The cognitive and neuroanatomical correlates of multitasking

Abstract

Patients who show the “strategy application disorder” can show deficits restricted to situations requiring multitasking, but the precise neuroanatomical and cognitive correlates of this problem have been rarely investigated. In this study, 60 people with circumscribed cerebral lesions and 60 age- and IQ-matched controls were given a multitasking procedure which allowed consideration of the relative contributions of task learning and remembering, planning, plan-following and remembering one’s actions to multitasking performance. Lesions to the left posterior cingulate and forceps major regions gave deficits on all measures except planning. Remembering task contingencies after a delay was also affected by lesions in the region of the left anterior cingulate, and rule-breaking and failures of task switching were additionally found in people with lesions affecting the medial and more polar aspects of Brodmann’s areas 8, 9 and especially 10. Planning deficits were associated with lesions to the right dorsolateral prefrontal cortex (RDLPFC). A theory of the relationships between the cognitive constructs underpinning multitasking was tested using structural equation modelling. The results suggest that there are three primary constructs that support multitasking: retrospective memory, prospective memory, and planning, with the second two drawing upon the products of the first. It is tentatively suggested that the left anterior and posterior cingulates together play some part in the retrospective memory demands, while the prospective memory and planning components make demands on processes supported by the left areas 8, 9 and 10 and the RDLPFC respectively.

Introduction

“We had many systems failures and they were in need of your constant attention. Many days I’d start an experiment in the morning to get it running and then I’d run over and help hacksaw through a pipe and plug the ends and then run back to my experiment. I’d have three or four watches on with alarms set to different things that I’d have to run back to. So I was multitasking in order to try to get everything accomplished”. Jerry Linenger, US astronaut, describing life aboard the Mir space station (BBC T.V. “Horizon” programme, 23 April 1998).

Life aboard the Mir space station must have been cognitively highly demanding. Not only did the astronauts have a host of routine functions to perform, but they also had a series of experiments to carry out in their limited time aboard, with each experiment varying in importance, complexity, and completion time. Additionally, the personnel on this particular mission had to deal with a host of unforeseen and potentially life-threatening occurrences. Each of these important incidents would take time from their planned schedule, requiring reassessment of achievable goals, re-prioritization and rescheduling.

Most of us have encountered situations that will have made similar cognitive demands to the one encountered by Jerry Linenger, albeit less frenetic and less grave. For instance, consider the demand of preparing a formal meal for a large number of guests. Such a task requires considerable planning and preparatory actions: deciding upon a menu; purchase of the individual constituents; consideration of amounts, guest preferences, and of the feasibility of the cooking process and the probability of failure (based on the cook’s experience). All such factors will have to be evaluated before cooking even begins. Once it does, the cooking and preparation times of many different dishes will have to be dovetailed, unexpected occurrences (e.g. the lack of some small but important ingredient, the arrival of an early guest) must be dealt with, and then each dish needs to be monitored at each stage, left to cook, returned to at the appropriate time and so forth. The cognitive abilities which support these complex human activities are crucial to effective everyday existence: even in the context of good intellect, knowledge and recollection of his expected duties Jerry Linenger would have been little more than a liability to his crew if he had been unable to remember to return to an experiment once he had switched to another, or to appreciate the relative seriousness of unforeseen events. The present paper aims to investigate the cognitive processes underlying these behaviours, their relationship to each other, and to discover which parts of the brain are particularly involved in supporting them.

Recent evidence suggests that the cognitive processes underlying performance in these sorts of situations may be selectively impaired in some neurological patients. Thus Shallice and Burgess [74] reported three patients who had all suffered frontal lobe damage and who showed marked impairments in everyday life activities despite little or no detectable impairment on a range of traditional neuropsychological tests, including those of executive abilities previously shown to be sensitive to frontal lobe dysfunction. Indeed their performance on tests of general intellectual function was superior, and despite extensive examination none showed impairments on any formal test which was commensurate with their disability in everyday life.

The disability took specific forms. The patients were impaired primarily on tasks that involved prioritisation, organisation and execution of a number of different tasks within a given period. Shallice and Burgess [73], [74] described these tasks, rather inelegantly, as “multiple sub-goal scheduling” tasks. However Jerry Linenger’s term “multitasking” is more succinct and will be used here. Shallice and Burgess argued that a key component of successful performance in these situations was the ability to create and activate delayed intentions. The successful performance of the patients on the traditional tasks was explained as not being relevant to the patients’ deficits since they did not tap these particular functions, in contrast to everyday life activities such as shopping or preparing a complex meal, which often involve the prioritisation of competing demands and the creation, maintenance and activation of delayed intentions.

