Intravenous infusion practices across England and their impact on patient safety: a mixed-methods observational study

Intravenous medication practices

Infusion pumps are safety-critical devices in widespread use across many clinical areas: for pain management, delivery of antibiotics and chemotherapy, anaesthesia and fluid management. They are used in many clinical areas, from paediatrics to elderly care, in intensive care, in operating theatres and on general wards. Studies have found that infusion practices vary significantly both between and within hospitals. For example, in a series of observational studies, nurses in an oncology day-care unit were found to follow fairly basic procedures in setting up planned infusions,8 whereas nurses in an intensive care unit (ICU) routinely used advanced functionality, frequently setting up several pumps in parallel to deliver different medications. 9 There is an increasing drive towards standardising devices within organisations, intended to reduce the potential risks associated with staff using a range of different devices or devices configured in different ways, or being required to operate devices that they are not familiar with or have not been trained to use. However, not all clinical areas require the same functionality; for example, Carayon et al. 10 studied how nurses used infusion devices in different areas of the hospital. They compared the tasks actually carried out with the tasks as defined by ward protocol, identifying divergences in practice and highlighting ways in which those divergences increased overall system vulnerability.

Safety of intravenous medication administration

Intravenous medication is essential for many hospital inpatients. However, providing IV therapy is complex, and data suggest that errors are common. 11–13 In a systematic review of UK studies using structured observation of medication administration, errors were found to be five times more likely in IV than non-IV doses. 14 Internationally, published error rates vary from 18% to 173% of IV doses given, in studies using structured observation of medication administration. 15 An international systematic review estimated the probability of making at least one error in the preparation and administration of a dose of IV medication to be 0.73, with most errors occurring at the reconstitution and administration steps. 16 Although many of these errors do not result in patient harm, all can cause anxiety for patients and staff and can reduce patients’ confidence in their care. As a result, the administration of IV medication has been identified as a significant topic of concern by regulators, manufacturers and health-care providers. 3

The potential role of smart infusion pumps

To reduce errors associated with IV infusions, ‘smart pumps’ incorporating dose-error reduction software (DERS) have been widely advocated. 4,17,18 This software checks programmed infusion rates against pre-set limits for each drug and clinical location, using customisable ‘drug libraries’, to reduce the risk of infusion rates that are too high or too low. Limits may be ‘soft’ (in which case they can be over-ridden following confirmation by the clinician) or ‘hard’ (in which case they cannot). Smart pumps may be standalone or integrated with electronic prescribing and/or barcode administration systems, and usually allow administrative data, such as number and types of over-rides, to be downloaded for analysis. Although smart pumps were in use in 68% of US hospitals in 2011,19 a figure now likely to be much higher, their use is not yet as widespread in the UK. 5 Such technology can potentially identify and prevent some kinds of medication errors, but cannot prevent all possible errors. Smart pump use also comes at a cost, both financially and in terms of the changes needed to make their use effective. For instance, Husch et al. 6 carried out a hospital-wide point-prevalence study of errors in IV infusions using standard infusion pumps, and identified infusion rate errors in 37 cases (8% of all infusions) and wrong medication in 14 cases (3%). However, they estimated that only one of these errors would have been prevented using standalone smart pumps. More were judged to be potentially preventable if the pumps were integrated with other hospital systems, such as computerised prescriber order entry (CPOE) and barcode medication administration (BCMA). In a survey study involving 29 hospitals across Canada that either had implemented smart pumps or were in the process of doing so, Trbovich et al. 20 found that respondents did not take the steps necessary to realise the potential safety benefits of pumps; for example, they did not standardise drug concentrations, develop drug libraries, set dosing limits or monitor how pumps were used.

A recent systematic review identified 21 quantitative studies of smart pumps,21 the majority of which studied the over-rides recorded in the smart pump logs and/or used unreliable methods of identifying medication errors and adverse drug events, such as incident reports. The authors concluded that smart pumps can reduce but not eliminate error, and that the picture was far from conclusive. Furthermore, most studies were conducted in the USA; none was from the UK, where systems for prescribing and administering medication differ significantly from those in the USA. 22 For example, nurses play a more active role in preparing IV medication in the UK, and verbal orders are much less common. We therefore know little about the effect on patient safety of using smart pumps in general and nothing about their likely impact in the UK.

Assessing the severity of errors

As well as studying the prevalence of medication errors, their clinical importance must be taken into account when comparing drug distribution systems or assessing the effects of interventions. 23,24 Medication errors range from those with very serious consequences to those that have little or no effect on the patient. Assessing the clinical importance of errors therefore increases the clinical relevance of studies’ findings compared with studies based on prevalence alone. In many studies, actual error outcomes are not known, either because there is no longitudinal patient follow-up or because researchers intervene to prevent errors from causing patient harm. Methods of measuring the potential severity, or clinical importance, of medication errors are therefore needed to evaluate the effectiveness of interventions designed to reduce them.

A systematic review of methods for measuring the clinical importance of prescribing errors identified a wide range of available methods but no comparative studies. 25 No comprehensive review has focused on methods to assess clinical importance of medication administration errors. However, in studies included in a systematic review of the prevalence and nature of medication administration errors, Keers et al. 26 noted that the two most commonly cited severity assessment methods were the US National Coordinating Council for Medication Error Reporting and Prevention (NCCMERP) severity index,27 and Dean and Barber’s method. 28 The former is an ordinal scale designed to be used by local staff who would usually have knowledge of actual outcomes and other contextual information, and the latter is an interval scale used by experts using descriptions of the errors without knowledge of their outcomes. In Chapter 3, we describe a comparison of these two methods of assessment to shed further light on the strengths and weaknesses of each.

Systems of practice: intravenous medication administration documentation

Numerous policies and procedures have been put in place at every participating hospital with the aim of reducing the risk of harm during IV infusion administration, and our first analysis presented in Chapter 4 focuses on variations in policy and practice around IV infusion administration.

A 2007 Patient Safety Alert for England and Wales29 made recommendations to reduce errors in injectable medicines, including risk-assessing procedures and products, reviewing protocols, providing technical information and competency-based training, and conducting an annual medicines management audit. It highlighted how procedures should be:

Clearly documented, reflect local circumstances and describe safe practice that all practitioners can reasonably be expected to achieve.

p. 3. 29

© National Patient Safety Agency 2007. Copyright and other intellectual property rights in this material belong to the NPSA and all rights are reserved. The NPSA authorises healthcare organisations to reproduce this material for educational and non-commercial use

Before the ECLIPSE study, to our knowledge, no study had investigated whether or how health-care organisations had responded to this advice or how well these procedures are adhered to in practice.

There has been limited research into procedural and documentation deviations of IV infusion administration. Husch et al. 6 included procedural and documentation errors in their study of 426 IV infusions in a US hospital; this study found two of the most prevalent error types to be no rate on the additive label, affecting 46% of infusions, and patient identification (ID) issues, affecting 13% of infusions. Schnock et al. 7 used a similar method across 10 US hospitals to examine 1164 IV infusions and reported 60% of infusions with an additive label that deviated from policy, 35% of infusions where giving sets were not labelled according to policy, and 0.2% of infusions where the patient had no ID wristband. No previous study has investigated procedural and documentation deviations in the UK and none has explored the surrounding context or possible reasons for the discrepancies identified.

Systems of practice: nursing

Parts of this section are based on Vos et al. 30 © 2019 The Authors. Published by Elsevier Ltd. This is an open access article under the CCBY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

The second analysis presented in Chapter 4 focuses on the role of nurses as a source of system resilience. Nurses undergo training, education and lifelong learning in order to provide safe and high-quality care. 31–33 In nursing education, problem-solving is emphasised and is often presented as a systematic process in which one moves from assessment to eventually evaluating the effectiveness of an intervention. 33,34 However, daily nursing practice is often more complex than this linear picture suggests. The planning and implementation of nursing interventions, such as administering IV therapy, requires clinical judgement, and thus an interpretation of the situation by the nurse. 33,34 As a result, actions demand standard approaches to be modified or new approaches to be improvised according to the patient’s response. 34 This flexibility and variability in clinical work can result in a gap between ‘work-as-imagined’ (policy) and ‘work-as-done’ (practice),35–37 but can be paramount to achieving safer practice. 38

In the context of IV therapy, non-compliance with policy is a deviation. However, not all deviations from policy lead to negative consequences for patient care. 39 Larcos et al. 39 have proposed looking at ‘resilience’ in health-care practice to recognise flexibilities that might contradict policy. Examining the safety of IV therapy requires a broader view that also includes focusing on what goes right. The safety of IV therapy might also be influenced by clinical judgement, professionalism and knowledge.

Nurses are a key player in the delivery of high-quality IV therapy to patients. 40 Their decisions and actions contribute to the safety of IV therapy. However, it is unclear how this is accounted for in practice when measuring deviations from policy. One of our analyses (presented in Chapter 4) investigated those situations in IV therapy in which nurses are a source of system resilience in relation to quality and safety of IV practice through their actions and clinical judgement.

Systems of practice: layers of influence

The final analysis presented in Chapter 4 focuses on system adaptability. 41 It draws inspiration from the work of Brand,42 who argues that what makes an artefact adaptable is a design that differentiates between constructional categories according to their likelihood of change, coupling together elements that have the same rate of change into layers that are loosely coupled to other layers that have different change rates. This allows fast-changing layers to ‘slip past’ more stable layers, minimising the scale of impact of change, which makes the artefact easier and less costly to change to accommodate needs. This view is relevant to many designed artefacts that require adaptation and change during their useful lifespan to accommodate new or changing use requirements. This style of thinking can be adapted to view and assess the relative impact of an intervention in one layer on the functioning and performance of the entire artefact or system. The lower (more stable) layers are harder to modify than layers that are higher up in the hierarchy. Changes at the lower layers are also likely to have a greater impact than those at higher ones. The analysis reported in Chapter 4 identifies the layers for IV infusion administration.

Human factors and human error

Building on prior work on human error and IV medication administration, the ECLIPSE study was strongly informed by a human factors perspective. This approach is based on:

• understanding the factors that contribute to errors occurring

• focusing on what really happens in practice, which is often different from what is believed to happen

• taking a ‘systems’ view that puts the person at the centre.

There is an extensive literature on human error, much of it summarised by Reason. 43,44 In brief, Reason identifies two main types of error. The first is mistakes, which occur when people do not have all the information needed to choose an appropriate action, or fail to plan or solve a problem correctly; mistakes can often be reduced by improved training. The second type is slips and lapses, which occur when people intend to do the correct thing, but inadvertently do something else as a result of distraction, fatigue or similar; slips and lapses cannot be eliminated through training, but can often be reduced through the design of tools or systems. Reason points out that the obvious causes of errors often appear at the ‘sharp end’ in the localised actions of an individual (e.g. a nurse programming an infusion pump), but that there are typically many latent failures behind these active failures (e.g. shortage of staff, working practices, local culture). He also distinguishes between errors and violations, with the latter classified into three groups: routine violations (cutting corners), optimising violations (doing something for personal gain) and situational violations (necessary to get the task done – often called workarounds). He argues that different causes of errors and violations require different counter-measures to prevent them or mitigate their effects. Reason focuses on major incidents about which there is general agreement that something went wrong. However, this is not always the case; for example, Furniss et al. 8 document ‘unremarkable errors’: when things did not go entirely according to plan and yet there was no untoward outcome.

Karsh et al. 45 contrast various paradigms for improving patient safety, arguing that, historically, most approaches (such as that presented by Reason44) have focused on minimising harm (by imposing barriers to prevent errors or injuries), and that an alternative human factors perspective is needed. They frame this in terms of designing systems to improve performance, including minimising hazards, and argue that both perspectives (imposing barriers to minimise the chances of bad things happening while also engineering systems to optimise performance) are necessary to improve safety.

There is a growing recognition that formalised descriptions of practices focus on ‘work as imagined’ and often miss the nuances and factors that most influence safety in practice. Blandford et al. 35 discuss this in terms of ‘work as imagined’ (as defined in training or in policies) and ‘work as done’ (as observed in practice). Heeks46 described this as the ‘design–reality gap’, pointing out that systems that do not match their users’ needs provoke workarounds, inefficiencies and sometimes rejection.

