Background Acute care teams (ACTs) represent action teams, that is, teams in which members with specialised roles must coordinate their actions during intense situations, often under high time pressure and with unstable team membership. Using behaviour observation, patient safety research has been focusing on defining teamwork behaviours—particularly coordination—that are critical for patient safety during these intense situations. As one result of this divergent research landscape, the number, scope and variety of applied behaviour observation taxonomies are growing, making comparison and convergent integration of research findings difficult.
Aim To facilitate future ACT research by presenting a framework that provides a shared language of teamwork behaviours, allows for comparing previous and future ACT research and offers a measurement tool for ACT observation.
Method Based on teamwork theory and empirical evidence, we developed Co-ACT—the Framework for Observing Coordination Behaviour in ACT. Integrating two previous, extensive taxonomies into Co-ACT, we also suggested 12 behavioural codes for which we determined inter-rater reliability by analysing the teamwork of videotaped anaesthesia teams in the clinical setting.
Results The Co-ACT framework consists of four quadrants organised along two dimensions (explicit vs implicit coordination; action vs information coordination). Each quadrant provides three categories for which Cohen's κ overall value was substantial; but values for single categories varied considerably.
Conclusions Co-ACT provides a framework for organising behaviour codes and offers respective categories for succinctly measuring teamwork in ACTs. Furthermore, it has the potential to allow for guiding and comparing ACTs study results. Future work using Co-ACT in different research and training settings will show how well it can generally be applied across ACTs.
- Human factors
- Medical emergency team
- Patient safety
Statistics from Altmetric.com
Improving patient safety is a major concern in healthcare. There is abundant evidence that human factors play a central role in improving patient safety, especially in high-risk domains such as emergency medicine, surgery and anaesthesia.1–3 In particular, communication and coordination within these healthcare teams are critical processes that can either diminish or promote patient safety.1 ,4 Studying these team processes requires behaviour observation.5 ,6 Behaviour observation helps to identify which teamwork processes are associated with effective and safe performance.7 Compared with self-reports, behaviour observation allows for measuring actual team-level phenomena and dynamics that teams may not even be aware of and that may unfold over time. Thus, it provides us with new knowledge on what effective teams do differently compared with ineffective teams and how they do it.5 For example, behaviour observation studies on leadership in acute care have shown that leadership behaviour is positively associated with performance during critical and barely standardised situations, but negatively associated with performance during routine and highly standardised situations.8 Even during short lasting critical situations such as cardiac arrests, it is the appropriate timing of leadership behaviour that defines whether leadership is helpful or not (within the first 30 s after entering the scene).9 Likewise, behaviour observation research on mastering diagnostic challenges in acute care situations has shown that doctors can fall into four modes of problem-solving (fixated, stalled, vagabonding, adaptive)—only one of which is actually effective (adaptive mode).10 All of these findings are essential for understanding what characterises effective teamwork in high-risk healthcare environments. None of them could have been obtained without behaviour observation.
Using the genuine advantages of behaviour observation, patient safety research has investigated teamwork behaviours that are particularly effective for promoting team performance. Especially in the last decade, teamwork behaviours such as leadership, coordination and communication have been studied in a variety of high-risk healthcare contexts using behaviour observation.4 ,11–13 As one side effect, the number, scope and variety of behaviour taxonomies used for this research are growing. Even for observing acute care teams, almost a dozen taxonomies exist.7 ,13–22 This divergence of observation measures reflects the current divergence of the field but complicates the comparison and convergent integration of research findings.
In this paper, we present a theoretically as well as empirically driven framework for integrating teamwork behaviour taxonomies for acute care teams. This structure is intended to serve as meta-taxonomy that facilitates comparison, integration and convergence of findings from different observational studies.
To begin with, we will review the relevant literature on acute care team functioning and identify three methodological challenges. Second, we present the development of the Framework for Observing Coordination Behaviour in Acute Care Teams (Co-ACT) and the integration of two extensive behaviour taxonomies resulting in the suggestions of 12 succinct behaviour codes, which can serve as categories for observing acute care teams. Third, we present this taxonomy and the results of a reliability test of these codes and discuss the strengths, current limitations and respective further research needs. With this framework, we aim to foster research and training on teamwork in healthcare, eventually leading to improvements in clinical performance and patient safety.
