Interventions to reduce mortality from in-hospital cardiac arrest: a mixed-methods study

The nature of the problem and policy responses

Despite national guidance,1 regional collaborations2 and additional resources to facilitate change over the last two decades,3 data from the National Cardiac Arrest Audit (NCAA) show that a fourfold variation in in-hospital cardiac arrest (IHCA) incidence and survival exists between hospitals in England (D.A. Harrison, Intensive Care National Audit & Research Centre, 2013, personal communication). It is estimated that resuscitation teams are called to between one and five IHCAs per 1000 hospital admissions, equivalent to around 20,000 IHCAs in NHS hospitals in England each year, with survival to discharge around 15%. 4,5 Although unexpected cardiac arrests are relatively rare, with any ward likely to see only one or two cases per year,6 they may signal a failure to manage antecedent events. Reviews of these patients have shown that many exhibit clear signs of deterioration (physiological changes or level of consciousness) for up to 8 hours beforehand. 7,8 This is despite a further 100,000 ward patients being visited annually by outreach teams with advanced critical care skills because of concerns for their clinical state. One-third of these patients will have died within 30 days of this contact, indicating the high-risk nature of their conditions. 9 Unchecked deterioration, even if not leading to death, is costly for both patients and the NHS, creating a need for extra treatment, prolonging lengths of stay and increasing risk of exposure to hazards, such as drug side effects and hospital-acquired infections.

Since the late 1990s, evidence has been emerging that has demonstrated that patients admitted to the intensive care unit (ICU) from the ward have a worse outcome than those coming straight from the operating theatre or the emergency department. 10,11 Furthermore, up to 25% of ICU patients die following discharge from the ICU. These findings were seen as an indication of the suboptimal care of critically ill patients on the wards. The Department of Health and Social Care (DHSC)’s review of critical care provision in 200012 reported that problems existed with the skills and knowledge of ward staff and their access to advice and support in the care of acutely ill patients. These problems were compounded by increasing workload pressures. In 2005, the National Confidential Enquiry into Patient Outcome and Death (NCEPOD) report An Acute Problem?13 found that in many hospitals, junior doctors continued to be largely responsible for providing emergency care on the ward, with inadequate consultant input, and that communication failures between teams contributed to delays in escalating care to a higher level. The report estimated that 21% of admissions to the ICU from general wards could have been avoided with better care. A further NCEPOD report based on a review of a sample of deaths associated with IHCAs14 found that a lack of early assessment by consultants, inadequate management and monitoring plans, a lack of decision-making around appropriate ceilings of care or suitability for resuscitation and failure to escalate care in response to deterioration continued to contribute to these events.

Other far-reaching changes affecting hospitals in recent years have challenged the pace of improvement in the quality of care of acutely ill patients. Most significantly, the implementation of the European Working Time Directive15 has led to a reduction in the number of hours that junior doctors are able to work. There have been concerns that senior doctor coverage has not expanded to adequately fill the gap. The introduction of shift systems as part of these changes has also had an impact on teamwork and continuity of care. 16 Difficulties recruiting and retaining both medical and nursing staff and the increasing use of agency staff and health-care assistants (HCAs) to deliver key clinical activities has further exacerbated the situation. 17–19 As the elderly population expands, many of whom are living with multiple long-standing illnesses, there is mounting pressure on hospital beds. This leads to some patients being cared for on wards that are not suitable to meet their needs and growing numbers of acutely ill patients being cared for on the wards as they wait for a bed to become available in the ICU. 20,21

In response, the DHSC’s 2000 report Comprehensive Critical Care12 set out a strategy to provide an integrated hospital-wide approach to critical illness, in effect breaking down the walls between ICUs and the wards, thus ensuring wider access to critical care expertise through support and collaboration. This policy led to the introduction of nurse-led critical care outreach teams (CCOTs). 22 The task of these teams was to strengthen critical care skills amongst ward staff and provide support to ward staff in managing patient deterioration. In doing so, it was hoped that unplanned ICU admissions would be reduced and that there would be better outcomes for patients discharged from the ICU to the wards. They were also to have a role in promoting the uptake of track-and-trigger systems (TTSs), scoring systems based on physiological measures to detect patients at risk and trigger appropriate responses. 23 In 2007, the National Institute for Health and Care Excellence (NICE) issued guidance on the care of the acutely ill patients, reiterating the need for CCOTs and the widespread use of TTSs. 24

With recent evidence that failure to recognise and respond to deteriorating patients accounts for around one-third of avoidable mortality within hospitals,25,26 this issue has continued to be prioritised in efforts to improve patient safety across the NHS. 27 Recent initiatives include the promotion of the situation, background, assessment, recommendation (SBAR) tool by the NHS Institute for Innovation and Improvement. 28 This is one of a number of structured handover tools designed to improve the quality of communication of urgent patient information among clinicians. A national, standardised TTS [the National Early Warning Score (NEWS)] was introduced by the Royal College of Physicians (RCP) in 2012. 29 The RCP and the NCEPOD have called for consultant assessment shortly after admission and adherence to policies related to treatment limitation and ‘do not attempt cardiopulmonary resuscitation’ (DNACPR) decisions to ensure that cardiopulmonary resuscitation (CPR) is not undertaken on those patients unlikely to benefit. 14,30

Interventions aimed at identifying and responding to patient deterioration

DeVita et al. 31 developed the concept of the rapid response system composed of two parts. The afferent arm is focused on the recognition of patient deterioration through the collection and interpretation of abnormal physiological parameters and triggering an appropriate response to that deterioration. The efferent arm brings staff with critical care skills to the patient’s bedside in a timely way. This system requires knowledgable staff, who understand the implication of any changes in a patient’s level of consciousness or physiological status, to accurately convey that concern to secure the appropriate response. Ensuring an effective system requires ongoing monitoring of performance and improvement activities when failures are detected. 31 Staff education, TTSs, tools for structured communication between staff and rapid response teams (RRTs), an international generic term for critical care-trained response teams that incorporates UK outreach teams, are the four key interventions that need to be in place to identify and respond to deterioration.

Life-support training, which focuses on preparing staff to deal with cardiopulmonary arrest, is mandatory for most NHS clinical staff,32 but formal courses that take a wider perspective on the knowledge and skills needed to identify patient deterioration and provide early intervention are also widely available. 33 Research has suggested that such training can influence detection of deterioration,34 early management by ward staff35 and the likelihood of calling for help if needed. 36

Track-and-trigger systems are based on the association between abnormal physiological findings, poor outcomes and the fact that early intervention can prevent serious consequences, such as cardiac arrest. 37 Such systems let staff ‘track’ a range of physiological parameters including respiratory rate, blood oxygen levels, temperature, blood pressure and heart rate to create scores. These scores then ‘trigger’ specific actions, which might include increased frequency of monitoring or a request for further support. They are seen as an effective way of monitoring the condition of acutely ill patients and their increase in use over the past 20 years has seen a large and diverse number of tools in use across the NHS. 38 Addressing concerns around the lack of consistency across hospitals, and even within hospitals,39 the NEWS has been accepted across the country since 2013, following evaluations that showed that it performed at least as well as, and often better than, the scoring systems already in place. 40 Hospitals are increasingly adopting electronic TTSs to replace paper-based systems. Electronic systems can be designed to counter known problems with paper-based TTSs by mandating entry of a full set of patient observations, accurately calculating scores and automatically sending an alert to an appropriate responder when a particular score threshold is met. 41,42 Many of these recently introduced electronic TTSs use the NEWS as the system for score calculation.

Clinician-to-clinician handovers represent a high-risk area for patient safety if key information about a patient’s risk of deterioration is omitted. 43 There are a number of standardised communication tools, the use of which is designed to increase the effectiveness of communication between clinicians, especially when information has to be passed between different professional groups. 28 A RCP survey in 2010 identified a range of tools in place across NHS hospitals, with some hospitals not using any. 44 SBAR, originally designed for use in the military, organises the information to be transferred into four parts: (1) situation – a description of what is happening now and where, (2) background – the patient’s past medical history, (3) assessment – an overview of the patient’s current condition including physiological measures and (4) recommendation – what help is needed and when. 45 This has become the main tool promoted for use in the NHS. 28

The trend for establishing designated RRTs began in Australia in the 1990s as an expansion of the role of the cardiac arrest team, following the recognition that earlier interventions could improve patient outcomes. 46 Subsequent to the publication in 2000 of Comprehensive Critical Care,12 funding was made available in England to establish CCOTs, the majority of which were nurse led. This was followed by the establishment of RRTs in Scandinavian countries and the USA. 31 The roles of these teams and their composition varied within and across countries, often in response to organisational needs. 47 In Australia, Sweden, Denmark and the Netherlands, the majority of teams called to attend patients actively deteriorating on the wards are doctor led and in the USA they may be nurse or doctor led. 48

NHS acute hospital trusts in England have a number of teams that respond to emergencies. Some have a doctor-led medical emergency team (MET) that attends sudden and overwhelming clinical events (e.g. massive gastrointestinal bleeding or an epileptic seizure), a doctor-led resuscitation cardiac team for cardiopulmonary arrests and a nurse-led outreach team responding to ward patients at earlier stages of deterioration. Unlike outreach teams that have staff members permanently allocated to the team, METs and resuscitation teams tend to be formed from ‘on-call’ staff who come together as a team on a particular day or weekend as part of their ‘on-call’ duties.

In general, the METs and resuscitation teams could be described as reactive to adverse events, such as acute collapse or IHCA. Following the original objectives set out for these teams, nurse-led outreach teams usually play both a proactive role in preventing deterioration (through education and support of ward staff, promoting the use of TTSs and actively identifying high-risk cases) and a reactive role by attending to deteriorating patients when alerted by ward staff. By intervening at an earlier stage of deterioration, evidence suggests that outreach teams are likely to have a greater impact on reducing cardiac arrests and unplanned admissions to the ICU. 49 Given that deterioration in some ward patients may indicate proximity to the natural end of their lives, part of this impact may be attributable to the opportunity it provides team members to interact with patients and their relatives to identify goals of care. As a consequence, treatment limits can be agreed, preventing patients from receiving inappropriate CPR when it is felt it would provide little benefit. 50 International evidence is emerging that this is an expanding role for both doctor-led and nurse-led RRTs. 51,52 In view of the fact that no recommendations were made as to the structure of CCOTs in Comprehensive Critical Care53 it is not surprising that wide variation in service models have been reported across hospitals. Some teams play a larger role in the implementation of TTSs and the education of ward staff, whereas others spend a greater proportion of their time providing clinical care for patients on the ward.

The relationship between interventions and outcomes

Despite concerted national efforts to address the issue of ‘failure to rescue’ deteriorating patients, variation in IHCA rates and survival persists. This suggests that either the effectiveness of these recommended interventions is not as great as anticipated or problems with implementation limit their effect. At face value, a rapid response system would seem to be an appropriate intervention to address delayed response to deterioration. However, its components are each complex interventions in themselves, and implementation may not lead to the predicted outcome because of the possibility of unanticipated consequences. 54 This is evident from the inconsistent findings of evaluations of even the most scrutinised interventions, such as RRT (both doctor led and nurse led), with the early promising results from single-site studies not replicated in larger trials. 55,56 The growing literature on complex interventions also points to the importance of implementation fidelity if such interventions are to achieve predicted outcomes in different settings. 54,56 Variation, such as in outreach team composition or roles, the proportion of patients they are called to see, the levels of uptake of TTSs and the process of making DNACPR decisions will lead to differences in outcomes. 53 The importance of implementation fidelity is apparent in one local collaborative (University College London Partners Deteriorating Patient Collaborative) that reported significant reductions in IHCA incidence through the introduction of standardised approaches to the management of acutely ill patient across the patient pathway in 13 hospitals (J Welch, University College London, 2013, personal communication).