Workarounds have been studied extensively in various areas of health care. For example, Koppel et al. 47 provide an account of workarounds in BCMA, identifying 15 categories of workarounds and 31 categories of reasons for people employing workarounds; although workarounds are typically used to get work done in a timely way, they risk compromising patient safety and highlight issues in the broader system of care (in this case, the way in which BCMA has been implemented). Similarly, Debono et al. 48 present a systematic review of workarounds reported in 58 studies relating to nurses’ workarounds in acute-care settings; they highlight the paradox that workarounds can often simultaneously enable and compromise patient care.

Looking beyond the individual interacting with a device or other technology, a human factors systems approach places the person (the user of technology), the equipment and the task within the physical environment and the organisational setting, and accounts for behaviours within that context. This is exemplified in the work of Carayon et al. ,49 who adopt these concepts as foundational for the Systems Engineering Initiative for Patient Safety (SEIPS) model for reasoning about patient safety. They argue that:

Human Factors Engineering interventions that do not consider issues across the whole system, including organisational factors, are unlikely to have significant, sustainable impact on patient safety and quality of care.

Carayon et al. 49

Carayon et al. present SEIPS as an approach for analysing a work context in terms of patient safety. We adopt this approach in one of the analyses reported in Chapter 7.

A complementary perspective that has also been widely applied for understanding health-care systems and health technology is that of distributed cognition. 50 One methodology for applying the theory of distributed cognition, Distributed Cognition for Teams (DiCoT),51 involves analysing a system in terms of physical structure, information flow, how artefacts support cognition, social structures and how a system has evolved over time. DiCoT has been applied to understanding infusion pump use in various settings including intensive care9 and an operating theatre. 52 This perspective informed our data collection in phase 2.

Safety I and Safety II

Traditionally, the literature on safety has focused on barriers to bad things happening (‘Safety I’). As summarised above, it has now been recognised that safety cannot be achieved only by creating barriers, but that performance must also be improved (‘Safety II’). Vincent and Amalberti53 propose a continuum of ways to keep a system safe, or optimising performance. At one end of the continuum they place ultra-adaptive performance, which embraces risk and relies on the skill of an individual; at the other end they place ultra-safe practice, which relies on standardised practices, checklists and policy, based on an expected standard of performance for each role, and limiting scope for individual excellence. Between these extremes they place high-reliability systems that invest trust in the team and manage risks by focusing on team performance and learning. Ultra-safe practices embody a Safety I culture, and high-reliability systems embody a Safety II culture. Although ultra-adaptive systems have a place in health care overall, they rarely play a substantial role in IV medication administration.

We aimed to incorporate a Safety II approach in interpreting our findings. This approach moves away from the traditional focus of classifying all deviations as errors, eliminating error and blaming the human for unreliable processing. Instead, it encourages one to think about deviations in terms of performance variability, how to understand and manage this variability, and that the human component can make positive contributions to safety. 38,53 Safety II maintains that mechanistic performance is inadequate because complex systems are open to surprises, often run in degraded states, and have competing goals and conflicting pressures. So, not all deviations are errors, humans can be a source of success and failure, and errors and discrepancies might point to system issues.

Safety II played an important role in the planning and conduct of our research: (1) it helped us understand that we were sampling everyday work and its consequences; (2) it encouraged us to take a more nuanced approach to any deviations from the medication ordered, classifying some as discrepancies rather than all being errors, as in previous research; (3) it has made us mindful of system resilience, such as where some deviations added to safety; and (4) it encouraged us to engage with local rationality and challenging trade-offs. Safety II advises that one should attend to where the system goes right as well as where it goes wrong. After all, often when safety is working well, it can appear that nothing is happening: there is the absence of incident. However, Safety II defines safety as the presence of something, something that is constantly created, so it makes sense to also look at practices that can potentially contribute to safety that might otherwise be overlooked.

Safety II recommends looking at ‘positive deviations’ in the system. However, ‘positive deviance’ is an established approach in health care that investigates excellence in performance. 54 This involves measuring performance and sampling those practices that are the highest performing. We did not take this approach. Instead, we learnt in earlier phases of our research that clearly defining and measuring IV infusion administration performance is challenging. Data from phases 1 and 2 pointed to something that was more variable and multilevelled than we had previously expected, with multiple contextual dependencies. Following this, therefore, we reviewed the literature on complex adaptive systems (CAS) to better understand how to interpret our findings in terms of complicated and complex systems.

Complexity science and complex adaptable systems

With hindsight, it is now evident that, in framing the ECLIPSE study, the perspective taken was of IV infusion administration as a relatively simple system, centred around the infusion pump. From that perspective, it made sense to ask simple questions around whether ‘smart pumps’ were safer than traditional pumps or gravity feed, and what best practice is in this area. However, our findings cannot readily be distilled into such simple questions or their answers. In the final months of the project, we therefore turned to richer theories that illuminate, and enable new interpretation of, our findings. In particular, prior research on complexity science and CAS provides promising avenues for future research and for interpretation of our existing data.

Complexity science is the science of systems that are difficult to define in simple or structured terms. 55 For example, Holland56 describes CAS as comprising a large number of agents that are diverse in both form (e.g. people and computer systems) and capability (e.g. multidisciplinary teams).

Plsek and Greenhalgh57 argue that behaviour emerges from such systems without being designed ‘top down’ through the local actions and interactions of agents within the system and that it is not generally possible to address problems in such systems by reducing them into a set of simple problems that can be addressed independently. Plsek and Greenhalgh57 propose trying multiple approaches and gradually evolving improved systems over time using techniques such as the plan-do-study-act (PDSA) cycle of quality improvement. 58 Plsek59 also argues that it is not possible to design CAS in detail, but that what needs to be set up are the conditions under which desired outcomes are more likely.

Complex systems are commonly contrasted with simple or complicated systems. A simple system is well understood, it is possible to apply evidence-based best practice, incoming information can be categorised, and an appropriate response can be generated; team members may need to co-ordinate activity, but there are no tight interdependencies. In complicated systems, cause–effect relationships are understood, but may be at-a-distance (in time or space); longitudinal studies may be needed to establish relationships; and team members need to co-operate to achieve desired outcomes. By contrast, in complex systems, cause–effect relationships are non-linear and there are many agents with different roles and relationships, making a conventional style of analysis that decomposes a system into independent subsystems impossible. A complex system includes both simple and complicated problems, but is not reducible to them. Glouberman and Zimmerman60 note that health-care systems are often managed and analysed as if they are complicated when in fact they are complex, and that managers address problems with approaches that are based on rational planning; these do not work as expected because they fail to take account of the complexity of the system.

The same point is made by Braithwaite et al. ,61 who argue that, to make ideas ‘stick’ in health care, it is essential to exploit people’s natural enthusiasms and networks, and that clinicians will ‘become more involved in promoting safer and better care if invited rather than compelled’. Braithwaite et al. 62 argue that, for CAS:

Despite the potential for unpredictability, non-linearity, and even messiness and chaos, there are patterns, behaviours, structures and routines which together define the system, and guide behaviour within it.

Braithwaite et al. ,62 p. 13

The examples they share are largely drawn from large-scale complex systems such as hospitals, and their focus is on the roles of people (particularly networks of people) and the diffusion of practices, with little attention paid to the design of tools or protocols.

Van Beurden et al. 63 summarise the Cynefin framework, which has four principal domains: simple, complicated, complex and chaotic. A fifth domain (disorder) applies when a problem is not well enough understood for it to be classified as one of the first four. They argue that, to effect positive change, it is necessary to develop ‘probes’ so as to ‘sense’ what configurations are most likely to succeed, and then ‘respond’ in ways that amplify effects. The Cynefin framework can be used to describe transitions between different domains as people find out more about what they are dealing with, either in discovering previously unknown information that increases complicatedness and complexity, or by discovering relationships and patterns that can simplify how one perceives the system. 64

Sittig and Singh65 introduce an eight-dimensional model to support reasoning about the design, deployment and evaluation of health information technology (IT) systems based on the perspective of health care as comprising CAS. The eight dimensions are hardware and software infrastructure; clinical content; the user interface; people such as end-users and developers; workflow and communication; internal organisational features such as policies and cultures; external factors such as regulation; and measurement and monitoring (which implicitly includes learning). These eight dimensions serve as a checklist, or set of probes, to ensure appropriate coverage of considerations in analysing the performance of technology in a complex adaptive health-care system.

These ideas of simple, complicated and complex systems, with their different properties and modes of problem-solving, have been applied in various areas of health care and health technology. For example, Greenhalgh et al. 66 have developed a framework to account for the (non-)adoption and spread of digital health interventions based on two orthogonal scales: the simplicity/complexity of the problem and domains of health concern (e.g. the health condition, the technology, the users, the organisational context). Glouberman and Zimmerman60 analyse the Canadian Medicare system as a complex system. Begun et al. 67 take a similar approach to analysing innovations in care delivery in the USA. Sittig and Singh65 focus specifically on health IT systems (e.g. electronic health records). No previous research has explicitly studied DERS or IV infusion administration as a complex system.

Study design

The study employed a mixed-methods approach bringing together complementary quantitative and qualitative methods. A quantitative point-prevalence study, using observational methods, was used to document the prevalence, types and clinical importance of errors associated with the infusion of IV medication, that is, ‘what’ and ‘how many’ deviations occurred in practice. Qualitative debriefs and focus groups with relevant hospital staff helped explain the findings, in other words ‘why’ the deviations occurred in practice. The design of the point-prevalence study was based closely on that used in a similar US multicentre study that studied general medical, general surgical, medical intensive care and surgical ICUs. 7 This approach, originally developed by Husch et al. ,6 involves trained staff systematically comparing details of each IV infusion in progress at the time of observation with the medication prescribed, to identify any discrepancies. Within the ECLIPSE study, once a preliminary analysis of quantitative data had been conducted, debriefs and focus groups were held with key staff to reflect on the point-prevalence results and relate these to details of hospital IV practices.

Study setting, recruitment and sample selection

The study took place within acute hospitals across England. Expressions of interest were sought from English NHS hospital trusts through the National Institute for Health Research Clinical Research Network, NAMDET and contacts from a previous study. 5 Interested parties were then invited to complete an online survey to provide an overview of their hospital, capacity to take part, infusion pump types and practices. Eighteen NHS trusts responded to the survey, providing information about 26 potential hospital sites. We aimed to include a mix of hospitals that used smart pumps with DERS and those that did not. Few hospital trusts that made use of smart pump technology across all clinical areas responded to the survey, so they were approached separately to maximise the number of observations involving smart pump technology. Hospitals were chosen purposively, with the aim of representing maximum variation in terms of type, size, geographic location, potential indicators of patient safety such as being a Bruce Keogh Trust,69 hospital mortality indexes70 and media reports, and their self-reported use of infusion devices and smart pump technology. Thirteen acute hospitals, two specialist children’s hospitals and one specialist cancer hospital took part in the first phase of the ECLIPSE study. Appendices 1 and 2 summarise the recruitment process and characteristics of each participating trust.

We conducted observations in three clinical areas (general medicine, general surgery and critical care) in 13 hospitals; in eight of these we also conducted observations in paediatrics and oncology day care. Two specialist children’s hospitals collected paediatric data only. One further trust collected oncology day-care data at three hospital sites.

Medication infusions included any medication, fluids, blood products and nutrition administered via an IV infusion, including patient-controlled analgesia (PCA). This slightly extends the focus of previous studies,6,7 which focused on medication and fluids (which are also legally classed as medications), to include other IV administrations. We aimed to include observation of 2100 infusions in total across all study sites to give a confidence interval around a 10% overall error rate across hospitals and clinical areas of 8.7% to 11.3%. 6,17

The debriefs were attended by the two observers and staff who helped to organise the research at the site, such as the local co-ordinator and the local principal investigator. A focus group was held at each site; these were attended by different hospital staff (including ward managers, senior nursing staff, patient safety specialists, medical electronics personnel, trainers, those with responsibility for procurement, and senior managers), so they varied in type and number of attendees. A member of the research team facilitated the debrief and focus group sessions.