Behavioural science applied to acute care teams
Teamwork represents a major factor contributing to medical errors, in surgery second only to lack of competence.1 ,4 ,23–25 This raises the question as to which best practices of teamwork allow healthcare teams to work effectively, to prevent errors and to improve patient safety. Research investigating this question looks at ACTs, that is, healthcare teams in high-risk, dynamic disciplines such as emergency medicine, surgery and anaesthesia. These teams represent action teams; that is, teams of highly skilled specialists who work together for brief performance events that require flexibility and improvisation in an unpredictable context, often under high time pressure and with unstable team membership.11 ,26–28 Consequently, teamwork in action teams only rarely focuses on long-term goal achievement or on team building. Instead, teamwork mainly serves the purpose of coordination during brief performance sequences, for example during the induction of general anaesthesia.11 ,27 ,29–31 Particularly during critical, rapidly deteriorating situations (eg, unexpected ‘cannot ventilate, cannot intubate’ crisis during induction of general anaesthesia), team coordination behaviour is the most central coordination tool available in addition to respective treatment algorithms—and even algorithms require coordination with respect to who will perform which task.
As behaviour observation allows for investigating what effective teams do differently compared with ineffective teams and how they do it,5 it is the method of choice when identifying teamwork behaviours that are effective in such intense situations. Using behaviour observation, specific teamwork behaviours could be associated with high clinical performance in various scenarios such as an induction to general anaesthesia, the management of a cardiac arrest or the handling of a difficult airway situation.2 ,8 ,9 ,29 ,32–34 These studies have demonstrated that teamwork behaviour is no triviality and is indeed critical for performance, and provided the equally important empirical basis for training interventions. However, as an undesirable side effect, the number and scope of behaviour taxonomies are increasing, making comparison and integration of research findings—and thus drawing convergent conclusions—increasingly difficult. Current and future research on teamwork in healthcare is at risk to become too divergent and the co-existence of a variety of behavioural taxonomies impedes a coherent research approach. The aim of the current study is to provide a remedy for this situation and to make a step towards convergence of research findings. In order to advance applied behavioural research aiming at improving patient safety by investigating best practices of teamwork, it is time for converging research findings and promoting a shared research language by developing and using a behaviour framework. This framework needs to be based on theory as well as empirical evidence. If behaviour observation were more strongly based on theory and empirical evidence, this would also help to meet three methodological challenges that come along with aiming to identify and promote best practices of teamwork processes in ACTs.
Three methodological challenges
Reviewing the landscape of current literature on ACT functioning, it becomes apparent that there are three challenges involved in studying teamwork by means of behaviour observation (table 1).
The first challenge refers to identifying the optimal balance between specificity versus generalisability and represents a variant of the classic bandwidth-fidelity dilemma.35 Researchers must decide whether to investigate team processes from a general perspective using methods that capture all teamwork behaviours simultaneously or whether to focus on one single process (eg, closed-loop communication) to explore it in detail.
The second challenge refers to deciding whether to rate the quality or describe the occurrence of teamwork behaviour. For example, further data analysis possibilities and limitations will be greatly influenced by whether researchers evaluate how well closed-loop communication was performed using a behaviour-anchored rating scale ranging from 1 to 5 or whether they measure when, by whom, to whom and with which words closed-loop communication was performed.
The third challenge refers to the linking of research findings with team training content. As different labels for the same behaviour hamper the development of a common language of teamwork, the translation of research findings into straightforward training programmes is equally impeded. The continuous transfer, update and feedback, between research and training, needs to be promoted by shared language of relevant teamwork behaviours.
These challenges illustrate that future research on effective teamwork in ACTs needs a theoretically and empirically backed-up framework of relevant teamwork behaviours (table 1). Such a framework would guide observational ACT research in a way that those behaviours which are measured are indeed critical for performance,36 and would allow for directly comparing and integrating findings and advance our understanding of ACT functioning and effectiveness. We will show that there is sufficient theory and empirical evidence for developing a content-valid framework. Using this theory and empirical evidence we have developed Co-ACT, the Framework for Observing Coordination Behaviour in Acute Care Teams. Co-ACT can direct future observational research on ACTs by pointing out which behaviours to observe.