Shallice and Burgess were by no means the first or last to have reported this “strategy application disorder” (see also [65], [72]). Eslinger and Damasio [37], Goldstein et al. [46] and Duncan et al. [31] have also reported other striking cases with similar patterns. These authors have offered a variety of explanations for their patients’ patterns. Goldstein et al. [46] gave an analysis in terms of the model of Shallice and Burgess, but Damasio and his colleagues [26], [28], [70] have explained their patients’ problems in terms of the failure of a system which operates to signal the possible deleterious consequences of a course of action (the “somatic marker hypothesis”). Duncan et al. [31] took yet another view, characterising the patient’s deficits in terms of “goal neglect” (see also [30], [32]) a phenomenum related to Spearman’s ‘9’ [76].

These views are probably complementary rather than in competition. Thus Shallice and Burgess [72], [73], [75] explained the patient’s problems as deficits in cognitive control systems, using the framework of Norman and Shallice [61]. This model has little to say about the social consequences of these deficits however, and there are similarities between the concept of “temporal” or “intention” markers used by Shallice and Burgess ([72], [73]; see [16] for further details), and that of “somatic markers” used by Damasio and his colleagues (e.g. [29]). Overall there is no direct disagreement between the two explanations, which can perhaps best be seen as seeking to explain different aspects of these patients’ problems. Similarly, Duncan’s [30], [32], [33] “goal neglect” account shares similarities with both these views in that it postulates, as they do, the failure to respond to a signal or cue as an important factor in the patients’ presentation. It seems likely therefore that it may be possible to find an explanation which might integrate these apparently conflicting views.

The basic empirical findings concerning these patients are however somewhat more agreed. Many of the more recent studies have used as least one of the two measures developed in the Shallice and Burgess [72] study (the “six element” and the “multiple errands” tests), and all cases to whom they have been administered have failed at least one of them. The multiple errands test (MET) is a real-life task based around a shopping precinct. Participants are given some money and an instruction sheet and asked to buy various items, find out certain information, and be at a certain location at a specific time, whilst observing a number of rules such as “you must not enter a shop other than to buy something” which emphasises the planning and prospective memory demands of the task. The second task, the six element test (SET), was designed to tap a subset of the same cognitive components, but do, so under more controlled and easily quantifiable conditions. Subjects were presented with three subtasks; each split into two sections. They were told that they were not permitted to carry out the first section of a given subtask followed by the second section of the same task, and that earlier items within a task scored more points than later ones. They were told that otherwise they were free to organise their efforts in any way they saw fit with the overall objective being to score as many points as possible within a permitted period.

Shallice and Burgess’s patients showed a variety of errors on the MET, tending to break the rules, leave items unfinished, forget to carry out the prospective memory items, and make several departures from social convention (e.g. clambering up onto a display of fruit outside a grocer’s shop to peer through the shop window). The patients’ failures on the experimental task (the SET) were more easily characterised. They consisted of a low number of subtasks attempted and/or an abnormal number of rule-breaks. This pattern occurred in the context of a normal work-rate in the patients. Of course these test results may have been epiphenomena as regards their everyday behaviour disturbances, but the circumscribed nature of their problems together with the face validity of one of the tests in particular (the MET) argued against such a view.

More recently Burgess et al. [10] provided further empirical evidence that patients with these sorts of everyday life problems tend to perform poorly specifically on these multitasking tests, and also that the cognitive process(es) damaged in these patients may usefully be regarded as constituting a discrete cognitive system. Carers or relatives of a group of ninety-two mixed aetiology neurological patients were given a questionnaire (the DEX; Burgess et al. [12]) which asked them to rate the frequency of occurrence of twenty of the most common dysexecutive symptoms in the patients they knew well. When the results were subjected to factor analysis (orthogonal rotation), five factors were selected; “inhibition” (deficits in response suppression and disinhibition); “intentionality” (everyday deficits in planning and decision-making); “executive memory” (e.g. confabulation, preservation); and two purely affective factors — positive and negative affective changes. A range of neuropsychological tests was also administered to the patients, which allowed for examination of the relationships between the scores for these behavioural symptom factors and individual test psychometric performances. It is the second factor, “intentionality” which is of interest here. Of all the tests given, which included measures of intelligence, memory, language, perception as well as ten measures of executive function only the six element test (SET) was significantly related to the factor scores for “intentionality” (r=0.46; P<0.001 criterion). These results provide confirmation of the relationship between performance on the SET and everyday situations involving planning and organisation in a much wider range of patients. Furthermore, they also provide further evidence that the processes supporting complex goal-directed behaviour may operate independently from other executive functions.