Data collection

Identifying and assessing deviations

The observers at each site recorded any deviations from a prescriber’s written or electronic medication order, the hospital’s IV policy and guidelines, or the manufacturer’s instructions. This included administration of medication to which the patient had a documented allergy or sensitivity, but did not assess other aspects of the clinical appropriateness of the medication order. We also collected data on policy violations and procedural or documentation deviations that may increase the likelihood of medication administration errors occurring. We specified four types of procedural and documentation deviations a priori: (1) giving sets not labelled appropriately; (2) documentation of administration inaccurate or incomplete; (3) infusion additive labels missing, incomplete or incorrect; and (4) patient ID wristbands missing or with incorrect, illegible or missing information. We identified two further types of deviation where policies varied among trusts during the debriefs and focus groups: (5) prescription and administration of IV flushes, and (6) procedures for the double-checking of medication. Finally, we encouraged observers to record any other irregularities, anomalies or workarounds related to the administration. Some of these were grouped together for analysis and formed new categories. Table 1 presents definitions of deviation types.

Data management and analysis: point-prevalence study

Data cleaning and classification was a highly iterative process as data collectors and the team worked through the many ambiguities and contextual factors that shaped practice.

Data on all deviations were adjusted in response to the debriefs and focus groups held at each trust. Researchers within the research team then collated and analysed these data together. This multisite analysis led to further adjustments in the quantitative data to ensure that they were as consistent as possible between trusts, so that similar deviations that had been rated differently were made consistent. To help with this process, clinicians within the research team reviewed deviations that local observers found difficult to categorise, similar deviations of similar type coded differently by observers and all deviations rated as category D or above. Minor changes were made to classifications of type and severity as needed. Examples of some of the classification heuristics developed within the team and discussions around particular observations are included in Appendix 4.

Error and discrepancy rates were calculated as the proportion of infusions with at least one error or discrepancy, using total opportunities for error (total number of doses administered, plus any omitted doses) as the denominator. Results were presented according to overall error and discrepancy rates, and individual types of errors and discrepancies, grouped into medication administration deviations and procedural/documentation deviations. Variations in deviation rates between clinical areas, delivery modes and infusion types were explored descriptively with their 95% confidence intervals, and chi-squared tests where appropriate. Qualitative data were analysed inductively.

Supplementary analyses

Two supplementary analyses of the data were conducted, comparing the NCCMERP method of assessing severity with an alternative approach, and comparing the findings from England against broadly equivalent data gathered in the preceding US study. 7

Developing accounts in terms of systems of practice

Three complementary analyses, drawing on the phase 1 qualitative data, were conducted to better understand the causes underpinning the quantitative data.

Analysis of the errors identified in phase 1 to explore the role of smart pumps

The quantitative data set from phase 1 provides the opportunity to explore the potential role of smart pumps in preventing or contributing to errors. We therefore conducted a supplementary analysis of these data to identify:

• the different types of errors identified in infusions given via smart pumps and those given using other types of pump

• for errors identified in infusions that were not given via a smart pump, the extent to which these errors would have been likely to have been prevented using a smart pump

• for errors in infusions that were given via a smart pump, the extent to which the use of a smart pump may have contributed to the error.

Each error of NCCMERP severity index category C or above was reviewed by two clinical members of the ECLIPSE study research team (a pharmacist and a nurse) to make a judgement as to whether the error would have been expected to have been prevented using a smart pump or whether a smart pump contributed to the error. Each error was reviewed and classified in discussion between the two researchers.

The following assumptions were made:

• If using a smart pump, it was assumed that a suitable drug library entry was available and that staff would select and use the correct entry, and that any alerts arising as a result of soft limits would not be over-ridden if they were alerting to an error.

• The following errors would not be prevented using any kind of smart pump:

  • omission errors, including those due to unavailability of pumps or giving sets, pumps not functioning or staff not knowing how to use the equipment

  • errors concerning patient ID

  • errors involving the administration of expired medication

  • labelling errors

  • delay in starting or completing the infusion.

• The following errors would not be prevented using a standalone smart pump but may be preventable if smart pumps are integrated with electronic prescribing administration systems:

  • giving medication without a corresponding medication order

  • wrong rate errors where the rate was incorrect in relation to what was prescribed for that particular patient, but where the rate administered was within the usual range for the drug concerned

  • wrong rate errors concerning much lower (rather than higher) rates than those prescribed, in drugs for which very low rates are sometimes used in clinical practice.

• For errors involving infusions of drugs or other fluids given via gravity feed where the rate was incorrect because the clamp was set incorrectly, it was assumed that these would be prevented by any pump (not necessarily a smart pump).

For the remaining errors (other wrong rate errors, dose/volume errors, wrong medication, concentration errors, and any other errors), each was reviewed individually with a judgement made on a case-by-case basis. Errors were classified using one of two sets of categories depending on whether or not a smart pump had been in use at the time of the error, as shown in Table 3.

To increase the rigour of this analysis, members of an expert panel were then invited to assess the preventability of a 25% random sample of the errors detected, stratified by smart pump use, traditional pump use or gravity administration. The panel members were a self-nominated group of 13 drug library experts who either had implemented or were in the process of implementing smart infusion devices in their organisation; they had previously been identified through the medication safety officer network for England to form a National Drug Library Expert Group.

All 13 experts were invited to participate by e-mail. Those who agreed (seven experts) were sent an anonymised sample of the data (listing whether a smart infusion pump, a traditional pump or gravity administration had been used and a description of the error) and requested to complete their assessment of preventability using the same predetermined categories (see Table 3). Participants were also requested to provide any comments or narrative about the errors listed, as well as comments on the assumptions made and the process of assigning categories. Agreement among the panel and between the panel and the research team was assessed descriptively.

Analysis of incidents reported to the National Reporting and Learning System

Having explored the potential role of smart pumps in preventing or contributing to administration errors within the ECLIPSE study point-prevalence data, most of which related to relatively minor errors, we took the opportunity to gain a complementary perspective through analysing data reported to the NRLS. All health-care organisations within England and Wales should report patient safety incidents to the NRLS. Although there is known to be considerable under-reporting in all incident-reporting systems, it is generally believed that serious errors are more likely to be reported. Although it was not part of our original protocol, we decided to conduct this supplementary analysis to complement the ECLIPSE study point-prevalence data. Our objectives were to identify more serious errors involving IV infusions and then to comment on the extent to which these may have been prevented by use of a smart pump or, for those that were given by a smart pump, the extent to which the smart pump may have contributed.

We requested data from the NRLS relating to incidents involving infusion pumps. Following approval by the NRLS and signing of a confidentiality agreement, we obtained anonymised reports for all incidents reported to have occurred between 1 January 2005 and 31 December 2015 that contained search terms relating to infusions or infusion pumps in any of the following fields: ‘description of what happened’ (field IN07), ‘action preventing reoccurrence’ (IN10), ‘apparent causes’ (IN11) and ‘device name’ (DE03). A full list of the search terms used is provided in Appendix 5.

From the data obtained, we selected those incidents rated as being of ‘moderate’ severity or above (moderate, severe or resulting in death; Table 4 provides definitions) that occurred in hospital inpatient settings, excluding operating theatres. We then removed all those relating to subcutaneous or other non-IV infusions; those where an IV infusion pump was mentioned in the report but the patient safety incident was unrelated; those that related to different types of pump, such as intra-aortic balloon pumps; and any others judged to be irrelevant.

All reports were reviewed by two clinical members of the ECLIPSE study research team plus a third expert medication safety officer with experience of setting up drug libraries. The team discussed each case and made a judgement as to whether the error would have been expected to have been prevented using a smart pump or, for those that were given by a smart pump, the extent to which the smart pump may have contributed.

The same assumptions were made as in the section above, and the remaining errors (other wrong rate errors, dose/volume errors, wrong medication, concentration errors and any other errors) reviewed individually, with a judgement made on a case-by-case basis.

A separate subanalysis was conducted for those errors involving the wrong rate, dose or volume, using the same approach, as these are arguably those most likely to be preventable with a smart pump. To put these data into context, we then estimated the number of infusions given each year in the English NHS. This was done by identifying several sources of relevant data [National Patient Safety Agency (NPSA; now incorporated with NHS improvement) Safer Practice Notice 01: Improving Infusion Device Safety,74 electronic prescribing data from one study site, and IV giving set purchasing data from a second site] and estimating the average number of IV infusion administrations per bed per annum from each of these sources, and then multiplying by the number of beds in English hospitals. 75

Study design

Phase 2 was a qualitative study based on observations and interviews in five hospitals. Observations included staff administering IV medication and setting up pumps, supplemented by interviews with staff to further understand why certain errors occur within each context. Data gathering and analysis were driven from a human factors and sociotechnical system perspective. As described above, this emphasises how the safety and performance of a system is an emergent property of the ways the people, policy, practices, artefacts and equipment combine within it. The analysis focused on the causes of error, the need for any workarounds in practice, and identifying best practices in safe IV medication administration.

The patient’s role is often neglected in medical care;76 we addressed this by interviewing patients about their IV infusion experiences. These patient-centric findings have the potential to inform practice through better consideration of people’s requirements and preferences.

Study setting, recruitment and sample selection

The phase 1 results were used to identify hospitals and clinical areas for more detailed study in phase 2. These were selected on the basis of the most practically and theoretically interesting, for example where different kinds of errors or practices had been found. We also wanted to maintain a geographic spread and a mix of different smart and non-smart pump use from different manufacturers. Five different hospitals, from five different trusts, were included in phase 2, each with two different clinical areas:

• site D – adult critical care and oncology day care

• site G – general medicine and general surgery

• site H – general medicine and oncology day care

• site I – paediatric critical care and paediatric general surgery

• site K – general surgery and paediatric general surgery.

Data gathering took place between October 2016 and December 2017.

In terms of sampling, there were three main strands of data collection at each phase 2 site:

• Ten days on the wards work shadowing staff, conversations and observations of IV infusion preparation and administration. It was expected that about 40 IV infusions would be observed at each hospital site, which could include bolus doses.

• Ten interviews with patients to get their perspective on the quality and safety of IV administration. Ethics approval to interview paediatric patients and/or their parents was not sought, which left a total of 35 interviews with adult patients.

• Ten interviews with staff to discuss IV administration practices, discrepancies, errors, etc. These were stakeholders with some managerial responsibility for IV infusions at the trust, for example the local principal investigator, medical device safety officer, medication safety officer, medical device trainer, someone involved in the procurement of IV pumps, relevant senior nursing staff, relevant ward managers, relevant medical electronics personnel and relevant patient safety specialists.

Data collection

Data collection focused on:

• establishing how IV infusion preparation and administration worked on the ward, including what triggered this process, what process was followed, what equipment was used, how it worked within the physical space and what issues there were

• social interactions around technology, including whether patients were involved in the set-up of IV administrations (e.g. what they were told about their medications and how alarms were responded to)

• what trade-offs and workarounds staff employed to achieve their goals

• how different technologies such as pumps, barcode readers, CPOE and vital signs monitoring devices were used together, and how each contributed towards patient safety.

Different modes of data collection were used to collect these data, as follows.

• Mode 1: scheduled interviews with clinical staff. This involved semistructured interviews with clinical staff addressing issues and activities to do with the administration of IV infusions. Interviews were audio-recorded. Document C on the project web page (www.journalslibrary.nihr.ac.uk/programmes/hsdr/1220927/#/; accessed 1 March 2019) presents the topic guide used.

• Mode 2: informal conversation with staff between tasks and during quieter periods (recorded as field notes). This involved talking to staff when there were naturally occurring breaks and lower quantities of work. This focused on clarifying observations the researcher had made.

• Mode 3: interviewing patients at the bedside. This involved semistructured interviews with patients around issues and activities to do with the administration of IV infusions. Written consent was obtained. The interview was audio-recorded where possible (n = 24); otherwise, notes were taken (n = 11). All data were anonymised for analysis and subsequent reports. Document D on the project web page (www.journalslibrary.nihr.ac.uk/programmes/hsdr/1220927/#/; accessed 1 March 2019) presents the topic guide used.

• Mode 4: photographs of medical devices and their situations of use were taken. These facilitated discussion in the phase 2 workshop with representatives from participating sites, but were not a direct focus of analysis.

