of serious complications or deaths. Treatment success was 90% in the i.v. arm and 84.8% in the oral arm. Re-admission was required in five episodes (7.6%) and deemed unrelated to the episode of NS in one. It was recognised that these results were obtained within a single specialist centre but that the findings supported undertaking a multicentre RCT evaluating oral antibiotics and early discharge. 36

A Cochrane review37 of oral versus i.v. antibiotics for NS, evaluating 22 trials comprising 3142 neutropenic episodes in 2372 patients, concluded that it is not likely that significant differences exist in treatment failure or mortality rates between oral antibiotic and i.v. antibiotic strategies. There was a trend towards more adverse events (AEs) in patients receiving oral antibiotics, typically gastrointestinal events, which did not necessitate treatment discontinuation. The majority of studies did not utilise any formal risk stratification tools but excluded high-risk patients with acute leukaemia, haemodynamic instability, evidence of organ failure or localising signs of infection.

The Cochrane review therefore broadly supported the early use of oral antibiotics in low-risk NS, but it was noted most trials were small in sample size, often single-centre and with methodological concerns and so a robust recommendation for upfront or early oral antibiotic therapy could not be made. It was suggested that ‘the combination of a quinolone and a second drug active against Gram-positive bacteria (for example ampicillin-clavulanate) seems prudent’. 37 This group also recommended that this therapeutic approach should be formally evaluated in patients with low-risk NS. The NICE GDG also considered oral antibiotic therapies but were unable to make a specific recommendation given variation in local microbiological resistance patterns and variation in use of prophylactic antibiotics. 3

Outpatient management of low-risk NS

The NICE GDG reviewed the evidence for inpatient versus outpatient management of NS and concluded outpatient management can be considered for selected low-risk patients, taking into account their individual clinical and social circumstances. 3 Although the metaregression undertaken by the GDG suggested early discharge (before 24 hours) may be associated with increased likelihood of re-admission or therapy change, the quality of evidence supporting outpatient management was low to moderate. The available data were limited by a lack of reporting of key outcomes such as critical care admission or clinically documented infection and a very low event rate for adverse outcomes including death. 3 Similarly, there is negligible literature relating to impact on quality of life for different models of care, including immediate use of oral antibiotics and non-admission to hospital, with a single study suggesting role function improved more for inpatients than home care patients but that emotional function declined with hospital admission. 38 It was therefore proposed that if a short period of hospital admission was found to be safe and effective for selected patients with NS, it could provide considerable improvements in quality of life and health resource usage.

Rationale for the trial

NICE therefore recommended that a randomised trial should be undertaken to evaluate the effectiveness of switching from i.v. to oral antibiotics within the first 24 hours of treatment in patients receiving i.v. antibiotics for NS. The early switch to oral antibiotic therapy in patients with low-risk NS (EASI-SWITCH) trial was developed in response to this recommendation and a commissioned call from the UK National Institute for Health and Care Research (NIHR) Health Technology Assessment (HTA) programme to address this evidence gap. It aimed to establish the clinical and cost effectiveness of an early switch to oral antibiotics 12–24 hours after i.v. antibiotic treatment commences in low-risk cancer patients with NS.

Acute oncology service development

While prospective RCTs evaluating upfront or early oral antibiotics remain lacking from publication of the NICE guidance, there have been significant service developments in response to acute care pressures and increasing demands due to increases in cancer incidence and available treatments with the aim of developing novel models of care for meeting cancer patients’ complex needs. 39 Ambulatory care has been introduced in acute medical and elderly care NHS settings with growing interest in developing care pathways in cancer services in recent years. This has included management of low-risk NS although it has only been implemented by a limited number of UK cancer centres. 40,41 Some have reported results from longitudinal patient series. Marshall et al. 42 reported a series of 100 patients from a large UK tertiary cancer centre over a 2-year period with NS who were assessed and given a first dose of i.v. antibiotics then managed on an ambulatory low-risk NS pathway. Patients were stratified using MASCC score and National Early Warning Score (NEWS) score and following observation for at least 4 hours were discharged for outpatient follow-up (repeat clinical assessment and routine bloods) within 48 hours. 42 Six of the 100 patients (8.8%) required re-admission within 7 days, typically with positive blood cultures, but none required critical care support. Brunner et al. 43 reported a low-risk NS ambulatory care pathway case series from another UK centre. One hundred and twenty-three patients presented with NS over a 2-year period, 41% of whom were deemed low risk based on MASCC score. Of these, 24 were managed on the ambulatory care pathway with a first dose of i.v. antibiotic and discharge with oral antibiotics and proactive telephone follow-up. A further 24 patients were admitted but had early discharge. Again, no serious complications occurred and the re-admission rate was 10%. However, despite the investment in establishing the ambulatory care pathway, approximately 80% of patients with NS were still admitted. Similarly, other international centres have reported real-world data where only a minority of patients are managed on ambulatory pathways or considered for same-day discharge. 44 For example, in a large US emergency department (ED) only 5% of NS patients were discharged home, with most low-risk patients admitted for inpatient antibiotics. 45 A subsequent large-scale review of approximately 350,000 US ED visits with NS confirmed this finding, with 94% of visits resulting in hospitalisation. 46 Cost analysis data from real-world data sets are also lacking.

Trial design

EASI-SWITCH was a UK prospective Phase 3, randomised, open-label, non-inferiority trial to evaluate whether early switch to oral antibiotics is non-inferior to standard care in adult patients with cancer with NS at low risk of complications.

The main aim was to determine the clinical effectiveness of early switch to oral antibiotics 12 to 24 hours after commencement of empirical i.v. antibiotics compared to standard care, which comprises continuation of i.v. treatment for at least 48 hours, based on treatment failure rate. Treatment failure was defined by a composite measure incorporating a number of important clinical outcomes assessed at day 14 of follow-up.

The trial included an embedded pilot study across four UK sites in order to test the recruitment and adherence assumptions which had informed the trial design.

Trial objectives

Research hypotheses

1. Early oral switch (within 12–24 hours after commencing i.v. antibiotic therapy) in cancer patients with low-risk NS is non-inferior to standard care (continuation of i.v. antibiotic therapy for at least 48 hours).

2. The incremental cost effectiveness of early oral switch is significant compared to standard care.

3. AEs are comparable between the two study arms.

4. Patients’ preference will be for early oral switch.

Study conduct

Patient information and consent

Potentially eligible patients were those who had commenced treatment with i.v. antibiotics for NS. Patients were identified at each study site daily through local acute admission/handover processes dependent on the unscheduled care admission pathways at site. Patients meeting these criteria were discussed with their treating physician on that day prior to enrolment to confirm their agreement to patient participation. This also provided an opportunity to confirm that their treating physician would be willing to follow the treatment strategy outlined in either arm of the trial. Patients were approached by a member of the research team and a patient information sheet was provided. Patients were given time to review the patient information sheet although this time period was < 24 hours given the acute care setting and timing of the intervention.

As enrolment was occurring at ward level and patients had already been initiated on treatment, patients being approached were clinically stable and viewed as competent to give informed consent in this setting. Patients who were unable to give informed consent, for any reason, were not recruited. Patients who indicated they were unwilling or unable to make a decision within the 24-hour time period were not recruited. Regulatory approvals were obtained for patient-facing materials additional to the patient information sheet to be used at site to make patients aware of the trial. These included a summary information sheet about the trial that could be included in the standard SACT patient education materials about NS and a poster to be displayed in SACT clinics and treatment units. All of these materials were prepared in collaboration with the trial patient representatives.

Informed consent for participation was sought from patients by appropriately trained research nurses (RNs) and medically trained investigators at site supported by the site principal investigator (PI) and local infrastructure. If patients required any further clarification about the risks and benefits of participation, this was provided by other research team members or an independent senior physician (one nominated in advance at each trial site). The PI (or designee) taking informed consent was required to have completed GCP training, be suitably qualified and experienced and be delegated this duty by the PI on the delegation log.

Screening and randomisation procedures

Electronic trial screening and recruitment logs, submitted by sites to the clinical trial unit (CTU) on a monthly basis, aimed to capture all patients who received a patient information sheet and whether they proceeded to consent and randomisation. Research teams were asked to provide a reason for non-participation if patients were not recruited.

After informed consent was obtained and eligibility was confirmed, participants were allocated to intervention or standard care groups using an automated randomisation system (sealed envelopes). Blocked randomisation with randomly permuted block sizes was used and a 1 : 1 allocation ratio. There were no factors for stratification. Access to the randomisation sequence was restricted and not accessible to site staff who enrolled patients or assigned interventions. Only the allocation of the intervention was blinded. As this was a pragmatic trial, it was felt that blinding clinical teams, researchers and trial participants to the intervention would limit the ability of the trial to measure the impact of the intervention on routine care pathways. Additional support from this approach came from our patient and public involvement (PPI) representatives, who viewed that participants would be highly likely to reveal their treatment allocation during discussion with healthcare providers and outcome assessors, making it unlikely these groups could remain blinded.

Trial treatment

Patients eligible to participate were aged over 16 years, receiving SACT for a cancer diagnosis and were receiving standard-dose i.v. piperacillin/tazobactam or meropenem as initial antibiotic treatment for suspected NS for < 24 hours. Patients were only permitted to be enrolled in the trial on one occasion in line with consensus guidelines. 47 All protocolised antibiotics were considered to be investigational medicinal products (IMPs) for the purpose of the trial: co-amoxiclav 500 mg/125 mg film coated tablets; ciprofloxacin 250 mg, 500 mg, 750 mg film coated tablets; meropenem 1 g powder for solution for injection or infusion and tazocin 4 g/0.5 g powder for solution for infusion.

Patient population

Patients commenced on i.v. antibiotics within 24 hours after starting treatment for low-risk NS were recruited from sites across England, Scotland, Wales and NI, comprising both large cancer centres and cancer units, to ensure that the sample is broadly representative of patients developing NS in the UK.

Withdrawal of consent

Participants were able to withdraw consent to participate in the trial at any time. If the participant withdrew consent during protocolised treatment, no further treatment within the trial was given and the clinician responsible for their care determined the safest and most appropriate continued management plan. A withdrawal of consent form identified which parts of the trial the patient wished to withdraw from: protocol-specified antibiotic therapy; future data collection (all data collected or data collected at day 14 and/or day 28 follow-up). Participants could be withdrawn from the study at the discretion of the investigator if any safety concerns arose.

Data management

Adverse event reporting

Directly observed or patient-reported AEs that were not related to underlying medical conditions were recorded by the site PI or designee. AEs clearly related to SACT administration (such as peripheral neuropathy) were not required to be recorded; however, if an AE could be due to SACT, NS or antibiotic therapy (such as fever or gastrointestinal symptoms), then it was required to be recorded. Initially the AE reporting period for the trial was from enrolment until 28 days after randomisation. This was amended subsequently to 14 days from enrolment until 14 days after randomisation in recognition that antibiotic AEs generally occurred within this time frame and that patients were typically receiving a further course of SACT within the 14–28 day window, resulting in AEs that were more likely to be SACT-related or a new episode of NS rather than related to the episode of NS that had resulted in trial entry or antibiotic therapy.

The PI (or designee) was required to make an assessment of expectedness of any AE deemed possibly, probably or definitely related to any of the trial IMPs based on the relevant Summary of Product Characteristics (SPCs).

AEs related to IMP exposure were deemed ARs. ARs were classified as expected (consistent with IMP side effects listed in the SPC) or unexpected (not consistent with the SPC).

Protocol amendments

All amendments to the trial protocol, patient information sheet, informed consent form and other key documents were submitted to the relevant regulatory authorities for approval prior to implementation. Local research and development approval was also obtained at each site. Version 2 of the trial protocol was the protocol approved for use as trial commencement and subsequent substantial protocol amendments are summarised below.

Embedded pilot study

The trial contained an embedded pilot study involving four UK sites to test the recruitment and adherence assumptions underpinning the study design. Progression to the full study was based on the following criteria:

1. Recruitment rate:

   1. progression without major modification if at least 75% of target reached

   2. progression with addition of further trial sites if between 50% and 75% of target reached

   3. progression unlikely if < 50% of target reached – discussion with trial oversight committees and funder required.

2. Adherence to protocol-specified intervention:

   1. progression without major modification if at least 75% adherence in both trial arms

   2. if adherence was between 50% and 75% of target, progression would be supported by a detailed analysis of the process and decision-points that led to non-adherence and a recognised strategy to address this identified

   3. progression unlikely if < 50% adherence in either arm.

3. Separation:

   1. separation in terms of the timing of antibiotic switch of at least 24 hours between the trial arms to enable progression was required.