Subbe and Welch57 have described a ‘chain of survival’ for the deteriorating patient. The chain relies on ‘high-quality recording of physiological signs; the education and mind-set of staff at the bedside to recognize pathological patterns; the reporting of abnormality to the response team; a timely and appropriate response by the latter’. 57 This model highlights the point that the impact of any one intervention is limited by any weaknesses at other points in the chain. For instance, TTSs rely on staff having the ability to effectively escalate concerns to avoid the TTS score being limited simply to monitoring a patient’s decline. Thus, the effectiveness of any intervention instituted to detect or respond to deterioration will, to some extent, depend on the thoroughness with which other interventions have been implemented. These interventions exist within a particular context that may also act as a barrier or facilitator to their effectiveness.

Attempts to estimate the impact of staff education, TTSs, tools for structured communication between staff, and outreach teams on IHCA rates and survival are complicated by the fact that a number of other mechanisms may affect these outcomes. For example, IHCA rates are also sensitive to changes in case mix. Older patients with multiple comorbidities are more likely have unshockable cardiac rhythms if they arrest, which increases the risk of death. 58 More older patients in the pool of those eligible for resuscitation will potentially reduce rates of IHCA survival. At the same time, more attention is being paid to improving end-of-life care and promoting early decision-making around goals of care and treatment limitations. 59 The appropriate use of DNACPR decisions to avoid futile resuscitation attempts is part of this effort. This will remove from the pool of patients who receive CPR those who are likely to have the poorest outcomes and potentially improve survival. In the UK, suitability for CPR is a clinical decision usually made by a patient’s consultant. National guidance exists to ensure good practice in this area and this is regularly updated to incorporate statutory or legal changes, such as the High Court ruling (‘the Tracey case’)60 in 2014, that ensured that patients in the UK must be consulted and informed of DNACPR decisions. 61 Little information is available about changes in rates of DNACPR decisions across the NHS over the recent years. Although an ageing population and improved practice in this area may have led to increased use, these increases may have been countered to some extent by the reluctance of consultants to discuss end-of-life issues with patients or relatives or by increasing pressure from families to continue active treatment. 62

The CPR process, both intra-arrest and post arrest, affects survival. 63 Standardisation of training, emergency calling systems and equipment have combined with improved management of arrest and post-arrest care to increase survival. The faster the response, the more likely a patient will still be in a shockable rhythm, such as ventricular fibrillation, when the resuscitation team arrives and this carries an improved survival rate. 6

Given that the use of different configurations of preventative interventions is well-established across the NHS, it has not been possible to evaluate the impact of each component using experimentation [i.e. a randomised controlled trial (RCT)]. However, the opportunity of natural experimentation, which exploits variation in health care, exists. 64 When combined with assessment of implementation fidelity,65 to determine which aspects of an intervention have been delivered, it will be possible to gain a better understanding of the association between different arrangements of interventions and IHCA incidence and survival. This study set out to test several hypotheses in order to make recommendations for best practice likely to have an impact on levels of avoidable serious harm and mortality. The hypotheses tested were as follows:

• TTSs (in combination with structured handover and education) will reduce IHCA rates by identifying deteriorating patients earlier while there is greater opportunity of reducing their risk of arresting.

• TTSs (in combination with structured handover and education) will have no association with survival following a cardiac arrest. It would be expected that the afferent (trigger) arm of the rapid response system would have little impact on survival compared with the efferent (response) arm.

• Outreach teams will reduce IHCA rates by reducing the risk of deteriorating patients arresting more than traditional patterns of response, such as ward-based staff.

• Outreach teams will increase survival following a cardiac arrest by increasing the application of treatment limitation orders, such as DNACPR decisions in deteriorating patients at high risk of an unsuccessful resuscitation.

Theoretical framework

The design of the study drew on three theoretical approaches. The first is natural experimentation, an approach that exploits variation in health-care service provision and outcomes to determine associations between different service models and those outcomes. 64 It is particularly useful when a range of services are already in place and it has become impossible to undertake randomised trials to determine which have the best outcomes. 66 Second, Rogers’67 ‘theory of diffusion transfer’ postulates that developing effective interventions is only the first step in improving outcomes. Aside from dissemination and adoption, interventions need to be implemented effectively and then sustained. The literature on complex interventions in health care indicates that implementation is a key determinant of discrepancies between expected and observed outcomes. 54 To address this, the third theory we employed was that of implementation fidelity (the degree to which programmes are implemented as intended68), which has received relatively little attention to date. 69 To identify how interventions were implemented in practice we used Carroll et al. ’s65 conceptual framework for implementation fidelity. The framework defines adherence in terms of intervention content, coverage, frequency of use and duration, all of which are influenced by four contextual moderators: (1) intervention complexity, (2) facilitation strategies, (3) quality of delivery and (4) staff engagement. Through understanding how interventions have been implemented in practice, we were able to go beyond what systems hospitals reported they have in place and identify key differences in how these interventions have developed and are currently organised and run. This information was used to support the development of our major quantitative analyses focused on associations between intervention types and outcomes and to determine explanatory factors for our findings.

Definitions

Our outcome measures, IHCA rates and survival, were obtained from NCAA. This was the first major study to be able to use data collected by this audit, which started in 2009 and was a collaboration between the Intensive Care National Audit & Research Centre (ICNARC) and the Resuscitation Council (UK). 61 The audit is now receiving reports from > 80% of hospitals in England, representative of the range of hospitals found in the NHS. The audit definitions70 for the outcomes were as follows:

• IHCA – all individuals, including neonates, who receive either chest compressions or defibrillation by a hospital-based resuscitation team in response to a 2222 call.

• Return of spontaneous circulation for > 20 minutes – all individuals receiving either chest compressions or defibrillation by a hospital-based resuscitation team who have a return of spontaneous circulation after a cardiopulmonary arrest for > 20 minutes. (Return of spontaneous circulation implies that the patient’s heart has begun to beat on its own again. This is a measure of short-term survival.)

• Survival to hospital discharge – all individuals receiving either chest compressions or defibrillation by a hospital-based resuscitation team who are alive at the time of discharge from hospital. This is a measure of intermediate-term survival.

We also measured longer-term survival including survival to 30 and 90 days post admission.

The NCAA audit, focusing only on those arrests that elicit a resuscitation team visit, provides an objective count of the majority of arrests, apart from those occurring in patients with DNACPR decisions in place and those occurring on coronary care units (CCU), high-dependency units (HDU) and ICUs where a resuscitation team is unlikely to be called. However, practice does vary across NCAA hospitals as to exactly which IHCA data are submitted, and 16.8% of the inpatient arrests in the audit do originate from CCUs (9.9%), HDUs (1.6%) and ICUs (5.3%), plus another 7.4% from theatres or imaging departments.

Work packages

The research was divided into four work packages:

Work package 1 – systematic literature review of evaluative studies of interventions designed to decrease the incidence of and improve outcomes from IHCAs.

• Research questions:

  • What was the scientific evidence for interventions designed to decrease the incidence of and improve outcomes following IHCA?

  • What are the essential components of effective interventions and key contextual factors that influence their impact (effectiveness)?

Work package 2 – developing and piloting a questionnaire for assessing the use of interventions designed to identify and respond to deteriorating patients in acute hospitals based on the combined findings from work package 1 and qualitative work in a sample hospitals.

• Research question:

  • How were interventions aimed at deteriorating patients implemented in practice (how did the organisation and contextual moderating factors vary across hospitals)?

Work package 3 – national survey of hospitals.

• Research question:

  • How much current variation was there in types of intervention across hospitals taking part in the NCAA and how had these changed over time?

Work package 4 – evaluating the impact of interventions through interrupted time-series and difference-in-difference approaches.

• Research question:

  • How was variation in interventions associated with variation in IHCA incidence and outcomes found in the NCAA?

Linkage across work packages

The principle of natural experimentation is to observe associations between differing arrangements of services and outcomes. To understand variation in service provision for deteriorating patients, we initially planned to summarise the policy drivers leading to the introduction or spread of the interventions and supplement this with a review of the evidence of effectiveness of interventions in relation to our outcomes of interest (IHCA and survival). From this review, we extracted the key components for each intervention that might be linked to positive outcomes to produce a typology.

Having established key components for each intervention, the next stage was to undertake qualitative semistructured interviews with frontline clinical staff and managers to understand how the services are being implemented on the ground in the NHS, how this implementation varies from the evidence-based typology and the nature of any contextual factors that might be acting as important barriers and facilitators to implementation.

Combining the findings from the literature review and the qualitative work enabled key variations in services across England, and other factors in the hospital environment that are potentially having an impact on patient outcomes, to inform the design of a questionnaire survey. The survey was the main mechanism through which we identified the current range of variation in services for deteriorating patients in NCAA hospitals and explored how these have changed over time.

The final stage saw the hospitals grouped by their differences in key service features, cross-sectionally and across time and patient-level outcome data for each hospital examined. In this way, we were able to establish the relationship between different service configurations and outcomes, and determine which variants produce the best outcomes.

Scope and search strategy for systematic reviews

Initial scoping searches were used to find the latest high-quality systematic review of each of the four categories of intervention to use as the foundation for our search strategy (see Appendix 1). A de novo search strategy was created for additional primary research published since the last high-quality systematic review. Underpinning these searches were four distinct population, intervention, comparison, outcome questions (Table 1), each tailored to the nature of the intervention and the available evidence.

Inclusion criteria common to all four interventions were (1) the implementation of a new intervention in the adult general inpatient environment, (2) interventional or observational studies with a comparison group, or a review of such studies, and (3) the assessment of outcomes. Intervention-unique criteria were also applied, for instance single parameter TTSs were excluded as they are not recommended by NICE and are therefore less likely to be used in modern clinical practice.

For all interventions, we excluded studies that were (1) entirely retrospective, (2) had no comparison group, (3) performed no statistical tests, (4) had an abstract only or (5) were already incorporated in a selected review paper. There were also exclusion criteria specific to each intervention, as listed in Table 2.

The strategies were finalised following consultation with a university librarian who was knowledgeable in search techniques for systematic reviews.

Searches were run in EMBASE and MEDLINE in October 2014 and updated in September 2017, with the expectation that this approach would capture the majority of studies of interest. A combination of Boolean operators, truncations, key words and Medical Subject Heading (MeSH) or Emtree terms were used (see Appendix 2). Results were limited to the English language and by date.

The sensitivity of each search was verified by checking its ability to find key papers identified during the scoping searches.

Selection strategy

Following removal of duplicate papers, they were divided between reviewers (CC, RZ and HH) for single title and abstract sifting. If a reviewer was uncertain about including a paper, a second reviewer’s judgement was sought and a consensus reached. The same process was applied to the full-text articles that remained after the initial sift. A detailed breakdown of each sift can be found in the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA)-style flow diagrams in Appendix 3.

Data extraction and quality assessment

Data items based on the Cochrane Handbook for Systematic Reviews of Interventions72 were extracted from each review (see Appendix 4). For published systematic reviews that had a broader focus, we restricted our data extraction to the primary research studies cited that were relevant for our aim. The number of included papers is indicated in the first column of each of the summary tables. The exclusion criteria were specific to each of the four categories of intervention (see Table 2).

The quality of each systematic review and primary research paper was assessed by a single reviewer using the NICE methodology checklists used in Clinical Guideline 5024 (CG50) (see Appendix 4). Primary research papers were assessed for selection bias, attrition bias, performance bias and detection bias (see Appendix 4). All papers were then allocated an overall rating based on the study type and quality assessment, founded on the system used by NICE in CG5024 (Table 3). This resulted in all systematic reviews being rated 2++. Most primary studies were rated 2– except for four73–76 rated 1– and two rated 2+. 77,78 The detailed tables containing extracted data are presented as Report Supplementary Materials 1–8.

Data synthesis and analysis

Given the heterogeneity of the studies, it was not possible to undertake a statistical summary; therefore, a narrative summary was undertaken. This started with tabulation of the study descriptions to identify commonalities and differences across the studies, along with consideration as to the weight of the evidence for each study based on the study design. This synthesis was summarised in tables reporting study outcomes throughout this report.