• Mode 5: observing ‘everyday’ practice at the wards. This involved observing staff and patients, their workflow and their use of medical devices. The focus was on IV infusions, but in a wider context.

• Mode 6: shadowing individual members of clinical staff. This involved observing the working practices of clinical staff, particularly focusing on tasks to do with administering IV infusions. This included calculating infusion rates, setting up devices, changing device settings, and performing the closing stages once a device had been finished with.

Extensive field notes were kept from the observational work. Photographs focused on physical spaces and the arrangement of equipment. Audio-recordings were transcribed and anonymised.

Data management and analysis

Consistent with common forms of qualitative data analysis (e.g. grounded theory77 and thematic analysis73), inductive methods were used to explore themes and patterns that emerged from the observational and interview data. Where applicable, we employed relevant theory to gain further insight and give theoretical weight to our analysis.

Data from patient interviews were analysed using a thematic analysis approach. The analysis consisted of a combination of deductive and inductive approaches, guided initially by the framework of the interview topic guide, but also allowing for the emergence of new themes.

The staff-related data were analysed using two complementary perspectives: SEIPS and CAS.

Public and patients

Two workshops involving public and patients were held during the project. The first was near the beginning of the project and comprised a 3-hour workshop involving nine people with experience of IV therapy in the hospital setting. 80 The first part explored their experiences of IV therapy, which informed the research questions that were used in the patient interviews in phase 2. The second part reviewed the patient-facing material for phases 1 and 2. The second workshop was conducted near the end of the project, when people with experience of IV therapy were invited to reflect and comment on our emerging findings. Patient representatives were also included on our advisory group and study steering committee.

Health-care professionals

There were four main ways that we engaged with health-care professionals to get feedback on emerging findings. The first was in the debriefs and focus groups in phase 1, which were much deeper and richer than expected, partly because we were presenting analysis that directly related to their work, so they could engage with it well and provide the local contextual information we were looking for. The second was in a workshop close to the end of the project, when representatives from the phase 2 sites were invited to reflect and comment on the emerging findings from the phase 2 analyses. This worked well, as sites could discuss areas of common interest, bringing different ideas and perspectives. The third involved 10 external talks to health-care professionals about our emerging findings at relevant conferences, meetings, webinars, etc., throughout the project. These were not as effective for engagement as we would have liked, but they contributed towards dissemination. The fourth involved our advisory group, whose membership comprised senior health-care professionals who were very knowledgeable and brought different perspectives to IV infusion administration practices. They provided expert guidance on the conduct and organisation of our research and on our emerging findings at different stages of the project.

Industry

We had planned to have closer engagement with infusion pump manufacturers towards the beginning of the project but our advisory group and study steering committee advised that this should be left until later in the project to counter any threat to the perceived independence of the study. In 2018, we co-organised two closed workshop with representatives of five key manufacturers, from which a confidential report has been produced. Through these events, we shared our findings and engaged in discussions with manufacturers and (hospital-based) user groups on their implications and future priorities in terms of technology developments and policies and practices to support safe IV medication administration.

Phase 1: data protection, informed consent and addressing observed errors

At a local level, the point-prevalence study was comparable to an audit of IV medication practice for each site. Written informed consent was obtained from ward managers by the local co-ordinator before including ward areas in the study. Patients and ward staff were offered an information sheet outlining the study. Feedback from a patient and public involvement workshop was incorporated into the patient information sheet and the training of data collectors in how to approach and inform patients and their visitors about the study. As per the protocol approved by the ethics committee, written consent was not required from individual patients or ward staff, and ward staff were not notified of the timing of data gathering, to minimise the likelihood of this influencing their behaviour. Results from individual sites were kept secure and anonymous to other sites.

Written informed consent was obtained from staff taking part in the qualitative interviews. Interviews were audio-recorded and stored on secure password-protected computers before leaving the site. Interviews were transcribed by a professional transcribing service, with due attention to confidentiality, and the transcripts were anonymised.

If an error with potential to cause harm was identified during the point-prevalence study, the local data collectors were instructed to inform the relevant nursing staff caring for the patient in a discreet manner so that remedial action could be taken in line with local procedures. Document E on the project web page (www.journalslibrary.nihr.ac.uk/programmes/hsdr/1220927/#/; accessed 1 March 2019) presents our protocol for dealing with suspected errors.

Phase 2: consent for observations and interviews

This qualitative study involved gathering information on the administration of IV infusions and related work practice through observations, interviews and discussion. There were no interventions or changes to normal medical practice, so there was little additional risk. Clinical staff on the wards were informed about the observations and verbal consent was sought before any close observations were undertaken. Patients on wards where observations were being made were informed of the study verbally and offered a patient information leaflet, as they were not directly involved. Patients and staff taking part in interviews gave written informed consent prior to data collection. Interviews were recorded when possible. In consultation with participating hospitals, it was agreed that it was not appropriate to interview paediatric patients about their IV therapy, and that it would be impractical to arrange interviews with their parents (given resource constraints), so ethics approval was obtained for interviewing adult patients only.

Medication administration deviations

Overall, 427 (21.3%) infusions involved at least one medication administration error (n = 211) or discrepancy (n = 257). The most frequent types of deviation concerned rates and unauthorised medications.

Procedural and documentation deviations

Overall, 961 infusions (47.9%) had at least one procedural or documentation error (n = 24) or discrepancy (n = 1219). Table 5 shows the frequency and severity of different types of procedural and documentation deviations. Non-compliance with hospital requirements for labelling infusion administration sets was most common. Procedural or documentation errors mostly involved unlabelled syringes, or infusions where the label was significantly inaccurate. For example, a patient prescribed 60 mg of pamidronate was being administered an infusion labelled as 30 mg, but staff confirmed that the patient had received the correct dose.

Commonalities and differences in findings

Table 9 shows a comparison between the USA and England of the different types of error detected.

Documentation errors were not recorded as a separate category in the US study. The figures for labels not being completed and tubing not being tagged according to policy can largely be accounted for by differences in policy between the two countries. There is a more detailed expectation of what should be included on labels in the USA, as specified by the Joint Commission,82 although these guidelines may be interpreted differently across different hospitals. In England, policies are written locally, so there is more substantial variation in policies across study sites.

Patient ID errors were higher in England than in the USA; the main contributory factor is likely to be the widespread use of BCMA in the USA, which makes it more difficult to administer medication without having formally identified the patient.

These findings relate to all deviations observed (some of which were classified as discrepancies in the English study, as discussed above). In practice, the vast majority of errors were classified as A–C according to the adapted NCCMERP severity index (see Table 2). Very few errors (n = 25; 0.79% of all infusions) were classified as D (‘An error occurred that would be likely to have required increased monitoring and/or intervention to preclude harm’) across both the USA and England. Across all 3172 observations, only two (one in the USA and one in England) were assessed to be level E (‘An error occurred that would be likely to have caused temporary harm’). None was rated F (‘An error occurred that would be likely to have caused temporary harm and prolonged hospitalisation’) or higher.

Table 10 summarises the error types for the more serious errors (categories D and E). If no in errors of a particular category occurred, that category is omitted from the table. This shows that the greatest proportion of category D errors featured gravity feed pumps running at the wrong rate. Of the category E errors, one was a delay (USA, smart pump) and one a concentration error (England, gravity administration). No errors were assessed as severe enough to have caused prolonged hospitalisation.

Overall, the commonalities across the US and English studies are much more striking than the differences, and the level of technological maturity makes little consistent difference across contexts.

This combined study draws on the observation of 3172 infusion administrations. Across those, as discussed by Lyons et al. ,68 the only clear conclusion is that medication administered by gravity feed is more likely to result in patient harm than medication administered via infusion pumps. Neither the US nor the English study identified any events in which patient safety was significantly compromised, resulting in death or serious harm. It is more likely that such incidents would feature using a different study method that focuses on reports of more serious incidents involving medication administration (see Chapter 5).

The dominant theme across contexts is the variability within and between sites, particularly relating to IV medication administration policy.

Nuanced differences in methods

As noted above, the process of comparing findings across the US and the English study contexts revealed a few unplanned variations in data-gathering methods. These are listed in Table 11.

Some of the variations are incidental: minor differences in the ways that the study protocol was interpreted across the two studies. For example, some ‘wrong rate’ errors were classified as ‘pump setting’ errors by the US team, where the pump setting was the cause of the wrong rate; conversely, the English team noted roller clamp positioning errors, which would have been classified by the US team as wrong rate errors.

Some variations are also a consequence of the use of different technologies. For example, some kinds of paper documentation omissions observed in England were blocked by CPOE/BMCA technology in the USA.

Differences in contextual factors

There were some key contextual factors across contexts and legislatures that shaped differences in methods and outcomes.

The first is the level of technological maturity across sites. In the USA, all sites had implemented electronic health records, BCMA and CPOE, and IV infusion/medication bags were routinely scanned before administrations. In the English study, some sites had implemented electronic health records, and some had implemented electronic prescribing and medication administration (EPMA). Few hospitals had implemented smart pumps with drug libraries across all clinical areas; more had implemented smart pumps in selected areas (most commonly critical care); others were using traditional pumps and gravity feed. The implementation of electronic health records and EPMA lags behind in England, although it is now progressing rapidly. 83

A second is the use of drug libraries. In the USA, vendors supply drug library packages, but there is nevertheless large variation across hospitals. 84 In England, sites that made use of a drug library reported having constructed it locally; this was typically a pharmacy responsibility.

At a national policy level, in the USA, Joint Commission guidelines drive policy; these guidelines are typically precise, although they may be interpreted differently by different hospitals. In England, the Care Quality Commission (CQC) expects every hospital to have relevant policies, but does not mandate their content.

Finally, in terms of local prescription policies, in the USA, fluids might be given without always requiring a repeated medication order, and KVO administrations may run without a medication order (standing hospital policy); verbal orders are discouraged by the Joint Commission but are not uncommon. In England, there is variation in local policy about what can be given without prescription, KVO practices are rarely reported, and verbal orders are very unusual.

Giving set labelling

Deviations in giving set labelling affected 26.8% of infusions. Deviations affected 16.7–100% of infusions that required a giving set label across all trusts (Figure 4). Deviations affecting giving set labelling were common at trusts D, G, H, K and P. Rates of deviation were affected by both the level of detail required by local policy and clinicians’ awareness of that policy. For example, at trust D, infusions in all areas except critical care were generally non-compliant; observers learned that their hospital policy required all IV giving sets to be labelled only in the closing stages of data collection, despite being asked to familiarise themselves with the relevant policy prior to data collection. In their debrief meeting, observers reported that this requirement was part of their peripheral cannula policy and that they had not previously been able to find it. At trust P, giving sets were labelled in critical care only when this was needed to differentiate between drugs, but the trust’s policy explicitly stated that all IV lines must be labelled with the time and date they were connected to the patient; none was labelled with this information. Trust K, which had the most comprehensive giving set labelling requirements, also had a high rate of non-compliance (83.5% of infusions); its policy required all IV lines to be labelled with the name and strength of the medicine, route of administration, diluent and final volume, patient’s name, expiry date and time, and name of practitioner preparing the medicine. Patient Safety Alert 20: Promoting Safer Use of Injectable Medicine29 does not specify whether or not giving sets need to be labelled.

FIGURE 4.

Variation in giving set label policy and deviations where a label is required among trusts. Reproduced from Furniss et al. 81 This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.

Reproduced from Furniss et al. 81 This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.

Documentation deviations

Deviations in documenting IV administration affected 16.7% of infusions, ranging from 4.6% to 36.5% across individual trusts (Figure 5). Failure to document the start time was the most common problem. Other less frequent, but potentially more troublesome, issues were discovered during the observations. In some cases, administration was not documented at all. In one case, 20 mmol of potassium chloride in 1 litre of 0.9% sodium chloride was prescribed, but the trust did not stock this formulation. Instead, staff administered two 500-ml infusion bags of 0.9% sodium chloride, with 20 mmol of potassium chloride in one of them. However, poor documentation meant that it was not clear that this prescription had been split across two bags and which was to be given first. Patient Safety Alert 20: Promoting Safer Use of Injectable Medicine29 recommends making a detailed record of the administration as soon as possible after that administration, but does not give more detailed directions.