The four-site pilot study was expected to run for 6 months between February and July 2016. Recruitment was < 50% target threshold for progression and the study was halted to review progress and proposals to address under-recruitment. This review identified the stringent eligibility criteria as a barrier to recruitment and not reflective of the routine management of patients with suspected NS. A protocol amendment (July 2016, described above) was submitted to address this. Recruitment was resumed with a 3-month extension to the pilot study (December 2016 to February 2017) with an improvement in recruitment rates (11 patients of an anticipated 13.5 recruited). On the recommendation of the HTA Programme Director, a larger pilot phase extension followed from April until November 2017. Progression with addition of new sites continued from this point until the end of the study.

Statistical analysis plan

Survey design and delivery

A series of surveys were undertaken to obtain stakeholder feedback on important questions relating to the trial. All electronic surveys were conducted using the SurveyMonkey tool (www.surveymonkey.co.uk). Ethical approval was not required for these projects as they sought only to define clinicians’ current standard of care and were categorised as service evaluation rather than research according to UK Health Research Authority definitions. All were designed pragmatically for data collection and to encourage responses. They were piloted informally by clinical and PPI colleagues at the lead site prior to wider dissemination to improve clarity and understanding but did not undergo formal reliability or validity testing. Participation was voluntary, with potential responders assured that no site or personal information would be publicly shared. Onward e-mail circulation to appropriate colleagues was encouraged to maximise responses, accepting this makes estimation of response rates challenging. At least one e-mail reminder was sent also to encourage responses, with all surveys open for completion for at least 8 weeks in total.

The analyses of the survey responses are presented descriptively, with percentages rounded to the nearest 1%. If the survey contained questions where respondents had the opportunity to leave comments, these are presented thematically, with representative illustrative quotes.

Site interviews

To understand factors impacting trial recruitment at sites, a series of purposeful semistructured interviews with key personnel at each site, including PIs and lead RNs, were conducted by the trial Clinical Research Fellow over a 2-month period (April to May 2018). Qualitative research has previously been demonstrated as one of the most helpful tools in identifying and overcoming barriers to recruitment. 48,49 The aim of the interviews was to explore clinician’s experience of recruiting and delivering the trial at their site and identify barriers to recruitment.

An initial e-mail invitation describing the nature and purpose of the interviews was sent to all PIs and lead RNs at all 12 open sites in March 2018. Participation was voluntary, and participants were reassured that no individual or site-specific information would be identified in summary reports. Interviews could be conducted either face to face or via telephone depending on participant preference. On receipt of a positive response, an interview was arranged and all were conducted in April and May 2018. Verbal consent was confirmed at the commencement of each interview. A short interview guide was prepared based on broad themes that had already emerged from previous discussions about the trial’s progress and difficulty recruiting. Participants were first asked in an open-ended question to simply comment on their experience recruiting to the trial and then highlight any challenges encountered. If not already discussed, they were then prompted to provide feedback on the trial’s screening and recruitment processes, including eligibility criteria. They were also prompted to comment on any issues encountered with local support for the study, research capacity, participant follow-up and data collection. Finally, interviewees were asked to highlight any strategies they felt had facilitated recruitment at their individual site. Interviews ranged from 15 to 50 minutes in length.

With the participant’s permission field notes were taken during the interview to capture key responses. A summary of the main findings was then verbally confirmed by the participant at the end of each interview to ensure content validation. Data saturation (i.e. no new information raised in later interviews) was reached during the interviews.

Thematic analysis of the qualitative data was then undertaken50 to generate a summary list of reported barriers to trial recruitment. The field notes for each interview were first scrutinised individually and organised into themes relating to barriers to recruitment. This systematic process was repeated for all interviews and then combined to produce a final group analysis summary list. Post analysis of the data, the co-chief investigators reviewed the data for completeness, reporting coherence between the data and reported themes to ensure robustness of the interpretation of the data.

Patient and public involvement

The trial has benefitted from having an experienced PPI member in the team, Mrs Margaret Grayson, from study conception to completion. The trial was developed in response to a commissioned call, meaning the main research question was pre-defined; however, input from the Northern Ireland Cancer Research Consumer Forum (NICRCF) via Mrs Grayson was that this was an important research question of value to patients where potential overtreatment and prolonged hospital admission may negatively impact upon quality of life. From this position, Mrs Grayson contributed to the trial design with particular input in the following areas: (1) helping to define the composite outcome measure and define secondary outcome measures important to patients and the health service; (2) defining the non-inferiority margin to incorporate patients’ views on acceptable trade-offs for treatment de-escalation; and (3) providing the patient viewpoint on the most appropriate method to obtain day 14 outcome data balancing the need for data quality with burden on patients. Additional support was given by the readers panel of the NICRCF at this stage through development and review of the trial protocol, the patient information sheet, the informed consent form and other patient-facing materials, for example brief summary flyers and posters for sites to use in SACT treatment units to make patients aware of the trial.

As the trial progressed, there was ongoing input from Mrs Grayson through both her membership of the TSC and participation in the TMG. Her input was critical when the overall study design was reviewed during the course of the trial, co-ordinating PPI opinion on the proposed changes and contributing to the final amended design. Through her linkages with the NICRCF and nationally (with the Independent Cancer Patient Voice group) she sought opinion on the ongoing importance of the research question and whether there was support for continuation of the trial with changes to the study design. As part of this process, she co-developed with the study team a PPI survey outlining proposed changes to the study design with particular focus on whether additional uncertainty around treatment effectiveness arising from a change in the non-inferiority margin and sample size would be acceptable to patients. She then collated responses, reporting to the study team the support of the majority of respondents and providing written and verbal communication with the funder in decision-making about these proposed changes.

The trial was also supported by additional independent PPI representation on the TSC and DMEC. Input from these representatives in review of trial data and progress was critical at decision-making points relating to trial progression and potential amendments to study procedures. Dissemination of results is an ongoing area of activity for all of the PPI team trial team members.

Pilot study aim

The aim of the internal pilot study was to carefully evaluate the recruitment and adherence assumptions underpinning the main study design. The main parameters of interest were:

• recruitment rates

• adherence to the protocol-specified treatment

• separation in terms of timing of the antibiotic switch between the two arms.

These criteria were set to guide trial progress and inform the procedures to be utilised in the delivery of the main trial.

Initial pilot phase (February to July 2016)

The internal four site pilot study was expected to run for 6 months. It commenced on the 17 February 2016, when Site 1 opened and completed on the 21 July 2016. Progress is summarised in Table 4.

Recruitment of 31 patients was originally projected if all four sites had been open to recruitment for the full 6 months. This was based on a recruitment rate of 1.7 patients per month per site, allowing for a 50% reduction in recruitment at each site during the first 3 months as sites became established. However, delays encountered at site level resulted in only one of the four sites being open for the full 6 month period. Taking this into account recruitment of 22 patients was anticipated, but only 7 were recruited. In view of this, the study was halted to enable review of screening and recruitment activity and recruitment target.

The original accrual target was based on historically published NS surveys and audit data from two sites, suggesting approximately 20 patients were admitted per month with NS in 2011 and 2013. This was consistent with the NICE guidance,3 which suggested active specialist units admitted at least 20 patients with NS per month. Allowing for exclusion of high-risk patients (approximately 30%51) and trial ineligibility it was assumed that 10 patients per month at each site would be eligible and recruitment of two of these appeared an achievable target.

On review of screening and recruitment logs the number of NS admissions was lower than expected. This was consistent with updated audit data from sites and may reflect changes in standard care relating to growing use of targeted therapies and immunotherapies in place of cytotoxic chemotherapy and continued improvements in the care of patients with suspected sepsis, including NS. However, even within this it was clear that the majority of patients screened for study entry were considered ineligible, as highlighted by activity at Sites 1 and 2 in the previous table where 1 of 32 patients and 2 of 26 patients were screened and recruited, respectively. The reasons for exclusion are summarised in Table 5; as expected around 30% of patients were high risk based on MASCC score, but for the remainder it seemed that the eligibility criteria did not reflect the patient population receiving treatment for NS in routine care pathways.

From discussion with the clinical and research teams at study sites, specific issues with the trial eligibility criteria mainly related to the stringent requirements for fever, neutropenia, use of other antimicrobials and organ function, summarised in Table 6.

Extended pilot phase I (December 2016 to February 2017)

A 3-month, four-site extension to the pilot study was delivered from 1 December 2016 to 28 February 2017, using the adjusted eligibility criteria, summarised in Table 8.

The expected recruitment rate was 1.7 patients per site per month, with again a 50% reduction permitted for the first 3 months of site opening. With recruitment on hold from 21 July 2016 until 1 December 2016 there was significant loss of momentum at sites, which was difficult to recover and combined with local site capacity issues resulted in a very short window of potential recruitment at two sites. Consequently, both of these sites failed to recruit during this extension phase, despite having been active in screening and recruitment during the first phase of the pilot study. The other two sites achieved an average monthly recruitment between the two sites of 1.8 patients per month, similar to the predicted recruitment rate, with review of screening data suggesting a positive impact of the adjusted eligibility criteria on recruitment. It was, however, again noted that the number of NS admissions remained lower than historical data, with an average of six admissions per month.

Extended pilot phase II (April 2017 to November 2017)

Due to the longer than anticipated pilot phase, under-recruitment was now inevitable if the project timeline was not modified, even if the number of recruiting sites for the main trial was expanded significantly from 12 to 20. To maximise the potential for successful completion of the trial, with the research question as fully addressed as possible, the preferred option was to increase the number of recruiting sites to 20 and extend the project timeline by approximately 1 year. To mitigate risk, it was agreed first to further extend the pilot phase until November 2017, aiming to open an additional seven sites during this period. This would allow an assessment of the potential recruitment at a broader selection of sites. A revised recruitment target was set of one patient per month per site to account for the lower frequency of neutropenic admissions.

A further 8 month period of recruitment to the pilot phase of the trial occurred between April and November 2017. Six additional sites were opened during this phase and a summary of recruitment by site is shown in Table 9. Even with the revised recruitment target of one patient per site per month, only 50% of the expected recruitment was met, with 24 patients recruited compared with the predicted 49 (average monthly recruitment rate across all sites was 0.4 patients).

Summary of pilot phase recruitment

By 30 November 2017, there had been 17 months of active recruitment to the study, with as planned 10 sites open to recruitment and an eleventh due to open shortly. Forty-two patients had been recruited compared with a revised target of 68 patients as summarised in Figure 1.

FIGURE 1.

Summary of recruitment activity at the end of the pilot phase (February 2016 to November 2017): total recruitment of 42 patients.

Adherence to the protocol-specified treatment

Adherence to the protocol-specified intervention was assessed by whether patients switched to co-amoxiclav and ciprofloxacin, 12–24 hours after starting i.v. meropenem or piperacillin tazobactam i.v. therapy for at least 5 days total treatment. Adherence to the standard care arm required receipt of at least 48 hours of i.v. piperacillin/tazobactam or meropenem. The pre-specified threshold for adherence of 75% in both arms was achieved with adherence 86% and 83% in the intervention and control arms, respectively.

Separation between trial arms

The final criteria for trial progression was evidence of adequate separation in terms of the timing of antibiotic switch of at least 24 hours between the trial arms. The time on i.v. antibiotics was calculated for each patient and then the mean calculated for each trial arm and the difference between the two arms assessed. The mean duration of i.v. antibiotics was 19 hours in the intervention arm and 48 hours in the control arm; hence, mean separation in terms of timing of antibiotic switch was acceptable at 29 hours.

Summary of embedded pilot study results

The internal pilot phase of the EASI-SWITCH trial demonstrated that while recruitment was challenging there were no other concerns relating to treatment failure, protocol adherence or safety concerns. Based on recruitment between April and November 2017, it was likely that on average each site would only be able to recruit one patient every 2 months. The pilot phase had demonstrated that recruitment to the original proposed sample size of 628 patients for the main trial was not going to be achievable. Recruitment had originally been scheduled to complete at the end of December 2018. With 10 sites open and recruiting on average one patient every 2 months, based on current recruitment activity total recruitment was estimated to complete at approximately 100 patients. This would result in a significantly underpowered study, with results unlikely to have any significant impact and the evidence gap for an early oral switch identified by NICE remaining unanswered. It was clear important assumptions underpinning the trial design would have to be urgently reviewed to decide about whether to progress beyond the pilot phase to main trial delivery.

Determining continued importance of the research question

In view of the recruitment difficulties and the longer than anticipated pilot phase, an updated understanding of how low-risk NS patients were currently being managed across the UK was critical to assess the continued importance of the research question and assess whether there remained equipoise between the trial arms.

Between January and June 2018, NS management policies were reviewed and a survey of practice was undertaken nationally. Local NS policies from throughout the UK were sought via an e-mail request distributed to acute oncology nurse specialists within United Kingdom Oncology Nursing Society (UKONS) and EASI-SWITCH team members. Policies were reviewed for compliance with NICE recommendations, with particular focus on those advocated as key priorities for implementation in the guideline and their approach to low-risk management.

