Download as pdf or txt
Download as pdf or txt
You are on page 1of 12

Neuropsychology Copyright 2007 by the American Psychological Association

2007, Vol. 21, No. 2, 212–223 0894-4105/07/$12.00 DOI: 10.1037/0894-4105.21.2.212

Changes in Attention and Information-Processing Speed Following Severe


Traumatic Brain Injury: A Meta-Analytic Review
Jane L. Mathias and Patricia Wheaton
University of Adelaide

Deficits in attention are frequently reported following severe traumatic brain injury (TBI). However,
methodological differences make it difficult to reconcile inconsistencies in the research findings in order
to undertake an evidence-based assessment of attention. The current study therefore undertook a
meta-analytic review of research examining attention following severe TBI. A search of the PsycINFO
and PubMed databases spanning the years 1980 to 2005 was undertaken with 24 search terms. Detailed
inclusion and exclusion criteria were used to screen all articles, leaving 41 studies that were included in
the current meta-analysis. Weighted Cohen’s d effect sizes, percentage overlap statistics, and confidence
intervals were calculated for the different tests of attention. Fail-safe Ns were additionally calculated to
address the bias introduced by the tendency to publish significant results. Large and significant deficits
were found in specific measures of information-processing speed, attention span, focused/selective
attention, sustained attention, and supervisory attentional control following severe TBI. Finally, age,
education, and postinjury interval were not significantly related to these deficits in attention.

Keywords: attention, information-processing speed, severe traumatic brain injury, meta-analysis, review

Supplemental materials: http://dx.doi.org/10.1037/0894-4105.21.2.212.supp

Traumatic brain injuries (TBIs) are often associated with a that is relevant to a task while ignoring irrelevant and distracting
number of cognitive deficits, the most frequent of which are information (van Zomeren & Brouwer, 1987, 1992). Divided at-
disorders of memory and attention (Oddy, Coughlan, Tyerman, & tention refers to the ability to simultaneously process more than
Jenkins, 1985). Although a large number of studies have investi- one source of information at a time (Williamson, Scott, & Adams,
gated deficits in attention following TBI, the findings are incon- 1996), and sustained attention refers to an ability to continuously
sistent with respect to the nature and extent of these impairments. maintain attentional focus under conditions of low stimulation and
Moreover, there is considerable debate about whether deficits in over a prolonged period of time (Williamson et al., 1996). Finally,
attention reflect a discrete cognitive deficit or whether they are the supervisory attentional control facilitates the strategic allocation
result of a more pervasive reduction in the speed with which of attentional resources in demanding situations in order to opti-
information is processed (Brouwer, Ponds, Van Wolffelaar, & van
mize performance (van Zomeren & Brouwer, 1994). In addition,
Zomeren, 1989; Brouwer, Withaar, Tant, & van Zomeren, 2002;
reductions in information-processing speed can have pervasive
Ponsford & Kinsella, 1992; Rios, Perianez, & Munoz-Cespedes,
effects on cognitive functioning, including attention.
2004).
Whereas evidence for a general reduction in speed of informa-
There are a number of different components of attention that can
be affected by TBI. One of the most basic of these is the ability to tion processing following TBI has consistently been found (Brou-
orient attention, which involves spatially directing or allocating wer et al., 1989, 2002; Felmingham, Baguley, & Green, 2004;
attention to sensory information (Hills & Geldmacher, 1998). Ponsford & Kinsella, 1992), specific deficits in the orienting of
Another aspect of attention is attention span, which refers to the attention (Cremona-Meteyard, Clark, Wright, & Geffen, 1992),
amount of information that can be processed at one time (Lezak, focused/selective attention (Chan, 2000; Chan, Hoosain, & Lee,
Howieson, & Loring, 2004). Selective or focused attention, on the 2002; Mangels, Craik, Levine, Schwartz, & Stuss, 2002; Vakil,
other hand, involves an ability to selectively attend to information Blachstein, & Hoofien, 1991), divided attention (Leclercq et al.,
2000; Lengenfelder, Schultheis, Al-Shihabi, Mourant, & DeLuca,
2002; Park, Moscovitch, & Robertson, 1999), sustained attention
(Chan, 2000), and supervisory attentional control (Rios, Perianez,
Jane L. Mathias and Patricia Wheaton, School of Psychology, University & Munoz-Cespedes, 2004; Spikman, van Zomeren, & Deelman,
of Adelaide, Adelaide, South Australia, Australia. 1996) have also been reported. Yet other studies have failed to find
Preliminary findings from this study were presented in May 2005 at the evidence for the existence of specific deficits (e.g. Bate, Mathias,
6th World Congress of the International Brain Injury Association, Mel- & Crawford, 2001a; Brouwer et al., 1989; Felmingham et al.,
bourne, Victoria, Australia. This research was partially funded by Grant
2004; Ponsford & Kinsella, 1992). This causes problems for cli-
207711 to Jane L. Mathias, Erin D. Bigler, Stephen C. Bowden, Nigel R.
Jones, and D. James Taylor from the National Health and Medical Re-
nicians when they are trying to determine which aspects of atten-
search Foundation of Australia. tion to assess and what measures to use when assessing the
Correspondence concerning this article should be addressed to Jane L. attention of a patient who has sustained a TBI.
Mathias, School of Psychology, University of Adelaide, Adelaide, South Inconsistencies in the research findings may, in part, reflect
Australia 5005, Australia. E-mail: jane.mathias@adelaide.edu.au methodological differences among studies with respect to the