Development of Co-ACT
The development of Co-ACT involved three steps which are illustrated in figure 1. In Step 1, we selectively reviewed the literature on action teams in high-risk work environments with respect to theoretical concepts and empirical findings and developed the basic structure of the framework choosing a model of team coordination in healthcare action teams27 as a theoretical foundation. In Step 2, we aimed at construct validating this structure by comparing it with two similar but more extensive taxonomies14 ,15 and by integrating their codes into the Co-ACT framework. In addition, using ACT literature we also integrated those behaviours that were empirically shown to be positively associated with team performance into the new framework (see online supplementary table S1).3 ,14 ,15 ,29 ,32 ,34 ,38–42 ,45 This resulted in an initial map of the four quadrants including the codes found in the literature assigned to the respective quadrant. In Step 3, based on the idea that Co-ACT should provide a tool for pragmatic and feasible coding, we merged the existing codes into three categories per quadrant (see online supplementary table S1).13 ,46 ,47 This process involved extensive debate among the authors.
Testing suggested Co-ACT codes for reliability
The 12 suggested categories of the final version of the Co-ACT framework were tested for reliability using two videotapes of complete inductions of general anaesthesia at a large teaching hospital. The data have been used in previous studies.29 ,34 Local institutional ethics committee approval as well as written consent by all participants and patients was obtained prior to data collection. The two anaesthesia teams consisted of one anaesthesia resident and one anaesthesia nurse. One consultant anaesthetist was available on stand-by. Video and vital parameter recordings were obtained using a set-up allowing synchronised recording and DVD playback of video, monitor and ventilator data. Co-ACT codes were applied using Interact software (Mangold International GmbH, Arnstorf). The inductions lasted about 20 min. Parsing (ie, defining the coding units) and coding were performed simultaneously: Watching the videotape, the coders marked the beginning (eg, defined the onset time when a nurse anaesthetist began providing information without request) and ending of a code (eg, defined when the nurse finished proving information) for each anaesthesia team member. The resulting coded events contained information about the Co-ACT's categories (eg, providing information without request), its timing (eg, beginning, end and duration; for example 3.5 s) and the acting person (eg, nurse anaesthetist). Two coders independently applied Co-ACT for coding the two videotapes and each coder spent between 20–25 h coding the behaviour of all team members on both videotapes. In accordance with others, the following procedure was applied48: After the first coder had coded the complete videotape using Interact, the first author made a copy of the respective Interact file which included only the coding units, but not the codes. She gave this file to the second coder, who encoded the predefined coding units with Co-ACT. This procedure was applied in order to avoid a confounding reliability testing of the taxonomy with that of the parsing.49
Results of literature review on coordination behaviour in ACTs
Theoretical conceptualisation of coordination behaviour in ACTs
For ACTs, which exist for brief performance sequences only and therefore rely on minute-by-minute coordination, literature on teamwork theory suggested that looking at two distinct team coordination dimensions is most promising: The task-specific coordination content (ie, action vs information management) and the coordination mode (ie, explicit vs implicit).50–53 For example, the model of team coordination in healthcare action teams proposes four styles representing the different combinations of these two dimensions (figure 2): (1) Explicit action coordination includes communication that aims at explicitly, unambiguously and directly coordinating joint actions, for example by giving instructions. (2) Implicit action coordination includes coordination-related behaviour and communication that facilitates action coordination via mutual anticipation, for instance by offering assistance without request. (3) Explicit information coordination includes communication that aims at explicitly managing information transfer and processing in order to gain mutual understanding, for instance by requesting information. (4) Implicit information coordination includes coordination-related behaviour and communication for tacitly managing team information processing, for example by providing task-relevant information without request.27 As most of the research on ACT functioning looks at teamwork with respect to explicit and implicit action and information coordination,3 ,9 ,29 ,32–34 ,45 ,54 this model provided an optimal theoretical basis for structuring, comparing and integrating existing behaviour taxonomies that are currently used for analysing the relationship between teamwork and performance in ACTs.