Rapid response teams

Six systematic reviews incorporated relevant studies from Australia, North America and Europe, from 1950 to 2015 (Tables 4 and 5). The reviews combined evaluations of both medically led and nurse-led teams, with studies focused on the latter being in the minority.

The first review contained two RCTs and 16 uncontrolled before-and-after studies. 24 One RCT was a large cluster trial of a medically led RRT alongside the use of an early warning score (EWS) and an education intervention. 82 They found no statistically significant impact on the cardiac arrest rate, unplanned ICU admissions or hospital mortality. In contrast, a smaller RCT83 reported a significant reduction in hospital mortality. The review authors suggested intervention variability as one possible explanation for the result in the larger cluster RCT because, although the calling criteria were uniform, the RRT makeup was not. In addition, they noted that the sample size was potentially inadequate. The results of the 16 uncontrolled before-and-after studies were mixed. Although a majority found a significant decrease in cardiac arrest rate and unplanned ICU admissions, less than half found a significant reduction in hospital mortality.

Chan et al. 56 also found no overall effect on hospital mortality in their meta-analysis of 11 mixed-quality studies, but they did find a 34% reduction in rates of cardiac arrest outside ICUs (although less impressive in the high-quality studies). A reduction in cardiac arrest rate was also reported by Massey et al. 84 Five out of eight studies (all single-centre observational studies) reported a significant reduction in arrests, ranging from 13% to 50%. Unlike the earlier reviews, four of the studies also reported a reduction in hospital mortality. Despite this, the authors concluded that, overall, the evidence for the effectiveness of the RRT was inconclusive, suggesting underuse of the teams as a possible explanatory factor.

McNeill and Bryden48 tackled the issue of intervention heterogeneity by dividing their review into teams that were medically led (20 studies) and those that were not (23 studies). They reported that METs reduced cardiac arrest rates and hospital mortality and, in addition, unplanned ICU admissions decreased. The evidence on hospital length of stay (LOS) and ICU mortality was unclear. The evidence for non-doctor-led teams was not as strong. Across 17 uncontrolled before-and-after studies, eight found reductions in cardiac arrests, five in mortality and one in ICU admissions. Three interrupted time-series studies found a reduction in cardiac arrest rates but no impact on mortality. A single-site, stepped-wedge RCT83 did show a significant reduction in mortality (48%) but validity was weakened by randomisation of only 2903 out of 7450 patients declared as eligible.

A systematic review by Winters et al. 79 reported a reduction in cardiac arrests and hospital mortality, although the latter was statistically significant in only seven studies, Maharaj et al. 47 conducted a meta-analysis of 29 studies that reported a decrease in cardiac arrest rates and in mortality. The authors found considerable heterogeneity for both outcomes, reflecting the variation in results across the included studies. Only three of the included studies were non-doctor led.

Four additional studies (one stepped-wedge RCT, one controlled and two uncontrolled before-and-after study) not included in the six reviews are some of the largest reported. Two studies evaluated medically led teams,77,80 and two evaluated nurse-led teams. 76,81 All studies also introduced either a NEWS or altered the escalation criteria for an existing scoring system. Both of the uncontrolled before-and-after studies reported a reduction in cardiac arrest rates and one also found a reduction in mortality. In contrast, both the RCT and controlled before-and-after study reported no change in cardiac arrest rates or hospital mortality, which they attributed to underuse of the RRT. 76,77

Track-and-trigger systems

There were three systematic reviews, starting with NICE in 2007,24 which included five studies83,85–88 of introducing aggregate weighted scoring systems (Tables 6 and 7). The one RCT reported a significant reduction in hospital mortality (but in combination with a RRT and continuing education). 83 One small, single-site before-and-after study confirmed a reduction in mortality but without a concurrent RRT,88 another reported no impact on mortality. 86 Only one of the three studies that assessed unplanned admission to critical care found a reduction,86–88 and only one of four studies83,86–88 (a RCT) reported a reduction in LOS. 83

A 2013 review by McNeill and Bryden48 reported decreases in cardiac arrest rates, unplanned ICU admissions and hospital mortality. In addition, they concluded that TTSs improved documentation of vital signs. This was based primarily on a before-and-after study that replaced a single parameter score with an aggregate scoring system plus an education programme. This review found that LOS was unchanged. The latest review, a 2014 study by Alam et al. ,89 was limited to studies of poor methodological quality, with heterogeneous study populations. The impact on cardiac arrest was inconsistent and of six papers assessing mortality, only two reported a reduction. One study reported decreased unplanned ICU admissions but increased HDU admissions.

We identified seven additional studies (five uncontrolled before-and-after studies, one interrupted time-series and one cluster RCT) of which four assessed a form of Modified Early Warning Score (MEWS). 75,90–92 One cluster RCT and three before-and-after studies reported improved recording of one or more vital signs. 75,90–92 De Meester et al. 91 also reported decreased hospital mortality, incidence of reoperation and LOS, but there were no changes in trends for cardiac arrest rates and hospital mortality found in the interrupted time-series study. 78

Less encouraging were the results of a large multinational before-and-after study that examined the introduction of electronic vital signs monitors. 93 The monitors used in every site were the same, but the warning score and escalation were based on each hospital’s pre-existing policy. There was no impact on cardiac arrest rates, frequency of RRT calls or hospital mortality. There was some international variation: US hospitals reported a decrease in LOS and an increase in the proportion of RRT call patients transferred to a greater-acuity ward. This suggested that caution is needed when transferring results between countries. These findings were echoed in the 2016 Australian interrupted time-series study. 78

McDonnell et al. 94 focused on the impact of introduction of a TSS on the knowledge and confidence of nurses. There were improvements in the number of staff concerns, reported levels of experience and knowledge, and staff confidence in recognising deterioration, reporting abnormal observations and in knowing who and when to contact for help. However, these improvements were small and the improvements were greater among non-registered nurses. In addition, confidence in asking senior staff to attend, a key part of the chain of survival, did not increase.

Standardised handover tool

The majority of studies in the four systematic reviews found that handover tools were associated with an improvement in at least one outcome (Tables 8 and 9). 95–98 Examples of improvements included decreases in the number of dropped tasks, the number of omissions, the number of preventable adverse events and LOS. However, there were also a small number of studies that reported an increase in the number of unexpected changes in care and the number of errors attributed to handover. 110,111 The most common benefit in the Robertson et al. 96 review was improved information transfer. Few studies considered outcomes, but four found a decreased LOS and two reported a decrease in adverse events.

In the 11 additional studies identified, three used the SBAR approach: Cornell et al. 106 found no significant difference in the time taken to complete shift reports, although there was an increase in the percentage of time spent on shift report tasks; Moseley et al. 100 found no improvement in 12 out of 16 measures, although medical residents reported an improvement in whether or not all the important data were transmitted during sign-out; and De Meester et al. 102 reported an increase in the number of unplanned ICU admissions and a decrease in hospital deaths (although concurrent educational interventions may have contributed).

Of the other eight studies, five tested a bespoke electronic handover tool. In the only study to use a parallel control group, users were more likely to report inclusion of key patient information and to be confident about the quality of the handover they received. 99 Ahmed et al. 101 also found an improvement in data transfer, the biggest improvement being the recording of senior review. Similarly, Graham et al. 104 reported improvements in the quality of written accounts of five out of six measures. Yazici et al. 103 noted an improvement in five quality markers, but this was sustained to 9 months for only three of them. Gonzalo et al. 108 also found limited benefits after 12 months, with no improvement in eight measures of communication and a worsening in communication of results of completed studies/consults.

The other three studies were of non-electronic handover approaches. One reported improvements in all self-reported measures of knowledge (patient diagnoses, hospital course, active concerns and treatment plans), as well in the overall helpfulness of the handover. 105 Another found that the proportion of missing data decreased for 18 out of the 36 criteria, including fall risk, pain status and home medication. 107 The third reported an 8% decrease in adverse events and a decrease in LOS across an orthopaedic service. 109

Continuing education

The systematic review by Liaw et al. 112 included three before-and-after evaluations113–115 of educational programmes aimed at doctors and nurses (Tables 10 and 11). One reported improved confidence across several measures, including recognising critical illness, keeping patients alive, remembering essential life-saving procedures and seeking out senior staff for help. 113 Of the other two studies, one reported no change in mortality or staff awareness of patients at risk on general wards 5–9 months post intervention,114 whereas the other reported a reduction in unplanned admissions to ICU and in hospital deaths. 115 However, in the latter, the educational intervention was accompanied by greater use of a MET review, more vital signs documentation and a new TTS. Connell et al. 116 included 13 studies evaluating the impact of educational programmes on doctors and nurses and all except one study showed either a change in behaviour or increased knowledge and confidence post training. This review also included the before-and-after study by Fuhrmann et al. 114 showing no change in hospital mortality.

A subsequent interrupted time-series analysis observed reductions in unsatisfactory pain score charting and observations (in medical, but not surgical patients) following an educational intervention. 117 A before-and after study found nurses’ knowledge of failure to rescue events increased after they were exposed to a deteriorating patient simulation scenario with a high-fidelity mannequin. 118 Of the three most recent studies, two were RCTs that evaluated a web-based training programme focused on improving the care of acutely ill patients74,119 and one was a small group teaching session. 73 Liaw et al. 74 looked at the impact of a web-based tool on qualified nurses in Singapore and found changes in knowledge and increased monitoring of both heart rate and respiratory rate. Johnston et al. 73 found improved assessment, communication and non-technical skills, along with improved identification of errors.

Main findings

Of the four interventions, the strongest evidence for impact on IHCA rates and survival exists for RRTs in which studies tended to have larger sample sizes and be focused on a smaller number of patient outcomes. There is evidence that RRTs reduce cardiac arrest rates and, to a lesser extent, mortality. The evidence was weaker for nurse-led teams. The introduction of TTSs has also been shown to be effective in improving the recording of one or more vital signs, although the sustainability of the impact is unclear, as is the impact on patient outcomes.

Studies evaluating standardised handover tools, which have been weaker methodologically, have mostly reported improvements, in particular in information transfer. Whether or not this translates into better patient outcomes is not known; despite the majority of studies showing a positive benefit, most of the studies were of poor quality and were small, uncontrolled before-and-after studies. 95–98,100–109

Continuing education is the most challenging of the four interventions to assess in terms of the quality and quantity of evidence available. There is some evidence of improved knowledge and self-reported confidence in a range of skills, as well as in process measures, such as frequency of observations, but the evidence for improved patient outcomes is weak.

Limitations

As other reviews of these topics have found,24,47,48,56,79,84,89,95–98,112,116 there is a mixed picture with regard to study results in all four interventions. There are a number of possible explanations for this. First, there is considerable diversity in the details of the interventions being employed within each category: RRTs varied in their calling criteria, composition and the hours they covered; TTSs varied in the parameters they included, the thresholds for activation and the responses they triggered; continuing education interventions had different curricula and teaching methods; and structured handover tools varied in the information being conveyed.

Second, in many studies the use of concurrent interventions creates a challenge; for instance, implementation of RRTs was associated with the introduction of new TTSs. Such an approach is pragmatic from the perspective of service providers as it tackles several aspects of the chain of survival at once; however, it’s difficult to attribute impact to any one of the interventions as each may have contributed all or nothing to the result.

Third, there was variation in the way interventions were implemented. For instance, Simmes et al. 120 reported that their MET was not consulted prior to a serious adverse event in 50% of cases, even though abnormal vital signs were observed and should have been activated. A study of the same team would have returned different results had it been consulted 100% of the time. Many factors may have modulated the implementation of the interventions. Some studies had educational packages to accompany the roll-out of their RRTs or TTSs, while others made no mention of this. In addition, the enthusiasm and teaching aptitude of the trainers in the educational interventions will likely have been key to their success or failure. Participants will have also influenced implementation; for example, 50% of the learners in Schubert’s118 study had just finished a night shift when they started the course, which could easily have affected their motivation and capacity to learn. There are also broader contextual factors at play, such as the support of hospital managers for the intervention or the financial support for its roll-out. Another potential moderating factor is the country the intervention was set in. Most of the studies took place in Australia, the UK and the USA, which have different health-care systems with different staffing, working hours and access to ICU beds. Evidence to support the idea that the country of use could be introducing variability in results comes from Bellomo et al. ’s93 study of electronic TTSs in Australia, the USA and Europe. They found that the American hospitals produced significantly different results on several key outcomes, including hospital LOS and the percentage of patients transferred to a greater-acuity ward following the RRT being called.