212
SEVERE TRAUMATIC BRAIN INJURY AND ATTENTION 213

aspects of attention that are examined (e.g., focused, divided reductions in speed of information processing largely account for
attention), the tests that are used, and the samples that are inves- the deficits in attention following severe TBI, then measures that
tigated. A wide variety of tests have been used to measure the tap processing speed should be amongst the measures with the
different aspects of attention, making it difficult to directly com- largest effect sizes. This study also examined the extent to which
pare the findings of numerous studies even when they are exam- a number of methodological variables (age, education, postinjury
ining the same aspect of attention. For example, researchers have interval) were related to deficits in attention following severe TBI
claimed to measure speed of information processing using the in order to determine whether studies that differ methodologically
Trail Making Test (Azouvi, Jokic, Van Der Linden, Marlier, & could validly be combined.
Bussel, 1996) and numerous reaction time tasks (Felmingham et
al., 2004; Leclercq et al., 2000; Ponsford & Kinsella, 1992; Spik- Method
man, Deelman, & van Zomeren, 2000), whereas focused attention
Literature Search and Inclusion Criteria
has been assessed using visual search tasks (Schmitter-Edgecombe
& Kibby, 1998), the Stroop Test (Ponsford & Kinsella, 1992), and A comprehensive search of the PsycINFO and PubMed elec-
the Trail Making Test (Felmingham et al., 2004). Divided atten- tronic databases from January 1980 to November 2005 was un-
tion, on the other hand, has been assessed with choice reaction dertaken in order to identify published articles that examined
time tasks (Ponsford & Kinsella, 1992), dual tasks (Leclercq et al., attention following severe TBI. The key search terms that were
2000), and the Paced Auditory Serial Addition Test (Chan et al., used to capture relevant articles are provided in Table A of the
2002), and supervisory attentional control has been assessed with supplemental material for this study. In addition, the bibliogra-
the Wisconsin Card Sorting Test (Rios et al., 2004), the Tower of phies of these articles were examined, and manual searches of a
London test (Shallice, 1988), and the Six Elements Test (Chan, number of journals were completed to identify any other relevant
2002). Not only do these studies use different measures, but they articles (refer to Table B of the supplemental material for further
also examine different aspects of attention, which makes it difficult details).
to determine whether certain types of attention are more affected For a study to be included in the current meta-analysis, it had to
than others and whether particular measures are more sensitive to fulfill the following criteria: (a) was published in a journal; (b) was
these deficits. Moreover, it is not possible to directly compare the published in English; (c) had a control group (age-, gender-, and/or
magnitude of the findings derived from different measures and education-matched); (d) was not a case study; (e) had a TBI group
different studies unless the scores are standardized. that had sustained a severe nonpenetrating injury (i.e., Glasgow
Variability among studies with respect to the postinjury interval, Coma Scale score ⱕ 8, loss of consciousness ⬎ 24 hr, and/or
age, education level, and premorbid intellectual abilities of partic- posttraumatic amnesia ⬎ 24 hr); (f) had no participant with a
ipants further adds to the difficulty in reconciling disparate find- known prior history of head injury, motor, visual, or language
ings. Impairments in cognitive function, for example, reportedly impairments, psychiatric illness, or substance abuse; (g) adminis-
resolve more rapidly in the first 6 months after an injury, and this tered tests of attention to both groups (i.e., excludes studies using
rate of recovery is affected by age (Williamson, Scott, & Adams, only electrophysiological measures); (h) reported the results in a
1996), making it important to consider postinjury interval and age format that enabled the calculation of effect sizes (e.g., means,
when comparing the findings of different studies. In addition, standard deviations, t tests, or F ratio from a one-way analysis of
education and premorbid intelligence are thought to influence variance [ANOVA]); and (i) had participants between 16 and 65
postinjury performance (Rios et al., 2004), suggesting that these years of age. Where age ranges for the patient or control groups
variables should also be taken into account when evaluating the were not included, a study was deemed eligible if the mean age ⫾
research literature. Similarly, the use of different patient groups, one standard deviation fell within this range. When demographic
such as consecutive hospital admissions or symptomatic referrals, details were not reported for the control group, the study was
can affect the likelihood of finding deficits in attention following included if the patient group met the age criteria.
TBIs. Finally, sample size has an impact on whether group differ- The literature search initially identified 2,971 potentially rele-
ences are statistically significant. Given the large variability in the vant studies. A preliminary application of the inclusion criteria to
sample sizes that have been investigated and the failure by most the information provided in the titles and abstracts of these articles
investigators to provide effect size calculations, which measure the yielded a total of 105 studies. Application of the inclusion criteria
magnitude of group differences independently of sample size, it is to the full-text versions of these studies revealed that 64 of these
difficult to know whether the statistically significant or nonsignif- did not meet one or more of the inclusion criteria, leaving 41
icant results are partly an artifact of sample size. However, critical studies.
reviews of the research literature rarely acknowledge the impor- When conducting meta-analytic reviews, it is important to en-
tance of this variable and are not in a position to formally evaluate sure that a study’s participants are independent of those used in
the impact of sample size, or any of the preceding variables, on the other studies, as the computational procedures assume that studies
research findings. A quantitative review of the research literature is are independent of one another (Hunter, Schmidt, & Jackson,
therefore needed to address these issues. 1982; Rosenthal, 1995). If the same participants contribute more
The current study provides a meta-analysis of research that has than one score to the calculation of an effect size when studies are
examined deficits in attention following severe TBI in order to combined, it can distort the magnitude of this effect (Rosenthal,
consolidate this literature. Specifically, we sought to identify the 1995). Thus, two publications by Bate and colleagues (Bate et al.,
measures of attention that are most sensitive to deficits in attention 2001a; Bate, Mathias, & Crawford, 2001b), which reported sepa-
following severe TBI. We hypothesized that if, as has been sug- rate findings for the same participants, were combined and treated
gested (Felmingham et al., 2004; Ponsford & Kinsella, 1992), as one study. In addition, there were two publications that de-
214 MATHIAS AND WHEATON

scribed more than one study where all or most of the participants Effect Size Calculations and Analyses
were different (Park et al., 1999, Experiments 1 and 2; Ponsford &
Cohen’s d effect sizes were calculated for the current meta-
Kinsella, 1992, Experiments 1 and 3). Both of these publications
analysis, following the method described by Zakzanis (2001).
were treated as two studies. In total, 41 studies were included in
Cohen’s d is not influenced by sample size and measures the
the current meta-analysis (references for studies included in this
difference between two means divided by the pooled standard
meta-analysis are marked with an asterisk in the reference list). A
deviation (Zakzanis, Leach, & Kaplan, 1999). If means and stan-
total of 48 different tests of attention were used by two or more
dard deviations were not provided for a study, we converted t
studies.
values and one-way F statistics to d using the formula set forth by
Zakzanis (2001). Probability values were not used to calculate
Data Collection and Preparation effect sizes in the current meta-analysis, as they confound the
magnitude of an effect size with sample size (Hunter et al., 1982),
In some cases, participants were retested (e.g., longitudinal
such that a p ⬍ .05 may overestimate an effect size in a large
studies measuring recovery or studies assessing the efficacy of
sample and a nonsignificant result that is assigned a p ⬎ .05 will
rehabilitation), necessitating the use of a number of criteria to
systematically underestimate the size of an effect in a small sample
determine what scores would be included from these studies (Lip-
(Rosenthal, 1995). The different tests of attention were not equally
sey & Wilson, 2001). Specifically, where tests were administered
represented across studies; therefore, the use of p values in the
on more than one occasion, only data from the first session were
calculation of effect sizes may have disproportionately disadvan-
used (to minimize practice effects). Where a patient group was
taged infrequently used tests, particularly where small samples
tested twice but only compared with a control group on the second
were used.
occasion (e.g., Azouvi et al., 1996), the first set of scores for the
Effect sizes were calculated in a multistage process. The first
patient group was compared with those of the control group in
stage involved calculating effect sizes for each score of every test
order to equate the practice effects for the two groups. Moreover,
that was used by a study. If a study provided multiple scores for a
a test was omitted from analysis if it was not adequately described
test, and these were to be combined for the purposes of this
and a reference was not provided, thereby precluding determina-
meta-analysis, an average effect size for that test was calculated.
tion of whether the test assessed attention or some other cognitive
The effect sizes obtained from different studies were then averaged
function.
to evaluate the combined findings. As the reliability of an effect
Some minor transformations to the published data were neces-
size is affected by the size of the sample from which it is derived,
sary before it could be analyzed. First, where standard errors were
it is important to weight effect sizes before they are combined
reported, these were converted to standard deviations in order to
(Lipsey & Wilson, 2001). Whereas weighting by sample size is
calculate effect sizes. Second, the data for a number of descriptive
one option, Hedges and Olkin (1985) have established that the
variables (e.g., postinjury interval, duration of posttraumatic am-
inverse variance (i.e., the inverse of the squared standard error)
nesia, loss of consciousness) were transformed to ensure a com-
provides the best weight because it provides a better measure of
mon scale of measurement. Finally, in Experiment 1 of Park et al.
the precision of an effect size (i.e., larger standard errors are
(1999), the scores for one test had been divided into three time
associated with less precise effect sizes). An overall weighted
intervals to examine changes in performance over time. These mean effect size for a test was therefore calculated by weighting
scores were averaged in order to calculate a single effect size for each of the effect sizes from individual studies by the inverse of its
this test. In a second experiment by Park et al. (1999), data were variance and then averaging them (dw). All effect sizes were
reported for three blocks of trials for each of three intertrial calculated in such a way that a positive Cohen’s dw indicated that
intervals (10, 20, and 60 s) for the Paced Auditory Serial Addition patients with severe TBI were more impaired on tests of attention
Test in order to test a hypothesis that was specific to that study (a than participants who had not suffered a TBI. According to Co-
total of nine scores were reported). These scores were averaged for hen’s criteria (Cohen, 1977, 1992), a small effect was defined as
the present purposes. d ⱖ .2, a moderate effect as d ⱖ .5, and a large effect as d ⱖ .8.
Each test of attention was categorized into one of seven aspects To put this into perspective, a medium effect size of .5 indicates
of attention in order to organize the data provided by these studies. that the means of the two groups differ by half of a standard
Consistent with common theoretical and clinical distinctions in the deviation.
attention literature, the following categories were used: informa- A percentage overlap (%OL) score was also calculated in order
tion-processing speed, orienting of attention, attention span, fo- to provide a measure of the extent to which the scores from the two
cused/selective attention, divided attention, sustained attention/ groups overlap. The %OL is inversely related to the effect size and
vigilance, and supervisory attentional control. As there were in- was calculated using the table provided by Zakzanis et al. (1999).
consistencies among authors with respect to how the different An effect size of zero (d ⫽ 0) is associated with 100% overlap in
measures were classified, Lezak (1995) and Spreen and Strauss the scores of the two groups, whereas an effect size of 4.0 is
(1998) were consulted in order to provide a framework for cate- associated with only 2.3% overlap in scores, indicating that the
gorizing tests. This system of categorization was adopted to sim- groups perform very differently on that measure. Confidence in-
plify the presentation of data and is not intended to imply that there tervals for the effect sizes were additionally calculated using the
is agreement amongst researchers about the specific aspects of method described by Lipsey and Wilson (2001) to determine
attention that are assessed by these tests or that these tests provide statistical significance. If the confidence interval for an effect size
pure measures of a construct. does not span zero, the true population effect size differs signifi-
SEVERE TRAUMATIC BRAIN INJURY AND ATTENTION 215