It is likely that a wide range of factors, including awareness of guidance, personal treatment preferences and clinical experience, will influence individual compliance with local policies when delivering NS care. A complementary electronic survey aimed to more fully reflect clinicians’ daily clinical practice, attitudes and preferences when managing NS. The survey link was disseminated via the UKONS, Royal College of Pathologists’ electronic newsletter and the clinicians involved in the EASI-SWITCH trial with onward dissemination to colleagues encouraged. In the survey, respondents were encouraged to reflect on their individual routine practice, rather than what their local institution’s NS policy or national guidelines might recommend.

A review of standard practice, completed for NICE in 2012, suggested approximately a third of 51 English and Welsh centres surveyed included the option of empirical oral antibiotics for lower-risk adult patients, with the intention of immediate discharge or earlier discharge compared with i.v. inpatient antibiotics. 3 It was unknown if this remained the status quo in the years following the publication of NICE guidance, with no national data available regarding centres’ standard approach to low-risk NS patients.

An updated understanding of management of low-risk NS patients was essential to assess whether there remained equipoise between the trial arms. If clinical practice had already largely shifted towards the intervention arm of EASI-SWITCH, there may be less impetus to continue to dedicate patient time and resources towards the trial. An updated audit of paediatric NS policies in 2015 had shown a lack of adherence, however, to NICE guidance, with continued variability in diagnostic criteria and management. 52 The impact of the NICE recommendations on adult NS management was unknown and it was unclear whether the publication reduced variation in practice for adult cancer patients.

A total of 53 adult NS policies were returned and reviewed: six cancer network policies and 47 acute trust policies, with representation from all four UK nations. As the request for policies was often forwarded to colleagues within a healthcare trust, it was not possible to determine an overall response rate. Most policies (94%) provided advice for both oncology and haematology patients. All policies, which were dated, had been updated post 2012, with no policy obviously issued before the NICE guidance was published and 57% specifically referenced that guideline.

A total of 235 responses were received from the electronic survey of individual practice. After excluding those containing only demographic data or minimal responses, there were 217 evaluable surveys. Again, there was representation from all four nations, with 69% of respondents based in England. The majority (72%) worked in oncology (predominantly as oncology consultants, acute oncology specialist nurses and advanced nurse practitioners), 23% worked in haematology and the remaining 5% were consultants mainly based in EDs or acute or general medicine. Forty-three per cent worked in university hospitals or regional cancer centres, 52% worked in acute hospitals with SACT day case units on site, and the remaining 5% worked in smaller acute hospitals which did not deliver SACT.

Of major interest was understanding current practice regarding routine risk stratification and any specific low-risk protocols or practices. The policies varied in their adoption of the NICE recommendations focused on assessing patients’ risk of septic complications and in decision-making about when, or if, to switch to oral antibiotics and about hospital discharge.

When practice was reviewed in 2012 by the NICE GDG, approximately a third of adult policies included the option of empirical oral antibiotics for lower-risk patients, with the intention of immediate discharge or earlier discharge compared with i.v. inpatient antibiotics (NICE30). In contrast, approximately half of current polices (53%) encourage identification of low-risk patients, with potential discharge on oral antibiotics within 48 hours after hospital admission. Almost half the policies (45%) suggest stratification based on MASCC score. Risk stratification is often only performed by senior oncology/haematology doctors or acute oncology specialists, if available.

Only 40% of clinicians reported routinely risk-stratifying patients within 24 hours of presentation. The MASCC score was most commonly used (by 70%), with the remaining 30% of clinicians using a local institutional risk-scoring system, a Modified Early Warning Score (EWS; Subbe et al. 53) or the Clinical Index of Stable Febrile Neutropenia (CISNE; Carmona-Bayonas et al. 54).

A wide range of approaches to early oral antibiotics and outpatient management are described in policies and by clinicians. Only 9% of policies clearly describe the option of starting empirical oral antibiotics for low-risk patients from the outset and most of these admit patients for 24–48 hours before discharge. Similarly, only a small number of survey respondents (5%) suggested they would start upfront oral antibiotics rather than i.v. antibiotics initially for low-risk patients. A further three haematology consultants suggested this would be considered for patients with difficult i.v. access or who were in the terminal phase of illness. Most policies do not define a period of observation following the start of oral antibiotics, although 13% of policies specify at least a 24-hour admission to hospital. The most common ‘low-risk’ oral antibiotic regimen is co-amoxiclav (625 mg three times daily) and ciprofloxacin, (either 500 mg or 750 mg twice daily; 16 of 20 policies specifying a regimen) for 5 to 10 days in duration.

A number of the policies referred to the extensive additional clinical criteria that the American Society of Clinical Oncology (ASCO) suggest should exclude patients from outpatient management, including ensuring patients do not live alone, they have access to a telephone and transport and can return to hospital promptly. 7,29

NICE commented in 2012 that the criteria for considering either stopping or switching from i.v. to oral antibiotics greatly varied. This work showed continuing variation in ongoing inpatient management of uncomplicated NS. Sixty-eight per cent of policies provided advice about when it might be suitable to switch from i.v. to oral antibiotics, in addition to guidance about early oral antibiotics. The majority (57%) recommend considering a switch from i.v. to oral antibiotics at 48 hours. The European Society for Medical Oncology (ESMO) guideline offers further advice about a range of factors to consider at 48 hours for switching to oral antibiotics and duration of treatment, which have been incorporated into many policies. 28 This includes whether the patient is still febrile and the neutrophil count is > 0.5 × 10⁹/l.

One hundred and sixty-three survey respondents provided some indication as to when they would routinely switch low-risk uncomplicated patients to oral antibiotics and discharge home. These questions were not answered by clinicians who treat with oral antibiotics from the outset. Thirty-seven per cent reported switching to oral antibiotics typically within 24 hours of admission, 44% within or at 48 hours and 20% after 48 hours of admission. Various standard times to discharge after switching patients to oral antibiotics were also reported, ranging from immediate discharge in 27% to after at least 24 hours of observation in 53%.

It was evident that clinicians generally do not recalculate MASCC scores after admission. They confirmed, however, that the most important criteria, apart from initial risk scores, for determining when to switch to oral antibiotics are duration of apyrexia (70%) and if symptoms are improving (62%). Over half the respondents consider the patient’s neutrophil count (56%) and availability of microbiology results (52%) as important. Fewer consider the duration of i.v. antibiotics the patient received (31%) and expected duration of neutropenia (12%).

Survey respondents highlighted common reasons preventing early discharge of low-risk patients before 48 hours of hospitalisation (see Table 10), most commonly a clinically well patient experiencing ongoing fever.

It would therefore seem that despite the introduction of NICE guidance risk stratification is not routine practice for many NHS centres or clinicians. A small number of specialist teams are prescribing upfront oral antibiotics for low-risk patients, but this practice appears limited. There has been some increase in the number of policies which include the option of discharging low-risk patients on oral antibiotics before 48 hours but approximately half of policies still advise initial i.v. antibiotics and subsequent oral switch after 48 hours or later. Even more clinicians, approximately two-thirds, in their routine practice choose to switch from i.v. to oral antibiotics at 48 hours or later. It is therefore evident that for at least half of centres and clinicians surveyed routine practice is in accordance with the standard care arm in the EASI-SWITCH trial. There has not yet been widespread adoption of risk stratification tools or routine early oral antibiotics in low-risk patients. A number of factors appear to have slowed progress, including the limited evidence base highlighted in the NICE guidance and in some cases dissatisfaction with current risk stratification tools, which has resulted in centre, clinician and patient hesitance about early oral switch approaches.

The majority of policies encourage initial investigations and empirical antibiotic regimens that accurately reflect NICE recommendations and therefore the initial management approach assumed in the EASI-SWITCH trial. Eighty-five per cent of polices promote initial beta-lactam monotherapy, compared to only 36% in NICE’s 2012 review. There has been a general reduction in the routine use of empirical aminoglycosides and glycopeptides for patients with central lines: 13% of current policies include gentamicin as part of an empirical dual antibiotic regimen, compared with 63% in 2012. Four of the current policies suggest initial doses of gentamicin but that these only be continued if there are signs of severe sepsis or for haematology patients. Local organism antibiotic resistance patterns are likely to be a contributing factor for those policies where gentamicin is routinely administered to all patients and is permitted under NICE guidance. Only 4% of current policies suggest empiric use of glycopeptides if a central line is present but a line infection not suspected, compared to 15% in 2012. Clinical practice therefore supported the standard care i.v. antibiotics being utilised in the EASI-SWITCH trial.

Finally, comparing diagnostic thresholds to the NICE guidance, only 19% of policies and 24% of survey respondents define neutropenia as a neutrophil count of ≤ 0.5 × 10⁹/l as suggested by NICE, with most employing a broader definition. While there has been a reduction in policies simply defining neutropenia as < 1.0 × 10⁹/l (51% compared with approximately 70% of adult policies in 2012) this is offset by those that include patients whose neutrophil count is < 1 × 10⁹/l and expected to fall. More policies now encourage starting empirical treatment in patients who have had a single temperature ≥ 38°C (78% compared with 50% in 2012), in keeping with NICE criteria. Similarly, a patient presenting with a single temperature of 38°C or above would meet the temperature criteria used by 96% of clinicians and just over a third (36%) of these clinicians would start empirical treatment if lower temperature thresholds were recorded (e.g. single/sustained temperature of ≥ 37.5°C/≥ 37.7°C). The importance of considering NS in patients who are generally unwell, irrespective of their temperature, was stressed in the majority of policies, including prompts for signs and symptoms such as rigors, altered mental state and haemodynamic instability. Overall the review of current policy and practice supported the proposed adjustments to the eligibility criteria to the trial.

Review of assumptions underpinning study design

The review of current NS practice confirmed the continued importance of the research question nationally and the potential impact the trial could have on routine NS management in the UK. A careful review of the assumptions underpinning the original trial design and sample size calculation was therefore undertaken to determine whether the delivery of a smaller study could still have a meaningful clinical impact.

On initial design, the trial was powered at 90% and with a one-sided 97.5% CI, to detect non-inferiority of the early antibiotic intervention within a margin of 10%, assuming a 15% treatment failure rate in the standard care arm. This would require 269 patients per group. Allowing for a 5% dropout rate and up to 10% crossover from control to intervention, 314 participants would be required for each group, resulting in an overall target sample size of 628 patients.

The assumed 15% treatment failure rate for patients receiving standard care was derived from the three studies thought to best reflect the proposed control arm in relation to the populations included and duration of i.v. antibiotic treatment administered. 36,55,56

The pilot results raised no concerns regarding treatment adherence, separation between the treatment arms or the observed treatment failure rates in each arm. It was therefore presumed that the assumptions relating to those sample size parameters remain valid but, in retrospect, the choice of non-inferiority margin and statistical analysis may have been too stringent for a pragmatic trial where treatments were being used within their licensed indications or established clinical practice and where the risk of treatment failure was unlikely to result in serious risk to patients (in particular critical care admission or death).

Progression to main trial

In addition to this stakeholder support, the study sponsor and HTA programme director were also supportive of continuation with the proposed revisions in study design. Based on the recruitment during the pilot phase, the revised target recruitment rate was 0.5 patients per site per month. Accrual of the revised sample size of 230 based on this target would be achieved by increasing the number of trial sites (to a total of 20) and the duration of recruitment to allow for the halt during the pilot phase and its subsequent prolongation.

Screening and recruitment post pilot phase

The internal pilot phase of EASI-SWITCH highlighted lower than anticipated recruitment rates, even with the revision of eligibility criteria and associated positive impact on recruitment. Further revision to study design was deemed acceptable to the trial stakeholders with the aim of delivering a clinically meaningful study. On review of updated audit data from two study sites, it was clear there was a sustained downward trend in NS admissions, consequently a revised recruitment target of 0.5 patients per site per month was set with an accompanying increase in number of recruiting sites to 20.

Post pilot phase, the EASI-SWITCH trial continued recruitment for another 24 months. In total, 19 out of a planned 20 sites opened to recruitment, which were a combination of cancer centres and units. This included 14 sites in England, 3 in Northern Ireland, 1 in Scotland and 1 in Wales (see Appendix 3, Table 37). In April and May 2019, 4 sites requested closure due to inactivity, leaving 15 active sites at the time of trial closure.

Identification of interested and appropriate sites was challenging. A range of methods were employed to promote the trial including the NIHR clinical research networks, relevant professional groups and national meetings. NIHR regulations around TSC and DMEC independence precluded participation of some interested sites because of associations with TSC and DMEC members. Table 11 summarises reasons for site non-participation.