Fourth, the majority of studies used an uncontrolled before-and-after design. There is, therefore, the risk of confounding affecting the results. The majority of these studies provided no reassurance that confounding had been adequately taken into account.

It is also possible that our own methodological approach, such as restricting searches to just two databases (MEDLINE and EMBASE), may have led to the exclusion of some relevant research studies. However, the impact was mitigated by using the most recent rigorous published systematic literature review as a foundation for our review across three out of the four interventions.

Implications

Although the existing literature provides a fairly consistent picture of the benefits of each of the four types of intervention, larger studies conducted over a longer period of time, to allow the intervention to become established and to assess sustainability, are needed. A core set of outcomes based on the chain of survival should be agreed and applied to improve interstudy comparability. There is also a need to define precisely what intervention is being evaluated to facilitate meaningful pooling of the evidence. For instance, RRTs could be described in terms of four dimensions: (1) team makeup, (2) hours of operation, (3) services provided and (4) ICU admission rights. This would reduce the problem of variability when synthesising results from different studies and would allow meaningful comparisons and conclusions to be drawn. Finally, it is unlikely that any single intervention would perform in the same way when put in place in hospitals with different cultures or resources for the intervention’s roll-out. This interplay between context and the success or failure of these interventions has not been addressed sufficiently. Given the heterogeneity of the interventions and the contexts in which they were implemented, plus the fact that many studies did not provide an adequate level of detail of either the intervention or the context, it was not possible to create a typology based on this literature alone. This required qualitative work within NHS hospitals, the findings from which are presented in the following chapter.

Study setting

The semistructured interviews were conducted in NHS acute hospital trusts.

Sample

Purposive sampling was used to identify a range of hospitals of different size, location, teaching status and foundation versus non-foundation status in order to maximise the likelihood of understanding the full range of interventions that have been used and the ways in which they have (or have not) been implemented. The principal investigator contacted 13 trusts across England, who SG members knew to have a diverse range of services, and invited them to take part in the study (Table 12). A senior clinician at each trust was asked to identify up to five staff as key personnel involved in deteriorating patient care. Potential interviewees were sought at different levels throughout the trust, encompassing strategic-level staff (e.g. medical directors), staff involved in planning patient care (e.g. consultants in the ICU), members of outreach teams and ward-based clinical staff (junior doctors and nurses) involved in day-to-day patient care and monitoring.

After securing NHS R&D approval, those members of staff identified were approached via e-mail to take part in an interview. All interviews were conducted by members of the research team (CC, RZ and HH), the majority being face to face and 15% being via telephone. Prior to interviews, written consent was obtained after fully informing participants of the objectives of the study and providing an information sheet. Interviewees were assured that neither their name nor hospital would be recognisable in the presentation of any analyses. In total, 60 people agreed to be interviewed (Table 13). Interviews lasted between 20 and 90 minutes, depending on staff time available, and were recorded as detailed field notes with accompanying audio-recording. Interviews followed a topic guide developed before and during pilot interviews.

Topic guide

The topic guide was developed based on the findings from the literature review, as well as on issues related to implementation identified from descriptive papers in the field, recommendations in guidelines and the expertise of the Project Management Group (PMG) and other contacts. Carroll et al. ’s65 framework for implementation fidelity guided the design of questions focused on how far actual interventions adhered to their intended designs. Box 1 shows an example of how the guide was applied in practice to one such intervention: the outreach team.

The interviews principally assessed service configurations for TTSs (e.g. the NEWS), outreach teams, handover procedures and practices, education provision and any contextual factors such as leadership and management, organisational culture, staff engagement and communication, strategic goals and resourcing, monitoring, and evaluation affecting those services. In addition, other important moderating factors, such as changes in end-of-life care practices, (e.g. use of DNACPRs) were covered. The topic guides for interviews with outreach team staff and strategic-level staff are found in Report Supplementary Material 9 and Report Supplementary Material 10. Questions also evolved based on topics raised through the course of the interviews. The questions asked varied by interviewee depending on their area of expertise; for example, strategic staff were more likely to know about quality improvement incentives, such as the Commissioning for Quality and Innovation (CQUIN) payments scheme.

Data analysis

All interviews were typed up as in-depth field notes. An initial thematic analysis was undertaken to group the data by broad themes, concepts and emergent categories. 121 A further in-depth thematic analysis of the data was undertaken to classify the findings into pragmatic categories, which were used to produce the main findings, after discussion with key members of the research team.

Deteriorating patients in context

‘Deteriorating patients’ were characterised as patients with worsening physiological measures (e.g. blood pressure, heart rate, temperature or respiratory rate) who were in need of escalation to a higher level of care to prevent adverse outcomes. Generally, interviewees reported that deterioration is identified through observations of physiological measures and their appropriate interpretation. In the last decade, much work has been undertaken to improve the care of acutely ill patients in hospital in order to prevent deterioration and adverse outcomes. Interviewees were familiar with local and national initiatives aimed at improving the care of hospitalised patients, such as the implementation of a NEWS, funding for outreach teams/specialist nurses, sepsis action plans or CQUIN payments. Some interviewees stated that a great deal of work had been carried out in the preceding 5 years and improvements in care had been observed, but acknowledged that more work is needed in the future.

Service configurations

Other moderating factors

Strengths and limitations

The study had a large sample size and included staff with a range of professional roles who were closely associated with delivering care to acutely ill patients. We were therefore able to ensure that we captured information on how services for deteriorating patients varied within and across trusts and to gather as wide a range of views on contextual facilitators and barriers as possible. The hospital sites were also varied in terms of geographical location, size and teaching status.

The study has several limitations. Interview data were primarily collected to inform survey development rather than as an in-depth qualitative study, with interviews typed up as field notes as opposed to being transcribed verbatim. As interviewees were informed about the purpose of the interviews this may have led to an element of social desirability bias in some responses, but generally we found that staff were more than willing to tell us about the problems they faced in implementing services. Only one researcher, who had not collected the original data, undertook the analysis. As a consequence, there is a risk that interpretation of these data does not wholly capture the meaning intended by the interviewees. However, the findings were circulated to experts in the area for external validation. Furthermore, the need to cover a number of different interventions and contextual factors that might affect their effectiveness did not allow in-depth questioning to better establish deeper beliefs and attitudes. In addition, any references made to changes over time will be subject to the accuracy of recall.

Implications

Given that our literature review did not provide a clear steer as regards the composition of effective interventions to detect and respond to patient deterioration, semistructured interviews with hospital staff were required to explore how these interventions have developed on the ground. We found that the majority of hospitals had adopted the NEWS. Such standardisation has the potential to improve outcomes as staff across the NHS become familiar with its use. However, key contextual factors that may prevent attainment of these outcomes were reported by staff, including the delegation of observations to less-qualified nursing staff and the modification of the tool to address concerns about its sensitivity in certain subgroups of patients. Outreach teams are a much less standardised intervention, with different levels of skill across the team, hours of coverage and autonomy. Their ability to work system-wide on improving the care of acutely ill patients through educating ward staff to care for such patients and ensuring correct use of tools, such as the NEWS, is limited by increasing hands-on clinical demands and, in some cases, under-resourcing. Handover tools and formal educational courses covering acute care are also becoming increasingly standardised. Most hospitals report promoting the SBAR tool and have ALERT™ or similar courses. However, neither of these things is a guarantee that improved communication or clinical skills are seen in practice. The effectiveness of these interventions will also depend on how long they have been running, the degree of educational support and promotion of staff engagement, and ongoing quality assurance. Members of the outreach team are often charged with these duties but may face challenges in finding the time to undertake them. Fulfilment of this role is more likely if there is board-level prioritisation and support.

Although these interventions have been implemented, other changes in the system have occurred that modify their impact. The increasing emphasis on improved end-of-life planning and protection of patients from interventions that may be of little benefit to them, such as resuscitation, has resulted in improved awareness of the need for timely DNACPR decisions. Outreach teams have gradually become more involved in this decision-making. Removing substantial numbers of patients who are likely to have poor outcomes from CPR might be expected to reduce cardiac arrest rates and improve survival.

Our analysis identified key intervention components to explore further in the survey. These elements included the range of TTSs in use, their spread across wards, the main groups using them and how well they were completed; the composition of outreach teams, their roles and autonomy, their involvement in education and the impact of work pressures; the types of handover tools in use and approaches to promoting use; and contextual factors affecting both interventions and outcomes, such as the implementation of DNACPR decisions. Our findings indicated that formal education courses did not vary substantially across hospitals; therefore, no candidate questions on these interventions were put forward for the survey.

Pilot survey

An initial consultation with staff at the ICNARC was undertaken to determine the design features of previous surveys that they felt had resulted in a good response rate. Examples of other surveys that investigated NHS service configurations were also reviewed.

The semistructured interviews reported in Chapter 3 helped hone questions related to each intervention, serving to clarify how they have been implemented (e.g. structure, staffing, coverage, etc.). Contextual questions critical to the delivery of services were based on responses from interviewees as to which factors were most important in the delivery of their services such as any governance arrangements, communication processes, resources, etc. A list of candidate questions was collated and circulated to the PMG and SG for initial comment. Subsequently, each question was analysed by the PMG according to a range of criteria, including the amount of likely variation between hospitals, overlap with similar questions, the likelihood that answers would be available from respondents and the extent to which questions capture the intensity of service provision. Redundant questions were discarded. The survey was piloted with six clinical staff in different hospitals to test face and content validity after which revisions were made to ensure any ambiguity was removed from the questions and that instructions were clear. Questions related to formal educational courses were dropped as they did not provide enough variation in response between respondents.

Sampling frame

The NCAA staff sent the questionnaire to their lead clinical contact (usually the lead resuscitation officer) at each participating acute general hospital (excluding specialty hospitals). These contacts were asked to complete all questions within their remit and then forward the survey to another member of staff best placed to complete other questions. We suggested that this would probably be the lead for the outreach team.

Survey design

The survey comprised mainly closed questions, but space was left for comments that allowed respondents to provide textual clarifications. It was divided into sections covering the following areas:

• detection of deteriorating patients – type of TTS in use, ward coverage and compliance and staff groups responsible for taking observations

• response to deteriorating patients (focused on RRTs that have designated team members who respond to deterioration on the ward and not teams that respond to emergency/2222/cardiac arrest calls. The generic term RRT was used in this section instead of outreach team because we wished to capture all teams fulfilling this role if a hospital had more than one and because some hospitals call their outreach teams by different names) – RRT composition, roles, hours of coverage, calling frequency of various staff groups and level of demand on the service

• ceilings of care – use of TEPs and DNACPR

• handover – handover tools in use and methods to promote use

• contextual factors – use of overspill wards and agency staff

• data collected for NCAA – estimates of how many cardiac arrests were submitted to the audit and interventions to improve case ascertainment.

It was created in SurveyMonkey (SurveyMonkey Inc., Palo Alto, CA, USA) to allow electronic data entry but a Microsoft Word (Microsoft Corporation, Redmond, WA, USA) version was also available on request (see Report Supplementary Material 11). A survey guide (see Report Supplementary Material 12) was sent out with the initial mailing and this contained an e-mail and telephone number should respondents have queries not addressed by the guide.

Data collection

A personalised e-mail was sent to each NCAA contact containing the link to the questionnaire on SurveyMonkey accompanied by the guide and the date of the first deadline, which was in 4 weeks’ time. For each of the first 4 weeks, those that had completed the survey were entered into a prize draw for a £50 gift token. Follow-up e-mails were sent weekly to non-respondents for the first 4 weeks up to the deadline and then weekly for a further month. One reminder e-mail came with an endorsement letter from the chair of the Resuscitation Council. From this point on, non-respondents were followed up individually by e-mail and telephone. Messages advertising the survey were circulated in the National Outreach Forum Newsletter and through the personal contacts of the SG. If no contact could be made with a hospital, an e-mail was sent to the medical director asking for help with facilitating a response.