cantly from zero, indicating that there is a significant difference multiple regression) is less affected by the problems of heterosce-
between the performance of the TBI and control groups on that dasticity (i.e., where the distribution of sample sizes is skewed)
test. and multicollinearity (i.e, where predictor variables are correlated
One criticism of meta-analytic studies is that there may be a bias with one another) than are bivariate correlations and other methods
toward publishing studies with significant findings, which would of analyzing moderator variables. A WLS multiple regression was
mean that an analysis of published research does not include therefore also conducted to analyze the effects of these three
nonsignificant findings that have not been published (referred to as moderator variables. Standard errors and confidence intervals for
the file-drawer problem), thereby inflating the resulting effect sizes the results of the WLS multiple regression were adjusted according
(Rosenthal, 1995). In order to address this problem, we calculated to the method outlined by Hedges and Cooper (1994). This regres-
a fail-safe N (Nfs) for each effect size using the procedure de- sion analysis was performed on the data from the 22 studies that
scribed by Lipsey and Wilson (2001). This statistic provides an provided details for all three moderator variables.
estimate of the number of unpublished studies with nonsignificant
findings that would be needed to call the current findings into Data Interpretation
question, enabling the reader to evaluate the likely impact of the
file-drawer problem (Zakzanis et al., 1999). The larger the Nfs Weighted mean effect sizes (standard deviation, minimum,
statistic, the more confident we can be of the current findings. maximum) for the seven categories of attention and for each
The original versions of tests were often modified for use in a individual test of attention were calculated, as were Nfs, %OL, and
study. One difficulty that occurs when tests and their adaptations confidence intervals (CIs). The conclusions drawn from this study
are treated separately is that the effect size for a measure may be are based on the combined interpretation of these different statis-
based on only one study. The purpose of a meta-analysis, however, tics. From a neuropsychological perspective, a clinician would be
is to analyze data derived from multiple studies (Zakzanis et al., more confident that a severe TBI has impacted on attention when
1999). Thus, where a test and an adaptation of it were deemed to there are large group differences (Cohen’s dw) and, consequently,
be similar, they were grouped together and labeled using a com- there is limited overlap in the performance of the two groups
mon term, (e.g., number and letter cancellation tasks were labeled (%OL). Clinicians would also be more confident in using a mea-
“cancellation tests”). This reduced the number of tests under sure to assess the attentional deficits of a person who had sustained
consideration to 23. a severe TBI if the CI did not span zero (i.e., effect size is
An additional issue that warrants consideration is whether there significant) and it was unlikely that unpublished findings could
are methodological differences among studies that may affect the draw the current findings into question (i.e., large Nfs). For the
findings and, therefore, the effect sizes derived from these studies. purpose of this meta-analysis, it was decided that in order for a
One solution is to treat these methodological differences as mod- finding to be suggestive of compromised performance following a
erator variables that may contribute to variations in the results severe TBI, it must meet the following three criteria: (a) have a
found by different researchers (Hunter et al., 1982; Zakzanis et al., large effect size (i.e., d ⱖ .8) and, consequently, a limited degree
1999). As meta-analyses necessarily involve combining the results of overlap in the scores obtained by severe TBI and control
of studies that differ methodologically, moderator analyses enable participants, (b) have a confidence interval that did not span zero
an evaluation of whether it is valid to combine the results of these (N.B., CIs are affected by N and variance), and (c) have an Nfs
studies (Sharpe, 1997). Between-studies variation in age, educa- score that was large enough to make it unlikely that there were that
tion, and postinjury interval, therefore, were examined in order to number of unpublished studies with nonsignificant findings in
determine whether these variables influenced the magnitude of the existence. As different tests were used with varying frequency, it
group differences in attention. Estimated premorbid IQ was not was decided that the Nfs should at least be greater than the number
examined for this purpose, as there is increasing evidence to of published studies that had used a particular test (i.e., Nstudies).
suggest that measures of premorbid IQ may be affected by severe
TBI (Freeman, Godfrey, Harris, & Partridge, 2001; Mathias, Bow- Results
den, & Bigler, in press; Morris, Wilson, Dunn, & Teasdale, 2005;
Participants
Riley & Simmonds, 2003). In order to analyze these moderator
variables, it was necessary to calculate a mean age, educational Data from a total of 1,651 participants from 41 studies contrib-
level, and postinjury interval for each study. For age and educa- uted to the current meta-analysis, comprising 823 severe TBI
tion, this was done by averaging the data from the TBI and control patients (71% male, 29% female) and 828 control participants
groups for that study (e.g., MTBI⫹control age). We assessed the (68% male, 32% female). Demographic details for individual
relationship among these moderator variables and the mean effect studies are summarized in Table C of the supplemental material for
size for the studies that reported these data. All 41 studies provided this article. Although 7% of studies reported recruiting a selected
information relating to participants’ age, 29 provided educational sample of TBI patients (i.e., patients complaining of attentional
data, 32 provided postinjury intervals, and 22 studies provided problems), most did not report this information (93%). As Table 1
details for all three moderator variables. Pearson correlation coef- shows, fewer than half of the studies reported estimates of pre-
ficients (rs) were calculated to examine the relationship between morbid IQ for the TBI and control groups (NTBI ⫽ 7, Ncontrol ⫽ 10)
these three moderator variables and effect sizes. or specific injury severity data for the TBI patients (i.e., Glasgow
Although correlations are frequently used in the analysis of Coma Scale, duration of posttraumatic amnesia, loss of conscious-
moderator variables, Steel and Kammeyer-Mueller (2002) have ness; Ns ⫽ 11 to 19). Moreover, only 32 studies (78%) reported
demonstrated that a weighted least squares (WLS) multiple regres- the postinjury interval. When the ages, educational levels, and
sion (using the inverse of the sampling error variance to weight the premorbid IQ estimates for those studies that provided this infor-
216 MATHIAS AND WHEATON

Table 1
Demographic Details for the Traumatic Brain Injury (TBI) and Control Groups and Injury Data for the TBI Group
TBI Controls

N N
Variable (studies) M SD Range (studies) M SD Range

N 41 20.1 10.8 6.0–51.0 41 20.2 9.7 6.0–44.0


Age (years) 41 31.6 23.4–55.5 37 31.0 19.5–51.6
Education (years) 29 12.8 9.5–16.2 29 13.3 10.7–16.5
Estimated premorbid IQ 7 101.2 86.1–112.0 10 107.6 101.1–114.0
GCS score 14 4.7–6.6
PTA (duration in days) 19 42.4 30.2 6.0–125.0
LOC (duration in hours) 11 863.3 1,373.7 8.25–4,937.5
Time since injury (days) 32 1,178.0 1,151.7 59.9–4,015.0

Note. GCS ⫽ Glasgow Coma Scale; PTA ⫽ posttraumatic amnesia; LOC ⫽ loss of consciousness.

