Statistics from Altmetric.com
The hospital work environment is complex with many competing demands on doctors’ and nurses’ time and cognitive capacity. Clinicians use various strategies to handle workload demands,1 two of which we examine in detail: task-switching and multitasking. Both may be used in response to a stimulus, for example, being asked a question. A clinician can respond by pausing the current task to answer the question or continuing the task while concurrently answering the question, that is, he or she can multitask.
Task-switching is a strategy to deal with what is variously defined as an interruption. Interruptions are known to be an implicit part of clinical work.2–7 Evidence of their cognitive burden has been documented in experimental psychology8–10 and aviation,11 where associations between interruptions and increased error,9 stress9 ,11 and task completion time8–10 have been reported. Such outcomes in the clinical setting could have severe repercussions, and among studies examining interruptions to clinical work there are several reporting negative effects.12–14
Despite this evidence, the relationship between interruption and error in hospitals appears complex. Not all studies of this relationship found an association,15 and several review papers discuss the overly simple assumption of a straightforward cause and effect pathway.16 ,17 Interruptions can have a positive effect on patient safety and on efficiency.16–18 Also, some studies suggest that clinicians use strategies like task-switching to deal with competing workload demands.1 ,19 Indeed, there are many factors in the clinical environment that cause a broad spectrum of workflow discontinuities, from a simple question to a medical emergency, and the way in which clinicians manage these is not well understood. While multitasking has also been implicated as a source of cognitive load and a cause of error in other professions,20 ,21 it too is central to clinical workload management, yet few studies have examined the role and impact of this strategy in detail4 ,22 whether negative23 or otherwise.
Thus, it is important to understand the role of task-switching and multitasking in workload management and what drives these work patterns. We aimed to provide a detailed characterisation of clinicians’ work practices by addressing three main objectives: (1) to describe rates of task-switching and multitasking, (2) to identify factors that are associated with clinicians using one strategy over the other and (3) to examine to what extent strategy is determined by external as opposed to individual factors.
Data were derived from three studies conducted in a 400 bed metropolitan public teaching hospital in Sydney, Australia. The first dataset relates to observations of clinicians’ work within the emergency department (ED) which had an average of 75 presentations per day. Participants comprised seven interns, 22 residents, six registrars and five physicians.24 The other two datasets relate to doctors6 and nurses25 on four general medical and surgical wards (table 1). Of the ward doctors, 19 were interns, 20 residents and 18 registrars. Nurses comprised 19 registered nurse (RN) new graduates, 21 enrolled nurses, 13 year 2–4 RNs, 28 level 5+ RNs and 23 clinical nurse consultants and specialists. Both ward-based studies combine data from two periods of observation. Observation sessions occurred on weekdays between 07:00 and 19:00 h.
Observations were carried out using the Work Observation Method By Activity Timing (WOMBAT).26–28 This is a modified time and motion approach that collects information about work tasks, including task-switching and multitasking. The method involves observers shadowing participants while they are carrying out their normal work activities for 1–2 h at a time and recording every task performed in that time via a handheld computer. Information about multiple dimensions of work tasks is collected including, what task, with whom and using what information tools (eg, use of a phone, computer). All tasks are automatically time-stamped on data entry. If two or more tasks are observed to occur simultaneously they are recorded as multitasking, while if a task causes the premature cessation of another, this is recorded as task-switching. Figure 1 shows an exemplar screenshot of the WOMBAT data collection tool. Human research ethics approval was obtained from the study hospital and the University of New South Wales for all studies.