FIGURE 5.

Policy deviations relating to documentation of medication administration. Reproduced from Furniss et al. 81 This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.

Reproduced from Furniss et al. 81 This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.

Additive label deviations

Deviations in recording the required details on the additive label affected 10.9% of all infusions. Deviations affected 3.3–74.0% of infusions that required an additive label across trusts, and policy requirements affected different proportions of infusions at different trusts (Figure 6). For example, at trust J, 78% of infusions were standard fluids with no additives, and these did not require a label. Furthermore, not all trusts specified the information required on additive labels in the relevant policy but there seemed to be an implicit expectation in all trusts that nurses should complete all parts of the additive labels. Most additive label deviations were considered low risk by observers and focus group participants, such as missing batch numbers for licensed non-biological medications. As with giving set labelling, trust K had a high deviation rate; its written policy required the most information to be documented on its additive labels: patient name, ward/clinical area, drug, final concentration and volume, administration rate, total amount of drug added to the syringe or bag, batch number and details of the medication added (diluent, date prepared, time prepared, expiry date, expiry time and route of administration). This was more detailed than the Patient Safety Alert 20: Promoting Safer Use of Injectable Medicine29 recommendations of name of medicine, strength, route of administration, diluent and final volume, patient’s name, expiry date and time, and the name of the practitioner preparing the medicine.

FIGURE 6.

Variation in additive label policy and deviations where a label is required among trusts. Reproduced from Furniss et al. 81 This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.

Reproduced from Furniss et al. 81 This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.

Trust B’s policy required nurses to record the batch number on additive labels. However, nurses in the focus group raised concerns about the utility of doing this, for example for short infusions that would be thrown away after 20 minutes. They suggested that a better place to record batch numbers (if necessary) would be in the patient’s medical records, where this information would be permanent. One nurse suggested that some medications come with a removable batch number sticker that could be stuck in the patient’s notes. Trust D focus group participants said that they had no detailed additive label requirements written into policy, did not expect the batch number to be commonly completed on the label, and wondered if the labels should be redesigned without the section for batch numbers.

Potentially significant deviations included an additive label only marked ‘DEX’, which referred to dexamethasone but could have been confused with dextrose or other drugs; and a completely unlabelled syringe of fentanyl in a syringe driver. Other additive label deviations initially were suspected to be medication errors but, on further investigation, found to be solely documentation issues. For example, observers found a 1000-mg bottle of paracetamol infusing into a patient prescribed 675 mg. However, the nurse reported that they had removed 325 mg before setting up the infusion, so the patient would receive the correct amount. The observers pointed out that the bottle was not labelled to indicate that this had been done, but the nurses said that it was usual practice to remove the excess dose and not label these changes on that ward. Patient Safety Alert 20: Promoting Safer Use of Injectable Medicines29 makes no recommendations for the process of removing excess dose but does recommend that labels are used for medicines prepared in clinical areas and that detailed records of administration are made.

Patient identification deviations

The percentage of infusions with a patient ID deviation varied between clinical areas: general surgery (2.5%), critical care (2.5%), general medicine (5.1%), paediatrics (9.9%) and oncology day care (10.3%). Patient Safety Alert 20: Promoting Safer Use of Injectable Medicines29 recommends that patient ID and details are checked in accordance with local policy. NPSA, Safer Practice Notice 11: Safer Patient Identification85 recommends that all hospital inpatients in acute settings wear ID wristbands.

The deviation rate relating to ID wristbands was 5.8% overall, ranging from 0.0% to 16.9% across trusts (Figure 7). Trust F was fully compliant, which may be because it had prioritised this area and had been auditing this practice prior to our study. Trust C was the only trust whose policy stated that patients receiving IV infusions in oncology day care were not required to wear ID wristbands. At trust B, the oncology day-care manager reported ongoing problems with ID wristband compliance in oncology day care, although local policy required these to be worn. She perceived that it was difficult to change staff behaviour and reported technical problems with the printer used to print the wristbands.

FIGURE 7.

Policy deviations relating to patient ID. Reproduced from Furniss et al. 81 This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.

Reproduced from Furniss et al. 81 This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.

Variability in intravenous flush policies

Most trusts had a patient group direction (PGD) that allowed nurses to administer small-volume flushes (e.g. 1–20 ml of 0.9% sodium chloride) without a patient-specific medication order. However, oncology day-care units sometimes used larger flushes that would need a separate prescription if they fell outside the limits of the PGD (e.g. up to 250 ml or 500 ml across a series of infusions). However, at trust K’s oncology day-care unit, these larger volumes were administered without a prescription or a PGD; this practice was deemed acceptable by the trust’s haematology and oncology care oversight groups. Trust D had an electronic prescribing system that automatically included larger flushes in its chemotherapy regimens, although one flush was observed running but was missing from the medication order.

The issue of whether to flush the whole giving set or just the IV access device arose at a number of trusts. For example, at trust P, a nurse had prepared a 100-ml bag of 0.9% sodium chloride that was not prescribed, and was beyond the 20 ml PGD limit, to flush between giving omeprazole and furosemide infusions. The nurse intended to flush the whole giving set to ensure that the whole dose was administered and to avoid manipulating the connection with the access device. This was noted as unusual practice by trust P observers, who reported that the first giving set would usually be detached, the access device flushed and then a new giving set connected for the second drug. However, some focus group participants recognised that this may lead to partial infusions as some of the dose would remain in the giving set. Patient Safety Alert 20: Promoting Safer Use of Injectable Medicines29 recommends flushing the access device before and after administration.

Variability in double-checking policies

Patient Safety Alert 20: Promoting Safer Use of Injectable Medicines29 recommends double-checking systems (e.g. an independent check or smart pump technology), but it does not go into detail about how these should be done. Different single, double, independent and second checking procedures were required for IV infusions at different trusts. Trusts A and O explicitly permitted single checking for IVs except for specified high-risk drugs, specific situations and controlled drugs. For example, trust O’s policy required staff to double-check before administering chemotherapy. By contrast, trust G’s policy required staff to double-check all stages of preparation and administration from cupboard to bedside, although there was acknowledgement in the focus group that this was not always practical. The wording of double-checking policies at some trusts implied that this was ‘required’, whereas others seemed more flexible, with wording such as ‘where possible’. During the focus group, nurses and pharmacists at trust P recognised that the wording of their policy was ambiguous. The policy was intended to mean that a second clinician signs to confirm that the right patient is receiving the right drug with the correct pump settings, whereas the nursing staff who attended the focus group thought that the second signatory only confirms the contents of the bag or syringe. Trust P’s pharmacy staff also wanted to move away from the concept of a ‘second checker’, as this terminology suggested that it could be a less important and only confirmatory role, and move towards a ‘second administrator’ who was equally accountable and would be expected to carry out a thorough independent check. Trust I was the only trust to have a separate detailed appendix to its main policy that specified what an independent double-check involved.

Summary

Almost one in two infusions had at least one deviation from local policy. Adherence to procedural requirements varied markedly, and it was difficult to make comparisons across sites due to wide variation in local policies, as recently reported in the USA. 7 Most participating trusts had a lower prevalence of deviations relating to additive labels and patient ID than in the hospital studied by Husch et al. ,6 and fewer additive label and giving set labelling deviations than those reported by Schnock et al. 7 We found that the prevalence of deviations involving patient ID wristbands was higher than 0.2% of infusions, as recorded by Schnock et al. 7 The inclusion of oncology day-care units and paediatrics in our study exacerbated this difference because of their deviation rates are higher than those of other clinical areas.

Anticipatory resilience

Larcos et al. 39 define anticipatory resilience as proactively making a decision or taking a course of action that has an expected consequence in a given situation. Such situations were behind some of the deviations discussed in the debriefs and focus groups. For example, nurses were reported as taking action to save time on low-risk infusions so that they could allocate more time to high-risk infusions (e.g. chemotherapy). In other cases, nurses reported making changes to infusion rates if this was judged to be in the best interests of the patient, which could have been classed as a deviation in our quantitative study. Participants suggested that prescribing an infusion with the infusion rate expressed as a range might provide more flexibility to respond to patients’ conditions where appropriate:

You may had it on neonatal intensive care unit or someplace but we do it on ITU [intensive therapy unit] we get a range of an infusion, we’d get one to ten depending on what the patient’s needs are. [. . .] The prescription doesn’t have to get rewritten and the nurse can play around with the medication in the middle.

Site F

Some trusts had detailed policies stating how IV giving sets should be labelled and which information should be included; others did not have such policies in place. Practice in terms of giving set labelling also depended on the ward and the type of infusion (intermittent vs. continuous). Critical care units were often mentioned as areas where nurses used giving set labelling to make practice safer even if there was no requirement to do so:

. . . I think you find on ICU that it is made standardised practice, to have a line tagging in place. [. . .] And it is our policy in critical care, and we always train new nurses to label all the lines. [. . .] Because obviously, wherever you got multi-lumen central lines, etc., etc., you know, obviously you’ll five lines hanging off it [. . .] It’s difficult to identify which is which where it gets near to the patient, so obviously at that point, line tagging is absolutely essential to see what’s on what line going into the central line, etc. [. . .]. However, if there was a patient come to the ward who just wanted infusion by link to a single peripheral cannula, then there probably isn’t such the same kind of need or ethos around line tagging [. . .]

Site D

Where the labelling of giving sets was not common practice, participants revealed that this did not necessarily mean that no information was available about when the giving set needed to be changed. For example, nurses might hand over these details:

Basically there’s other systems for monitoring lines, like [name] has just described, which are preferred to necessarily tagging the lines. So as part of the critical care daily check, lines, as well as the things that they talk about, not just at nursing handovers, but, actually, as part of the main medical ward round, so it’s a discussion point for the care of that patient, and therefore the information is better in the electronic system which is what they’ll be referring to during the ward round [. . .]

Site N

Occasionally, proactive practice when setting up infusions could also cause deviations. For example, the use of blood giving sets for other types of infusion could be considered a deviation, but this may reflect nursing staff anticipating that patients may urgently need blood products.

Responsive behaviour

Responsiveness is defined as reacting effectively when a situation changes. Nurses reported not following a medication order exactly if it was not possible to administer the medication as prescribed. For example, when a prescribed volume of fluid (such as 1000 ml) was not available, nurses described modifying their practice based on what was available (such as giving two bags of 500 ml). This ensured that the patient still received the prescribed treatment, although the medication order was not always updated to reflect what was actually given. Nursing staff perceived that this would result in delays and so they had to make a choice between interrupting or delaying treatment until a valid prescription was available and starting treatment without the medication order precisely matching what was given:

So they did only have 500 mls available? So they needed to give a litre, so they split the prescription, not the prescription, but they split where they signed by 500 mls and they gave 500 mls, so they signed it twice, 500-ml bags against the prescription that is 1 litre. It’s not an error but there’s a potential . . .

Site B

Frequently, patients’ conditions fluctuated, requiring nursing staff to think critically about the medication prescribed. In these cases, improving the safety of IV therapy might mean that a patient no longer needed the drug or amount that was initially prescribed. Safer IV practice in these instances would require nurses to adjust their plan according to the situation, based on their clinical judgement and preferably informing the prescriber, but those actions could lead to the reporting of a deviation (particularly if there was a delay in updating the documentation):

[. . .] the prescription there for [noradrenaline] for example, but if it’s not required we stop it. So in that situation it could be a discrepancy as well because the medication order is signed to be given, but we’re not giving it because of clinical judgment.

Site B

Patients’ clinical conditions also affected nurses’ judgement about how to interpret the prescription, for example by adjusting the concentration of a drug solution:

I think what happened is patient had deteriorated, so rather than getting loads and loads of fluid through the noradrenaline they increased the concentration of the noradrenaline. So rather than giving an 8-mg syringe driver they gave 16-mg syringe driver. So with regards to the prescription, it was still within that . . . um range, but they just increased the concentration of the medication to give them less fluid and so the medication doesn’t run out as quick. So they’re still getting the same dose, but just a more concentrated form [. . .]

Do doctors always prescribe this when it’s double strength or triple strength?

No.

So that’s often a decision that the ITU [intensive care unit] nurse makes, changing the concentration, but the patient’s still getting the same amount of drug.