Using best practice in maximising responses, we expected to achieve a response rate of 80%. This target response rate was based on experience from surveys administered by ICNARC in two of their earlier National Institute for Health Research-funded studies. The Critical Care Outreach Services survey (SDO/74/2004)150 achieved a response rate of 80% and the Critical Care Modernisation survey (SDO 08/1604/133)3 achieved a response rate of 84%. These high rates reflected the long-standing relationship between ICNARC and the hospitals participating in its various audit programmes.

To increase the validity of questionnaire responses, follow-up telephone calls were made to seek clarification if responses were unclear.

Data analysis

Survey responses were exported from SurveyMonkey to a Microsoft Excel^® 2016 spreadsheet (Microsoft Corporation, Redmond, WA, USA) for checking and cleaning. Data were then imported into Stata^® version14 (StrataCorp LP, College Station, TX, USA). Continuous variables were categorised and reported as percentages. Differences in hospitals’ current provision of care were illustrated using bar and pie charts. Changes in hospitals’ historic provision of TTSs and outreach were plotted graphically by quarter to illustrate trends in these variables between 2009 and 2015.

Response rate

The survey was sent to 171 acute general hospitals taking part in NCAA and 139 responded (81.2%). The non-respondents did not differ from the respondents in terms of location and length of time in NCAA. There was a low response to the questions about the use of overflow wards and agency staff, and those that did respond indicated that their answers were their own estimates. It was decided to drop these questions from the analysis.

Characteristics of hospital interventions

Track-and-trigger systems

The majority of NCAA hospitals operated a paper-based TTS in 2009 (94.1%) (Figure 1). These were a range of systems including the MEWS and a hospital’s own scoring systems. The NEWS was introduced in late 2012. Following its introduction there was a decline in the use of other paper-based systems and by the end of 2015, 48.7% of hospitals indicated use of a paper-based NEWS.

Only a handful of hospitals had an electronic system in 2009 (1.7%), but this had grown to 29.4% in 2015; the majority of these systems are also based on the NEWS (67.5%).

Over the index period (2009–15), 21.8% of hospitals made no changes in their TTSs, 58.8% made one change, 18.5% made two changes and 0.8% made three changes. A total of 11.8% of hospitals reported operating dual systems at some point over this time period.

FIGURE 1.

Changes over time in TTSs (2009–15). q, quarter.

Compliance, a composite variable comprising full and accurate completion of observations on the TTS and escalation of patients in line with local policy, has shown a gradual increase in both reporting within hospitals (from 65.5% to 85.7%) and the proportion of hospitals reporting compliance rates of > 90% (from 15.1% to 41.2%), with an accompanying fall off in rates of compliance below the level of 75% (from 18.5% to 9.2%) (Figure 2).

FIGURE 2.

Compliance with TTSs between 2009 and 2015. q, quarter.

Hospitals were asked to provide a breakdown by estimated percentage of which professional groupings (registered nurse, student nurse, HCA) predominantly take observations on the wards and complete the TTS. The predominant group was the one that took at least 10% more of the observations than the next highest group. The mixed category was allocated if the difference in proportions taking observations was < 10%. Figure 3 shows the predominant staff group completing the TTS across responding hospitals.

FIGURE 3.

Dominant professional groups taking observations in 2015.

Outreach teams

Over the 5-year study period, there was a trend towards hospitals setting up outreach teams and increasing the hours that the team covered (Figure 4). The number of hospitals without outreach fell from 26.1% to 17.6%; those providing only a Monday to Friday service fell from 16.8% to 5.0%. There was a slight increase in teams that covered 7 days per week, but daytime shifts only, from 32.8% to 38.7%, but the largest increase was in the proportion of teams that were offering a 24/7 service, which increased from 24.4% to 38.7% over the study period.

FIGURE 4.

Outreach team coverage from 2009 to 2015. q, quarter.

Outreach teams were typically composed of a mixture of grade 6 and 7 nurses, and 43.7% of teams included a grade 8 nurse. Only 4% of outreach teams had doctors with protected clinical time dedicated to the team. A total of 20% had input from a consultant, usually from the ICU, who mainly had an administrative and training role for one or two sessions per week. The most common additional team members in the 16.6% of outreach teams that included other health professionals were physiotherapists.

Nearly all of the teams provided step-down care for patients who had recently been transferred from the ICU to the wards (97.3%), as well as undertaking assessment of deteriorating patients on the ward, which included taking arterial blood gases (95.5%) (Figure 5). Most teams also played a role in end-of-life decision-making and the consideration of DNACPR decisions (92%). Two-thirds of teams reported involvement in the monitoring of patients for acute kidney injury (64.9%). Just under half of teams had capacity for independent prescribing on the team (48.6%). A total of 51.9% of teams had some protected time for teaching or quality improvement hours timetabled into their monthly schedule; 14.8% had > 10 hours.

FIGURE 5.

Roles of outreach teams in 2015. AKI, acute kidney injury.

Liaising with an ICU consultant (44.3%) or the patient’s own consultant (31%) were the main referral routes that teams used when arranging an admission to the ICU (Figure 6). A minority of teams (7.3%) could make direct referrals.

FIGURE 6.

Primary referral routes into the ICU used by outreach teams in 2015.

Nearly all of the teams reported that the staff who made the most requests to outreach were registered nurses (99%) and junior doctors (77%). Half of the teams also identified other doctors, including consultants, as frequent callers (45.7%) (Figure 7).

FIGURE 7.

Staff members who made calls to the outreach team in 2015.

When outreach teams were asked whether or not demand regularly exceeds supply across the week, as an assessment of strain on the system, nearly two-thirds responded that this was the case for both weekday (60%) and weekend (58%) shifts (Figure 8). A total of 57.6% of hospitals with outreach teams said that normal service was suspended on ≥ 14 days across the previous year, and that this was usually because of inadequate staffing levels.

FIGURE 8.

Reporting of demand regularly exceeding supply across the week in 2015.

Handover systems

A total of 87.7% of hospitals reported using SBAR as their main standardised communication tool. A range of approaches to promote and encourage its use were reported, the most popular being training courses (96.1%), mention at staff induction (84.7%) and integration into local policy (77.7%) (Figure 9).

FIGURE 9.

Approaches used by hospitals to promote the use of the SBAR communication tool in 2015.

Contextual factors

Two-thirds of hospitals reported reviewing the case records of > 80% of patients who experienced an IHCA in an effort to learn from the care that these patients had received.

Efforts to ensure that patients who are unlikely to benefit from CPR do not under go CPR still appear to have some way to go, with reports of only half of those eligible patients actually having a plan in place that outlines appropriate levels of intervention in the case of deterioration (TEPs) (Figure 10). Moreover, half of the survey respondents estimated that > 50% of their patients who experience an IHCA should have had a DNACPR decision in place before the arrest occurred (Figure 11).

FIGURE 10.

An estimate of the percentage of patients eligible for a TEP who actually had one in place in 2015.

FIGURE 11.

An estimate of the percentage of patients experiencing an IHCA who should have had a DNACPR decision in place in 2015.

Strengths and limitations

The number of hospitals in the NCAA has been growing steadily since its inception, and around 70% of English NHS acute hospitals were participating in the audit in 2015 when we conducted our survey; the response rate of 81% suggests that our findings are representative of English acute hospitals. There are a number of limitations that need to be considered. We were unable to determine if survey respondents differed from non-respondents for any characteristics other than geographical location and the length of time they had participated in the NCAA. Non-responding hospitals may have differed systematically with regard to the intensity of their interventions. The survey was limited in size to promote a good response rate. It was therefore not possible to explore all aspects of the interventions and the contextual factors that have influenced their implementation. For example, we did not collect data on the ‘dose’ of outreach teams, a factor that may be an important element of team effectiveness. 47,156 We also did not gather information on other differences in service configuration that may determine effectiveness, for instance what response systems were being used in hospitals that reported having no outreach team. This limitation might have had a greater impact on findings for outreach teams than those for TTSs. Moreover, respondents found some of the contextual questions hard to answer because of a lack of direct access to the required information. This led to more missing data for these questions. We required staff to report on changes across a 5-year period, which opens up the possibility of recall bias. For organisations with significant staff turnover, some respondents may not have been working at the hospital when changes were initiated. This bias was potentially reinforced by our limiting of respondents to two staff, neither of whom were front line or ward based, who may have held different perspectives.

Data sources

Five data sources were used for the analyses:

1. The organisational survey of NCAA hospitals – data on the systems that hospitals have in place to identify and care for deteriorating patients as described in Chapter 4. Data from this source are linked to other data sources by hospital and by time period (quarter).

2. The NCAA data on IHCAs – data on IHCAs collected in the NCAA and including dates, times and locations of IHCAs, sociodemographic data and outcomes. Data are recorded for each arrest and are linked by hospital, patient and time period to other data sources.

3. Hospital Episode Statistics episodes for patients in the NCAA – data describing episodes of admitted patient care for patients in the NCAA were provided by NHS Digital. 157 Linkage of NCAA and HES data was undertaken by NHS Digital based on personal data provided by the NCAA. Unique encrypted patient identifiers for successfully matched patients were provided to allow patient-level linkage to other data sources. An additional year of data prior to the start of the NCAA was requested to allow measurement of comorbidity using a modified version of the Charlson Comorbidity Index. 158 Linkage to other data sources was by hospital, time period and patient.

4. Hospital Episode Statistics for patients admitted to hospitals participating in the NCAA – a second, larger data set of HES episodes of admitted patient care was provided by NHS Digital157 to allow the number of hospital admissions by quarter to be calculated for each NCAA hospital. Owing to the huge volume of admissions and episodes since 2008/09, these data were restricted to episodes for ordinary admissions to NHS trusts with hospitals participating in the NCAA. These data therefore excluded certain types of admissions such as day cases, regular attenders and some maternity admissions. An additional prior year of data for each trust was provided to allow measurement of comorbidity. Linkage to other data sources was by hospital, time period and patient.

5. Office for National Statistics mortality data159 – these data were provided by NHS Digital to allow measurement of 30- and 90-day mortality following IHCAs. Data were linked to other data sources by patient.

Data from the NCAA and the smaller HES data set (NCAA patients only) were linked to identify the HES episode that included the IHCA and, if different, the first episode of the hospital admission that includes the episode with the IHCA. The location and time of day of the arrest was used to determine if IHCAs occurring on the date of admission could be treated as post admission. Any IHCAs located in emergency departments were not linked. Some admissions with IHCAs involve transfers between hospitals in the same NHS trust; these admissions were attributed to the hospital where the IHCA took place. If an admission involved multiple IHCAs in different hospitals, then the admission was split to create a separate ‘admission’ for each hospital.

Inclusion criteria and data linkage

The eligibility of hospitals was first determined by identifying those hospitals that had completed the survey and those hospitals participating in the NCAA. A hospital was deemed to have completed the survey if responses were provided for a minimum of four consecutive quarters for three measures. The measures were the hours covered by the outreach team (no team, Monday to Friday days only, Monday to Sunday days only or full 24/7 coverage), the use of TTSs (no system, NEWS/NEWS-based) and the type of TTS (paper-based or electronic). A hospital was eligible if a minimum of three consecutive quarters of participation in NCAA coincided with complete survey data for those quarters. Participation in the NCAA was deemed to start from the first complete quarter of data collection.

Eligible hospitals were linked to data on hospital admissions from HES in order to identify all the ordinary admissions by hospital and by quarter and to derive the denominator for estimating IHCAs. Linking by hospital involved identifying hospitals from five character provider codes and site of treatment in HES. Relevant codes were identified from the linked NCAA and HES episode data and the database of NHS organisation codes. The proportion of missing or generic codes were reviewed for each hospital; NHS trusts or time periods where admissions could not be reliably attributed to a NCAA hospital were excluded. The number of admissions by quarter for each hospital was derived using the episode end date of the first episode in the admission. These were reviewed and explanations sought (HES data quality notes, hospital restructuring) for substantial changes in order to identify and exclude periods of under-reported hospital activity.