In total, 827 patients were screened for trial entry. The main reasons for exclusion (in order of frequency) were patients not fulfilling the diagnostic criteria of NS, a diagnosis of acute leukaemia, deemed at high risk of complications using the MASCC score (score < 21), history of penicillin (or other IMP) allergy, the physician in charge was not willing to support entry into either trial arm or not treat with CSFs, signs of severe localising infection present or the patient had already received > 24 hours of i.v. antibiotics. In total, 129 participants were recruited out of the target sample size of 230. In November 2019, the TSC recommended trial closure due to under-recruitment. A summary of recruitment activity across the trial phases is provided in Table 12 and Figure 2.

FIGURE 2.

Cumulative actual recruitment vs. predicted recruitment for the EASI-SWITCH trial (April 2017 to November 2019 inclusive). Total recruitment of 129 patients (November 2019). With the adjusted recruitment rate of 0.5 patients per site per month from January 2018 predicted recruitment was 216 patients by November 2019, based on the number of open sites.

Identification of barriers to recruitment

In view of the lower than anticipated target population, there was a need to understand the barriers to recruitment experienced by sites as the trial progressed and identify potential solutions in order to maximise the likelihood of successful trial delivery. This was particularly important in recognition of the narrow window of opportunity for patients to be recruited, whereby screening and recruitment both had to occur within 24 hours of commencing i.v. antibiotics for NS.

To understand factors impacting recruitment at sites, two approaches were taken as the trial progressed post pilot phase. These were (1) semistructured interviews with key site personnel and (2) an investigator survey of barriers to recruitment.

Additional strategies to enhance recruitment

As the trial progressed and recruitment failed to meet expectations, a pragmatic shift in recruitment strategies occurred, with additional approaches added to try and address the modifiable barriers identified. As previously discussed, revision of eligibility criteria was made during and after the pilot phase to ensure the study was pragmatic and reflected the patient population routinely treated for NS. Further minor adjustments were made based on investigator interview feedback in May 2018, including: (1) recruitment of patients given additional systemic antibiotics at the time of presentation (such as a single dose of gentamicin) provided it was clinically appropriate to discontinue these at the time of enrolment, to include patients presenting to non-specialist units; (2) addition of suspension oral antibiotics to protocolised antibiotic therapy for patients unable to swallow tablets; (3) clarification of use of short-acting CSF to include patients who had a course of short-acting prophylactic CSFs prescribed as part of their SACT regimen but continuing to exclude those commenced on CSF as treatment for the presenting episode of NS.

In response to feedback about clinical workload, AE reporting was simplified, reducing the reporting period from 28 days to 14 days, with reporting of AEs beyond the 14-day primary outcome follow-up felt unlikely to add anything additional, in terms of understanding AEs related to the early switch antibiotic intervention. Shortening the reporting period was felt to benefit research teams, as it was not uncommon for patients to be re-admitted to hospital again within the current 28-day reporting period, once they had received a further cycle of chemotherapy. These hospital admissions were unrelated to the patient’s EASI-SWITCH episode and treatment but required research teams to report them as a serious AE within 24 hours of becoming aware. By shortening the reporting period to just the primary follow-up period of interest, it was hoped it would allow teams to prioritise time for recruiting new patients by reducing the intensity of patient follow-up required.

In the electronic survey distributed in March 2019, investigators were also asked to share any strategies they had found useful at their site in facilitating recruitment, so they could be shared with other sites. All comments related to optimising awareness of the trial, with clinical staff first assessing and managing new admissions and trying to create a suitable pool of staff who could provide consistent screening of patients and perform research-related tasks.

Main trial progress

While revisions to the eligibility criteria and trial design appeared to have a positive impact on recruitment initially, it is apparent from the recruitment graph in Figure 2 that, by the end of 2018, recruitment was consistently falling behind target and appeared to be plateauing thereafter, suggesting achieving even a smaller study in a timely manner would be challenging. Consequently, the DMEC recommended study closure on grounds of under-recruitment in November 2019. Despite clinician and researcher support for the study, persistent challenges to recruitment in relation to the numbers of potential patients and the challenges of identifying and enrolling patients within a narrow time window following admission remained. It is also possible that as time progressed, the variation in equipoise between trial arms demonstrated in the investigators’ survey (March 2019) in comparison to the site team interviews (April 2018) became more prominent and was potentially accompanied by changes in clinicians’ treatment approaches.

Recruitment

One hundred and twenty-nine patients were enrolled in EASI-SWITCH when trial closure was recommended by the DMEC in November 2019. In total, 827 patients were screened for study entry; the most common reasons for screen failure are described in the previous chapter and listed in Appendix 5. Recruitment numbers by site are summarised in Table 16 with a list of sites provided in Appendix 3, Table 37.

Consolidated Standards of Reporting Trials flow diagram

In total, 129 patients were randomised from 19 sites. Sixty-five patients were randomised to the early switch intervention and 64 to the standard care control arm (see Figure 3, CONSORT diagram). Of these, 125 patients were included in the primary ITT analysis (comprising 61 patients allocated to intervention and the 64 patients allocated to control). Four patients were excluded (all allocated to intervention) for the following reasons: one patient was randomised in error, one withdrew from continued participation and two patients were lost to follow-up. Twelve patients included in the primary ITT analysis were excluded from the PP analysis; consequently, the primary PP population comprised 53 patients in the intervention group and 60 in the standard care group. The reasons for exclusion of the eight patients in the intervention group were: premature discontinuation of antibiotic treatment (n = 4), interruption to antibiotic treatment involving at least two consecutive missed doses (n = 2) and insufficient initial i.v. antibiotic treatment (n = 1). All four excluded patients in the standard care group had received < 48 hours of i.v. antibiotic treatment.

FIGURE 3.

CONSORT diagram.

Baseline characteristics

Table 17 describes the patient, tumour and treatment characteristics in the intervention and control groups. As anticipated, only a minority of participants had a haematological malignancy (6.2%) in comparison with solid tumours (93.8%), of which the most common cancer type was breast cancer (54.6%). The majority of patients were receiving treatment in the neoadjuvant or adjuvant setting (60.4%) rather than palliative setting (27.9%). In general, baseline characteristics were well balanced across the intervention and control groups in relation to cancer type, anticancer treatment and symptoms and signs of infection (including fever, absolute neutrophil count and MASCC score). Use of prophylactic Granulocyte colony stimulating factor (GCSF) support was noted to be higher in the standard care arm (34.4%) in comparison to the intervention arm (15.6%).

Treatment after trial entry

All patients (n = 64) randomised to the standard care arm commenced allocated study treatment; however, four patients did not complete allocated treatment, receiving < 48 hours of i.v. antibiotic therapy. Two patients randomised to the early switch intervention (n = 65) did not proceed to allocated study treatment due to ineligibility (n = 1) and withdrawal of consent (n = 1). Therefore, 63 patients in the intervention arm commenced allocated treatment; however, protocolised treatment was not completed in eight patients for a number of reasons. These included premature switch to oral antibiotics (after < 12 hours of i.v. antibiotic therapy; n = 1); late switch to oral antibiotics (after 24 hours of i.v. therapy; n = 2); completion of < 5 days antibiotic treatment in total (n = 1) and a combination of late oral switch (after 12 hours) and short duration (< 5 days) of treatment in total (n = 2).

The mean duration of i.v. antibiotic therapy (and SD) was 54.8 hours (24.2) and 22.6 hours (26.9) in the standard care and early switch intervention groups, respectively.

No patients in the standard care arm were lost to follow-up for the day 14 primary outcome measure. Three patients is the intervention arm were lost to follow-up at day 14; however, it was still possible to determine the primary outcome for one of these patients, who had a treatment failure before the day 14 time point.

In total, of 65 patients allocated to intervention, 61 were included in the primary outcome ITT analysis and 53 in the PP analysis. Of the 64 patients allocated to standard care, all were included in the ITT analysis and 60 included in the PP analysis.

Protocol deviations

In total, 93 protocol deviations were reported. The most common reason for these was assessments taking place outside of schedule (45.2%). These typically related to the timing of study questionnaires at baseline, day 14 and day 28 of follow-up. This might be anticipated, particularly for the follow-up questionnaires, given this patient population are likely to have multiple hospital attendances related to continued SACT or management of complications arising from treatment or their underlying disease. Protocol deviations by number and type across treatment groups are listed in Table 18.

Treatment outcomes

Primary outcome

Both ITT and PP analyses were conducted for the primary outcome measure of treatment failure. It was pre-specified that equal weighting would be given to both analyses such that a definitive conclusion of non-inferiority required both analyses to concur. A one-sided 95% confidence limit placed about the difference in treatment failure rate was compared to the non-inferiority margin of 15%. Conclusion of non-inferiority required that the upper bound of the one-sided 95% CI placed about the difference in treatment failure rates (intervention minus control) should not exceed 15% (see Figure 4). The 95% one-sided confidence limit was derived from the upper bound of a two-sided 90% CI. The Pearson chi-square test was used to test significance of observed differences between trial arms. Table 19 presents the number (%) and the differences in proportions (95% CI) in relation to the primary outcome in the ITT and PP analyses.

FIGURE 4.

Forest plot of differences (90% CI) by analyses performed.

TABLE 19 - Analyses for the primary outcome in the ITT and PP populations

	Standard care (n = 64)	Intervention (n = 61)	Risk difference (90% CI)	p-value
Treatment failure ITT (N = 125)
Yes	9 (14.1%)	15 (24.6%)	0.11 (−0.01 to 0.22)	0.14
No	55 (85.9%)	46 (75.4%)
Treatment failure PP (N = 113)
Yes	8 (13.3%)	9 (17.0%)	0.04 (−0.07 to 0.148)	0.59
No	52 (86.7%)	44 (83.0%)

In the ITT analysis, 9 of the 64 patients (14.1%) in the standard care arm met the primary end point of treatment failure, compared with 15 of 61 (24.6%) in the early oral switch arm. Based on the pre-specified 15% non-inferiority margin, the risk difference of 10.5% (one-sided 95% CI −1% to 22%; p = 0.14) was such that the intervention was not found to be non-inferior to standard care in the ITT population.

The result of the PP analysis was not consistent with the ITT analysis. In the PP population, 8 of the 60 patients (13.3%) in the standard care arm met the primary end point of treatment failure, compared with 9 of 53 (17%) in the early oral switch arm. The risk difference of 3.7% (one-sided 95% CI −7% to 14.8%; p = 0.59) was such that the intervention was found to be non-inferior to standard care in the PP population.

Non-inferiority for the early switch intervention could not be concluded in the absence of concurrence of both the ITT and PP analyses and, even if concurrence had been demonstrated, findings would remain limited due to relative under-powering of the analyses. The main constituents of the composite primary outcome measure that accounted for patients reaching the treatment failure end point by day 14 were persistence/recurrence of fever and/or physician-directed escalation from the protocolised antibiotic regimen. None of the treatment failure events recorded in either arm were attributable to the need for critical care support or death before day 14. Tables 20 and 21 present the constituents of the primary outcome measure in the ITT and PP populations, respectively. The number of patients for whom each component of the composite outcome measure was available is expressed, by trial arm, for that measure and the p-value given is from significance testing of observed differences between trial arms using the Pearson chi-square test.

TABLE 20 - Constituents of the composite primary outcome measure leading to patients reaching the treatment failure end point in the ITT population

	Standard care (n = 64)	Intervention (n = 61)	Difference (95% CI)	p-value
Treatment failure ITT
Yes	9 (14.1%)	15 (24.6%)	0.105 (−0.010 to 0.221)	0.14
No	55 (85.9%)	46 (75.4%)
Persistence/recurrence of fever (T ≥ 38°C) after 72 hours of i.v. antibiotic initiation	n = 62	n = 60
Yes	4 (6.5%)	10 (16.7%)	0.102 (−0.010 to 0.215)	0.08
No	58 (93.6%)	50 (83.3%)
Physician-directed escalation from protocol-specified antibiotic treatment
Yes	6 (9.4%)	10 (16.4%)	0.070 (−0.047 to 0.187)	0.24
No	58 (90.6%)	51 (83.6%)
Day 14: CCU admission	n = 62
Yes	0 (0.0%)	0 (0.0%)
No	62 (100%)	61 (100%)
Day 14: Re-admission to hospital	n = 62
Yes	2 (3.2%)	2 (3.3%)	0.001 (−0.062 to 0.063)	0.99
No	60 (96.8%)	59 (96.7%)
Day 14: Patient survival status
Alive	64 (100%)	61 (100%)
Deceased	0 (0.0%)	0 (0.0%)

TABLE 21 - Constituents of the composite primary outcome measure leading to patients reaching the treatment failure end point in the PP population

	Standard care (n = 60)	Intervention (n = 53	Difference (95% CI)	p-value
Treatment failure
Yes	8 (13.3%)	9 (17.0%)	0.036 (−0.075 to 0.148)	0.59
No	52 (86.7%)	44 (83.0%)
Persistence/recurrence of fever (T ≥ 38°C) after 72 hours of i.v. antibiotic initiation
Yes	4 (6.8%)	6 (11.5%)	0.048 (−0.060 to 0.156)	0.38
No	55 (93.2%)	46 (88.5%)
Physician-directed escalation from protocol-specified antibiotic treatment
Yes	5 (8.4%)	5 (9.4%)	0.011 (−0.094 to 0.116)	0.84
No	55 (91.7%)	48 (90.6%)
Day 14: CCU admission	n = 58
Yes	0 (0.0%)	0 (0.0%)
No	58 (100%)	53 (100%)
Day 14: Re-admission to hospital	n = 58
Yes	2 (3.5%)	2 (3.8%)	0.003 (−0.066 to 0.073)	0.93
No	56 (96.6%)	51 (96.2%)
Day 14: Patient survival status
Alive	60 (100%)	53 (100%)
Deceased	0 (0.0%)	0 (0.0%)

A post hoc exploratory analysis using generalised estimating equations (GEE) to account for possible clustering of observations within participating centres produced a risk difference (RD; 95% CI) of 0.106 (−0.032 to 0.244); p = 0.13, indicating no significant difference in risk of treatment failure when examining study site as a random effect.