Site F

There were also other situations in which deviations were identified as a result of nurses acting in the patient’s best interests. These included occasions when one of several infusions was delayed to avoid risks of administering incompatible infusions via the same IV access, when verbal orders were given, or when nurses had requested that the doctor make changes to the documentation but this could not be done:

. . . we saw somebody put up a fluid and there was no prescription for it [. . .] well they’d increased the fluids overnight, overnight there’s skeleton staff, [name] had requested a prescription to be written, but it hadn’t. [. . .] at the moment the surgical team were in theatre, there’s nobody to come and write a prescription [. . .] so what do you do, do you just not put up the fluid that the child needs?

Site I

In addition, patients might have been away from the ward to undergo tests, which led to nurses having to ‘catch up’ with infusions. In situations like this, nurses worked to provide safe IV practice while dealing with competing priorities, all of which could have an impact on the patient’s condition:

So that then you would delay the next dose . . . [. . .] But, you know, then you still try and work to try and get it back on track otherwise you end up totally out of sync. One dose is going to have to be given a little bit earlier then just to get it back on track.

Site C

Prescribing errors were not the focus of our study. However, our data suggest that nursing staff often improved the safety of IV practice by questioning prescriptions if they thought that a mistake might have been made:

[. . .] it’s very clear in our policy that if it doesn’t match [prescription and drug library], then, speak to your ward pharmacist. Speak to a pharmacist or speak to the prescriber to challenge that prescription. It’s very clear in our policy that we don’t just hope for the best.

Site K

Past experience

Past experience is defined as drawing on existing knowledge to influence the sequence and nature of work activities. 39 As well as the current situation in which nurses were working, their approach to the safety of IV therapy seemed to differ according to their familiarity with the infusion, their training and their previous experience.

Often, different medication charts and prescribing systems were used in the same trust, leading to deviations being reported because the observers could not find the relevant chart. If patients had transitioned through the hospital, they might have arrived from another clinical area with IV fluids still running. Nurses had to scrutinise a patient’s documents to find the original prescription in order to provide safe IV therapy:

And you have so many different charts, as well. For instance, you get a kid from [unclear], they have a different drug chart in A&E [accident and emergency], and he goes to [another ward] and gets a different drug chart. Gets worse and goes to ICU, has another different drug chart. Or goes to theatre then to ICU, and you’ve got some drug that are written on some piece of paper in illegible writing. [. . .] You can’t find it. You’re relying on what people have told you beforehand.

Site L

Intravenous medication started in another clinical area might have been prescribed on a different document and/or not yet re-prescribed on the relevant documentation for the new area. If medication was not yet re-prescribed, nurses were left to decide between not providing further IV therapy, and continuing the medication and requesting a new medication order later:

[. . .] what we do have to be careful is where we have different prescribing systems in operation and we’re transferring patients. Quite often we get people saying, for example on acute medical unit that patients haven’t been prescribed fluids in A&E [accident and emergency], but they’re running. And in fact they’re on the central alerting system and not on the orange drug chart. And the patient will get transferred from a paper ward to an electronic with fluids running. The prescription is on the paper but there is no active prescription on electronic, so it looks like that’s not an authorised medication. But in fact it’s an existing prescription on a different prescribing system.

Site K

Workarounds

Our findings suggested that some policies might need to be revised to better align with practice. For example, although verbal orders were not permitted, staff often acknowledged that practice deviated from policy in this respect:

Our medicines policy is perhaps a bit naive in saying we should not do verbal orders. Which is fundamentally what it says at the moment. And then perhaps we do need to go back to revisit where verbal orders are taken, which would be additional, you know.

Site D

Nurses sometimes worked around policies that were perceived to be inefficient or unworkable:

Often we’ll walk around and say can we give another fluid bolus, and we’ll walk off because we’re busy, and we’ll say that’s allowable for two boluses, and then needs a formal signature. And then two further boluses can be given like that, repeatedly. Because we were just finding that [. . .] for the nursing staff to hunt us down to try and get a signature would consume a lot of their time.

Site H

Staff at some trusts had created a workaround for the administration of small-volume flushes such as 10 ml of 0.9% sodium chloride, as staff did not feel it was realistic for these to be individually prescribed. Whereas some trusts used PGDs to allow nurses to administer flushes to pre-defined groups of patients, and some used pre-filled syringes licensed as medical devices rather than medicines, others reported it to be ‘common practice’ to give flushes that were neither prescribed nor covered by a PGD:

Yes. It happens across the trust. [. . .] Because, the trust doesn’t want to pay for the expense of medical devices because obviously it costs [. . .]

Site D

Practically, the number of prescriptions that would have to be written for saline flushes, we did look at this would it be possible to have a patient group [direction], but then we got medical devices that are used for saline flushes and you’ve got saline flushes, so do we have a PGD just for those times that you don’t use a medical device.

Site I

The use of additive labels to specify anything added to an infusion was noted as a mechanism to improve the safety of IV therapy. Some trusts used pre-printed additive labels with blank boxes that had to be completed by the nurse. However, additive labels were not always completed in accordance with the policy. Several reasons were given for incomplete or missing labels, including saving time on low-risk infusions and the relevance of some of the requested information, particularly batch numbers:

So, the batch number should go on the documentation that you keep, whereas putting a batch number on a label that you’re putting a bag up, you give it an hour, then you throw it in the bin. What is the point in that, surely?

Some of my infusions are 10 minutes. So, by the time you . . . Through two pages’ worth of paperwork to get a 10-minute bag up . . .

Site B

In summary, in common with previous studies on workarounds,47,48 it was found that nurses introduce workarounds – many of which then become established practice – to facilitate their work, and often to deal with deficiencies in the broader system of care. Although these are commonly in the patients’ best interests in the short term, many may compromise patient safety in the longer term.

Nurses performing risk assessments

Although incomplete additive labels were not deemed good practice, nurses reported this as a way of freeing up time to allow them to focus in greater detail on higher-risk infusions. On occasion, nurses did not label giving sets, even when there was a policy requiring this. The perceived risk of not labelling giving sets differed depending on the type of infusion (e.g. replacement fluid vs. a specific drug):

I don’t know if they tag all of the fluids necessarily. They tag all of the infusions of drugs because they want to make sure that when they change a syringe or the vial that they’re connecting the right to the right. But I don’t know if they would do it necessarily with IV fluids, which is probably something we need to pick up.

Site P

Although most participants acknowledged the importance of patients wearing ID wristbands for safe IV therapy, it was perceived that some situations did not allow patients to wear them (e.g. emergency) or made it difficult for them to be issued (e.g. broken printer, limited availability of staff). In these instances, nurses might have to make a decision between breaking the policy stating that an ID wristband was necessary for treatment and not giving a treatment that a patient might urgently need.

. . . there were times in [emergency department] when you’d come for [resuscitation] and you wouldn’t get your ID bands.

Site N

In practice, the risks associated with missing patient ID wristbands were felt to differ depending on the patient. Specifically, risks were perceived to be lower for inpatients who were alert and able to confirm their identity verbally, for ‘known’ oncology outpatients, and for patients receiving fluids rather than other medications:

It all depends what they receive, if they just have fluids it [ID band] probably doesn’t matter, yes, whether it’s the right patient or the wrong patient, unless they have sort of cardiac failure, etc., but you don’t want to give the chemotherapy to the patient or things like that, or any medication which has more implications.

Site B

However, as the following quotation illustrates, such risk assessments may not be appropriate, given that IV iron is also a relatively high-risk medication:

[. . .] we almost never make a chemotherapy error that I know of but we’ll make errors with other things, you know. Because, again, that, sort of, attention to detail and the seriousness of doing, like, you probably would never make an error with insulin, or, you know . . . Because you, you can’t . . . You just can’t make errors with that stuff, but you can make an error with an iron infusion. [. . .]

Site B

Summary

This section has reported our findings related to nurses as a source of system resilience in relation to IV therapy. This may account for some of the more minor deviations identified in our quantitative study that resulted from judgements made by nurses with the aim of improving the safety of IV therapy, often in the context of limited resources and conflicting pressures. These findings also illustrate how the safety of the system appears to rely in part on individual strategies and behaviours that contribute to system resilience.

Types of errors in smart pumps, traditional pumps and gravity administration

We first conducted a descriptive analysis of the types of medication and documentation errors occurring in those infusions given via smart pump, those given via traditional types of pump and those given via gravity feed. These data are summarised in Table 9 and suggest that documentation errors were less likely with smart pumps, but that labels not being completed or giving sets not being tagged according to policy were more likely in infusions given via smart pumps. It is not clear why this would be the case, but the most likely reason is more stringent policies in trusts using smart pumps (so more opportunities for deviations from those policies). Rate errors were more common in infusions given via gravity, as might be expected, but there was no difference between smart and traditional pumps (see Table 9). Delays in administration were also more common in infusions given via gravity. Patients not wearing ID bands were more likely when infusions were given via traditional pumps, most likely to be a result of the lower prevalence of ID bands in oncology day-care units, where traditional pumps were routinely used. Incomplete or delayed completion of the infusion was less likely in infusions given via smart pumps and, as might be expected, errors involving roller clamps were more common in those given via gravity feed.

Role of smart pumps in preventing or contributing to errors

We then assessed the preventability of these medication errors using smart infusion pumps, or the potential contribution of smart pumps where these were used.

Including the medication administration errors (n = 211) and the miscellaneous errors (n = 5), there were 216 errors of severity level C and above identified across the 16 organisations. Of these 216 errors, 157 were in infusions not given via a smart pump. The remaining 59 were in infusions given via a smart pump.

Preventability of those 157 errors identified in infusions that were not given via a smart pump is summarised in Table 14.

The two errors judged to be possibly preventable with a standalone smart pump, depending on the limits set in the drug library, were as follows:

• Dexamethasone was administered over 11 minutes, but the IV guide and prescription both stated that dexamethasone in 100 ml of saline should be given over 15 minutes. This would depend on how the limits were set on a smart pump and whether or not this rate discrepancy would create an alert.

• Hartmann’s solution was given at 7.5 ml/hour but a rate of 125 ml/hour was prescribed. This would depend on whether or not ‘too low’ rate limits were set on a smart pump – which is unlikely, as even very low rates are sometimes required in clinical practice.

An analysis of the role of smart pumps in preventing or contributing to errors identified in infusions given via a smart pump is summarised in Table 15.

Both of the errors to which the smart pump was judged to have contributed were piperacillin/tazobactam being given over 1 hour instead of the recommended 30 minutes, as the drug library did not allow a 30-minute infusion.

The error that may have been prevented by different limits related to atracurium of 500 mg/50 ml prescribed at 0–5 ml per hour being given at 10 ml/hour in a critical care area.

Expert panel assessment

Seven out of the 13 drug library experts approached agreed to participate. A total of 58 errors were sent, of which 33 involved the use of traditional pumps, 16 involved the use of smart pumps and nine involved gravity administration. Many of the panel members did not differentiate between standalone smart pumps and those linked in a closed-loop manner to an electronic prescribing system in their responses.

Of the 33 errors in infusions given using traditional pumps, the overall view of the panel was that 24 (73%) would be preventable using a closed-loop smart pump or a standalone smart pump with suitable limits, and that nine (27%) would not. This contrasts with the research team’s assessment for these 33 errors, that 16 (48%) would be preventable using a closed-loop smart pump with details of the medication order populated from an electronic prescribing system, and 17 (52%) would not be preventable with any type of smart pump.

For the 16 errors given via smart pumps, the overall view of the panel was that nine (56%) would have been prevented using a closed-loop system, in five (31%) cases the smart pump had no effect on preventing or causing the error, and in two (13%) cases different limits would have prevented the error with a standalone smart pump. The smart pump was considered to have contributed to the error in only one instance, by only one of the judges, a case involving administration of morphine in the incorrect diluent. The research team’s assessment was that 11 (69%) would have been preventable only with a closed-loop system and that in five (31%) cases the smart pump had no effect.

Of the nine errors in gravity administrations, the panel’s overall view was that seven (78%) would have been prevented using a smart pump, one (11%) would have been prevented with any pump and one (11%) would not have been prevented. The research team’s view was that six (67%) would have been prevented by any pump, one (11%) would have been prevented with a closed-loop smart pump, one (11%) was not applicable (an infusion of blood) and one (11%) would not have been prevented with any kind of pump.