An additional criterion for including hospitals or specific time periods for hospitals was a requirement that there should be a minimum of two eligible quarters of data before and after any quarter that includes a change in one of the variables of interest, for example when a hospital switches from a paper-based to an electronic TTS. Quarters were dropped until this criterion was met.

The application of the above criteria identified the hospitals and time periods that were eligible for analysis and created a data set of survey variables and individual hospital admissions linked by hospital and by quarter. Data on IHCAs were then linked to hospital admissions using the unique episode key for the first episode of an admission in the previously linked NCAA and HES data. IHCAs were ineligible for linkage and inclusion if the IHCA (or the first IHCA in a series) occurred in the emergency department, the IHCA was a visitor or staff member rather than a patient at the time of arrest or no matching ordinary hospital admission was identified. Multiple IHCAs within an admission were dealt with by retaining NCAA case mix and outcome data for the first IHCA and creating a new variable for the number of IHCAs for each hospital admission.

Identification of variables for case mix and outcomes

Case mix for all hospital admissions included sociodemographic and clinical variables from HES. Sociodemographic characteristics comprised age group (in 10-year age bands), sex, ethnicity (white, black/black British, Asian/Asian British, any other and missing/not reported) and deprivation decile group using the Index of Multiple Deprivation. 160 Other case mix variables comprised emergency/non-emergency admission and main reason for diagnosis (circulatory, respiratory, pregnancy-related or other). Comorbidity was also captured using a modified Charlson Comorbidity Index score derived from the index admission and prior hospital admissions. A hospital-level variables, based on the proportion of admissions with a main diagnosis of atherosclerotic heart disease (International Classification of Diseases, Tenth Revision, code I259),161 was used to capture the extent of specialisation of cardiac services among individual NHS trusts and hospitals.

Case mix for analysis of survival of IHCAs also used variables collected in the NCAA. These were the length of hospital stay prior to the 2222 call, the location of the cardiac arrest, the presenting or first documented rhythm and the reason for admission to hospital. In addition, case mix used the interactions between the location of the cardiac arrest and the presenting rhythm.

The outcome for the analysis of IHCA rates was the count of eligible IHCAs occurring during the hospital admission. Survival for all hospital admissions was defined as being discharged alive from that hospital admission. Survival following IHCA was measured in three ways. Survival for 20 minutes was defined as the return of a spontaneous circulation that was sustained for > 20 minutes in a patient who was attended by the hospital-based resuscitation team (or equivalent) and who received chest compression(s) and/or defibrillation in response to the 2222 call. Survival from the date of IHCA to 30 and 90 days was identified using date of death from ONS mortality data and last known date alive was identified using HES admissions data.

Derivation of intervention variables

The organisational survey collected quarterly data on TTSs and outreach from 2009 to 2015 to capture changes in provision within hospitals as well as differences between hospitals. Three primary variables were derived from these data:

1. Use of NEWS/NEWS-based TTSs or non-NEWS TTSs – all hospitals reported using some form of TTS. Any TTS defined as NEWS, based on NEWS or locally modified NEWS was included as a NEWS/NEWS-based TTS. All other TTSs were included as non-NEWS TTSs.

2. Paper TTSs or electronic TTSs – all hospitals indicated if their TTS was paper-based or electronic. Hospitals that used both paper and electronic TTSs, for example, during a period of transition to an electronic TTS, were categorised according to the predominant system in the relevant quarter.

3. No outreach team, non-24/7 outreach team or 24/7 outreach team – the survey of hospitals captured historic information on whether or not a hospital had an outreach team and if it operated during the week, at the weekend and during the night. Few hospitals had an outreach team that did not operate at the weekend so three categories of outreach were defined. An outreach team was categorised as 24/7 if it operated on days and nights for 7 days a week. All other outreach teams were categorised as non-24/7 RRTs.

Three secondary variables describing a hospital’s current (at the time of survey in 2015) systems were derived from survey data. These variables do not vary with time, and so are able to capture only those differences between hospitals:

1. Outreach team features: this variable was scored between 0 and 5 based on positive responses to the following – (1) the outreach team included at least three core functions from ICU/HDU step-down care, independent non-medical prescribing, obtaining and interpreting arterial blood gases, and initiating DNACPR discussions with the clinical team and responding to acute kidney injury alerts, (2) the outreach team made direct admission to the ICU, (3) no temporary outreach team suspension in the last year, (4) demand for outreach was not regularly exceeded and (5) doctor or minimum of 10% senior nurse membership of the outreach team. A score of zero was recorded if there was no outreach team.

2. Handover system features: this variable was scored between 0 and 8 based on positive responses to the following – (1) staff induction explicitly advised on handover tool use, (2) posters on wards to encourage use, (3) use is mandatory for calls to the outreach team, (4) tool reminder notices are adjacent to telephones, (5) a written local policy advises on use, (6) use of tool is explicitly mentioned on patient observation charts, (7) stickers exist for use in patients’ notes and (8) staff training courses advise on use.

3. Contextual factors: this variable was scored between 0 and 6 based on positive responses to the following – (1) formal written TEPs are used, (2) at least 50% of patients eligible for a TEP had one in place, (3) IHCAs are subject to case record review, (4) at least 90% of case records are reviewed, (5) < 50% of IHCAs were judged as unsuitable for resuscitation after the event (should have had a DNACPR in place) and (6) < 20% of IHCAs were judged as unsuitable for resuscitation after the event (should have had a DNACPR in place).

Analysis

Descriptive analysis was used to show the distribution and changes in the provision of TTSs and outreach teams in hospitals included in the analysis, and the clinical and sociodemographic characteristics of hospital admissions and patients who have an IHCA. Annual trends in IHCA rates and survival were estimated using Poisson regression and logistic regression, respectively, with hospitals included as random effects and adjustment for case mix and seasonality.

Poisson regression was used to evaluate the impact of interventions on rates of IHCAs. The outcome was the count of IHCAs in the hospital admission. The majority of patients had no IHCA, whereas some had multiple IHCAs during their hospital stay. Hospitals were fitted as random effects. All models were adjusted for case mix from HES.

The analysis was conducted in four stages: (1) examining individual interventions in separate case mix-adjusted analyses of all hospitals, (2) examining a combination in interventions in a case mix-adjusted analysis of all hospitals; (3) examining individual and combined interventions in a case mix-adjusted analysis restricted to those hospitals that reported a change in the intervention during the study and (4) examining the impact of the three secondary variables.

In the first stage of the analysis, each intervention was evaluated individually after adjusting for case mix, temporal trend and seasonality. Two alternative ways of modelling the impact of interventions were specified: (1) a difference in levels and (2) a difference in slopes. 162 The intervention was included as a dummy variable to test for a difference in levels. Intervention–time (quarter) interactions were used to test for a difference in slopes.

Figure 12 shows how the two approaches (i.e. a difference in levels or in slopes) are modelled in hospitals that have a change an intervention during the study (e.g. changing from a non-NEWS to NEWS TTS). In these hospitals, the approaches are, in essence, interrupted time-series. A difference in levels models an immediate change in outcome following a change in an intervention. A difference in slopes models a change in an outcome’s trend over time following a change in an intervention and is the same as fitting a linear spline.

FIGURE 12.

Illustration of alternative approaches to modelling the impact of an intervention in a hospital reporting a change in the intervention. (a) Difference in level; and (b) difference in slope.

Figure 13 shows how the two approaches are modelled in hospitals that have no change in the intervention during the study (e.g. hospitals that only had a NEWS TTS or only had a non-NEWS TTS). A difference in levels models the average difference in outcome between hospitals with a NEWS TTS compared to those with a non-NEWS TTS. A difference in slopes models the average trend or slope in the outcome in hospitals with a NEWS TTS compared with the average slope in hospitals with a non-NEWS TTS.

FIGURE 13.

Illustration of alternative approaches to modelling the impact of an intervention in hospitals reporting no change in the intervention. (a) Difference in level; and (b) difference in slope.

An analysis of all 106 hospitals using either a difference-in-levels approach or a difference-in-slopes approach will therefore combine the effects of within-hospital differences (as shown in the relevant graph in Figure 12) and between-hospital differences (as shown in Figure 13) in the intervention. Thus, six separate models were used to estimate the individual impact of the three interventions (NEWS/non-NEWS TTS, paper/electronic TTS, and outreach team) using the two approaches (difference in levels, difference in slopes). To examine a difference in levels, the incidence rate ratios (IRRs) for the difference in levels were estimated with 95% confidence intervals (CIs). To examine a difference in slopes, the IRRs and 95% CIs were estimated for the slope for each category of an intervention (e.g. for paper TTS and for electronic TTS). IRRs and CIs were annualised to show the average yearly change in rates, and p-values were calculated to test for evidence for a difference in slopes. All models were adjusted for case mix, quarterly trend and seasonality, with hospitals included as random effects.

The second stage of the analysis involved selecting interventions and approaches for inclusion in a model that combined different interventions. Inclusion required evidence for an association between intervention and outcome at a p-value of < 0.1 from the first stage of the analysis. IRRs and 95% CIs were estimated after value of adjusting for other interventions included in the model, case mix, quarterly trend and seasonality, with hospitals included as random effects.

The third stage was to repeat the first two stages, after restricting the hospitals in the analyses to those that reported a change in intervention during the study, for example a hospital that switched from a paper to an electronic TTS. This was done to provide some assurance that any impact of an intervention observed in an analysis of all hospitals was also observed if restricted to those hospitals that reported a change in that intervention. The final stage of analysis involved examining the impact of the three secondary variables (outreach features, handover features and contextual factors) on IHCA rates. These were examined individually in case mix-adjusted models, with IRRs and 95% CIs estimated for a 1-point increase in the count. All analyses were repeated after restricting IHCAs to those occurring on wards. This was done to examine the sensitivity of the results using rates of ward-based IHCAs as the outcome, rather than all IHCAs in admitted patients attended by the resuscitation team regardless of the location of the arrest.

Poisson regression was also used to examine the impact of interventions on hospital survival for all hospital admission. An identical four-stage modelling approach was used, with adjustment for case mix, temporal trends and seasonality. Hospitals were modelled as random effects.

Logistic regression was used to examine the impact of interventions on four measures of IHCA survival: (1) return of a spontaneous circulation of > 20 minutes, (2) hospital survival (survival to discharge), (3) 30-day survival and (4) 90-day survival. Again, the four-stage modelling approach was used. Case mix also included variables from NCAA: the length of hospital stay prior to the 2222 call, the location of the cardiac arrest, the presenting or first documented rhythm and the reason for admission to hospital. In addition, case mix used the interactions between the location of the cardiac arrest and the presenting rhythm. Hospitals were included as random effects and the impact of interventions was measured as odds ratios (ORs) with 95% CIs.

Descriptive analysis

There were 106 hospitals eligible for analysis. The number of quarters of eligible data from the hospitals ranged from 3 to 22 between the final (October–December) quarter of 2009 and the first (January–March) quarter of 2015. The data for analysis comprised 13,059,865 hospital admissions from HES linked with 32,364 patients having 34,202 IHCAs from the NCAA database.

Table 14 shows the characteristics of all patients admitted to 106 hospitals included in the analysis, and the characteristics of those patients who had an IHCA. An IHCA was more likely in older male patients who have greater comorbidity and who entered hospital as an emergency admission. The mean age of IHCAs changed little over time, whereas there was a small increase in the mean age of all admissions. Trends in case mix of IHCAs over time indicated a small reduction in the proportion of females and the proportion of emergency admission compared with trends in all admissions.

The majority of the 32,820 IHCAs occurred on wards (21,595; 65.8%). Other locations included CCUs (10.3%), emergency admissions units (10.0%), critical care units (4.7%) and cardiac catheter laboratories (2.6%).