Secondary outcomes

Among patients for whom secondary outcome data were available, there was a reduction in median duration of inpatient admission. Time to fever resolution, re-admission to hospital to day 28, survival to day 28, or changes to the originally intended SACT regimen appeared similar between the two arms (see Table 22).

Subgroup analyses

Subgroup analyses were pre-planned in the ITT population in the following subgroups: (1) tumour type (solid tumour vs. haematological malignancy); (2) neutrophil count at randomisation (≤ 0.5 × 10⁹/l vs. > 0.5 × 10⁹/l) and (3) maximum temperature on the day of presentation (< 38°C vs. > 38°C). A post hoc subgroup analysis based on blood culture results (positive or negative) was also performed. The first pre-specified analysis was not undertaken due to the small number of patients with haematological malignancy enrolled in the study. Risk differences and 99% CI from the treatment × subgroup interaction models are presented in Table 23 for the remaining subgroups for the primary outcome of TFR. The p-values presented are from a global test for interaction and indicate no significant interactions.

Safety

Background

A within-trial economic evaluation was performed to assess the cost-effectiveness of early switch to oral antibiotics compared with usual care in the treatment of NS in patients with cancer. This included a cost-effectiveness analysis (CEA) consistent with the primary outcome measure to estimate the cost per treatment failure avoided at day 14 and a CUA to estimate the cost per QALY at day 14. The primary measure used in these analyses was the QALY, estimated from the EQ-5D-5L questionnaire administered at baseline and at day 14 of follow-up (either in person or using the validated telephone version). In recognition that the EQ-5D-5L only measures the potential effect on health of an early switch from i.v. to oral antibiotics and does not reflect patients’ preferences for non-health effects of the intervention, such as early discharge from hospital, a Patient Follow-up Questionnaire was used to collect additional information on this.

Methods

As the study treatments are for infection, not cancer, a short-term time horizon reduced the chance of costs and effects being contaminated by the impact of underlying disease or cancer therapy. A 14-day time horizon was adopted and the analysis was performed from a hospital perspective. Discounting of costs and outcomes was not necessary due to the time horizon being < 1 year. Patients’ use of hospital resources was collected over the study period on the case-report form using data from the day 14 interview and review of medical records, including treatments and medication received during the primary admission and any associated re-admissions. Costs were calculated by attaching appropriate unit costs from publicly available sources59–61 (e.g. Department of Health National Schedule of Reference Costs) and are listed in Appendix 6, Table 38. The final year of data collection was taken as the cost year (2018/2019).

For the CUA patients’ HRQoLs were measured at baseline and at day 14 of follow-up using the EQ-5D-5L (either in person or using the validated telephone version) and the 3L mapping function62 was used to convert responses into a single utility value, as currently recommended by NICE. 63 The area-under-the-curve method was used to estimate patient-specific QALYs accrued over the study duration. For the cost-effectiveness analysis, treatment failures were categorised as described in the previous chapter.

Since the EQ-5D-5L only measures the potential effect on health of an early switch from i.v. to oral antibiotics and does not reflect patients’ preferences for non-health effects of the intervention, such as early discharge from hospital, a Patient Follow-up Questionnaire was used to collect additional information on this.

Following database lock, individual patient data were evaluated to measure costs and QALYs related to early oral switch and standard care. Descriptive statistics were used to summarise the hospital resource use, associated costs and outcomes as means with 95% CIs for each group. Significance was judged where the CIs of differential means excluded zero or p < 0.05. The study was powered for the primary outcome of treatment failure but not for costs, QALYs or cost effectiveness.

The mean differences in costs, treatment failures and QALYs between groups were estimated and incremental cost-effectiveness ratios (ICERs) were calculated to estimate the cost per QALY gained (CUA) and the cost per treatment failure avoided (CEA) and the non-parametric bootstrapping was used to resample with replacement the cost and outcome data from the original sample to generate 1000 replicate ICERs. These were then plotted on the cost-effectiveness plane to display their joint distribution. Cost-effectiveness acceptability curves (CEAC) were constructed from the data by calculating the proportion of the ICER replicates which would be considered cost-effective at various thresholds of willingness-to-pay (WTP) for an additional QALY and to avoid a treatment failure. In general NICE consider interventions with an ICER of < £20,000/QALY to be cost-effective;64 however, no such threshold exists for treatment failures avoided.

The net monetary benefit (NMB) was also used to aid interpretation. The NMB is a summary statistic representing the value of an intervention in monetary terms when a WTP threshold for a unit of benefit is known. A positive NMB indicates the intervention is cost-effective. A range of threshold values were used for each analysis. All analysis was carried out using Stata 15/IC (StatCorp) for Windows^®.

Results

One hundred and twenty-six patients were eligible for the economic analysis; 62 in the intervention group and 64 in the standard care group. The results are presented in two parts: the CUA based on using the QALY as the outcome using complete data; and the cost effectiveness using treatment failures avoided, also only using complete data. Results using all available data are included in Appendix 8, Tables 39–41.

Cost–utility analysis

To maintain the correlation structure of the data the primary CUA analysis only included those participants with complete cost and QALY data, therefore if EQ-5D-5L data were not collected at baseline and day 14, or cost data were missing, they were excluded. These data were available for 110 patients (52 in the intervention group and 58 in the standard care group), with a higher proportion in the intervention arm having missing data (16% vs. 9%). Use of hospital services within 14 days of randomisation, including the primary admission and any subsequent re-admission(s), is presented in Table 25. The mean length of stay in participants’ primary admission was 0.52 days longer for standard care, as might have been anticipated from an early oral switch intervention. The mean length of stay during re-admission was less than half for standard care patients compared to those in the intervention arm. None of the differences between groups were statistically significant.

TABLE 25 - Hospital service use by group with complete cost and QALY data. Values are number (percentages) of patients using the service and mean (95% CI) use

Includes patients starting antibiotics in emergency department.

Service	Intervention (n = 52)		Standard care (n = 58)		Mean difference (95% CI)
	n (%)	Mean (95% CI)	n (%)	Mean (95% CI)
Primary admission
Ward days	52 (100.00)	2.62 (1.99 to 3.24)	58 (100.00)	3.14 (2.71 to 3.57)	−0.52 (−1.26 to 0.22)
Emergency-department attendancea	2 (3.85)	0.04 (−0.02 to 0.09)	2 (3.45)	0.03 (−0.01 to 0.08)	0.00 (−0.07 to 0.08)
Re-admission
Ward days	4 (7.69)	0.46 (−0.13 to 1.06)	2 (3.45)	0.19 (−0.08 to 0.46)	0.27 (−0.35 to 0.89)
Emergency-department visits	4 (7.69)	0.08 (0.00 to 0.15)	2 (3.45)	0.03 (−0.01 to 0.08)	0.04 (−0.04 to 0.13)
Outpatient visits	2 (3.85)	0.04 (−0.02 to 0.09)	2 (3.45)	0.03 (−0.01 to 0.08)	0.00 (−0.07 to 0.08)

The corresponding costs of these services, along with drug costs, are presented in Table 26. The additional time spent as an inpatient led to a mean difference in cost of £250. The mean cost of study drugs was over £30 higher in the standard care arm, due to the lower unit cost of ciprofloxacin and co-amoxiclav, and the difference was statistically significant. The mean total cost was £1726 in the standard care arm compared to £1565 for intervention, a mean difference of £161, which was not statistically significant.

TABLE 26 - Cost (£UK) of hospital service use by group inpatients with complete cost and QALY data. Values are mean (95% CI) costs

	Intervention (n = 52)	Standard care (n = 58)	Mean difference (95% CI)
	Mean cost £ (95% CI)	Mean cost £ (95% CI)
Primary admission
Ward days	1250.81 (951.16 to 1550.46)	1500.72 (1294.76 to 1706.67)	−249.91 (−603.17 to 103.35)
Emergency-department attendance	7.54 (−3.06 to 18.13)	6.76 (−2.73 to 16.24)	0.78 (−13.24 to 14.80)
Re-admission
Ward days	220.73 (−63.81 to 505.27)	90.70 (−37.14 to 218.55)	130.03 (−167.57 to 427.63)
Emergency-department visits	15.08 (0.39 to 29.76)	6.76 (−2.73 to 16.24)	8.32 (−8.61 to 25.25)
Outpatient visits	5.19 (−2.11 to 12.49)	4.66 (−1.88 to 11.19)	0.54 (−9.12 to 10.19)
Medication
Study drug	52.18 (44.00 to 60.35)	83.19 (68.68 to 97.70)	−31.01 (−48.00 to −14.02)
Concomitant medication	13.34 (2.33 to 24.35)	33.40 (2.02 to 64.77)	−20.06 (−54.44 to 14.33)
Total	1564.86 (1171.52 to 1958.21)	1726.18 (1453.16 to 1999.19)	−161.31 (−626.74 to 304.11)

Mean HRQoL scores are reported in Table 27. There were no significant differences in baseline or follow-up scores in EQ-5D-5L or visual analogue scale. Both measures showed increases in both groups by similar amounts, with the base figure being higher in the intervention arm for each method. The incremental QALY gain was small and not statistically significant: 0.002 QALYs is equivalent to 0.73 days in full health per year, but given that the follow-up period was 14 days, any difference would be small.

TABLE 27 - Health-related quality-of-life scores by group (patients with complete cost and QALY data). Values are mean (95% CI) scores

Variable	Intervention (n = 52)	Standard care (n = 58)	Difference (95% CI)
	Mean (95% CI)	Mean (95% CI)
EQ-5D-5L utilities
Baseline	0.78 (0.71 to 0.84)	0.74 (0.68 to 0.80)	0.03 (−0.05 to 0.12)
14 days	0.82 (0.77 to 0.87)	0.77 (0.72 to 0.82)	0.05 (−0.03 to 0.12)
EQ-5D VAS
Baseline	65.94 (59.99 to 71.90)	59.31 (53.97 to 64.65)	6.63 (−1.25 to 14.52)
14 days	76.12 (71.13 to 81.10)	69.98 (65.37 to 74.60)	6.13 (−0.58 to 12.84)
QALYs	1 (0.029 to 0.032)	0.029 (0.027 to 0.031)	0.002 (−0.001 to 0.004)

Although the differences are small, in cases such as this where the intervention is cost-saving and leads to more positive outcomes the intervention is said to be the dominant strategy (see Table 28). The ICER is not calculated as it would be a negative and would not convey any meaning. 65 Figure 5 demonstrates the result of the bootstrapped ICERs and the majority of points fall in the south-east quadrant, indicating that in the majority of cases the intervention is associated with cost savings and QALY gains, albeit small and not statistically significant. The CEAC in Figure 6 shows at a WTP threshold of £20,000/QALY there is a 78% probability the intervention is cost-effective compared to standard care. The NMB was positive for all WTP thresholds (see Table 29), indicating the intervention was cost-effective compared to standard care, while the CEAC constructed (see Figure 7) from this method showed a marginal difference in cost-effectiveness probability (79%).

TABLE 28 - Results of the cost–utility analyses at 14 days (1000 bootstrap samples)

nC, number analysed in standard care arm; nI, number analysed in intervention arm.

Values not presented for negative ICERs.