mation were compared, the TBI and control groups did not differ Inhibition Task, Auditory Odd-Ball Task), yielding a total of 14
significantly, suggesting that these groups were well matched on scores, that were used by two or more studies (see Table 2).
these variables. Moreover, 11 of these scores met the study criteria for differenti-
ating among samples. Specifically, the Stroop Test (speed and
Measures of Attention accuracy scores of the Word Reading, Color Naming, and Inter-
ference subtests), the Symbol Digit Modalities Test (accuracy
The weighted effect sizes (dw) for the individual measures of scores for the oral and written versions) and the Digit Symbol Test
attention (mean, standard deviation, minimum, maximum, 95% (accuracy score), an Inhibition Task (speed scores from the control
CIs, Nfs, %OL), broadly grouped by category of attention and rank and interference conditions), and the Trail Making Test (speed
ordered by effect size, are provided in Table 2 (refer to Table C in scores for Parts A and B, and ratio A:B) all demonstrated large and
the supplemental material for specific study details). Only effect significant effect sizes with high to very high Nfs.
sizes for tests that were used by two or more studies are reported Divided attention has been assessed using either an RT measure
here because effect sizes that are based on a single study are not under dual task conditions or the Paced Serial Addition Test.
thought to provide a reliable measure of group differences Whereas the effect sizes for these measures were significant and
(Rosenthal, 1995). This reduced the number of studies that were the Nfs statistics were high enough to suggest reasonable confi-
included in this part of the meta-analysis to 37. Measures of the dence in these findings, the mean effect sizes suggest more modest
orienting of visual attention were only used by one study and, deficits in divided attention following severe TBI than were evi-
consequently, are not reported here. dent on other measures of attention. Moreover, even though per-
Tests specifically targeting information-processing speed were formance on these tasks is likely to be affected either directly or
amongst the most commonly used measures in the studies ana- indirectly by processing speed, the deficits that were evident on
lyzed in this meta-analysis (refer to Table 2). A variety of simple these measures were smaller than those found on measures spe-
and choice reaction time (RT) tasks were used, which revealed cifically designed to target information-processing speed (i.e., sim-
significant group differences with very high Nfs values, such that ple and choice RT tasks)
persons who had sustained a TBI performed, on average, one Sustained attention or vigilance was assessed by visual search
standard deviation below their healthy peers. Moreover, process- and cancellation tasks in a total of eight separate studies. As seen
ing speed was found to be slower in response to both simple and in Table 2, both types of tests yielded large group differences,
complex (choice) tasks. Thus, as predicted, speed of information relatively low overlap statistics, CIs that did not span zero, and
processing is significantly impaired following severe TBI. large Nfs when the scoring system was based on a combination of
Attention span was also assessed by a large number of studies speed and accuracy or speed alone.
(N ⫽ 13), using a variety of tests that assessed digit, reading, word, Finally, there were no tests of supervisory attention control that
and letter span. Of these, only reading span (ability to retain the met the current study criteria for differentiating among groups
last word of increasing numbers of sentences), word span (where (refer to Table 2). Interestingly, commonly used tests, such as the
the stimuli are 2- or 3-syllable words), and digit span detected card sorting tests (e.g., the Wisconsin Card Sorting Test) and the
sizeable and significant effects with large Nfs values (Table 2). In Tower of London, which purportedly assess the executive aspects
contrast, the separate scores for forward and backward digit span of attention, showed only small to moderate, albeit mostly signif-
and letter span showed only small to medium effects. As shown in icant, effect sizes.
Table 2, there were specific (e.g., reading span, digit span) mea-
sures of attention span that appeared to be more affected by severe
Moderator Variables
TBI than others (e.g., alphabet span, digits forward and backward).
Focused or selective attention was the aspect of attention that Pearson correlation coefficients were calculated in order to
was most commonly assessed in the research literature. There were examine the relationship between variables that might be expected
six different tests (Stroop, Symbol Digit, Digit Symbol, Trails, to affect performance on the measures of attention, as measured by
SEVERE TRAUMATIC BRAIN INJURY AND ATTENTION 217

Cohen’s d effect sizes. The age, years of education, and postinjury In order for a test to be considered useful for the clinical
time interval for participants in those studies that reported these assessment of patients who have sustained a severe TBI, it was
data were therefore correlated with the mean effect sizes for these determined that a test must be able to reliably distinguish between
studies. Small and nonsignificant correlations were observed for these two groups (as indicated by large effect sizes and small
age (r ⫽ ⫺.19, n ⫽ 41, p ⬎ .05), education (r ⫽ .06, n ⫽ 29, p ⬎ %OL). There should also be a high degree of confidence in the
.05), and postinjury interval (r ⫽ ⫺.22, n ⫽ 32, p ⬎ .05), accuracy with which a measure distinguishes between the two
indicating that there was no relationship between these variables. groups (95% CIs), and the conclusions drawn from the research
A WLS multiple regression was additionally performed on the data literature must not be systematically biased by the tendency to
from the 22 studies that reported data for all three variables. The publish statistically significant findings (measured by the Nfs sta-
mean effect size for a study was the dependent variable, the three tistic). An analysis of the results for each of the different tests of
moderator variables were the independent variables, and the in- attention indicated 20 measures that yielded effect sizes meeting
verse variance was used as the weighting variable. The final model the study criteria (i.e., d ⱖ 0.8).
was nonsignificant ( p ⬎ .05) and accounted for only 17% of the When specific measures of information-processing speed were
variance. These analyses indicate that there was no systematic examined (i.e., reaction time tasks), we found that severe TBI
variation in performance on the different measures of attention as produced clear deficits in processing speed, with responses to
a consequence of differences in these methodological variables. simple and complex stimuli being comparably affected. This find-
ing suggests that the time taken to process even simple stimuli is
Tests Used by Only One Study affected by a severe TBI and that the contribution of this deficit to
performance on tests of attention that purportedly assess other
Whereas meta-analytic techniques are primarily used for con-
aspects of attention cannot be ignored. Indeed, in examining the
solidating the findings of a number of studies, it is also important
amount by which the effect sizes for other measures of attention
to identify measures that discriminate between groups but have
that involve a timed component (e.g., Digit Symbol, Stroop) ex-
only been used by one study. In doing so, lesser known but
ceed that for information-processing speed (Cohen’s d ⬇ 1), it is
potentially innovative measures can be identified with a view to
apparent that reduced processing speed may account for a large
encouraging the collection of additional data in order to adequately
component of these effects. The contribution of deficits in infor-
assess their usefulness. Effect sizes for measures that met the study
mation-processing speed therefore needs to be given careful con-
criteria (Cohen’s d ⱖ 0.8, CIs ⫽ 0, Nfs ⬎ Nstudies) but were based
sideration when assessing any aspect of attention. However, reac-
on only one study are therefore summarized in Table 3, where it
tion time tasks are not routinely used in clinical situations because
can be seen that there were an additional 22 measures that revealed
of a reliance on computer-administered stimuli, the need for com-
large and significant deficits in performance following severe TBI.
mercially available programs, and the limited availability of nor-
Indeed, some measures had very large effects, with very small
mative data. It may therefore be necessary to rely on one of the
%OL statistics (e.g., oral Symbol Digit Modalities speed and
other simple timed measures of attention (e.g., time to complete
accuracy scores, Pigache Attention Test, Short Stimulus Task,
Trails A) to tap into these deficits.
Random Number Generation task, Visual Nonsearch Condition of
With respect to attention span, there were one traditionally used
a Visual Search Task). Moreover, the large Nfs values, relative to
measure (e.g., Digit Span) and two less commonly used measures
the number of published studies that had used these measures,
(e.g., Reading and Word Span) that showed large group differ-
suggest that it is unlikely that these findings are merely a conse-
ences. All of these measures assess a person’s ability to hold
quence of publication bias.
increasing amounts of information in working memory, with the
main difference between tests being the stimuli (digits or words).
Discussion
In the case of the Reading Span task, participants are presented
Overall, the data for this meta-analysis were obtained from 41 with increasing numbers of sentences and are required to recall the
studies that examined the attention of 823 persons who had sus- last word of each sentence (Goldenberg, Oder, Spatt, & Podreka,
tained a severe TBI, together with 828 controls, using a large 1992; Schmitter-Edgecombe & Nissley, 2000). Word Span, on the
number of different tests, subtests, and scoring procedures (N.B., other hand, involves participants having to recall increasing se-
tests and close variations of tests were combined). On the basis of quences of 2- or 3-syllable words (Park et al., 1999). Although
the available data, the TBI and control groups appeared well these tests are not timed, information-processing speed is also
matched in terms of age, education, and premorbid IQ, suggesting likely to contribute to performance on these tasks.
that group differences in these variables were not contributing to Focused, or selective, attention was examined in a total of 23
the study findings. However, none of the studies reported depres- separate studies using 14 measures, 11 of which clearly discrimi-
sion scores, so it is not possible to determine whether poorer affect nated between groups. Commonly used measures, such as the
may have contributed to some of the deficits in attention that were Stroop Test, the Symbol Digit Modalities and Digit Symbol Tests,
apparent. Importantly, the current meta-analysis found that while a and the Trail Making Test, all showed evidence of impaired
wide variety of tests have been used in research that has examined performance following severe TBI, with some measures from
attention following severe TBI, there is considerable variation in these tests proving to be slightly better than others (e.g., the speed
the extent to which performance on these measures is affected by and accuracy score of the Stroop Word Reading subtest being
such an injury. There are also a number of infrequently used better than the same score from the Color Naming and Interference
measures that, if supported by additional empirical evidence, may subtests). Moreover, there was only one test of focused attention
prove valuable in the assessment of attention following severe that was discriminating but not well known, namely, the Inhibition
TBI. Task, which involves naming a target word in the presence of a
218 MATHIAS AND WHEATON