Operational definition of terms
The array of scenarios that sufficiently burden a clinician's cognition as to cause a change in work have been conceptualised in a number of ways.29 ,30 Given that the term interruption has been muddied by inconsistent use in the literature,31 we define an internally consistent set of terms which, as far as possible, are taken from the literature, but without use of the term interruption. A primary task1 ,17 ,32–34 may be subject to some form of trigger that elicits a change in the work process; for example, while documenting, a phone rings. A clinician may respond to the trigger by replacing the primary task with a new, or secondary,1 ,17 ,34 task (task-switching35); in our example, pausing documentation to answer the phone. Alternatively, a clinician may add a secondary task to the primary task (multitasking); for example, continuing to document while answering the phone. If the clinician ignores an attention-demanding prompt then the prompt is not considered a trigger as it does not result in workflow change. If action in response to a trigger is deferred via some short communication, this is still considered task-switching in itself. A trigger may be an externally or internally generated stimulus, but only internal triggers that result in multitasking have been recorded; for example, initiating a conversation while documenting. While in theory it may be possible to determine if a task-switch is due to an internal trigger, this is not captured in our data.
Comparison of task-switching and multitasking rates
In order to address the first aim, rates of task-switching and multitasking per hour were compared. Rates with 95% CIs were calculated by primary task type using separate Poisson regression models for each group.36 These modelled the rate of triggering events per unit time with trigger type (task-switch or multitask), task type and the interaction between the two as the covariates, and with time on task as the offset. If during a direct care task, for example, a clinician switched to a medication task, then the time spent on the direct care task was considered the exposure time for that trigger event and so contributed to the denominator of the direct care task-switching rate. Similarly, if an instance of multitasking was initiated during a direct care task, the duration of direct care contributed to the denominator of the direct care multitasking rate. Contrast statements were used to test the difference between task-switching rates and multitasking rates for each task type. The reporting of rates of task-switching and multitasking per unit time is common in the literature, but may not account for variation in the pace of work. Rates per task may be a reasonable alternative; however, defining the denominator for multitasking rates is non-trivial and rates of this form have not been included in this analysis.
Factors associated with clinicians’ use of strategy
In line with the second aim, we used logistic regression to determine factors associated with clinicians’ employment of task-switching (coded as 1) over multitasking (coded as 0) when triggered. The analysis used all instances of task-switching or multitasking, and included information on both the primary and secondary tasks associated with each trigger event. Random intercepts logistic regression was chosen to allow for correlation between outcomes within clinicians and wards, and to estimate clinician-specific rather than population-averaged effects. Potential covariates were: type and duration of primary task, tool involved in primary task (eg, computer), the type and duration of secondary task, tool and person involved in secondary task (eg, patient). Temporal factors included time of day, weekday and previous strategy used. Clinician seniority, years of experience and age were also potential predictors. The ED model considered the number of daily patient presentations as a proxy for clinical workload. A previous study of the ED data identified a temporal observer effect associated with the sole observer.37 The two ward-based studies each had several observers. Hence, an observer effect was considered as a potential covariate in each model.
A model for each group was built by purposeful selection38 with a significance level to retain covariates of 0.05. All models allowed for correlation within clinicians and the models for ward doctors and nurses also allowed for correlation within wards. Certain tasks may not be able to occur concurrently; however, the aggregation of tasks into task types (eg, direct care) means that all multitasking combinations between task types are able to occur. The definitions of each task type are given in online supplementary etable 1. These work task categories were developed through extensive piloting and consultation with clinicians26 ,39 and have been shown to be valid in other settings.28 Due to the challenge of interpreting interaction terms in a clinically useful way in this context, only first order terms were considered in the models.
Strategy determined by individual or external factors
The random intercepts models allow for variation between individuals due to unmeasured factors by allowing a different intercept value for each subject and we used this to assess the third aim of the study. The extent of this inter-clinician variation in strategy use is indicated by the variance between individual intercepts, after taking into account other factors in the model. This was done via a Wald test on the estimated intercept variance and its standard error. Significant intercept variance is evidence that there were other factors specific to individuals that influenced the use of strategy.
Among 8370 tasks observed in the ED, there were 1269 task-switches and 1942 instances of multitasking. For ward doctors, there were 795 task switches among 17 783 tasks in total, along with 6168 multitasking instances, and for ward nurses there were 28 809 tasks in total during which there were 800 task-switches and 4482 instances of multitasking (table 2).