The level of agreement among the expert panel assessors ranged from 57% to 100% per error; the median agreement was 71%. There was 100% agreement among panel members that 6 out of the 58 errors were possibly preventable using a smart pump, depending on the limits.

The narrative comments provided by the expert panellists were mainly statements of assumptions made when assessing preventability, although one of the panellists included observations about the types of errors that were included in the sample.

Generally, panellists based their decisions on a ‘best-case’ scenario, that is, the smart pumps were set up to optimal limits and functionality and drew data from the electronic prescribing system, with some different assumptions from those made by the research team. For example, one panellist stated that their assessment was based on the assumption that:

where present they are always used – there is no access to other infusion devices; that all drugs administered via a closed loop system are standardised concentrations of pre-prepared bar coded injectables, and that there are no facilities for the preparation of other doses in clinical areas (non-standard concentrations, hand-labelled etc.) i.e. they are ‘closed’ and strictly adhered to.

Panellist 3

There were contrasting views between panellists about the impact on of the type of device on infusions that exceeded the expiry or prescribed duration, or were not prescribed at all:

That the closed loop smart pump system included barcode scanning of medicines (this would have prevented all errors where the infusion was running without a prescription and those where the wrong drug/dose had been added to the bag).

Panellist 2

That the infusions within the electronic prescribing system (within the closed loop system) could be set up with warnings for infusion expiry and that the system would allow you to prescribe single bags of fluid and also prescribe fluid to run continuously (allowing the nurse to set up consecutive fluid bags).

Panellist 4

Smart pumps cannot address a prescription that has not been prescribed.

Smart pumps cannot reduce the likelihood that drugs and fluids are continued longer than is intended.

Smart pumps may have the capability to alarm up to 24hrs after an infusion has commenced as a reminder to change a syringe.

All panellists assumed that smart pumps were programmable with soft and hard limits, and would control the rate of delivery through the library or through linkage with an electronic prescribing system.

The panellists also noted that smart pumps were unlikely to prevent all infusion errors:

Delays in treatment cannot be remedied by smart pump technology.

Panellist 4

A great many of those errors are about the wider systems involved in IV medication use.

Panellist 5

Additional comments from panel members are presented below:

Interesting how few of the problems in [the] sample could have been mitigated by a smart pump . . . was it a truly random sample?

As in everything, there is often no one solution to a problem. i.e. smart pumps are not the whole answer.

It would appear from completing this that smart pumps plus closed loop electronic prescribing probably aren’t the entire solution either.

I had difficulty entering ‘Not Preventable’ on the table – this is probably a psychological issue I have in accepting that an error is ‘Not Preventable’ – there is always contributing factors therefore these must be able to be altered in order to prevent an error happening again- and that elusive ‘acceptable level of risk’ is forever changing.

Interestingly, the research team was more conservative in their assumptions about the extent to which a smart pump would be able to prevent errors, whereas the expert panel made more aspirational assumptions. Akin to a sensitivity analysis, if we assume that the reality lies between the two extremes, then this might suggest that 48–73% of the errors that occurred in traditional infusion pumps could be prevented with a closed-loop smart pump. As many of the panel members did not differentiate between standalone smart pumps and closed-loop smart pumps linked to electronic prescribing, it is not possible to provide a similar sensitivity analysis for standalone smart pumps.

When assessing preventability using smart pumps, neither the research team nor the expert panel considered the challenges of building drug libraries and associated hard and soft limits, and the associated disadvantages of selection errors or alert overload. A further limitation is that most of the expert panel members were at the early stage of smart pump implementation and may not have seen the full consequences (intended and unintended) of using smart pumps in clinical practice.

Errors in infusions that were not given via a smart pump

For those 115 errors reported in which it appears that the infusion was not given via a smart pump, the results are summarised in Table 16. Nine were omission errors due to pump unavailability or rate errors in infusions given by gravity and therefore could have been prevented by the use of any pump. For 59 errors, the infusion was given by a pump, but the error would not be expected to be prevented by a smart pump, such as wrong rate errors, whereby the rate was incorrect for that particular patient, but within the usual range for the drug concerned. It was noted that some errors were attributed to staff pressing the button (intended for the patient’s use) on PCA pumps and others were attributed to pump malfunctions. Overall, 35 errors were judged to have been potentially preventable with a standalone smart pump, depending on the limits set in the drug library. A further nine were potentially preventable with a closed-loop system in which the smart pump rate was set via an electronic prescribing system. In three cases, lack of detail or clarity in the incident report meant that we were unable to draw any conclusions. Equivalent data for the subset of errors involving the incorrect rate, dose or volume are also given in Table 16.

Analysis of errors identified in infusions that were given via a smart pump

For those eight errors reported in which it appears that the relevant infusion was given via a smart pump, the results are summarised in Table 17.

In seven cases it appeared that the use of a smart pump had contributed to the error; this largely concerned selection of an incorrect entry from the drug library. In one case, insufficient detail meant that we were unable to assess the case concerned. The seven cases in which the smart pump was judged to have contributed to the error are listed in Box 1. These comprised four ‘wrong dose’ errors and three ‘wrong rate’ errors. All were reported as resulting in moderate harm. In most cases, it seems that the error occurred as a result of the interaction between the user and the infusion pump, rather than either alone.

To estimate the number of infusions given each year in English NHS hospitals, we identified several sources of relevant data. First, NPSA’s Safer Practice Notice 01: Improving Infusion Device Safety74 refers to 15 million infusions being performed in the NHS every year, although it does not specify whether this relates to England, England and Wales, or the whole of the UK, and does not indicate the source of the data. If we assume that the data relate only to England, where there are 130,000 NHS inpatient beds,75 then this equates to 115 infusions per bed per annum. Second, an analysis of electronic prescribing data from Imperial College Healthcare NHS Trust (1500 non-critical care beds) suggested that somewhere between 1000 and 2000 doses are prescribed to be given by infusion in 1 day, which is equivalent to 533 infusions per bed per annum for non-critical care areas. Third, based on IV giving set purchasing data, one of our study sites estimated that 180,000 IV infusions are given each year within their 500-bed trust (including critical care), equivalent to 360 infusions per bed per annum. Taking the median figure of 360 infusions per bed per annum, we estimate that there are 45 million infusions given per year in English hospital inpatient departments. Our NRLS data comprised 123 reports relating to infusion pumps in the inpatient setting that were classed as moderate or above over an 11-year period, which equates to one report per 4 million infusions. We note that this estimate has numerous limitations in that the number of infusions is a broad estimate, and that the NRLS data include Wales as well as England. Nevertheless, it indicates that reported incidents relating to infusion pumps that are at least of moderate severity are very rare.

Summary

The under-reporting associated with all incident reporting systems means that it is not possible to draw firm quantitative conclusions from these data. We were also limited by lack of information in many of the reports, which made it difficult to determine exactly what had happened. In particular, very few reports specified that a smart pump was used; lack of denominator data means that it is not possible to conclude whether errors were less likely to be reported in infusions given via smart pump, whether smart pumps were not in common use during the period studied, whether some incidents did involve smart pumps but this was not specified in the report, or a combination of these points. It was also noted that, in many cases, the researchers questioned the severity rating given by those reporting the error, with nothing in the incident description suggesting actual patient harm for many of the reports. For example, for two of the four incidents reported to have resulted in death, there was nothing in the description to suggest death (or, indeed, any harm).

However, these data do suggest that up to one-third of NRLS infusion pump incidents of at least moderate severity would have potentially been prevented by a standalone smart pump, with use of a closed-loop system increasing this number to 38%. If analysis is restricted to errors involving the wrong dose, rate or volume, then a higher percentage of 41% were potentially preventable with a standalone smart pump, increasing to 51% with a closed-loop system, although these figures are all based on assumptions around the correct set-up and use of smart pumps and their associated drug libraries. In other cases, the use of a smart pump contributed to errors, usually as a result of selecting the wrong entry in the drug library. It is not possible to calculate a net benefit of smart pumps, as no denominator data are available for the numbers of infusions given by smart pump or other kinds of pump.

The work system

The detailed observations of IV infusion medication preparation and administration were arranged into the categories of organisation, physical environment, technology and tools, tasks and people.

Summary of consequences and opportunities for learning

As in phase 1 (see Chapter 4), time was identified as a resource to be managed, and also a force that shaped activities. The practice across many wards of prescribing infusions to start at particular times of day created two kinds of bottleneck. First, there was a bottleneck of IV medication preparation, which required staff to be engaged in a resource-intensive, safety-critical task that was concentrated into a short time, often also in a confined space. Second, this practice meant that there were times of peak demand for infusion devices, which were sometimes in short supply. Both of these bottlenecks meant that staff had to manage resources carefully, but this put stress on staff, stress that they often managed by deviating from local policy in order to manage patient safety and well-being (see also Chapter 4).

The fact that downloading pump log data on alerts and over-rides would have meant taking pumps out of active service, as well as taking up staff time, meant that pump logs were not, as currently implemented, a useful source for learning. They have potential to be a more useful source of learning if log data could be downloaded automatically (wirelessly) and unobtrusively, and if analysis of the log data could be done quickly and simply, with the results reported in a way that is immediately meaningful to staff in terms of areas for improvement.

Staff time was also a resource to be managed, and there was a necessary trade-off between a member of staff being available to work on the ward and that member of staff being upskilled through training. This influenced staff attitudes to incident reporting, in particular: they had to consider not only the direct time taken to report an incident (balanced against the perceived value of doing so) but also the knock-on time cost if it resulted in staff being required to undertake additional training in response.

Another factor that emerged repeatedly, particularly in staff interviews, was interruptions. As outlined above, this was an issue that needed to be managed in both preparation and administration of IV infusions. The strategy of wearing a particular coloured apron was not perceived as effective in limiting interruptions, and the design and use of many IV preparation rooms also limited their effectiveness as a space in which people could work without interruption.

Intravenous infusion administration as a complex adaptive system

Our analysis suggests that IV infusion administration is a CAS comprising nurses, pharmacists, doctors and other clinicians, regulators, drug and device manufacturers, incident reporting mechanisms, national policy and local procedures, stickers, trays and trolleys, and other factors and paraphernalia that shape the system. Different histories in hospital trusts and individual wards set the context for current and future actions, for example the risks for some issues may be more salient as a result of previous incidents at a particular hospital, or involving particular people. However, different subissues might be better associated with simple or complicated systems within this wider CAS. For example:

• A potentially simple issue is advice on labelling IV infusion giving sets. From phase 1 we found that some sites had a high rate of discrepancies because their local policy was to label every giving set. Some nurses felt that it was not worth labelling a giving set for an infusion that would last for only 10–20 minutes. There was broad agreement that there were two reasons for labelling giving sets: (1) when a giving set is to be used for more than 24 hours, the label can help track the time the giving set has been in use, and (2) when a patient has more than one giving set, the label can help differentiate which drug is being given via which giving set. Trusts A and O had local policy that advised labelling giving sets only for continuous infusions and when more than one infusion was in place. This wording better aligns local policy with nurses’ practical concerns, and the two reasons for labelling giving sets. This, therefore, seems a candidate issue for moving towards some form of best practice.

• A potentially complicated issue relates to site D’s ICU, which had arranged for propofol infusions to be pre-made so that nurses were saved the time of preparing them. Some critical care nurses said that it would also be helpful to have pre-made infusions of noradrenaline, but these infusions would not be purchased because of the cost. The matron reported that pre-made propofol was one of the most expensive drugs, and he thought that pre-made noradrenaline would cost more. He would have liked to make these and other changes, but the trust was already under budgetary pressures. Other trusts might have less severe financial pressures and be able to purchase more pre-made IV preparations. The matron would need to weigh these extra drug costs against the time it would save nursing staff and the added preparation accuracy that comes with pre-made preparations. He or she would also need to consider any additional risks, such as how different concentrations of pre-made noradrenaline are stored and labelled so that they were not confused, and how long these preparations are stable once they have been made up. This is a complicated problem, but the different issues are known and could be considered to find a good solution for this issue.