Figure 14 shows a general downward trend in the rate of IHCAs over time, with evidence of seasonality (higher rates in October to December and January to March quarters). Table 15 indicates a reduction of 6.4% per year once case mix and seasonality are allowed for. Note that some fluctuation in earlier quarters is attributable to there being fewer data and fewer hospitals from this time period.

FIGURE 14.

Trends in the crude rate of IHCAs attended by the resuscitation team in 13 million hospital admissions. q, quarter.

Hospital survival (discharge alive) for all hospital admissions is over 95% (Figure 15). The odds of survival is increased by 5.3% per year (see Table 14) between 2009 and 2015. Hospital survival for admitted patients who had IHCAs also rose, with an increase in the odds of survival of 4.8% per year (Table 15).

FIGURE 15.

Trends in hospital survival for all hospital admissions and for IHCAs attended by the resuscitation team. q, quarter.

Just under half of patients with IHCAs survived to 20 minutes (return of a spontaneous circulation of > 20 minutes) and the odds of surviving to 20 minutes increased, on average, by 7.0% per year after allowing for differences in case mix (Figure 16 and see Table 15). Similarly, the odds of surviving to 30 or 90 days increased, on average, by around 5% per year.

FIGURE 16.

Trends in survival at 20 minutes, and 30 and 90 days in IHCAs attended by the resuscitation team. q, quarter.

Association between interventions and in-hospital cardiac arrests

All hospitals in the analysis used some form of TTS. The majority (81%) operated a paper-based system throughout the study. A total of 18 (17%) hospitals switched from a paper system to an electronic system, and two hospitals operated an electronic system throughout the study.

During the study period, almost half of the hospitals changed to a NEWS or NEWS-based TTS from a non-NEWS system. There were 22 hospitals with no outreach team throughout the study and five that introduced a team during the study period. There were 29 hospitals with 24/7 outreach teams throughout the study and six hospitals that increased the hours of the outreach team to 24/7 during the study (Table 16).

The secondary variables of interest were outreach features, handover system features and contextual factors. There were 29 hospitals (including 22 with no outreach team) that reported no features. The median number of outreach features was one (interquartile range 0–3) and only one hospital reported all five. The median number of handover features was five (interquartile range 3–6), with a maximum of seven of eight features. The median number of contextual factors reported as positive was three (interquartile range 1–3), with 16 hospitals reporting none and only one hospital reporting all six.

Table 17 shows the associations between interventions and IHCAs in 13 million hospital admissions after adjusting for case mix, time trends and seasonality. There was evidence that use of a NEWS or NEWS-based TTS was associated with a lower rate of IHCAs compared with a non-NEWS TTS (IRR 0.925; p < 0.001). There was no evidence for a difference in annual trends in IHCA rates between NEWS/NEWS-based TTSs and non-NEWS TTSs (p = 0.193 for a difference in slopes). There was also evidence that an electronic TTS was associated with a lower rate of IHCAs compared with a paper TTS (IRR 0.923; p = 0.005). There was no evidence for a difference in annual trends between an electronic TTS and a paper TTS (p = 0.217 for a difference in slopes). There was no evidence that the rate of IHCAs differed for a non-24/7 outreach team compared with no outreach team (IRR 0.976; p = 0.554) or for a 24/7 outreach team compared with a non-24/7 outreach team (IRR 1.039; p = 0.305). There was evidence that annual trends in IHCA rates differed. The greatest reduction was observed for non-24/7 outreach teams (IRR 0.932 or 6.8% reduction in rate of IHCAs per year), which represented a difference in slopes in comparison with no outreach (p = 0.008) and with a 24/7 outreach (p = 0.035).

The three variables with evidence for an association were included in a single model in the next stage of the analysis (see Table 17). The lower IHCA rate associated with a NEWS or NEWS-based TTS remained (IRR 0.916; p < 0.001) after adjusting for the other two variables. Similarly, the lower IHCA rate associated with an electronic TTS also remained (IRR 0.924; p = 0.006) after adjusting for the other two variables. There was weaker evidence that the downward trend in the IHCA rate with a non-24/7 outreach team was greater than for no outreach team (p = 0.091) and for a 24/7 outreach team (p = 0.099).

The third stage of the analysis involved restricting the sample of hospitals to those hospitals that changed intervention during the study (Table 18). There were 52 hospitals that switched from a non-NEWS to a NEWS TTS, 18 that switched from a paper to an electronic TTS and 11 that changed their outreach team (five from no outreach team to a non-24/7 outreach team, and six from a non-24/7 to 24/7 outreach team). In the 52 hospitals that changed to a NEWS/NEWS-based TTS, there remained evidence for a lower rate of IHCAs after switching to a NEWS/NEWS-based TTS (IRR 0.904, p < 0.00, after adjusting for the other two variables). Similarly, in the 18 hospitals that changed to an electronic TTS there was still evidence for a lower rate of IHCAs after switching (IRR 0.884, p = 0.001, after adjustment for the other two variables). However, evidence for a difference in slopes associated with the outreach team was not observed in an analysis restricted to the 11 hospitals that reported a change in their outreach team. There was no longer any evidence for a stronger downward trend in the IHCA rate with a non-24/7 outreach team; the IRR for the annual trend in IHCA rates for a non-24/7 outreach team (IRR 0.969) indicated a smaller downward trend than for no outreach team (IRR 0.936, p = 0.360 for a difference in slopes) and a 24/7 outreach team (IRR 0.936, p = 0.426), although neither difference was statistically significant.

The final stage of the analysis was to examine the impact of the three secondary variables measuring outreach team features, handover features and contextual factors. These variables were added to the case mix-adjusted model that included the two statistically significant variables from Table 17 (NEWS/NEWS-based or non-NEWS TTS; paper or electronic TTS). There was no evidence for any association when each of the three secondary variables were included on their own or when all three were included in the same model (Table 19).

Sensitivity analysis was conducted to restrict the outcome (rate of IHCAs attended by the resuscitation team for all admitted patients) to IHCAs occurring on the ward. The restricted analysis produced similar results for TTSs to those shown in Table 17. In the analysis of combined interventions, the IRR for the impact of a NEWS/NEWS-based TTS was 0.901 (95% CI 0.858 to 0.944; p < 0.001), a marginally larger effect than the 0.916 shown in Table 17. The IRR for the impact of an electronic TTS was 0.869 (95% CI 0.809 to 0.933; p < 0.001), indicating a larger effect than the 0.924 in shown in Table 16.

The change in slopes for the impact of outreach team was also retained in the model, with IRRs for annual trends of 0.997 (95% CI 0.976 to 1.017) for no outreach, 0.951 (95% CI 0.936 to 0.966) for non-24/7 outreach and 0.957 (95% CI 0.940 to 0.975) for 24/7 outreach. Comparison of trends in the sensitivity analysis showed evidence for a decline in IHCAs with non-24/7 outreach compared with no outreach (p = 0.001 for difference in slopes), a difference which is less apparent in Table 17 (p = 0.091).

The impact of restricting the primary analysis to those hospitals that changed intervention during the study is shown in Table 18 and indicated consistent results for TTSs, but not for outreach. Similar results were produced by restricting the sensitivity analysis to those hospitals that changed intervention. Evidence of effects were observed for NEWS/NEWS-based TTSs in the 52 hospitals that changed to NEWS: IRR 0.895 (95% CI 0.844 to 0.949; p < 0.001). In the 18 hospitals that changed to an electronic TTS, the IRR was 0.845 (95% CI 0.764 to 0.934; p = 0.001). In common with the primary analysis, the apparent evidence for a difference in slopes for outreach was no longer observed in the 11 hospitals that changed their outreach during the study. As with the primary analysis, the IRR for no outreach strengthened to 0.935, and the IRR for non-24/7 outreach weakened to 1.021 (p = 0.068 for difference in slopes).

Association between interventions and survival in all hospital admissions

Table 20 shows the impact of interventions on the survival of 13 million hospital admissions to 106 hospitals after adjusting for case mix, time trends and seasonality. There was no evidence that use of a NEWS or NEWS-based TTS was associated with better survival compared with a non-NEWS TTS (IRR 0.9998; p = 0.838). There was evidence for a difference in annual trends in survival between NEWS/NEWS-based TTSs and non-NEWS TTSs (p = 0.012 for a difference in slopes), with a stronger trend observed with non-NEWS TTSs. There was no evidence that survival was improved with an electronic TTS (IRR 1.0010; p = 0.400), but there was evidence that the trend in improvement was stronger with a paper TTS (p = 0.013 for a difference in slopes). There was some evidence that the survival rate differed for a non-24/7 outreach team compared with no outreach team (IRR 0.9983; p = 0.056), but no difference between a 24/7 outreach team compared with a non-24/7 outreach team (IRR 0.9995; p = 0.525). There was no evidence that annual trends in survival rates differed (non-24/7 outreach team vs. no outreach team p = 0.394 and 24/7 outreach team vs. non-24/7 outreach team p = 0.270 for difference in slopes).

The three variables with evidence for an association were included in a single model in the next stage of the analysis (see Table 20). The impact on trends in the survival rate associated with a NEWS or NEWS-based TTS was no longer statistically significant (p = 0.122) after adjusting for the other two variables. Similarly, the change in trend associated with an electronic TTS was also no longer statistically significant (p = 0.361). There was suggestive evidence (p = 0.083) that a non-24/7 outreach team was associated with a lower rate of survival than no outreach team.

In the third stage of the analysis, the three variables included in the combined analysis in Table 20 were examined in the restricted samples of hospitals that changed intervention during the study. The results in Table 21 show no evidence of any statistically significant association for any of the interventions in these groups of hospitals.

In the final stage of the analysis, shown in Table 22, the inclusion of the three secondary variables found no evidence for any difference in survival with respect to outreach team features (p = 0.374), handover features (p = 0.315) and contextual factors (p = 0.214). Including all three together in the same model made little difference.

Association between interventions and survival in in-hospital cardiac arrests

The analysis of survival in the 32,820 IHCAs attended by the resuscitation team in 106 hospitals is shown in Table 23. For the first outcome, the return of a spontaneous circulation that was sustained for > 20 minutes, there was no evidence that either a NEWS/NEWS-based or electronic TTS was associated with a difference in survival or a difference in trends. Similar results were found for hospital survival, survival to 30 days and survival to 90 days. For outreach, there was no evidence for a difference in the odds of survival (difference in levels) in any of the four survival outcomes. However, there was some evidence that the annual trend in odds of survival for 20 minutes was greater with a non-24/7 outreach team (OR 1.059, 95% CI 1.028 to 1.091) than it was for no outreach team (OR 0.992, 95% CI 0.959 to 1.026; difference in slopes p = 0.013). However, this difference was not observed in the other three outcomes.

The second stage of the analysis involved examining the impact of the different trends in the return of a spontaneous circulation that was sustained for > 20 minutes in the restricted sample of 11 hospitals that changed their outreach provision during the study. The results were broadly consistent with those in Table 23, with a downward trend in those hospitals with no outreach team (OR 0.971, 95% CI 0.899 to 1.050) and an upward trend in those hospitals with a non-24/7 outreach team (OR 1.095, 95% CI 1.002 to 1.197). The statistical significance for the difference in slopes was p = 0.091. There was little evidence that the trend for 24/7 outreach (OR 0.985, 95% CI 0.890 to 1.091) differed from the trend for non-24/7 outreach (difference in slopes p = 0.204).

However, investigation of the impact of outreach in the 11 hospitals in terms of a change in levels (as opposed to a change in slopes) showed a reduction in survival following a change from no outreach team to a non-24/7 outreach team. This was observed for all measures of survival: 20-minutes survival OR 0.831 (95% CI 0.693 to 0.996; p = 0.045), hospital survival OR 0.687 (95% CI 0.540 to 0.874; p = 0.002), 30-day survival OR 0.664 (95% CI 0.522 to 0.846; p = 0.001) and 90-day survival OR 0.674 (95% CI 0.524 to 0.866; p = 0.002). There was no evidence for any difference in hospitals changing to 24/7 outreach from non-24/7 outreach. There was also evidence that the case mix of IHCAs changed following the introduction of outreach. The odds of a non-shockable IHCA increased in hospitals switching from no outreach team to a non-24/7 outreach team: OR 1.315 (95% CI 1.092 to 1.583; p = 0.004) after adjusting for all other case mix variables, excluding presenting rhythm.