According to Shallice and Burgess (e.g. [73]), critical aspects of the cognitive systems supporting the behaviours required by multitasking tests such as the SET were part of Shallice’s “supervisory attentional system” [71]. However, no assertion was made regarding the constructs underlying the different stages of behaviour (planning, prospective memory etc.), or how they might relate together. For instance, it could not be determined whether the same cognitive subsystems were used in the creation and the activation of intentions, or how these might relate to those used at the planning stages. Furthermore whilst the distinction between the supervisory and non-supervisory cognitive components received some prima facie support from the single cases and from traditional neuropsychology, a formal demonstration of the statistical plausibility of such ideas is not possible with a single-case design [9].

The current investigation aims to elucidate these relationships and to discover more precisely their neuroanatomical correlates. However an enquiry of this sort first requires some basic notion of the demands that might be made by multitasking. This will be outlined in the next section, where we take as a starting point the theoretical framework of Burgess and Shallice [13], [15], [72], [73], [74], [75].

Given the ubiquity of the behaviour, there are surprisingly few studies directly addressing the cognitive demands of human multitasking. However the most obvious candidate critical cognitive processes are those related to the realisation of delayed intentions (“prospective memory”), since most tests purporting to measure other putatively relevant functions (e.g. “planning” or “retrospective memory”) may be performed normally in patients with isolated multitasking impairments (e.g. [72]). It is this (i.e. prospective memory (PM)) which distinguishes a test like the SET from other experimental tasks involving the presentation of more than one task at a time. Thus whilst dual- or multiple-task paradigms also involve task switching (e.g. [2], [69]) they do not involve the deferral of task execution over lengthy periods of time (switches typically occur with an interval of a few seconds or less).

Similarly, the problem which researchers are trying to solve in prototypical multiple-task situations (e.g. air-traffic control [5], [80]) is how attentional resources are allocated in the face of competing demands [56], [57]. By contrast, multitasking tests like the SET are different in that it is axiomatic that the subject is concentrating on a foreground task; the PM cue does not necessarily interfere with performance of this foreground task; and the length of the retention interval precludes conscious and continuous intention rehearsal.

In addition, there are two other characteristics of the naturalistic multitasking situation being considered here which differ from experimental multiple-task paradigms [9], [72], [73], [74]. The first is that what constitutes adequate performance is not strongly signalled by external demands: the participants have to decide for themselves what is an acceptable target, and when they have reached it. The second is that the process of “marker creation” (i.e. the creation of an intention to do something at the future time) is largely self-initiated. Furthermore the participants decide when or under what circumstances they will realise their delayed intentions.

If this account is correct, the number of behavioural variables that one might consider as relevant to naturalistic multitasking are the following: three “retrospective” mnemonic variables (learning of task parameters, subsequent memory for them, and for what one has already done); a measure of the adequacy of a participant’s initial plan; and additionally, some measure of the functioning of prospective memory (PM) is necessary. Although far from being a straightforward measure of PM functioning, the ability to pursue a plan must require this component to some degree (at least in the case of multitasking situations having the characteristics described above). Finally, a measure of overall task performance is of course also required.

Investigation of the individual contributions made by each of these behaviours to overall multitasking performance is, however, not methodologically straightforward for reasons which are too complex to be fully explored here (see [9] for discussion). Briefly, however, there are two primary concerns. The first is that there may be some cognitive processes which are required only in the co-ordination of others, or are stressed only in situations where sub-tasks are combined (perhaps the simplest example would be a dual-task paradigm; see [52] for a similar argument). A related, but second, point is that there may be processes relevant to multitasking, which are specific to the ill-structured nature of the situation [43]. The solution to these problems that is adopted here is to study the behavioural and cognitive components contributing to multitasking whilst keeping the behavioural sequence intact, rather than study each component in isolation. In this way, the method used here is quite different from the inference method used in previous multitasking studies (e.g. [1], [20], [31], [37], [46], [72]).

However, this method raises its own problems. Firstly there is the matter of serial dependence [9]: performance at any stage of a behavioural sequence will be affected by what has gone before, so measures are not strictly independent. Measurement in these situations therefore becomes a matter of studying degrees of co-relation (or more properly, covariance) between stages rather than looking for absolute impairments as is typical in neuropsychology. This in turn raises a basic but important psychometric issue: if one wishes to measure success at stage B (say, plan-following) whilst taking into account success at stage A (say, the planning stage), measurement error introduced at stage A will be likely to be repeated at stage B, causing correlated error measurements. Unless these are accounted for, they may give the appearance of greater relation between stages than actually exists. For these reasons, the following investigation attempts to study the behavioural and cognitive components contributing to multitasking whilst keeping the behavioural sequence intact, and using an analytic method which is specifically designed to confirm theoretical models which contain casual links and allow for the consideration of correlated error variances: structural equation modelling (for descriptions of this method see [6], [54]).