Comparison of task-switching and multitasking rates
Overall, task-switching rates per hour were higher in the ED (6.0, 95% CI 5.7 to 6.4) than for doctors and nurses on wards (2.2, 95% CI 2.1 to 2.4 and 1.8, 95% CI 1.7 to 2.0). In contrast, multitasking rates per hour were lower in the ED (9.2, 95% CI 8.8 to 9.6 vs 17.3, 95% CI 16.9 to 17.8 and 14.1, 95% CI 13.8 to 14.5) (table 2). Rates for each task type also differed significantly for task-switching and multitasking rates within each group with the exception of ED rates of administrative tasks, answering a pager and supervision.
Factors associated with clinicians’ use of strategy
The logistic regression models used to address the second aim identified several variables common to all work groups that were significantly associated with use of task-switching over a multitasking strategy. These were: the type of both primary and secondary tasks, the duration of the secondary task and whether a patient or a phone was involved in the secondary task (table 3). An observer effect of some kind was present in all models. In the ED model, the three time periods for the sole observer were significantly different, while in the other two models there were significant differences between observers. ORs refer to estimated fixed effects. Higher odds of task-switching (OR>1) is equivalent to lower odds of multitasking and vice versa.
Clinicians in all work groups were most likely to switch tasks when triggered if the primary task involved documentation (table 3). Nurses were more likely to stop what they were doing during social activities compared with other tasks, but the opposite was true for both ED and ward doctors. The duration of primary task was only significant in the ward doctors’ model where the odds of task-switching when triggered increased the longer a clinician spent on the primary task (OR 1.25, 95% CI 1.16 to 1.34) after adjusting for the effect of task type. ED clinicians were more likely to multitask if triggered while on the phone (OR 0.55, 95% CI 0.32 to 0.97).
Doctors in the ED and on wards most frequently suspended their primary task to perform direct care or answer a page when triggered, and nurses were most likely to switch tasks to perform direct care or for professional communication. For all groups, documentation and medication secondary tasks were more likely to be multitasked than most other task types. Longer duration of secondary tasks was significantly associated with increased likelihood of task-switching, and this effect was greater among nurses than both doctor groups (nurses OR 1.66, 95% CI 1.52 to 1.81; ED doctors OR 1.20, 95% CI 1.10 to 1.30; ward doctors OR 1.20, 95% CI 1.11 to 1.30).
People involved in the secondary task, in addition to the observed clinician, had a significant effect on a clinician's strategy. Clinicians in all groups were less likely to task-switch when triggered if a patient was involved in the secondary task (ORs between 0.21 and 0.47). ED doctors were more likely to switch tasks when triggered if a nurse or doctor was involved in the secondary task (OR 1.37, 95% CI 1.04 to 1.81; OR 1.54, 95% CI 1.19 to 1.99). Among ward doctors, a secondary task involving another doctor meant a reduced likelihood of stopping their primary task (OR 0.58, 95% CI 0.46 to 0.71). Similarly among nurses, secondary tasks with another nurse were associated with reduced odds of task-switching (OR 0.54, 95% CI 0.44 to 0.66).
Among ED clinicians and ward nurses, strategy type was strongly associated with the strategy used for the previous trigger. Specifically, a clinician was more likely to switch tasks if the previous strategy was also a task-switch (ED OR 1.92, 95% CI 1.60 to 2.31; nurses OR 1.81, 95% CI 1.47 to 2.22). Time of day was a significant factor for ward doctors where the likelihood of task-switching was much lower at the end of the day (after 17:00) compared with all other times. Strategy use also varied by day of the week for nurses with task-switching more likely on Monday or Friday and less likely on Tuesday or Thursday. Seniority, age and clinician experience were not significant in any model.