	Mean incremental total health service costs (£) (95% CI)	Mean incremental QALY gain (95% CI)	Incremental cost-effectiveness ratioa
Primary analysis (nI = 52, nSC = 58)	−161.31 (−630.59 to 307.96)	0.002 (−0.001 to 0.004)	Dominant strategy
Sensitivity analysis – controlling baseline characteristics (nI = 52, nSC = 58)	−125.45 (−612.33 to 361.44)	0.001 (−0.000 to 0.002)	Dominant strategy
Sensitivity analysis – multiple imputation of QALY values (nI = 62, nSC = 64)	18.42 (−520.09 to 556.93)	0.001 (−0.001 to 0.004)	£11,437.45
PP analysis (nI = 45, nC = 54)	−319.70 (−795.55 to 156.16)	0.002 (−0.001 to 0.004)	Dominant strategy

FIGURE 5.

Cost-effectiveness plane for the primary cost–utility analysis showing 1000 bootstrapped replications of mean incremental costs and QALY gain and the willingness-to-pay threshold of £20,000/QALY.

FIGURE 6.

Cost-effectiveness acceptability curves showing the probability of intervention being cost-effective compared to standard care for the primary and sensitivity analyses using QALYs as the outcome.

TABLE 29 - Incremental net benefit at various willingness-to-pay thresholds per QALY gained

Willingness to pay per additional QALY	Incremental net benefit (mean, 95% CI)
£0	161.31 (−304.11 to 626.74)
£1000	162.83 (−303.16 to 628.81)
£5000	168.87 (−299.51 to 637.25)
£10,000	176.43 (−295.24 to 648.11)
£20,000	191.55 (−287.71 to 670.82)
£50,000	236.91 (−272.52 to 746.34)

FIGURE 7.

Cost-effectiveness acceptability curve showing the probability of intervention being cost-effective compared to standard care using the net monetary benefit method.

Cost-effectiveness analysis

Complete data were available for 124 patients (98%) for this analysis (100% standard care, 97% intervention), resulting in a slightly different population to the CUA. Table 30 shows the use of hospital resources. For those re-admitted to ward, the mean length of stay was more than double for intervention patients, 0.4 compared to 0.17 days, in part due to two of these patients both remaining hospitalised for 11 days. Notably in the intervention arm, one patient was transferred to critical care 5 days after starting antibiotics and was still there at the 14 day follow-up. This patient was excluded from the CUA as they did not complete the EQ-5D-5L at day 14.

TABLE 30 - Hospital resource use by group in patients (complete cost and treatment failure data). Values are number (percentages) of patients using the service and mean (95% CI) use

Service	Intervention (n = 60)		Standard care (n = 64)		Mean difference (95% CI)
	n (%)	Mean (95% CI)	n (%)	Mean (95% CI)
Primary admission
Ward days	60 (100.00)	2.78 (2.18 to 3.38)	64 (100.00)	3.17 (2.73 to 3.61)	−0.39 (−1.12 to 0.34)
Critical care days	1 (1.67)	0.15 (−0.15 to 0.45)	0 (0)	–	0.15 (−0.14 to 0.44)
Emergency department	2 (3.33)	0.03 (−0.01 to 0.08)	2 (3.13)	0.03 (−0.01 to 0.08)	0.00 (−0.06 to 0.07)
Re-admission
Ward days	4 (6.67)	0.40 (−0.11 to 0.91)	2 (3.13)	0.17 (−0.07 to 0.41)	0.23 (−0.32 to 0.78)
Emergency-department visits	4 (6.67)	0.07 (0.00 to 0.13)	2 (3.13)	0.03 (−0.01 to 0.08)	0.04 (−0.04 to 0.11)
Outpatient visits	2 (3.13)	0.03 (−0.01 to 0.08)	2 (3.13)	0.03 (−0.01 to 0.08)	0.00 (−0.06 to 0.07)

Costs of hospital resources and drugs are illustrated in Table 31. As was the case in the CUA, the difference in cost of study drug is statistically significant. Total costs were on average £22 higher in the intervention arm using these data – a difference that was not statistically significant.

TABLE 31 - Cost of hospital service use by group in patients (complete cost and treatment failure data). Values are mean (95% CI) costs

	Intervention (n = 60)	Standard care (n = 64)	Mean difference (95% CI)
	Mean cost £ (95% CI)	Mean cost £ (95% CI)
Primary admission
Ward days	1331.13 (1044.22 to 1618.03)	1516.95 (1307.80 to 1726.10)	−185.82 (−534.10 to 162.46)
Critical care	139.98 (−140.12 to 420.07)	0	139.98 (−128.25 to 408.21)
Emergency department	6.53 (−2.63 to 15.70)	6.13 (−2.46 to 14.71)	0.41 (−12.01 to 12.83)
Re-admission
Ward days	191.30 (−54.95 to 437.55)	82.20 (−33.52 to 197.91)	109.10 (−154.78 to 372.98)
Emergency-department visits	13.07 (0.33 to 25.80)	6.13 (−2.46 to 14.71)	6.94 (−8.09 to 21.97)
Outpatient visits	4.50 (−1.81 to 10.81)	4.22 (−1.70 to 10.13)	0.28 (−8.27 to 8.84)
Medication
Study drug	52.28 (44.83 to 59.73)	80.81 (67.45 to 94.16)	−28.52 (−43.94 to −13.11)
Concomitant medication	20.29 (3.54 to 37.05)	40.26 (7.84 to 72.68)	−19.97 (−56.83 to 16.89)
Total	1759.08 (1283.51 to 2234.65)	1736.68 (1466.92 to 2006.44)	22.40 (−510.16 to 554.96)

In contrast to the CUA, these data have shown the cost to be marginally greater in the intervention arm, and the outcome to be negative due to higher treatment failure rate (see Table 32), although no differences were statistically significant. When this occurs the intervention is dominated by standard care. Figure 8 shows a large percentage of the bootstrapped ICERs in the north-west quadrant with higher costs and negative outcomes. As before, negative ICERs have not been calculated.

TABLE 32 - Results of the cost-effectiveness analyses at 14 days (1000 bootstrap samples)

nC, number analysed in standard care arm; nI, number analysed in intervention arm.

Negative ICER values not presented.

Analysis	Mean incremental total health service costs (£) (95% CI)	Incremental treatment failures avoided (95% CI)	Incremental cost-effectiveness ratioa
Treatment failure avoided – primary analysis (nI = 60, nSC = 64)	22.40 (−508.92 to 553.71)	−0.09 (−0.23 to 0.05)	Dominated by standard care
Sensitivity analysis – controlling baseline characteristics (nI = 60, nSC = 64)	122.87 (−461.66 to 707.41)	−0.10 (−0.25 to 0.04)	Dominated by standard care
Sensitivity analysis – excluding patient who had prolonged ICU admission (nI = 59, nSC = 64)	−131.57 (−579.67 to 316.53)	−0.08 (−0.22 to 0.06)	£1650.22
PP analysis (nI = 52, nSC = 60)	−149.44 (−694.28 to 395.39)	−0.02 (−0.15 to 0.11)	£7285.40

FIGURE 8.

Cost-effectiveness plane for the primary cost-effectiveness analysis showing 1000 bootstrapped replications of mean incremental costs and treatment failures avoided.

There is no generally accepted WTP threshold for treatment failure avoided but the CEAC (see Figure 9) shows the probability of being cost-effective at varying amounts: at £1000 it was 36%, decreasing to 13% at £10,000. The CEAC does not cut the y-axis as some of the joint density includes cost savings (SE and SW quadrants of Figure 8) and asymptotes to zero because the density does not involve health gains. 66

FIGURE 9.

Cost-effectiveness acceptability curves showing the probability of intervention being cost-effective compared to standard care for the primary and sensitivity analyses, using treatment failure rate as the outcome.

When using NMB at varying thresholds of WTP, the result was positive (meaning the intervention is cost-effective) for values up to £240/treatment failure avoided, and negative from £250 and above (see Table 33). The CEAC using NMB data (see Figure 10) showed a 39% probability of the intervention being cost-effective over standard care at a WTP threshold of £1000/treatment failure avoided.

TABLE 33 - Incremental net benefit at various willingness-to-pay thresholds per treatment failure avoided

Willingness to pay for treatment failure avoided	Incremental net benefit (mean, 95% CI)
£0	22.40 (−510.16 to 554.96)
£200	3.86 (−515.83 to 523.55)
£250	−0.78 (−517.43 to 515.88)
£1000	−70.31 (−551.19 to 410.57)
£10,000	−904.68 (−2122.83 to 313.46)
£20,000	−1831.77 (−4384.27 to 720.73)

FIGURE 10.

Cost-effectiveness acceptability curve showing the probability of intervention being cost-effective compared to standard care using the net monetary benefit method.

Patient preferences

A total of 114 patients (60 in the standard care group and 54 in the intervention group) provided responses, with slightly fewer for probability-based questions. The overwhelming majority of patients (95%) in both arms were satisfied with the care received and level of hospital support, and were happy with their method of treatment, believing it to be effective (see Table 34). Approximately, 70% of patients in each arm disagreed that they were concerned their treatment would not work beforehand and a total of 16 patients were concerned about how treatment for NS would affect family and friends (intervention 15%; standard care 13%). Of these 16, 88% would be willing to accept the risk of re-admission (15%) for early discharge, and 81% said they would still accept the risk even if this doubled to 30%, indicating a strong preference to be discharged as soon as possible.

A greater proportion of those in the intervention arm thought it was important to be discharged 1–2 days earlier, 59% in comparison to 35% in the control arm (see Table 35). This shows a tendency for patients to choose the response in keeping with the treatment they actually received. Patients in both arms agreed it was important not to be re-admitted to hospital although with slightly more in the intervention group agreeing this (85% vs. 72%). The majority of patients in both groups agreed they would accept some risk of re-admission in order to be discharged early (72% in the intervention group and 65% in the control group). When asked which treatment they would choose, 71% of patients in the intervention arm said they would choose the early switch to oral antibiotics, compared with 46% of standard care patients, again suggesting allegiance to the treatment received.

When presented with the hypothetical scenario of risk of re-admission doubling from 15% to 30%, 74% and 64% of intervention and standard care patients respectively would still opt for early discharge.

There was some evidence of completion errors regarding these hypothetical risk-based questions observed in a small number of responses, showing some difficulty in understanding and interpreting these scenarios.

In both groups, the bulk of respondents did not receive home care on a daily basis or have dependents, as shown in Table 36. Of the 24 participants who stated they receive daily care at home, there was a strong inclination to get home: 22 (92%) would accept the risk to be discharged early; 19/22 (86%) would accept the higher risk of re-admission in order to be discharged early. There were seven patients who said they received daily care at home as well as having to care for children or dependents.

Conclusion

The primary CUA showed that although the early switch to oral antibiotics led to small gains in HRQoL and small cost savings the differences were not statistically significant. The probability of the intervention being cost-effective was highest for all WTP thresholds in the PP sensitivity analysis. Although two subsequent sensitivity analyses estimated the intervention as slightly more costly than standard care, the ICERs were favourable (well under £20,000/QALY). There was a higher number of treatment failures in the intervention group within the CEA, meaning the intervention was dominated by standard care. However, subsequent analysis removing a high resource outlier showed a small and not statistically significant cost saving for those receiving the intervention. This one patient, who had a prolonged ICU episode for an uncommon condition that was not related to treatment allocation in this trial, appeared to bias the costs allocated to the intervention arm. Given that this patient was not representative of patients with low-risk NS, it was felt appropriate to consider the effect of removing this patient’s data.

The results of the CUA and CEA are not consistent, which may reflect the difference in individual patient data used in the analyses as a result of missing data. While the CEA showed a greater number of treatment failures in the intervention arm, there was a small but not statistically significant gain in QALYs for this arm. This can be seen by the large number of bootstrapped ICERs in the NE and SE quadrants of Figure 5, suggesting the intervention did indeed impact on patient HRQoL, albeit in a small way. Interestingly the incremental QALYs went in the opposite direction to incremental treatment failures. Although the trial was unable to conclude that the intervention is non-inferior to standard care, the CUA showed both treatments had a similar impact on HRQoL. Therefore, although treatment failures as the primary outcome are clinically important and relate to the intended purpose of treatment for NS, these are not synonymous with HRQoL, as reflected in patient EQ-5D-5L responses, presenting a disparity between what is valued by clinicians and patients.

The aim of the patient preference questionnaire was to elicit preferences for non-health outcomes, and the responses showed the majority of patients were content with the treatment they received, regardless of the group they were randomised to. This suggests both treatment options for NS are acceptable to patients, and could support clinicians in helping patients make choices. Patients who were worried about how their treatment would affect family and those receiving care at home were found to be more prepared to accept higher levels of risk of re-admission in order to have an earlier discharge. These data were particularly revealing since they indicate that patients have a much higher tolerance for the possibility of treatment failure in order to enable early discharge for their primary admission.