Table 2
Weighted Effect Sizes for the Different Tests of Attention, Categorized by Category of Attention
Category of attention test N
(score)a (studies) M dw SD dw Min dw Max dw 95% CI Nfs %OL Study references

Information-processing speed
Choice reaction time (speed) 12 1.10 0.28 0.69 1.48 0.98–1.22 65 41 Felmingham et al., 2004; Ponsford & Kinsella,
1992, Exp. 1; Rugg et al., 1988; Rugg et
al., 1989; Sarno et al., 2003; Schmitter-
Edgecombe, 1996; Schmitter-Edgecombe et
al., 1992; Shum et al., 1990, 1994; Willison
& Tombaugh, 2006; Withaar & Brouwer,
2003; Zahn & Mirsky, 1999
Simple reaction time (speed) 9 1.07 0.68 0.72 2.98 0.94–1.20 47 41 Felmingham et al., 2004; Leclercq et al., 2000;
Perbal et al., 2003; Ponsford & Kinsella,
1992, Exp. 1; Sarno et al., 2003; Willison &
Tombaugh, 2006; Wilson et al., 1999;
Withaar & Brouwer, 2003; Zahn & Mirsky,
1999
Attention span
Reading span 2 1.01 0.04 0.99 1.04 0.62–1.40 9 45 Goldenberg et al., 1992; Schmitter-Edgecombe
& Nissley, 2000
Word span 2 0.81 0.29 0.61 1.02 ⫺0.02–1.64 7 53 Park et al., 1999, Exp. 1; Park et al., 1999,
Exp. 2
Digit span 6 0.80 0.42 0.08 1.25 0.52–1.08 23 53 Azouvi et al., 2004; Bate et al., 2001a, 2001b;
Leclercq et al., 2000; Rugg et al., 1993;
Schmitter-Edgecombe et al., 1992;
Schmitter-Edgecombe & Rogers, 1997
Alphabet span 2 0.57 0.37 0.39 0.91 0.03–1.11 5 62 Schmitter-Edgecombe & Rogers, 1997;
Schmitter-Edgecombe & Beglinger, 2001
Backward digit span 5 0.42 0.46 0.16 1.34 ⫺0.17–0.67 10 73 Azouvi et al., 2004; Bate et al., 2001a, 2001b;
Leclercq et al., 2000; Perbal et al., 2003;
Wilson et al., 1999
Forward digit span 5 0.25 0.14 0.07 0.39 0.00–0.50 5 79 Azouvi et al., 2004; Bate et al., 2001a, 2001b;
Leclercq et al., 2000; Perbal et al., 2003;
Wilson et al., 1999
Focused/selective attention
Stroop Word Reading (speed 4 1.45 0.38 1.20 2.05 1.15–1.75 28 30 Bate et al., 2001a, 2001b; Felmingham et al.,
and accuracy) 2004; Ponsford & Kinsella, 1992, Exp. 1;
Rios et al., 2004
Symbol Digit Modalities, 3 1.33 0.18 1.08 1.41 1.01–1.65 19 35 Bate et al., 2001a, 2001b; Ponsford &
oral (accuracy) Kinsella, 1992, Exp. 1; Schmitter-
Edgecombe & Beglinger, 2001
Symbol Digit Modalities, 3 1.31 0.09 1.23 1.40 0.99–1.63 19 35 Bate et al., 2001a, 2001b; Ponsford &
written (accuracy) Kinsella, 1992, Exp. 1; Schmitter-
Edgecombe & Beglinger, 2001
Digit Symbol (accuracy) 9 1.28 0.40 0.85 1.95 1.05–1.51 57 35 Rugg et al., 1988, 1989, 1993; Schmitter-
Edgecombe, 1996; Schmitter-Edgecombe et
al., 1992; Schmitter- Edgecombe & Rogers,
1997; Schmitter-Edgecombe & Kibby, 1998;
Schmitter-Edgecombe & Nissley, 2000;
Simpson & Schmitter-Edgecombe, 2000
Stroop Color Naming (speed 3 1.24 0.42 0.91 1.74 0.94–1.54 18 38 Bate et al., 2001a, 2001b; Ponsford &
and accuracy) Kinsella, 1992, Exp. 1; Rios et al., 2004
Inhibition Task, control 2 1.23 0.36 0.97 1.48 0.75–1.71 11 38 Ries & Marks, 2005; Simpson & Schmitter-
condition (speed) Edgecombe, 2000
Inhibition Task, interference 2 1.12 0.22 0.96 1.27 0.65–1.59 10 41 Ries & Marks, 2005; Simpson & Schmitter-
condition (speed) Edgecombe, 2000
Trail Making Test A (speed) 7 1.11 0.42 0.73 1.83 0.85–1.37 38 41 Azouvi et al., 1996; Brouwer et al., 1989;
Felmingham et al., 2004; Rios et al, 2004;
Simpson & Schmitter-Edgecombe, 2000;
Timmerman & Brouwer, 1999; Withaar &
Brouwer, 2003
Trail Making Test A:B ratio 2 1.04 1.00 0.65 2.06 0.59–1.49 9 45 Rios et al, 2004; Rugg et al., 1993
(speed)
Trail Making Test B (speed) 8 0.96 0.52 0.61 1.99 0.74–1.18 37 45 Azouvi et al., 1996; Brouwer et al., 1989;
Felmingham et al., 2004; Rios et al., 2004;
Simpson & Schmitter-Edgecombe, 2000;
Spikman et al., 2000; Timmerman &
Brouwer, 1999; Withaar & Brouwer, 2003
SEVERE TRAUMATIC BRAIN INJURY AND ATTENTION 219

Table 2 (continued)
Category of attention test N
(score)a (studies) M dw SD dw Min dw Max dw 95% CI Nfs %OL Study references

Stroop Interference (speed 4 0.87 0.39 0.79 1.58 0.65–1.09 16 48 Bate et al., 2001a, 2001b; Felmingham et al.,
and accuracy) 2004; Ponsford & Kinsella, 1992, Exp. 1;
Rios et al., 2004
Stroop Interference (speed) 2 0.36 0.22 0.10 0.41 ⫺0.02–0.74 3 73 Marsh & Knight, 1991; Spikman et al., 2000
Stroop Interference (error) 2 0.33 0.25 0.27 0.62 ⫺0.09–0.75 2 79 Marsh & Knight, 1991; Ponsford & Kinsella,
1992, Exp. 1
Auditory Odd-Ball Task 2 0.01 0.69 ⫺0.43 0.54 ⫺0.46–0.48 0 100 Rugg et al., 1988, 1993
(accuracy)
Divided attention
Reaction time dual task 3 0.70 0.32 0.46 1.04 0.47–0.93 10 57 Schmitter-Edgecombe & Nissley, 2000;
(speed) Spikman et al., 2000; Withaar & Brouwer,
2003
Paced Auditory Serial 5 0.60 0.30 0.28 1.09 0.45–0.75 14 62 Bate et al., 2001a, 2001b; Park et al., 1999,
Addition Test (accuracy) Exp. 1, Exp. 2; Ponsford & Kinsella, 1992,
Exp. 1; Spikman et al., 2000
Sustained attention/vigilance
Cancellation task (speed and 2 1.99 0.09 1.93 2.05 1.73–2.25 19 19 Geldmacher & Hills, 1997; Hills &
accuracy) Geldmacher, 1998
Visual search (speed) 2 1.78 0.74 1.46 2.51 1.59–1.97 17 23 Heinze et al., 1992; Schmitter-Edgecombe &
Kibby, 1998
Cancellation task (speed) 4 1.10 0.54 0.88 2.06 0.86–1.34 21 41 Hills & Geldmacher, 1998; Mattson et al.,
1994; Ponsford & Kinsella, 1992, Exp. 1;
Simpson & Schmitter-Edgecombe, 2000
Cancellation task (accuracy) 4 0.66 0.50 0.25 1.18 0.38–0.94 12 57 Bate et al., 2001a, 2001b; Mattson et al.,
1994; Ponsford & Kinsella, 1992, Exp. 1;
Simpson & Schmitter-Edgecombe, 2000
Supervisory attentional
control
Card-sorting tests (error) 6 0.76 0.39 0.25 1.40 0.54–0.98 22 53 Marsh & Knight, 1991; Rios et al., 2004;
Rugg et al., 1989; Schmitter-Edgecombe,
1996; Schmitter-Edgecombe & Kibby, 1998;
Simpson & Schmitter-Edgecombe, 2000
Card-sorting tests (accuracy) 6 0.60 0.52 ⫺0.07 1.38 0.37–0.83 17 62 Goldenberg et al., 1992; Marsh & Knight,
1991; Rios et al, 2004; Schmitter-
Edgecombe, 1996; Schmitter-Edgecombe &
Kibby, 1998; Simpson & Schmitter-
Edgecombe, 2000
Tower of London (speed) 2 0.37 0.72 0.09 1.12 0.07–0.67 3 73 Ponsford & Kinsella, 1992, Exp. 3; Spikman
et al., 2000
Tower of London (accuracy) 2 0.18 0.69 ⫺0.10 0.88 ⫺0.16–0.52 1 85 Ponsford & Kinsella, 1992, Exp. 3; Spikman
et al., 2000