Strategy determined by individual or external factors
For individual clinicians, the crude proportion of secondary tasks that were switched to ranged from less than 5% to over 80%. In each of the three models, after accounting for much of this variation through the model covariates, there was still significant variation between individuals according to the intercept variance assessed in relation to the third study aim (Wald test of intercept variance: p=0.04 for ED doctors and p=0.03 for ward staff). Inter-ward variation was zero for doctors and insignificant for nurses (p=0.58), indicating no evidence of difference in strategy use between wards within individual studies.
This study of over 1000 observation hours demonstrates that many factors affect clinicians’ workload management strategies. ED doctors switched tasks frequently in contrast to a predominance of multitasking among doctors and nurses on wards. This high task-switching rate among ED doctors is consistent with other studies.4 ,6 ,40 ,41 Increased task-switching frequency has been associated with work intensity42 ,43 and with error in the clinical setting.12 ,14 This suggests the elevated task-switching rate in EDs may be due to periods of higher work intensity which may increase error risk; although, the relationship among task-switching, work intensity and error risk is currently not well understood. In contrast, the rates of multitasking on wards were almost twice that of ED clinicians, suggesting that multitasking may be more easily carried out in settings of lower work intensity. This is consistent with the higher level of critical care provided in the ED compared with general wards. The inclusion of internally triggered multitasking may have increased multitasking rates and may partly explain why these rates are higher than overall task-switching rates in each work group.
The type of both primary and secondary task was a key factor in strategy use. In particular, documentation tended to be suspended in preference to most secondary tasks, but if documentation was the secondary task this was more likely to be performed in parallel with the primary task. Two studies of nurses’ work report that the priority of secondary tasks relative to primary tasks is a determinant of the strategy used, with relatively higher priority secondary tasks necessitating task-switching.1 ,19 An ED-based study also found that high priority triggers, such as calls to the patient resuscitation room, were more likely to cause task-switching than other less urgent triggers.44 In light of this, documentation may be largely treated as lower priority compared with other task types. Under this line of reasoning, the fact that the primary task was more likely to be paused if the secondary task was direct care suggests direct care was generally treated as higher priority. These effects were seen in all work groups, and this may imply a degree of prioritising certain tasks over others.
The strong association between strategy use and task type may also be due to specific task combinations being naturally more suited to task-switching or multitasking strategies.22 For example, while talking to a colleague, it is easier to simultaneously perform a routine manual task, but answering the phone would require task-switching. Further, particular tasks may not be able to occur concurrently (such as suturing two patients at once) and it is not known how the frequency of such mutually exclusive pairings may differentially affect the strategies that clinicians use.
The complexity, frequency and dissimilarity of secondary tasks can affect performance10 ,45 and likely influence which strategy a clinician employs. Although secondary task duration is unknown at the time a trigger occurs, it was used as a proxy for cognitive demand assuming that on average longer tasks are more demanding. According to multiple resource theory, secondary tasks consuming only some cognitive processing resources can be multitasked, while those consuming all resources must be switched to.17 ,46 Assuming then that secondary tasks causing task-switching are on average more demanding, then the observation that these tasks also tend to be longer suggests secondary task duration may indeed be a reasonable proxy for cognitive demand. However, without a direct measure of cognitive demand it is not possible to determine this with certainty. Although the number of daily presentations was not a significant factor in the ED, this measure may be too coarse to capture fluctuations in cognitive demand from one task to the next. The duration and complexity of secondary tasks have been associated with increased resumption lag, that is, time to reorientate when switching back to the primary task.15 ,47 Also, the longer the primary task is suspended the more likely that it will not be resumed at the correct place in the sequence or that it will not be resumed at all.16
In the ED, doctors were less likely to pause a task for a patient than for another doctor or nurse. This may reflect the higher priority assigned to interactions with colleagues, or that such interactions are more cognitively demanding in the critical care setting. A similar response to patients was observed in wards, but there was also a lower likelihood of stopping work when triggered by a fellow clinician. This may indicate a different mode of inter-clinician interaction in non-critical care settings.