A major strength of these analyses was the choice of perspective. By taking a hospital perspective, it meant there was the ability to collect daily data and details of any re-admissions using hospital records. The short follow-up duration also meant less of a burden for research staff and resulted in complete cost data for all patients. There was complete QALY data for 87% of patients and over 90% of patients provided responses to the patient preference questionnaire.

The health economics analysis had some limitations. Firstly, due to the challenges of collecting outcome measures from patient questionnaires versus routine data retrieval from hospital records, there was a higher number of patients with complete data for analysis in the CEA compared to the CUA (n = 124 and n = 110, respectively). The impact on staff time reduction for administration of oral versus i.v. antibiotics was not assessed, as given the time difference of only 24 hours, differences would be small. However, the additional time spent in hospital and associated costs per hospital bed already includes staff costs. The preferences questionnaire was administered mostly over the phone at day 14 (unless the patient had not been discharged or had a visit), and given the complexity of some of the questions, processing them could understandably lead to completion errors. To minimise error and increase acceptability, there is a need to aid patients’ understanding of risk. This may have wider implications in relation to the consent process in trials and practice, and considering whether or not risk has been satisfactorily explained to patients or those consenting on their behalf. Furthermore, it would be preferable if patients were asked for treatment preference before allocation as the results showed an allegiance to the assigned method and may have been biased by patients’ experience of treatment.

Aim of the study

The NICE guidance in 2012 provided the first UK consensus guideline for treatment of NS while also highlighting a number of outstanding questions regarding optimal management. NICE were unable to recommend switching from empirical i.v. to oral antibiotics at < 48 hours even in patients at low risk of infective complications because of uncertainty about whether this achieved comparable outcomes to standard practice of longer-duration i.v. antibiotics. To bridge this evidence gap, the EASI-SWITCH trial, commissioned by the UK NIHR, aimed to provide a definitive randomised trial evaluating the effectiveness of early oral switch in low-risk NS patients.

Main findings and interpretation of results

The within-trial review of NS management enabled an evaluation of current management of NS in the UK following publication of the NICE guidance, showing many local policies had been updated to accurately reflect many of the recommendations. Clinicians also reported standard clinical practice that appears to be compliant with many of the key aspects of the guidance but there appear to be no standardised sepsis diagnostic criteria in widespread use and instead there is a range of approaches to incorporating infectious/sepsis criteria into NS diagnostic and care pathways. There has been increased use of routine risk assessment and consideration of early outpatient oral antibiotics for low-risk patients but there appears to be a wide range of approaches to risk stratification, duration of treatment and discharge from hospital. Consequently, despite challenges with recruitment, there was support for continuation of EASI-SWITCH from clinicians across the UK due to its potential impact on current UK practice. For clinicians and centres that already considered early oral antibiotics and discharge it was felt the trial had the potential to validate their practice evolution and make a compelling argument for further acute oncology service investment and modernisation.

Despite this support, recruitment remained challenging, resulting in early termination of the trial. Multisite trial activity in the area of low-risk NS has been challenging to deliver in the UK. Prior to the EASI-SWITCH trial, the ORANGE trial (ORal Antibiotics for Neutropenic sepsis Giving Early hospital discharge), which opened in August 2007, aimed to evaluate immediate upfront oral antibiotic therapy in low-risk NS patients,67 but closed in 2009 due to poor recruitment (27 patients registered and 12 randomised). The trial team reported that recruitment of investigator sites and patients was extremely difficult, with sites unable to participate due to local admission and care pathways and conflicting clinical management and local antibiotic protocols. 68 While these barriers were overcome to some degree within EASI-SWITCH, similar themes emerged from the surveys and interviews undertaken during the trial and it is likely that a combination of these challenges and a shift in equipoise as the trial progressed towards upfront or early oral therapy for patients with NS resulted in the plateau in accrual seen prior to closure.

The consequence of the smaller than anticipated sample size is that non-inferiority for early oral switch cannot be proven. The conflicting results of the ITT and PP analyses for the primary outcome may reflect this but also highlight the potential for analysis of different patient data sets to deliver conflicting results. Because of this potential, we determined a priori that a firm conclusion on the non-inferiority of the intervention would be reached only if both the ITT and PP analyses were in agreement. While analysis of either population alone may lead to bias, there is particular potential for increased risk of bias in ITT analyses due to non-adherence to treatment or deviations from the protocol resulting in a higher likelihood of concluding non-inferiority incorrectly. 69,70 Within our data set, of the 11 ITT patients who were excluded from the PP population, 8 had been allocated to the intervention arm; of these, 4 had their antibiotic treatment stopped prematurely, 2 had substantial interruption to treatment (at least two consecutive missed doses), and 1 had less than the minimum of 12 hours initial i.v. treatment. The three excluded patients in the standard care arm had received < 48 hours of i.v. antibiotics. As the ITT population included these patients, who did not receive the complete protocolised intervention, an ITT analysis risks underestimating the efficacy of the intervention; by comparison a PP analysis may overcome this issue.

The trial was pragmatic in design; decisions such as non-blinding of patients and staff were guided by the opportunity to assess the effect of early switch on timing of hospital discharge as well as advice from our patient representatives that blinding was unlikely to be maintained in the course of patient care. Open-label treatment has the potential to introduce performance bias; in this instance it may be that physicians have a lower threshold for antibiotic escalation from oral therapy, resulting in treatment failure and risk of bias towards concluding inferiority of the intervention. Similarly, non-blinding of outcome assessment risks introduction of detection bias; however, the components of the composite outcome measure were largely objective measurements of treatment failure rather than subjective components or qualitative measures.

While non-inferiority of early switch could not be proven within this trial, it may be that an early switch approach results in a shorter initial admission for treatment of NS although with the trade-off that the risk of re-admission for treatment failure is more likely than with standard care therapy. Reassuringly, and accepting the smaller than anticipated sample size, the risk of treatment failure and requirement for antibiotic treatment escalation or re-admission with early switch was not associated with serious consequences in terms of critical care admission or death. AEs were generally as anticipated in this patient population and were reported at similar rates between the early oral switch and standard care groups. Overall, therefore, it would seem that early switch could be a safe management option for selected patients with low-risk NS.

Although the results did not permit a definitive conclusion on clinical efficacy of early oral switch in relation to treatment failure, there were small, albeit not significant, gains in HRQoL and patients found it an acceptable treatment approach. While the primary outcome measure of treatment failure is clearly an important efficacy outcome for clinicians and patients, the observed increased rate of treatment failure in the early switch arm was not reflected in the HRQoL results, highlighting the differences in outcome measures valued by clinicians and patients.

The results suggest that early oral switch may be a cost-effective approach within existing NHS NS care pathways. The primary CUA showed the early switch to oral antibiotics led to small, non-significant cost savings and improvements in HRQoL compared to standard care. The probability of the intervention being cost-effective was highest for all WTP thresholds in the PP sensitivity analysis. Although two subsequent sensitivity analyses estimated the intervention as slightly more costly than standard care, the ICERs were favourable (well under £20,000/QALY). There was a higher number of treatment failures in the intervention group within the CEA, meaning the intervention was dominated by standard care. However, subsequent analysis removing a high resource outlier showed a small, non-significant cost saving for those receiving the intervention. A major strength of these analyses was the choice of perspective. By taking a hospital perspective, it meant there was the ability to collect daily data and details of any re-admissions using hospital records. The short follow-up duration also meant less of a burden for research staff and resulted in complete cost data for all patients. However, due to the challenges of collecting outcome measures from patient questionnaires versus routine data retrieval from hospital records, there was a higher number of patients with complete data for analysis in the CEA compared to the CUA (n = 124 and n = 110, respectively), with complete QALY data for 87% of patients.

In addition to global HRQoL, a patient preference questionnaire was used to better understand the non-health outcome measures that are important to patients and their willingness to accept different treatment approaches in NS management. Over 90% of patients provided responses to the patient preference questionnaire, and the responses showed the majority of patients were content with the treatment they received, regardless of the group they were randomised to. Patients who were worried about how their treatment would affect family and those receiving care at home were found to be more prepared to accept higher levels of risk of re-admission in order to have an earlier discharge. This approach did have some limitations. The preferences questionnaire was administered mostly over the phone at day 14 (unless the patient had not been discharged or had a visit), and given the complexity of some of the questions, processing them could understandably lead to completion errors. To minimise error and increase acceptability, there is a need to aid patients’ understanding of risk. This may have wider implications in relation to the consent process in trials and practice, and considering whether or not risk has been satisfactorily explained to patients or those consenting on their behalf. Furthermore, it would preferable if patients were asked for treatment preference before allocation as the results showed an allegiance to the assigned method and may have been biased by patients’ experience of treatment. However, overall the data are revealing, indicating that patients have a much higher acceptance of the possibility of treatment failure in order to enable early discharge for their primary admission than might have been anticipated by clinicians.

Conclusion

A definitive conclusion about the clinical effectiveness of early switch to oral antibiotics in low-risk NS was unable to be reached from the study findings due to the limited sample size.

While non-inferiority of early oral switch could not be definitively concluded, it may be that early switch results in a shorter initial duration of hospitalisation for treatment of low-risk NS. However, it is uncertain whether there is a consequent increased likelihood of re-admission to hospital compared to standard care treatment.

Treatment failure or requirement for antibiotic escalation or re-admission was not associated with serious outcomes such as critical care admission or death. There was no unexpected or increased risk of AEs observed with early switch compared to standard care.

There were small, non-significant gains in HRQoL for early oral switch and patients found it acceptable. Patients, particularly those with caregiver roles, were more prepared to accept the higher levels of risk of re-admission in order to have an earlier discharge.

Given the findings, early oral switch may be an acceptable strategy for some patients who can adhere to such a regimen and would prefer reduced duration of hospitalisation while accepting a potentially increased risk of treatment failure, resulting in re-admission to hospital.

Implications for future research

EASI-SWITCH is the first UK multicentre RCT comparing early oral switch to the NICE recommendation of 48 hours i.v. antibiotic therapy in low-risk NS. Delivery of the trial has highlighted the difficulties with recruitment of patients to trials in the unscheduled care setting. Evidence supporting specific interventions to improve recruitment in this setting is lacking71 and further work evaluating strategies such as verbal provision of trial information and obtaining verbal consent is warranted across supportive care trials in cancer patients, not just in the setting of NS.

A number of specialist cancer centres in the NHS have already established acute oncology services and/or ambulatory care pathways to deliver early oral switch or upfront oral antibiotics. 39–41 This results in tension between delivering prospective trials evaluating management approaches for NS and accepting that current ambulatory care pathways or antimicrobial strategies utilising upfront or early switch to oral antibiotics are typically applied to highly selected patient populations. This shift in practice and our experience in delivering this trial, accompanied by the declining NS rates, suggest further randomised trials in this patient population will remain challenging to recruit to. However, it is clear there remains scope to optimise and unify management strategies for low-risk NS, in particular how best to identify patients suitable for a de-escalation or ambulatory care approach from initial presentation. While the MASCC score remains the most widely used within specialist cancer centres, others have reported that stratification tools such as MASCC score or CISNE are too cumbersome to be applied in real-life ED practice. 46 Other scores that encompass ‘clinical judgement’ are similarly felt less applicable in ED or acute medical settings than validated decision rules or algorithms. Further research should explore tools for patient stratification for low-risk de-escalation or ambulatory pathways, including use of biomarkers and/or point-of-care rapid viral, bacterial and fungal testing as an adjunct to clinical decision-making tools. This could include application to shorter-duration antimicrobial therapy in line with antimicrobial stewardship studies in other settings.

Evaluation of any future stratification tools, treatment strategies or ambulatory care pathways should include a formal economic analysis in addition to assessment of clinical efficacy. This is particularly important where novel algorithms or ambulatory care pathways are established but restricted to specialist oncology services rather than general acute care or ED settings.

The EASI-SWITCH trial has highlighted the importance of patients’ perspectives in setting research questions that encompass patients’ willingness to accept risk of treatment failure as a trade-off for less intensive treatment and shorter duration of admission. The results of the patient preference questionnaire could be used as a precursor for designing a future discrete-choice experiment to assess patient preferences for treatment for NS. This method is used in eliciting strength of preferences in health care by presenting hypothetical choices varying treatment options and levels of key attributes (e.g. life expectancy, side effects), and in doing so respondents make trade-offs. 72 They have been effectively used in other cancer populations73 and it has been long thought by experts that NICE should recommend their use in incorporating patient preferences,74 but while their US counterparts the FDA have endorsed their use, NICE have yet to do so.

Contributions of authors

Vicky Coyle (https://orcid.org/0000-0001-9566-4130) (Senior Lecturer in Medical Oncology), trial co-chief investigator, contributed to the concept and design of the trial, protocol development, recruitment, trial management and interpretation of results.