Empirical evidence of coordination behaviour in ACTs
Along with recent theoretical advancements, more and more empirical evidence of the importance of teamwork behaviours for performance of ACTs has been established (see online supplementary table S1). Examples of these behaviours are: team member monitoring, giving clear instructions and talking to the room. We have listed these behaviours representing the empirical foundation in online supplementary table S1 and assigned them to the four styles of ACT coordination (figure 2). For example, there was empirical evidence that talking to the room is important for ACT performance in command and control,39 medical emergencies3 ,55 and in anaesthesia.34 ,45 Likewise, there was growing empirical evidence on the relevance of speaking up behaviour in ATCs in aviation56 and healthcare such as the operating room38 and anaesthesia.57 ,58 That is, the studies listed in the fifth column of online supplementary table S1 provided the empirical basis of the framework of coordination behaviour in ACTs.
Summarising, there was ample theory and empirical evidence that allowed for developing the content-valid framework that is required for guiding observational research on ACT functioning. The model of team coordination in healthcare action team (figure 2) and the results of previous observation studies (see online supplementary table S1) provided the theoretical basis and empirical evidence, respectively, for the Co-ACT framework.
Co-ACT—an observation framework based on theoretical and empirical evidence of coordination behaviour in ACTs
The final Co-ACT framework consists of four quadrants (explicit action coordination, implicit action coordination, explicit information coordination, implicit information coordination) for each of which we suggest three categories (figure 3). Please see online supplementary table S1 for a detailed description of all categories including examples. Further codes (eg, acknowledgement) can be added as amendments.
Reliability of suggested Co-ACT codes
Overall, 266 events were coded. Analysis of Cohen's κ showed a mean value of κ=0.77, representing substantial strength of agreement.59 The category ‘information request’ (κ=1.00) achieved the highest and the category ‘planning’, which occurred least often (n=3), (κ=0.33) the lowest reliability (see online supplementary table S1). With the exception of ‘planning’ and ‘information evaluation’, κ values were higher for the explicit than for the implicit categories. However, κ values are affected by prevalence; they are at risk of being low despite high proportions of observed agreement due to a substantial imbalance in the concordance table's marginal totals, and conversely to be high due to an asymmetrical imbalance in marginal totals.60 We therefore report the observed proportions of positive (ppos) and negative (pneg) agreement in addition to the κ values (see online supplementary table S1).61
Our goal was to advance observation of ACTs for research and training purposes by proposing Co-ACT—a theoretically and empirically driven framework of coordination behaviour (figure 3). This framework aims at facilitating the observation of ACTs in a consistent manner and at comparing and integrating divergent findings from different ACT observational studies. Thus, Co-ACT provides a conceptual link between ACT teamwork behaviours and team performance. As an orientation guide in the current and future landscape of ACT observation research, it addresses methodological challenges of this field (table 1).
The development of Co-ACT is based on the integrated coordination model for action teams in healthcare27 and on empirical evidence of teamwork behaviours for ACTs. While developing Co-ACT, we integrated two previous behaviour taxonomies, a process that finally resulted in 12 categories organised along the basic, two-dimensional structure of Co-ACT. These categories should be considered as suggestions, which—based on our experience—can serve relatively well for measuring coordination behaviour in most clinical and simulated situations of ACTs. Of course, based on the specific research question, modifying some of the suggested categories would be appropriate. For instance, if the quality and impact of ‘speaking up’ behaviour were to be measured, refining Co-ACT's ‘speaking up’ category with respect to the speaking-up content (eg, suggestion vs opinion vs problem)62 would be useful. This refined category would retain its position within the Co-ACT framework, that is, it would still represent explicit action coordination behaviour.