Finally, both approaches (difference in levels and difference in slopes) were modelled together to examine the impact of outreach on the four measures of survival. No statistically significant associations were observed for an impact of outreach on any outcome (all p > 0.1). The magnitude of associations observed in previous analyses was also reduced. These findings support the general results in Table 23 that indicate a lack of evidence for an impact of TTSs or outreach on survival of IHCAs. The apparent better trend for improvement in 20-minute survival with non-24/7 outreach (see Table 23) was not supported in further analysis of the 11 hospitals that changed their outreach provision.

Secondary variables were examined in a case mix-adjusted analysis with no other interventions included (Table 24). There was some limited evidence that more outreach team features were associated with increased odds of 20-minute survival and hospital survival but this was not statistically significant at p < 0.05. There was no evidence that the number of handover features or the number of contextual factors were associated with differences in any of the four measures of survival.

Sensitivity analysis restricted to the smaller sample of 21,595 IHCAs attended by the resuscitation team that occurred on wards produced similar results to those from the primary analysis in Table 23. The ORs for NEWS/NEWS-based TTSs were larger but remained statistically non-significant for all survival outcomes. Survival was marginally better with NEWS/NEWS-based TTSs modelled as a difference in levels (ORs 1.028–1.105, p-values 0.182–0.574). The sensitivity analysis of annual trends also mirrored the primary analysis and produced results that marginally favoured NEWS for 20-minute survival (p = 0.055) and non-NEWS for other measures of survival (p = 0.178–0.495). Sensitivity analysis applied to electronic TTSs also produced similar results to those in Table 23. No statistically significant ORs were observed for a difference in levels (p-values 0.263–0.721) or a difference in slopes (p-values 0.425–0.914).

The primary analysis of survival indicated evidence for a difference in slopes (p = 0.013) favouring non-24/7 outreach (OR 1.059) compared with no outreach (OR 0.992). The sensitivity analysis produced similar results [OR 1.057 (non-24/7 outreach) and 0.992 (no outreach); p = 0.042]. No other comparisons of outreach were statistically significant in the sensitivity analysis (p-values 0.459–0.889).

The sensitivity analysis did produce different results to the analysis of secondary variables on survival shown in Table 24. This is most marked for the impact of outreach team features, with evidence for improved survival for all survival outcomes (Table 25). There is also some evidence that additional handover features are also associated with better odds of 20-minute survival. No significant results were observed for an association with contextual factors.

Strengths and limitations

This was a large study that included 106 hospitals with > 32,000 IHCAs in > 13 million hospital admissions. Potential confounding factors have been controlled for by adjustment for case mix, and temporal trends and seasonal differences have also been controlled for. The approach to the analysis investigated the consistency of results between the larger sample of 106 hospitals and the smaller groups of hospitals that changed their TTS or outreach provision during the study to ensure that findings are robust. This approach, together with the modelling of hospitals as random effects, provides some protection against unmeasured hospital-level confounding factors. However, there remains the possibility of residual confounding because of incomplete adjustment for patient case mix and other hospital factors that are time varying.

The study also relied on the accurate measurement of the interventions in the hospitals and the specification of the type of impact (difference in levels and difference in slopes) that these interventions might have. Mis-specification in either would generally reduce the study’s ability to detect a true impact of an intervention. The focus on three primary interventions reduced the risk of statistically significant findings arising by chance. However, the limited number of hospitals that changed their TTS and outreach provision during the study meant that the study lacked power to examine interactions between interventions and their impact on IHCA rates and survival. Under-reporting of attended cardiac arrests to NCAA will also reduce the power of the study to detect changes in outcomes. Our survey found that 13% of hospitals reported a case ascertainment of < 90%.

Objective 1: to design a typology of interventions based on previous research and an understanding of how interventions are implemented in England

Given the heterogeneity of the interventions and the contexts in which they were implemented, plus the fact that many studies did not provide an adequate level of detail of either the intervention or the context, it was not possible to create a typology of interventions designed to recognise and respond to patient deterioration based on this literature. Deeper understanding of how interventions were being implemented in practice was derived from qualitative work within NHS hospitals which, in turn, informed the information gathered by the survey.

Objective 2: to determine the association between different arrangements of interventions and in-hospital cardiac arrest incidence and outcomes

• Across 106 NCAA hospitals there was a reduction in IHCA rates of 6.4% per year and increased hospital survival after arrest rates of 4.8% per year.

• Use of the NEWS was associated with an additional 8.4% lower rate of IHCAs compared with pre-existing trends, and the introduction of an electronic TTSs was associated with an additional 7.6% reduction compared with pre-existing trends. Restricting the analysis to ward-based arrests lowered rates by 9.9% and 13.1% for NEWS and electronic TTSs, respectively.

• This study was not adequately powered to test the interaction between NEWS and electronic TTS interventions to determine if their individual associations with IHCAs were additive.

• Outreach teams, either non-24/7 or 24/7, were not associated with a reduction in IHCA rates.

• Neither NEWS nor electronic TTSs were associated with additional reductions in short (return of a spontaneous circulation for > 20 minutes), medium (survival to discharge) or longer-term (30- or 90-day) survival following an IHCA.

• Overall, outreach teams had no impact on short- or long-term survival over time.

• When analysis was restricted to ward-based IHCAs in 2015, increased intensity of outreach team implementation was associated with both better short- and long-term IHCA survival. Increased intensity of promotion of handover tools was associated with better short-term survival.

Objective 3: to identify service features associated with the lowest in-hospital cardiac arrest incidence and best outcomes

Our study set out to test a number of hypotheses of how interventions designed to detect and respond to patient deterioration on the ward might have an impact on the rate of IHCAs for which a resuscitation attempt was made and on the short- and longer-term survival of those patients. Continuing education interventions were dropped from this analysis after findings from the qualitative interviews and piloting of the survey suggested that there would be limited variability in provision across NCAA hospitals and across our index years. Following the survey, structured handover was also dropped from the time-series analysis as SBAR was in place in around 90% of hospitals. The hypotheses were revised to reflect the focus of the analysis on TTSs and outreach teams.

The use of TTSs will reduce IHCA rate by identifying deteriorating patients earlier while there is greater opportunity of reducing their risk of arresting.

Both NEWSs and electronic TTSs were associated with reductions in IHCA rates compared with pre-existing trends, but we were unable to confirm whether or not these separate associations were additive. This observation may be due to these two tools enabling the early identification of deteriorating patients. It has been established that physiological changes in heart rate, blood pressure, respiration rate and oxygen saturations precede most cardiac arrests and that these measures can predict cardiac arrest, admission to the ICU and mortality. 7,140 The TTSs promote the monitoring of physiological parameters, the calculation of risk-based scores based on these observations and provide graded guidance on action at particular score thresholds. A reduction in cardiac arrests may result from enabling ward staff to make early, simple interventions known to be effective in reversing deterioration, such as providing oxygen to a patient35,166 or by empowering the sharing of concerns about patients with appropriate responders. Such action may lead to an increase in the number of patients transferred to the ICU. The TTS is a system-wide intervention that has the potential to improve the performance of the afferent (trigger) arm of the rapid response system across the whole hospital. Introduction of the NEWS has standardised the TTSs used in the NHS and reduces some of the barriers to effectiveness found in previous studies, including unfamiliarity with the scoring system or response algorithm. 41 The introduction of an electronic TTS would be expected to counter known problems with paper-based TTSs, such as failure to undertake all required observations, miscalculation or the reluctance of ward staff to trigger the appropriate responses. 41,42

However, there may be other indirect mechanisms by which NEWSs and electronic TTSs may influence IHCA rates: they may change the number of patients who are eligible for resuscitation by encouraging ward staff to engage in the process of making DNACPR/end-of-life decisions as a result of recorded deterioration. Our study does not allow us to ascertain easily the contribution of such indirect mechanisms to the lowering of the IHCA rates. An analysis of changes in case mix between 2009 and 2015 in those patients undergoing resuscitation did not show any major differences across the years that might imply a substantial shift in the numbers of patients with poorer health becoming ineligible for resuscitation through the implementation of DNACPR decisions. Both direct and indirect mechanisms represent good care as long as they are appropriate responses for individual deteriorating patients.

• The use of TTSs will have no association with survival following an IHCA.

Our findings that indicate no association between TTSs and either short- or long-term IHCA survival support this hypothesis. This finding was not unexpected given that the response arm of the rapid response system and intra- and post-cardiac arrest care are likely to have a greater influence. 63

• Outreach teams will reduce IHCA rates by reducing the risk of deteriorating patients arresting more than traditional patterns of response from ward staff.

Our findings do not support this hypothesis. We found no association between either non-24/7 or 24/7 outreach teams and cardiac arrest rates. It had been expected that, through delivery of their key objectives of strengthening critical care skills amongst ward staff, supporting ward staff in the management of acutely ill patients and promoting the uptake of TTSs,12 outreach teams would be associated with a reduction in IHCA rates, with the greatest impact seen in hospitals with the most intensive services. Our findings may reflect that outreach teams form only part of the efferent or response arm of the rapid response system and that their individual effect cannot be separated from other response mechanisms, or that it does not particularly matter how the response is provided as long as the afferent arm is effective at identifying patients at risk and ensuring that the response is timely. Outreach teams may be making a greater contribution to the functioning of the afferent arm rather than the efferent arm through their education and support activities related to TTS implementation. However, the fact that configurations and roles of teams are still as variable as described in the findings of a previous national survey53 would also limit the strength of association found between these teams and IHCA rates at a hospital level. Furthermore, the nature of the introduction of the intervention across hospitals also affects the strength of association with outcomes. The faster and more widespread introduction of the NEWS provides a stronger signal compared with the more gradual transition to fully operational outreach teams, many of which were adaptions of local arrangements already in place.

These findings are not consistent with those from the majority of studies evaluating RRTs, which have found reductions in cardiac arrest rates. 47 However, given that the organisation and implementation of teams in these studies is heterogeneous and that their introduction is invariably associated with the simultaneous introduction of a TTS or a revision of the TTS response criteria, it is possible that the implementation of TTSs represent an alternative mechanism by which this effect on arrest rates is achieved. The evidence for nurse-led teams (as opposed to doctor-led teams) is mixed, with only half of the studies showing a positive association with IHCAs. 48 Most studies of RRT have an uncontrolled before-and-after design. The small number with more robust designs have shown a mixed picture with half of the studies showing a reduction in cardiac arrests (all nurse-led teams)163–165 and the other half showing no reduction. 76,77,82 None of these studies showed any change in hospital mortality. The largest of these, the MERIT study,82 was a cluster randomised trial of doctor-led RRTs in Australia that found no impact; however, there were major design limitations, not least the low number of team activations triggered (30%) despite a larger group of patients having clearly fulfilled the calling criteria.

• Outreach teams will increase survival following a cardiac arrest by increasing the application of treatment limitation decisions, such as DNACPR, in deteriorating patients at a high risk of an unsuccessful resuscitation.

Our findings from the time-series analysis did not support this hypothesis. We found that the presence of an outreach team was not associated with a change in short- or longer-term survival of patients who have had an arrest. In addition, there was no association found between outreach team presence and hospital survival. However, there was an association between increasing intensity of outreach team provision in 2015, as reflected by the team composition, roles, autonomy and ability to meet demand, and both short- and long-term survival. By playing a role in end-of-life care and the implementation of DNACPR decisions in patients for whom CPR would be futile, outreach teams can reduce the number of patients with the highest risk of death undergoing CPR. 50

MEDLINE

EMBASE

MEDLINE

EMBASE

MEDLINE

EMBASE

MEDLINE

EMBASE