The results identified several temporal factors related to management of triggers that have not previously been reported. ED clinicians’ and ward nurses’ strategies were strongly associated with the strategy used for the previous trigger. This suggests the previous strategy may somehow influence the current one or that triggers of a similar cognitive demand arrive in clusters. Prompted by the significant change in strategy use with time of day among ward doctors, subsequent analysis identified that nurses’ task-switching rates increased throughout the day, while ward doctors experienced a morning peak in multitasking rates, but a more constant rate of task-switching. Task-switching peaked in the ED in the late morning declining thereafter, while multitasking rates were steady. The significance of both previous strategy and time of day implies that sequencing and cumulative effects represent important factors in cognitive load and point the way for future investigation.
The significant intercept variance from the random intercept models suggests that there was individual variation in strategy use independent of external factors such as work tasks, people, tools and the hospital setting. This variation in individuals’ strategy use was also not associated with age, experience or seniority, and is consistent with findings that cognitive characteristics and personality affect individual prospective memory strategies.48 Addressing the third study aim, it appears that there are both external factors and factors related to the individual that determine the use of strategies. However, the observed inter-clinician variation cannot necessarily be attributed to differences in cognitive characteristics and may also be due to differing case mix or variation in specific tasks performed that are not captured by the relatively broad task categories. A study of paediatric nurses found that past experience with errors and the establishment of rigid routines are important clinician-specific factors that affect the way individuals manage triggers1 and such factors may also have influenced inter-clinician differences in this study.
Under the framework that triggers are a potential cause of error, a number of frequency reduction and impact minimisation strategies have been suggested. These include system focused changes like ensuring information or resources are available when and where required,2 ,16 and measures to prevent or minimise triggering during tasks of high working memory.17 For the latter strategy, tasks with a high propensity to receive triggers and with high risk of patient harm are obvious targets;31 for example, the implementation of ‘interruption free zones’ for preparing medications.49 Clinician-specific strategies that promote resilience to disruption have been proposed, including the use of resumption cues to ensure correct resumption of a task sequence, training staff to time triggers for task boundaries,9 ,50 and training clinicians for dealing with triggers to engender a sense of control and reduce stress.17 However, the use or effectiveness of such strategies is largely underexplored.51
Major strengths of this study include the large sample size and inclusion of data from different clinician groups. The fact that data were collected using the same methodology lends validity to inter-group comparisons. The significant observer effect was a limitation that was not anticipated given the high inter-rater reliability scores in each of the original studies based on univariate assessments. This indicates that a multivariate assessment of inter-rater reliability may be necessary to minimise any observer effects in future studies. In this study, the observer effects were mitigated by adjusting for these in the models. All observations were conducted in the same hospital, so the results may not necessarily reflect practices in all hospitals. The inclusion of internally generated triggers for multitasking but not task-switching was also a limitation. This is discussed above in relation to comparing task-switching and multitasking rates, while for the logistic models some unmeasured bias may exist if the proportion of multitasks that are internally triggered is dependent on other factors. The data do not capture all the complexity of the human interactions under study, but these unmeasured factors were largely accounted for by the random intercepts modelling approach.
We have conducted a large sample, multi-setting, observational study of the way clinicians manage their work. The use of task-switching and multitasking in response to triggers was associated with many factors related to the work tasks, the people and tools involved in those tasks, as well as factors related to individual clinicians. Despite clear differences in strategy use among ED doctors, ward doctors and ward nurses, clinicians in all settings appeared to prioritise certain types of tasks over others. The low priority given to documentation was consistent in all groups, as was the high priority given to direct care secondary tasks. These findings suggest that considerations of safety and workload may be implicit in task-switching and multitasking decisions. Although these strategies have been cast in a negative light, future research should explore their role in optimising competing quality and efficiency demands.
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 table
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.