Evaluation of Co-ACT with respect to reliability
We have applied the Co-ACT categories to coding teamwork during induction of general anaesthesia. Cohen's κ value for the overall taxonomy was substantial, but κ values and proportions of positive and negative agreement for single categories varied considerably. With the exception of ‘planning’ and ‘information evaluation’, reliability was higher for explicit than for implicit categories. This finding corresponds with findings indicating that implicit coordination categories are usually less reliable than explicit coordination categories and may reflect the very nature of implicitness—that is, being difficult to observe.14
Concerning problematic Co-ACT categories, ‘information-related talking-to-the-room’ and ‘information without request’—both implicit codes with low κ values—may indeed be difficult to discriminate from each other. Particularly in the setting of two-to-three-person teams, it may be hard to assess whether information was provided to the room at large or to a specific person. Furthermore, observing ‘monitoring’ is not an easy task, as it may be difficult to assess whether a team member is really observing his or her colleagues or just looking around; in particular in a clinical setting.
The explicit categories ‘planning’ and ‘information evaluation’ also had a low inter-rater reliability. Their low rate of occurrence may at least in part account for that result. Categories that already obtained satisfying reliability are: ‘instruction’, ‘speaking up’, ‘information request’, ‘information upon request’, ‘action-related talking-to-the-room’ and ‘gather information’.
Our procedure for testing the reliability of the taxonomy but not of the parsing (ie, defining coding units) aimed at avoiding a confounded measurement of both processes. One may argue that this procedure leads to overemphasising of the reliability, as during most event sampling procedures coding and parsing are performed simultaneously: if a coding system does not allow for defining clear parsing, the coding will inevitably be impaired. However, the procedures for parsing reported in the literature vary greatly for different kinds of event and time sampling, providing no gold standard for this process.3 ,9 ,48 ,63 ,64
Three reasons for using Co-ACT
The first reason for using Co-ACT is its specificity for ACTs. Addressing the specificity versus generalisability challenge of observation-based ACT research (table 1), it captures those behaviours that are particularly relevant for ACT performance (eg, team member monitoring) while disregarding behaviours that are relevant in longer-lived teams (eg, team building and development). The decision about which behaviours are relevant in ACTs was made based on teamwork theory and empirical evidence. Still, further research needs to show the validity of Co-ACT for being applied across ACTs.
In turn, Co-ACTs specificity could potentially allow for generalisability across tasks occurring within ACTs. For example, the intubation task, which we studied here for reliability testing, occurs during routine anaesthesia as preparation for scheduled surgery, and can occur in other ACT settings such as emergencies and intensive care. Hence, Co-ACT could allow for comparing and transferring research findings with respect to different ACT tasks. It could also allow for comparative studies with respect to different ACT disciplines. For example, one could investigate whether the effectiveness of specific behaviours such as speaking up varies between surgery and anaesthesia. Moreover, since situational and task characteristics define most of the team's coordination demands,52 ,55 one could investigate to what extent the relationship between ACT behaviour and ACT performance is contingent upon temporal and situational factors and how ACTs coordinate adaptively using Co-ACT. For example, implicit information coordination may be appropriate during routine situations but not during a complex crisis with ambiguous information.55 Also, in combination with measuring situational and organisational factors such as team psychological safety, Co-ACT could be used to investigate what facilitates the actual occurrence of safety-promoting behaviours such as speaking up.56 ,57 ,65
This leads to the second reason for using Co-ACT—its potential to serve as a point of reference. Due to its applicability within ACTs it could be used to translate findings from one setting to another as well as from research to training, which helps to deal with the challenge of linking ACT research findings and training content. For example, if using Co-ACT for observation-based research on determining how effective ACTs adapt their coordination behaviour from routine to crisis situations, these findings could straightforwardly be integrated into ACT training, again using the Co-ACT framework.
The third reason for using Co-ACT is that it allows for assessing the occurrence and timing of coordination behaviour, thus providing the basis for a detailed understanding of team interaction. Addressing the rating versus describing behaviour challenge of observation-based ACT research (table 1), this represents a significant advantage compared with existing static taxonomies of healthcare teamwork, in which the assumed quality of behaviours is rated or self-assessed as an overall score for a given episode. Co-ACT can be used to study the dynamics of the coordination process, revealing insights into the immediate functions of specific behaviours for the ongoing team interaction and performance. In particular, we consider the individual-level measurement that Co-ACT provides important for two reasons: First, interaction in ACTs is very dense and each team member is continuously engaged in coordination and/or task work—making team-level measurement almost impossible. Second, individual-level measurement allows for certain data analysis strategies on the team level such as role-specific lag sequential analysis, pattern analysis, network analysis and multilevel analysis, all of which transfer individual-level data to richer and more elaborate team-level data of team coordination than any team-level measurement would allow for.66
Future research needs
We hope that we have stimulated the use of Co-ACT for analysing coordination in ACTs. Although we regard Co-ACT as a promising method, future research is required to improve its quality and scope of applicability.