• A potentially complex problem is how to effectively learn from incidents. In the workshop for phase 2 sites, some participants reported that they were not satisfied with the quality of reporting of incidents, whereas other participants were generally happy. The form of the report was discussed, as was off-site access, and the benefits of giving fast, individual feedback to the person who completed the incident report. However, it was implied that a ‘culture of reporting’ was a prerequisite for coming to a meaningful understanding of errors and near-misses for learning. There were factors that may discourage staff from reporting errors, such as having to be retrained. One participant reported that a nurse requested an error not be reported because he or she was worried about the repercussions. On reflection, site staff who were not happy with the quality of their incident reports saw it as a simple issue caused by the design of the reporting forms. However, the site that was more satisfied seemed to treat this as a complex issue involving many other dependencies and a positive culture that affected staff motivations and behaviour.

There may also be different perspectives on the same issue, which depends on the scope of variables and the nature of relationships one is able or willing to consider. For example:

• Double-checking IV infusions could be considered a simple problem as no one is infallible and, intuitively, two checks are better than one. Some nurses reported positive reinforcement, as double-checks had caught errors they might otherwise have missed.

• Double-checking could be considered a complicated problem because it is not well defined. In phase 1, local policies and procedures varied, as did the staff’s perceptions. For example, the pharmacists who helped draft the policy at site P said that their intention was that nurses who double-checked IV infusions were signing that the right medication had been given to the right patient at the prescribed rate and time; this also included checking the pump settings. However, the nurses presumed that they were only signing that the preparation of the IV infusions was correct, perhaps encouraged by physically signing the additive label. Only site I had detailed advice on what was expected from an independent check. Such issues relate to known unknowns, which can be clarified, defined and tested to assess their effectiveness.

• Double-checking could be considered a complex problem because nurses adapt their behaviour depending on the informal risk assessments they perform in practice. For example, one nurse was observed casually asking another nurse to check an infusion of paracetamol that they had already hung up, whereas the same nurse carried out a more thorough check of an infusion they had not administered for a while. These decisions are also affected by the culture on the ward and by role models who influence what is seen as acceptable behaviour. Staffing levels also have an influence; for example, it is challenging when there are only two registered nurses on a night shift to carry out full independent checks together for 10–30 infusions. Furthermore, there is a risk that double-checking may cause more errors as a result of the first checker being less diligent because they expect their work to be double-checked. These situated adaptations, cultural influences and feedback loops are complex and hard to track.

Overall, IV infusion practice can be considered a CAS, which may have subcomponents that are influenced by the broader context.

Revealing informative variability by contrasting neighbours

Through the sites that we have engaged with, we have found that IV infusion practice in England is very variable both within and between trusts. Previous phases of our research have alluded to the different types of pumps, policies and procedures at different hospitals. However, variability also emerges from the different interactions that happen in different wards, clinical areas and hospitals. Some of this variability shows how the system, and the people in it, self-organise to try to improve quality and safety or to optimise other aspects of working practices. We illustrate some of this variability in our data by contrasting similar situations across different wards that were observed, or the same ward under different circumstances (e.g. different times or members of staff). We present this under five themes of social factors, artefacts (including technology, medications and equipment), physical space, information flow, and the wider organisational context.

One size does not fit all: context is everything

During the project we engaged with different clinical areas. At one level, IV infusion practice might not seem very different between these areas; after all, in each area a medication is prescribed, prepared and administered following the local policies and procedures. However, there are particular characteristics and contextual dependencies that affect emergent behaviour in the different clinical areas. We reflect on typical staffing levels, patient dependency, treatment timescales, key IV infusion practice characteristics, involvement of different clinicians, and key issues for each area in the following sections.

Implications for national policy

We identified a high degree of variability in trust policies to support IV administration, both within and between hospitals. Although some variation is likely to be appropriate based on the needs of different clinical settings, our findings suggest that there may be scope for greater standardisation to support consistent approaches. This would be likely to benefit patients who expect the same standards of care across the NHS, as well as staff who may move between hospitals.

• Our data suggest that national advice and guidance may be helpful in some areas, such as whether or not small-volume flushes need to be prescribed and, if so, how; labelling requirements for IV infusions and giving sets; and requirements around double-checking, if any, at each stage of preparation and administration.

In line with research in other areas of health-care practice, our qualitative data suggest that there is variation between hospital trusts in the culture of learning from errors and deviations.

• Our data suggest that practices around the preparation and administration of IV infusions would benefit from being included in ongoing efforts to encourage a positive culture around reporting and learning from errors.

• In line with earlier work on CAS in health care, our data show that IV infusion administration cannot be readily reduced to simple problems that have simple solutions (e.g. introducing smart pumps), and that a more nuanced approach to improving safety is likely to deliver greater benefits, including sharing best practices across and within NHS trusts. Examples of practices that had been adopted in some trusts are employing a pharmacy technician to help nurses with IV preparation and administration, introducing a linked pump and EPMA system, removing a radio from the treatment room and trying to improve learning mechanisms and a safety culture.

Use of infusion pumps and other technologies

Although smart pumps are often advocated as making IV infusions safer in health-care practice, our findings do not support this, with error rates similar for infusions given via smart pumps and those given via traditional pumps. Although smart pumps may have a role in preventing very rare and serious errors, there was no difference in the prevalence or types of error identified between infusions given via smart pumps and those given via traditional pumps in our study. Instead, our findings suggest that the use of any pump is safer than relying on gravity administration. Ensuring that sufficient numbers of traditional pumps are available, taking into account the peaks and troughs of demand at different times of day, may therefore represent a better use of resources than purchasing smart pumps. Our findings suggest that:

• gravity administration introduces vulnerabilities, particularly for safety-critical infusions

• a shortage of pumps forces staff to find workarounds, particularly when there are peaks and troughs of demand at different times of day

• there is not yet clear evidence that the costs of implementing standalone smart pumps, including the ongoing costs such as maintaining the drug library, always compare favourably with the costs of alternatives, such as traditional pumps or developing and maintaining integrated smart pumps (interoperating with EPMA, etc.), or with the costs of other patient safety initiatives.

When smart pumps are used, our data suggest a number of associated challenges. In particular, drug libraries are time-consuming to construct and difficult to manage, and were sometimes found to be out of date. In some cases, pump memories were insufficient to permit comprehensive drug libraries to be loaded. One-third of smart pumps observed in our study offered no advantage over standard pumps because of incomplete drug libraries. Additional problems were found to arise when a change in supplier resulted in changes in the strengths or volumes of the products concerned. Error prevention is likely to depend on exactly how drug libraries, and the associated limits for hard and soft alerts, are set up and used. Our findings also suggest that benefits and efficiencies may be possible through collaboration in setting up and sharing drug libraries, associated experiences and good practice.

• Our findings suggest that health-care organisations may benefit from collaborating, ideally on a national basis, to share practice in the set-up and use of drug libraries. Existing networks, such as those for medication safety officers or medical device safety officers, could support such an activity.

• Health-care organisations might review whether drug libraries should include all fluids or only those that are considered high risk (based on agreed criteria).

As well as the challenges with smart pump drug libraries as above, our observations and interviews highlighted some poorly defined user interfaces and displays on infusion pumps, which created challenges for staff in setting up and using them. We also found that when smart pumps were used, organisations were rarely downloading or analysing pump log data, as this was perceived as too difficult and/or time-consuming.

• Our findings suggest that future procurement decisions for IV pumps could usefully include user testing and comparing experiences at other hospitals as well as, for smart pumps, fully exploring memory capabilities for drug libraries, the ease with which these can be set up and updated, and the ease with which pump log data can be downloaded and analysed to deliver actionable insights.

An analysis of our phase 1 quantitative data and of data obtained from the NRLS suggests that many additional errors may be preventable by using systems in which smart pumps are integrated with EPMA systems. Site D’s ICU had such a system but we do not have enough data to determine whether it was statistically significantly different from ICUs that did not have this system. While cautious about making assumptions about the capabilities of integrated systems, and recognising that these systems are likely to bring their own challenges, it seems likely that the potential for smart pumps to improve patient safety may be more fully realised with integrated systems.

• There is some evidence to suggest that the use of systems integrating EPMA systems and smart pumps might be suitable in the English NHS. Further research is needed to evaluate whether or not, and under what circumstances, their potential is borne out in practice.

The advent of EPMA systems, while having many potential benefits, was identified as also introducing some challenges in allowing simultaneous access to the three elements required for safe administration: the medication order, the medication and the patient. With EPMA, details of the medication order may be on a computer that nurses cannot take to the patient’s bedside when infusions are being set up. A similar issue arises if a computer is not available in the room being used to prepare IV infusions. We also identified issues with interoperability, in that nurses sometimes had to ‘translate’ medication details between EPMA, smart pump libraries and the medication itself if each used different terminology.

• Our data suggest that health-care organisations using EPMA could usefully consider how best to make details of the medication order available at the bedside and in treatment rooms.

Local policy and practice

Our data suggest that procedural and documentation deviations sometimes represent discrepancies between official policy and everyday practice. In some cases, therefore, it may be more beneficial for managing risk to have policies that better reflect the realities of existing practice than to enforce compliance with policies that are not achievable in clinical practice. For example, policies allowing the administration of small-volume 0.9% sodium chloride flushes without a medication order in specific circumstances or for specific patient groups, an approach already in place in many trusts, could be introduced in hospitals where unprescribed flushes are currently accepted as standard practice by clinical staff but are not permitted according to local policy.

• There is evidence that IV infusion policies work best when they are aligned with the realities of practice. This may best be achieved by ensuring that they are written with input from practising clinical staff. One specific example is that realistic policies are needed in relation to small volumes of 0.9% sodium chloride as a flush to avoid these having to be prescribed individually.

We identified locally appropriate practices, such as greater involvement of pharmacy staff in preparing and administering IV infusions, and employment of a specialist infusion pump nurse to assist with managing and updating drug libraries, training and liaison with the pump manufacturer. Staff at some sites had interpreted current legislation and the guidance of professional bodies as precluding the administration of medication by health-care professionals other than doctors and nurses.

• Evidence from the ECLIPSE study highlights the potential value of greater sharing of good and novel practices around IV infusions for others to consider adopting or adapting. As above, this could be done through existing networks, such as those for medication safety officers or medical device safety officers.

As well as in their content, considerable variation was identified in terms of the style, length and readability of policy documents relating to IV infusions, many of which were difficult to locate on trust intranet systems and/or difficult to read. These problems were observed to lead to policy documents not being used or not being followed as intended.

• Finding from the ECLIPSE study suggest that health-care organisations would benefit from reviewing how policies and associated guidance documents are accessed, indexed, structured and written, with input from end-users to ensure that they are useful, usable and used in practice.

• It might be valuable to develop and disseminate standards on best practice in developing policy documents and making these easily accessible to staff.

We noted that the physical space in which health-care professionals work shapes their activities. For example, we identified instances when rooms used to prepare infusions were multipurpose and cramped, and had doors that did not work.

• Findings from the ECLIPSE study suggest that when wards are renovated or new facilities are designed, work practices might be better supported if these are taken into account in the design in order to aid access to all equipment needed while minimising distractions during the preparation and administration of medication.

Patient involvement

Although patients generally felt that their care in relation to IV infusions was safe, some suggested that their role in providing key information was not always drawn on by health-care professionals. Frequent pump alarms were a source of anxiety as well as interrupting patients’ sleep, adversely affecting patients’ experiences. Although inpatients are generally advised not to interact with infusion pumps, the resulting irritation from frequent alarms means that some do interact with them. Some patients also expressed an interest in being able to view what medications they are prescribed and why, which may be more challenging to provide with EPMA than with paper medication charts. Evidence from our study indicates that there are potential benefits from:

• Ensuring that patients are informed and involved with their infusion therapy to the extent that they wish to be, and that the patient is recognised as a key source of information about their previous treatments and any associated problems.

• Informing patients of the purpose of pump alarms and what to do when the alarms sound. There may also be benefits in reviewing the settings of pump alarms so that these are configured in a way that is appropriate for the patient.

• Considering how to best support patient involvement and access to information about their inpatient medication, particularly for hospitals using EPMA.

Unauthorised medication/no documented medication order