Caroline Forde (https://orcid.org/0000-0001-5021-9973) (Clinical Fellow in Medical Oncology) was the trial clinical fellow and contributed to trial design, protocol development and day-to-day trial management and undertook the review of NS practice as the study progressed.

Richard Adams (https://orcid.org/0000-0003-3915-7243) (Professor of Clinical Oncology and Cancer Clinical Trials) contributed to the trial design, protocol development, patient recruitment and interpretation of results.

Ashley Agus (https://orcid.org/0000-0001-9839-6282) (Senior Health Economist), trial health economist, contributed to trial design and development and led the concept and design of the health economics evaluation including data analysis, interpretation of results and writing of Chapter 6.

Rosemary Barnes (https://orcid.org/0000-0002-0574-8670) (Professor of Medical Microbiology) contributed to the trial design, protocol development and interpretation of results.

Ian Chau (https://orcid.org/0000-0003-0286-8703) (Consultant Medical Oncologist) contributed to the trial design, protocol development, patient recruitment and interpretation of results.

Mike Clarke (https://orcid.org/0000-0002-2926-7257) (Professor of Research Methodology and Director MRC Methodology Hub), trial methodologist, contributed to the trial design, protocol development and interpretation of results.

Annmarie Doran (https://orcid.org/0000-0002-9802-008X) (Trial Coordinator) contributed to protocol development and study set-up and was responsible for trial co-ordination and day-to-day management.

Margaret Grayson (https://orcid.org/0000-0002-1861-8541) (Chair Northern Ireland Cancer Research Consumer Forum), trial PPI co-investigator, contributed to the trial design, trial development and management, review of patient-facing materials, lay communication and co-ordinated PPI opinion on revision of the trial design.

Danny McAuley (https://orcid.org/0000-0002-3283-1947) (Professor of Intensive Care Medicine) contributed to the trial design, protocol development and interpretation of results.

Cliona McDowell (https://orcid.org/0000-0002-7644-7197) (Senior Statistician), trial statistician, contributed to development of the protocol and statistical analysis plan and confidential data analysis for the DMEC and led the analysis of primary and secondary outcome data.

Glenn Phair (https://orcid.org/0000-0001-8137-0959) (Health Economist), trial health economist, contributed to the health economic data analysis, interpretation of results and writing of Chapter 6.

Ruth Plummer (https://orcid.org/0000-0003-0107-1444) (Professor of Experimental Cancer Medicine) contributed to the trial design, protocol development, patient recruitment and interpretation of results.

Dawn Storey (https://orcid.org/0000-0002-6546-3183) (Consultant Medical Oncologist) contributed to the trial design, protocol development, patient recruitment and interpretation of results.

Anne Thomas (https://orcid.org/0000-0003-1998-3134) (Professor of Cancer Therapeutics) contributed to the trial design, protocol development, patient recruitment and interpretation of results.

Richard Wilson (https://orcid.org/0000-0001-8018-7730) (Professor of Gastrointestinal Oncology) contributed to the trial design, protocol development, patient recruitment and interpretation of results.

Ronan McMullan (https://orcid.org/0000-0001-6760-6072) (Professor of Medical Microbiology), co-chief investigator, contributed to the concept and design of the trial, protocol development, trial management and interpretation of results.

Data-sharing statement

Requests for access to de-identified patient data can be made by researchers to the chief investigators, and data will be shared subject to any constraints. Requests for access should be accompanied by a proposal describing the aims and scope of the research, details of the data requested and data analysis plan. Proposals will be considered by the Trial Management Group, co-investigators and sponsor who will make a decision regarding data access. A data-sharing agreement will be signed between the researchers, principal investigators and sponsor.

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 is important that there are safeguards to make sure that they are 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 manuscript presents independent research funded by the National Institute for Health and Care 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, 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, the HTA programme or the Department of Health and Social Care.

Summary of trial and context provided to PPI representatives

Neutropenic sepsis (NS), or infection that develops in the setting of a low white blood cell count, is a complication of chemotherapy treatment in patients with cancer. The ability to identify patients at low risk of complication from infection offers the opportunity to avoid unnecessary treatment and hospital stay which disadvantages both patients and the NHS.

We are carrying out a trial of patients at low risk of complications, to evaluate whether changing from intravenous (i.v.) to oral antibiotics on the first day of treatment is clinically and cost-effective in comparison with standard longer-duration i.v. antibiotics. The trial will study two patient groups: one group will switch from i.v. to oral antibiotics 24 hours after i.v. treatment commences; in the other group, patients will receive standard care i.v. antibiotic for at least 48 hours with subsequent antibiotic continuation or switch to oral therapy at physician discretion based on their usual practice. We will then compare the groups to see how effective the two different treatment strategies are.

The outcome that we will use to assess how well each treatment strategy works is ‘failure of treatment’ and we will measure this on day 14 of follow-up. This is defined by the occurrence of one of several undesirable outcomes that are all of importance to patients. These include: recurrence of fever; re-admission to hospital; antibiotic side-effects; quality of life and death. However, the most common way that patients’ treatment fails is that their fever comes back and they may need to be admitted to hospital again; by comparison, death is uncommon in patients with NS who are at low risk of complications (the group we are studying).

Question for PPI consideration: how many patients to study?

We have designed this trial based on the findings of other small studies. These indicate that it is reasonable to assume that early switch to oral treatment would have similar effectiveness to continuing on i.v. treatment for longer. We are thinking about the number of patients we might need to include in a large trial to confirm these findings. For example, with only one patient in each group, even if both had successful treatment, we would not be confident that this would be reproduced if all patients received this treatment across the NHS. By comparison, if we studied 10,000 patients in each group and found both treatments to be the same, we would be highly certain that this would be reproduced in NHS practice – but the trial would be impossible to do and cost tens of millions of pounds.

We are not trying to show that early oral switch is more effective than usual i.v. antibiotic treatment but are focusing on whether it is not much less effective treatment. This type of trial design is a ‘non-inferiority’ design in which there are statistical rules and tests to show how certain we are that the new treatment being tested is not much less effective than the usual treatment. Statisticians express how certain they are about study results using a 95% confidence range either side of the result. This means that researchers have very high certainty (95% certainty) that the true effectiveness of treatment lies within that range.

We are asking you to help us agree how wide this confidence range should be – or in other words what level of trade-off between the benefits of shorter duration of treatment and the risk of treatment failure is acceptable for patients. We also need to balance having enough patients to obtain meaningful results and not having so many patients that the study is not achievable.

We propose setting the limit of the 95% confidence range for this at 15%. This means that, if we show that both early oral antibiotic switch and longer courses of i.v. treatment effectively treat 85% of patients, we would have very high certainty (95% certainty) that the early oral switch approach had a ‘true’ effectiveness of between 70% and 100%. In other words, the study might conclude that the two treatments are similar but acknowledge that early oral switch is at least 70% effective while standard i.v. treatment is 85% effective.

In other words, for every 100 patients treated with an early oral switch strategy, all would benefit from early discharge home. However, while 15 would be expected to need re-admission (whether they had early switch or not), we would be acknowledging that early oral switch could lead to an extra 15 patients needing re-admission (in comparison with those who receive prolonged i.v. treatment in the first instance). Therefore, the trade-off we are asking you to consider is whether the possibility that an extra 15 patients may need re-admission is out-weighed by the advantage of the 100 patients receiving oral switch and being discharged early.

The main question is whether this possible difference in effectiveness between the two treatment options is acceptable. We had initially planned the study to assess whether the difference in effectiveness was no more than 10%; however, we have found that recruiting the number of patients needed to do this makes the study infeasible. Therefore, we are considering making a change to assess whether the difference in treatment effectiveness is no more than 15%, since that requires fewer patients and makes the study achievable.

We need to decide whether it is better to continue the study using the 15% confidence range or abandon the study altogether.

The main question is whether in your opinion it would still be useful to do the study if the confidence range is increased from 10% to 15%. In other words, do you think patients would be similarly convinced that switching to oral treatment earlier (to allow them to go home sooner) is a good idea if the additional risk of needing re-admission may be up to 15%, rather than 10%?

Given that the advantage of early oral switch is likely to be early discharge from hospital and the main consequence of the treatment being ineffective is re-admission we would like you to consider whether this trade-off would be to patients’ advantage overall.

2. PPI responses to sample size/NI margin question

Input from a range of PPI representatives was sought on this proposed change via our PPI co-applicant, Mrs Margaret Grayson, who distributed a summary of the study design and study progress including the rationale for reviewing the sample size and specifically the NI margin. The information was distributed to the Northern Ireland Cancer Consumer Forum and the Independent Cancer Patient Voice (a national independent patient advocate group which has the support of the professional members of the NCRI breast cancer Clinical Study Group).

Respondents were posed the question that if the NI margin was increased to 15% whether this level of trade-off between the benefit of less intensive treatment and the risk of treatment failure would be acceptable to patients; comments on the change were also welcomed.

There were 21 replies received and the majority of these supported this change (n = 19) with additional comments received from seven of these respondents included below. One respondent did not support this change and one was uncertain (comments from both included below).

PPI comments:

• We all know that there is a finite amount of funding and that the professionals are trying to do the best for all within these limits. The fact that those who would require re-admission are being closely monitored is very positive so having considered the points made in your paper I feel that 15% re-admissions would be an acceptable level.

• Very interesting study and I think well worth doing. I believe that for most patients going home early is key. Recovery at home is almost always preferably to being in hospital and I would definitely accept the very small extra risk providing good home back-up is in place.

• I would recommend this study even if the risk of re-admission may be up to 15%. Having personally seen this condition with my daughter when she had become neutropenic when she was being treated for breast cancer. I believe she would have much preferred the oral method of treatment and getting home sooner. As in her case she had to spend 2/3 days in hospital. She did not like this. But I must add as only one person’s experience she may or may not have had to go back into hospital perhaps to have i.v. treatment. Personally as her mother and carer I can see that it would be a very good study.

• This seems like a hard decision as it boils down to either doing the study or not doing the study. I wonder why recruitment was such a problem for the original study. I suppose we just have to accept that the study in its original form is not possible.

• It would certainly be advantageous for patients to be discharged early as no one wants to spend any more time than necessary as an inpatient. My husband hated it and was always agitating to get home! However, doubling the percentage of patients that then need re-admitted, seems, on the face of it, rather a high number. Do these patients then need more protracted treatment, will their sepsis be more difficult to treat, are they at risk of becoming more seriously ill? In essence, how much additional danger are patients facing if such a change in treatment is implemented? Does it mean that we would be content for 30% of patients to need re-admission? Sorry, I seem to be asking questions rather than giving an opinion – but this is what is going around in my mind as I read the description. Maybe it’s only after a study that such questions can be answered. For that reason alone, I would tend to be supportive of the trial going ahead, though with reservations. I would expect that, should the results begin to show a greater risk to patients, that the study would be stopped. Sorry if these thoughts are not particularly helpful.

• I think it would be useful to proceed with the trial if the confidence range is increased from 10% to 15%. Switching to oral treatment earlier will allow the patient to go home earlier and this in itself could aid recovery (psychological benefit) and reduce the risk of infection (less exposure to a wider population at home than in hospital). I would assume patients on oral treatment will be closely monitored at home with an identifiable easy/quick pathway for re-admission and i.v. antibiotics if necessary, as opposed to being processed through A&E. The patient’s home situation would be conducive to enabling close monitoring for indicators of infection occurring (maybe the presence of a carer?) and that there are no concerns regarding the home situation itself or quality of care at home. My feeling is that this study would be worthwhile.

• Margaret, I have read this and think that, on balance, it IS worth while making the change and trying for the lower level but I honestly feel that there is really a decision for the researcher and the ethics committee rather than a PPI decision.

• As you know I’m a fan of EASI-SWITCH, and believe that even with the increase in confidence range it is worth continuing the Trial.

• Very difficult. Clearly good to reduce the amount of i.v. chemotherapy, but what would spook me is the, albeit remote, possibility of death (certainly an ‘undesirable outcome’), for the sake of cutting the stay in hospital by a day. But I can see that for someone with family responsibilities, the reduction in the time in hospital might outweigh that risk. I don’t think that the switch to 85% confidence, rather than 90%, would make any difference – but is it not possible to continue with the trial and as long as the risks are clearly explained, give patients the choice? If the take-up is low, then the trial would have to be abandoned.

• I am not convinced that the gains to be had here are worth the trial. I would also want to see a health economic argument as re-admission to hospital may be more costly than reducing the initial stay by a day. But then this may be one of the trial outcomes. People who have sepsis are often very poorly for a long time and don’t always fully recover. I would want the treatment that’s going to give me the best chance of recovery.