First, future research is required with respect to improving the reliability of some of the suggested categories as they are behaviour measures that are frequently used. Categories that achieved low κ values (eg, ‘planning’, ‘monitoring’) may have to be defined more unambiguously, providing better behavioural anchors and examples. Reliability should be tested in other settings (eg, time sampling; live observation). Methodical advances are necessary to better capture implicit coordination in teams and investigate whether high-quality measurement of implicit coordination will require a combined assessment of observable behaviours and underlying team knowledge.67
Second, with respect to Co-ACTs generalisability across ACTs, future research should test the taxonomy in a variety of ACTs of different sizes performing different tasks. This will be essential for validating Co-ACT for ACTs in other healthcare settings such as surgery or emergency care. Findings from these studies could then be used to compare different medical teams in terms of their coordination requirements. This comparison would enhance our understanding of the similarities and differences between ACTs and, in turn, facilitate interdisciplinary cooperation.
A third need for improvement refers to the analysis of Co-ACT's feasibility, which was not tested in this study. Feasibility may be best studied by researchers and subject matter experts who were not involved in the development of the method, allowing for a more independent evaluation. For example, studies are needed to test how much training is required for applying Co-ACT and whether Co-ACT is equally suitable for being used by researchers and trainers. Although we assume that the simple two-dimensional structure of Co-ACT in combination with the limited number of suggested categories should facilitate its use, this needs to be explicitly tested. Likewise, Co-ACT has so far been used only for coding videotaped data but not for live coding. As other taxonomies used for live observation include more codes than Co-ACT, we assume that Co-ACT has the potential to provide the observer with a manageable set of codes for live behaviour observation.17 However, for more definite conclusions on the feasibility of Co-ACT it will be essential to test it comprehensively for live coding without videotapes.
Despite those future research needs, we believe that applying Co-ACT can serve as a framework that facilitates the comparison and integration of findings from different observational studies and that its suggested 12 behavioural categories are useful for observing teamwork behaviour in ACTs. We wish that Co-ACT's intended broader applicability would foster the exchange between researchers in different healthcare domains (eg, surgery and anaesthesia) as well as between researchers, training experts and practitioners. Beyond the scope of the Co-ACT taxonomy we hope to stimulate the scientific discussion on behavioural observation methods and the potential usefulness of behaviour frameworks that are theoretically and empirically founded.
We would like to thank Gudela Grote and Donat R Spahn for their support of this project, Barbara Künzle, Enikö Zala-Mezö and Johannes Wacker for their help with collecting the data used here, Silja S Sollberger and Raphael Agosti for their support in video editing and data coding. We thank Johann C. Weichbrodt for his helpful comments on an earlier version of Co-ACT. We also thank Margarete Boos, Carl Schick and Nadine Bienefeld for their helpful comments on an earlier version of this paper.
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
Files in this Data Supplement:
- Data supplement 1 - Online supplement
Contributors TM was responsible for the original study idea. MK and MJB developed Co-ACT and all authors contributed to its revision. MK was responsible for data analysis. MK and MJB drafted the manuscript and all authors contributed to its revision.
Funding This research was funded by the Swiss National Science Foundation (SNF 100013-116673/1; Principal Investigator: Gudela Grote, Co-Investigators: Donat R Spahn, Tanja Manser). The views in this article represent the opinions of the authors and do not necessarily reflect the views or position of the Swiss National Science Foundation.
Competing interests None.
Ethics approval Institutional Ethics Committee of Canton Zurich, Switzerland.
Provenance and peer review Not commissioned; externally peer reviewed.
Data sharing statement The raw data obtained in this study are the property of ETH Zurich and the Institute of Anaesthesiology of the University Hospital Zurich.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.