Note. N ⫽ number of studies contributing to the effect size; dw ⫽ weighted effect size; min ⫽ minimum effect size; max ⫽ maximum effect size; Nfs ⫽
fail-safe N; %OL ⫽ percent overlap between traumatic brain injury and control groups; CI ⫽ confidence interval.
a
Although a number of studies measured the orienting of attention, no measure was used by more than one study, necessitating the exclusion of these data
for this analysis.

distractor word, with or without interference (Ries & Marks, 2005; of focused/selective attention described above reflect either the
Simpson & Schmitter-Edgecombe, 2000). In contrast, neither the time taken to complete a task (e.g., Trails A and B speed) or the
speed nor the accuracy scores of the commonly used Stroop number of correct responses given under time-limited conditions
Interference subtest, which involves naming the ink color (e.g., (e.g., Symbol Digit Modalities Test accuracy); therefore, it is
blue) in which a color word (e.g., red) is printed, proved to be as highly likely that reduced processing speed contributed to the
discriminating as a score that combined speed and accuracy on this impaired performance of the TBI group on many of these mea-
same subtest. Moreover, the Stroop Interference subtest was not as sures.
discriminating as the Stroop Word Reading or Color Naming Although the most commonly used measures of divided atten-
subtests. This finding is somewhat counterintuitive as the Interfer- tion did not prove very discriminating, when measures that had
ence task is more likely to draw on selective attention than are only been investigated by one study were additionally considered,
simpler tasks, such as Color Naming or Word Reading. It is it became apparent that there are a number of measures that may
possible that the latter two measures proved better because they prove useful if further empirical evidence replicates these findings.
tapped the speed-of-processing deficits that are also apparent after These measures assess dual task performance in a variety of
severe TBI (Bate et al., 2001b; Felmingham et al., 2004; Ponsford conditions, ranging from a simulated driving situation (Withaar &
& Kinsella, 1992; Rios et al., 2004). In fact, all of the 11 measures Brouwer, 2003) to Posner’s Covert Orienting of Attention Test
220 MATHIAS AND WHEATON

Table 3
Weighted Effect Sizes for the Different Tests of Attention Used by Only One Study, Where Cohen’s d Was ⬎ 0.8
N M Cohen’s
Category of attention test (score) (studies) dw 95% CI Nfs %OL Study reference

Focused attention/selective attention


Symbol Digit Modalities, oral (speed and accuracy) 1 2.71 1.52–3.90 13 9 Ponsford & Kinsella, 1992, Exp. 1
Pigache Attention Test (error) 1 2.31 1.65–2.97 11 13 Keller et al., 1995
TEA Map Search (speed and accuracy) 1 1.25 0.74–1.76 5 35 Bate et al., 2001a, 2001b
TEA Telephone Search (speed and accuracy) 1 1.18 0.67–1.69 5 38 Bate et al., 2001a, 2001b
Go–no-go task (Go accuracy) 1 0.87 0.22–1.52 3 48 Rugg et al., 1989
Divided attention
Tracking dual task (speed and accuracy) 1 1.49 1.05–1.93 6 29 Withaar & Brouwer, 2003
Random number generation dual task (speed) 1 1.32 0.83–1.81 6 35 Leclercq et al, 2000
Tone monitoring dual task (speed) 1 1.14 0.73–1.54 5 41 Schmitter-Edgecombe, 1996
Phenome detection dual task (error) 1 1.04 0.54–1.54 4 45 Bate et al., 2001a, 2001b
COAT valid cue dual task (speed) 1 1.03 0.74–1.32 4 45 Bate et al., 2001a, 2001b
COAT neutral cue dual task (speed) 1 0.97 0.68–1.25 4 45 Bate et al., 2001a, 2001b
COAT invalid cue dual task (speed) 1 0.83 0.54–1.11 3 53 Bate et al., 2001a, 2001b
Paced Auditory Serial Addition Test (error) 1 0.83 0.35–1.31 3 53 Ponsford & Kinsella, 1992, Exp. 1
Sustained attention/vigilance
Visual Nonsearch (speed) 1 1.73 1.51–1.95 8 25 Schmitter-Edgecombe & Kibby, 1998
Visual Search (accuracy) 1 1.67 1.13–2.21 7 25 Schmitter-Edgecombe & Kibby, 1998
Supervisory attentional control
Short Stimulus Task shifting cost (speed) 1 2.02 1.12–2.92 9 19 Stablum et al., 1994
Random number generation (speed) 1 1.84 1.10–2.58 8 23 Leclercq et al., 2000
Random-number generation (error) 1 1.43 0.50–2.36 6 32 Azouvi et al., 1996
Tower of London (speed and accuracy) 1 1.00 0.48–1.52 4 45 Goldenberg et al., 1992
Orienting of attention
COAT valid cue (speed) 1 1.03 0.21–1.35 4 45 Bate et al., 2001a, 2001b
COAT neutral cue (speed) 1 0.92 0.57–1.27 4 48 Bate et al., 2001a, 2001b
COAT invalid cue (speed) 1 0.84 0.55–1.12 3 53 Bate et al., 2001a, 2001b

Note. N ⫽ number of studies contributing to the effect size; dw ⫽ weighted effect size; Nfs ⫽ fail-safe N for effect size; %OL ⫽ percent overlap between
traumatic brain injury and control groups; CI ⫽ confidence interval; TEA ⫽ Test of Everyday Attention; COAT ⫽ Covert Orienting of Attention Test.

(COAT; Bate et al., 2001a). Once again, the contribution of racy with which participants completed the Tower of London.
deficits in information-processing speed to performance on these Notably, neither the speed nor the accuracy with which the Tower
measures must also be considered, as performance is frequently of London test was performed discriminated between the TBI and
measured directly in terms of speed (e.g., the COAT) or accuracy control groups in the two studies that used these scores. Other
under time-limited conditions (e.g., the Paced Auditory Serial card-sorting tests that have often been used to tap into this aspect
Addition Test; Gronwall, 1977). of attention, such as the Wisconsin Card Sorting Test, also proved
Following severe TBI, patients also appear to experience con- less successful in discriminating between samples, although mod-
siderable problems in sustaining focused attention over a period of erate group differences were still apparent.
time when they are assessed on tasks that require them to search Finally, the orienting of visual attention, which involves the
for a visual target that is embedded amongst other stimuli (e.g., allocation of visual attention in space (Bate et al., 2001a), has not
visual search tasks, letter and number cancellation tasks). Notably, been widely examined in severe TBI samples. Consequently, there
whereas visual search and cancellation tasks are described as were no COAT measures that had been used by more than one
measuring sustained attention by Lezak (1995), both tasks are study. When those measures that had only been examined by one
performed under time-limited conditions and are, therefore, also study but revealed large effect sizes were additionally considered,
likely to be affected by reductions in information-processing it was apparent that there are a number of simple COAT measures
speed. that may prove useful for measuring this aspect of attention,
Deficits in the executive aspects of attention (supervisory atten- namely RTs to valid, neutral, and invalid cues (Bate et al., 2001a).
tional control), on the other hand, were not apparent on any of the As with the previous measures, the contribution of information-
measures that were used by two or more studies (i.e., card-sorting processing speed to performance needs to be considered, as these
tests, Tower of London) but were evident on four measures (ob- are timed measures.
tained from three tests) that were used by only one study. Specif- With respect to participant characteristics that may act to mod-
ically, there were large and significant differences on a task that erate the impact of TBI on attention, the data indicated that age,
required participants to make one of two responses (motor or education, and postinjury interval did not significantly contribute
auditory) depending on the stimulus (Short Stimulus Task shifting to the variability in effect sizes, suggesting that it is acceptable to
cost), a random number generation task in which participants were combine the results of studies that differ on these methodological
required to generate numbers without repeating sequences (speed, variables. However, these findings should be viewed with caution,
accuracy scores), and a score that combined the speed and accu- as the data on which these analyses were based were less than
SEVERE TRAUMATIC BRAIN INJURY AND ATTENTION 221

optimal. First, not all studies provided the requisite data, thereby tively. These requirements are essential to the evaluation of an
reducing the sample size for this analysis. Second, and probably individual study and to the integration of findings in the form of a
more importantly, mean effect sizes were calculated for each study meta-analysis and, if consistently provided, will raise the quality of
for these analyses. Thus, the effect sizes for all of the individual the research in this field with minimal additional data. Specifically,
measures that were used by a given study were averaged. If a study research into deficits in attention (or other cognitive functions)
yielded both small and large effect sizes for different measures, following TBI needs to provide the following information: (a)
these were combined to calculate a mean effect size, which was details of the age, educational level, injury severity (Glasgow
then correlated with the moderator variables. A more useful anal- Coma Scale, loss of consciousness, and/or posttraumatic amnesia),
ysis would involve analyzing each measure separately by corre- postinjury interval, affective state (e.g., depression), and source of
lating the effect sizes from all of the studies that used a particular participants (e.g., consecutive hospital admissions, symptomatic
measure with the age, education, and postinjury interval data for referrals for neuropsychological assessment), and the numbers
those studies. However, this was not possible as the maximum involved in litigation; (b) means, standard deviations, and effect
number of studies that used any given measure was 12. When
sizes for both significant and nonsignificant findings; and (c) clear
those studies that did not provide data for the moderator variables
descriptions of the study measures (or a test reference) and any
were excluded, the sample size was reduced to an unacceptably
modifications that have been made to the test. Many studies do not
low number. Thus, the current moderator analysis was necessarily
currently provide all of this information.
limited.
Other limitations to this study include those that are generally In conclusion, following a severe TBI, patients appear to exhibit
associated with meta-analytic procedures. Specifically, it is possi- a range of attentional deficits that are broadly subsumed under the
ble that there were unpublished studies with nonsignificant find- categories of information-processing speed, attention span, selec-
ings (possibly due to small sample sizes) or articles published in tive/focused attention, sustained attention, and supervisory atten-
other languages that may have been excluded from the current tion control. Moreover, performance deficits of one standard de-
analyses. While there is no obvious solution to this problem, the viation or more below scores of their healthy peers are apparent on
calculation of fail-safe statistics assists the reader in evaluating the a range of commonly used measures, including visual search tasks,
degree of confidence that can be placed in the current findings. In cancellation tasks (e.g. the Ruff 2 and 7 Test, letter and word
addition, it is possible that the search terms were not exhaustive cancellation tests), the Stroop Test (Word Reading, Color Naming,
and that some relevant articles failed to be identified. Although a and Interference subtests), the Symbol Digit Modalities Test (oral
relatively large number of terms were used, electronic searches and written versions), the Digit Symbol Test, and the Trail Making
only identify matches on the basis of words contained in the title, Test (Part A). However, the fact that many of these tests yield
abstract, or keyword listings. Manual searches of journals indexes scores that reflect the time taken to complete a task or the number
and bibliographies were therefore undertaken to reduce the likeli- of correct responses within a fixed time interval suggests that the
hood of missing relevant studies. contribution of reduced processing speed to performance on tests
Finally, it should be noted that a number of studies that were of attention must be given careful consideration. Finally, although
included within the current review used sophisticated experimental there is some suggestion that differences in patient variables (e.g.,
designs to investigate the specific nature of the deficits in attention
age, education, postinjury interval) do not significantly contribute
following TBI. In particular, Ries and Marks (2005); Schmitter-
to the extent of these deficits, this finding should be viewed with
Edgecombe and Kibby (1998); and Simpson and Schmitter-Edge-
caution, as it was not possible to analyze the moderating influence
combe (2000) used a variety of experimental conditions in order to
of these variables on specific measures.
isolate some of the cognitive processes that contribute to attention,
including speed of information processing. To this end, these
researchers examined interaction effects and within-group planned References
comparisons, in addition to the simple group effects (TBI vs.
controls) that were the subject of the current meta-analysis. Inter- *References marked with an asterisk are included in the meta-analysis.
action effects result from the relationship between two or more *Azouvi, P., Couillet, J., Leclercq, M., Martin, Y. Asloun, S., & Rous-
independent variables and a dependent variable and are not reduc- seaux, M. (2004). Divided attention and mental effort after severe
ible to a standardized difference between two groups (Tilley, traumatic brain injury. Neuropsychologia, 42, 1260 –1268.
1996). Thus, it was not possible to include effect sizes to summa- *Azouvi, P., Jokic, C., Van Der Linden, M., Marlier, N., & Bussel, B.
rize the results of these complex experimental designs in the (1996). Working memory and supervisory control after severe closed-
current meta-analysis. Instead, only the results of simple group head injury. A study of dual task performance and random generation.
comparisons on the control tasks were presented for present pur- Journal of Clinical and Experimental Neuropsychology, 18, 317–337.
poses. This research is, however, important as it provides a more *Bate, A. J., Mathias, J. L., & Crawford, J. R. (2001a). The covert orienting
detailed and experimentally controlled examination of attention of visual attention following severe traumatic brain injury. Journal of
and may eventually lead to the development of better tests of Clinical and Experimental Neuropsychology, 23, 386 –398.
*Bate, A. J., Mathias, J. L., & Crawford, J. R. (2001b). Performance on the
attention for use in clinical practice.
Test of Everyday Attention and standard tests of attention following
In addition, whereas a meta-analysis is ultimately limited by the severe traumatic brain injury. The Clinical Neuropsychologist, 15, 405–
quality of the data that it is provided by the primary research, it 422.
also provides an opportunity to specify some of the information *Brouwer, W. H., Ponds, R. W. H. M., Van Wolffelaar, P. C., & van
that is needed by the scientific community in order to adequately Zomeren, A. H. (1989). Divided attention 5 to 10 years after severe
critique the findings of a study, either qualitatively or quantita- closed head injury. Cortex, 25, 219 –230.
222 MATHIAS AND WHEATON

*Brouwer, W. H., Verzendaal, M., van der Naalt, J., Smit, J., & van Lengenfelder, J., Schultheis, M. T., Al-Shihabi, T., Mourant, R., & DeLuca, J.
Zomeren, A. H. (2001). Divided attention years after severe closed head (2002). Divided attention and driving: A pilot study using virtual reality
injury: The effect of dependencies between the subtasks. Brain and technology. Journal of Head Trauma Rehabilitation, 17, 26 –37.
Cognition, 46, 54 –56. Lezak, M. D. (1995). Neuropsychological assessment (3rd ed.). New York:
Brouwer, W. H., Withaar, F. K., Tant, M. L. M., & van Zomeren, A. H. Oxford University Press.
(2002). Attention and driving in traumatic brain injury: A question of Lezak, M. D., Howieson, D. B., & Loring, D. W. (2004). Neuropsycho-
coping with time pressure. Journal of Head Trauma Rehabilitation, 17, logical assessment (4th ed.). New York: Oxford University Press.
1–15. Lipsey, M. W., & Wilson, D. B. (2001). Practical meta-analysis (Vol. 49).
Chan, R. C. K. (2000). Attentional deficits in patients with closed head Thousand Oaks, CA: Sage.
injury: A further study to the discriminative validity of the Test of Mangels, J. A., Craik, F. I. M., Levine, B., Schwartz, M. L., & Stuss, D. T.
Everyday Attention. Brain Injury, 14, 227–236. (2002). Effects of divided attention on episodic memory in chronic
Chan, R. C. K. (2002). Attentional deficits in patients with persisting traumatic brain injury: A function of severity and strategy. Neuropsy-
postconcussive complaints: A general deficit or specific component chologia, 40, 2369 –2385.
deficit? Journal of Clinical and Experimental Neuropsychology, 24, *Marsh, N. V., & Knight, R. G. (1991). Relationship between cognitive
1081–1093. deficits and social skill after head injury. Neuropsychology, 5, 107–117.
Chan, R. C. K., Hoosain, R., & Lee, T. M. C. (2002). Talking while Mathias, J. L., Bowden, S. C., & Bigler, E. D. (in press). Is performance on
performing a task: A better attentional performance in patients with the Wechsler Test of Adult Reading affected by traumatic brain injury?
closed head injury? Journal of Clinical and Experimental Neuropsychol- British Journal of Clinical Psychology.
ogy, 24, 695–704. *Mattson, A. J., Levin, H. S., & Breitmeyer, B. G. (1994). Visual infor-
Cohen, J. (1977). Statistical power analysis for the behavioral sciences. mation processing after severe closed head injury: Effects of forward
New York: Academic Press. and backward masking. Journal of Neurology, Neurosurgery and Psy-
Cohen, J. (1992). Quantitative methods in psychology: A power primer. chiatry, 57, 818 – 824.
Psychological Bulletin, 112, 155–159. Morris, P. G., Wilson, J. T., Dunn, L. T., & Teasdale, G. M. (2005).
Cremona-Meteyard, S. L., Clark, C. R., Wright, M. J., & Geffen, G. M. Premorbid intelligence and brain injury. British Journal of Clinical
(1992). Covert orientation of visual attention after closed head injury. Psychology, 44, 209 –214.
Neuropsychologia, 30, 123–132. Oddy, M., Coughlan, T., Tyerman, A., & Jenkins, D. (1985). Social
*Felmingham, K. L., Baguley, I. J., & Green, A. M. (2004). Effects of adjustment after closed head injury: A further follow-up seven years
diffuse axonal injury on speed of information processing following after injury. Journal of Neurology, Neurosurgery and Psychiatry, 48,
severe traumatic brain injury. Neuropsychology, 18, 564 –571. 564 –568.
Freeman, J., Godfrey, H. P. D., Harris, J. K. J., & Partridge, F. M. (2001). *Park, N. W., Moscovitch, M., & Robertson, I. H. (1999). Divided atten-
Utility of a demographic equation in detecting impaired NART perfor- tion impairments after traumatic brain injury. Neuropsychologia, 37,
mance after TBI. Journal of Clinical Psychology, 40, 221–224. 1119 –1133.
*Geldmacher, D. S., & Hills, E. C. (1997). Effects of stimulus number, *Perbal, S., Couillet, J., Azouzi, P., & Pouthas, V. (2003). Relationships
target-to-distractor ratio, and motor speed on visual spatial search quality between time estimation, memory, attention and processing speed in
following traumatic brain injury. Brain Injury, 11, 59 – 66. patients with severe traumatic brain injury. Neuropsychologia, 41,
*Goldenberg, G., Oder, W., Spatt, J., & Podreka, I. (1992). Cerebral 1599 –1610.
correlates of disturbed executive function and memory in survivors of *Ponsford, J., & Kinsella, G. (1992). Attentional deficits following closed-
severe closed head injury: A SPECT study. Journal of Neurology, head injury. Journal of Clinical and Experimental Neuropsychology, 14,
Neurosurgery and Psychiatry, 55, 362–368. 822– 838.
Gronwall, D. M. A. (1977). Paced Auditory Serial Addition Task: A *Ries, M., & Marks, W. (2005). Selective attention deficits following
measure of recovery from concussion. Perceptual and Motor Skills, 44, severe closed head injury: The role of inhibitory processes. Neuropsy-
367–373. chology, 19, 476 – 483.
Hedges, L. V., & Cooper, H. E. (1994). The handbook of research syn- Riley, G. A., & Simmonds, L. V. (2003). How robust is performance on the
thesis. New York: Russell Sage Foundation. National Adult Reading Test following traumatic brain injury? British
Hedges, L. V., & Olkin, I. (1985). Statistical methods for meta-analysis. Journal of Clinical Psychology, 42, 319 –328.
Orlando, FL: Academic Press. *Rios, M., Perianez, J. A., & Munoz-Cespedes, J. M. (2004). Attentional
*Heinze, H.-J., Munte, T. F., Gobiet, W., Niemann, H., & Ruff, R. M. control and slowness of information processing after severe traumatic
(1992). Parallel and serial visual search after closed head injury: Elec- brain injury. Brain Injury, 18, 257–272.
trophysiological evidence for perceptual dysfunctions. Neuropsycholo-
Rosenthal, R. (1995). Writing meta-analytic reviews. Psychological Bul-
gia, 30, 495–512.
letin, 118, 183–192.
*Hills, E. C., & Geldmacher, D. S. (1998). The effect of character and
*Rugg, M. D., Cowan, C. P., Nagy, M. E., Jacobson, I., & Brooks, D. N.
array type on visual spatial search quality following traumatic brain
(1988). Event related potentials from closed head injury patients in an
injury. Brain Injury, 12, 69 –76.
Hunter, J. E., Schmidt, F. L., & Jackson, G. B. (1982). Meta-analysis: auditory “Oddball” task: Evidence of dysfunction in stimulus categori-
Cumulating research findings across studies (Vol. 4). Beverly Hills, CA: sation. Journal of Neurology, Neurosurgery and Psychiatry, 51, 691–
Sage. 698.
*Keller, I., Schlenker, A., & Pigache, R. M. (1995). Selective impairment *Rugg, M. D., Cowan, C. P., Nagy, M. E., Miner, A. D., Jacobson, I., &
of auditory attention following closed head injuries or right cerebrovas- Brooks, D. N. (1989). CNV abnormalities following closed head injury.
cular accidents. Cognitive Brain Research, 3, 9 –15. Brain, 112, 489 –506.
*Leclercq, M., Couillet, J., Azouvi, P., Marlier, N., Martin, Y., Strypstein, *Rugg, M. D., Pickles, C. P., Potter, D. D., Doyle, M. C., Pentland, B., &
E., et al. (2000). Dual task performance after severe diffuse traumatic Roberts, R. C. (1993). Cognitive brain potentials in a three-stimulus
brain injury or vascular prefrontal damage. Journal of Clinical and auditory “oddball” task after closed head injury. Neuropsychologia, 31,
Experimental Neuropsychology, 22, 339 –350. 373–393.
SEVERE TRAUMATIC BRAIN INJURY AND ATTENTION 223

*Sarno, S., Erasmus, L.-P., Lipp, B., & Schlaegel, W. (2003). Multisensory Steel, P. D., & Kammeyer-Mueller, J. D. (2002). Comparing meta-analytic
integration after traumatic brain injury: A reaction time study between moderator estimation techniques under realistic conditions. Journal of
pairings of vision, touch and audition. Brain Injury, 17, 413– 426. Applied Psychology, 87, 96 –111.
*Schmitter-Edgecombe, M. (1996). Effects of divided attention on implicit Tilley, A. (1996). An introduction to psychological research and statistics
and explicit memory performance following severe closed head injury. (3rd ed.). Brisbane, Australia: Pineapple Press.
Neuropsychology, 10, 155–167. *Timmerman, M. E., & Brouwer, W. H. (1999). Slow information pro-
*Schmitter-Edgecombe, M., & Beglinger, L. (2001). Acquisition of skilled cessing after very severe closed head injury: Impaired access to declar-
visual search performance following severe closed-head injury. Journal ative knowledge and intact application and acquisition of procedural
of the International Neuropsychological Society, 7, 615– 630. knowledge. Neuropsychologia, 37, 467– 478.
*Schmitter-Edgecombe, M., & Kibby, M. K. (1998). Visual selective Vakil, E., Blachstein, H., & Hoofien, D. (1991). Automatic temporal order
attention after severe closed head injury. Journal of the International judgment: The effect of intentionality of retrieval on closed-head-injured
Neuropsychological Society, 4, 144 –159. patients. Journal of Clinical and Experimental Neuropsychology, 13,
*Schmitter-Edgecombe, M. E., Marks, W., Fahy, J. F., & Long, C. J. 291–298.
(1992). Effects of severe closed-head injury on three stages of informa- van Zomeren, A. H., & Brouwer, W. H. (1987). Head injury and concepts
tion processing. Journal of Clinical and Experimental Neuropsychology, of attention. In H. S. Levin, J. Grafman, & H. M. Eisenberg (Eds.),
14, 717–737. Neurobehavioural recovery from head injury (pp. 398 – 415). New York:
*Schmitter-Edgecombe, M., & Nissley, H. M. (2000). Effects of divided Oxford University Press.
attention on automatic and controlled components of memory after van Zomeren, A. H., & Brouwer, W. H. (1992). Assessment of attention.
severe closed-head injury. Neuropsychology, 14, 559 –569. In J. R. Crawford, D. M. Parker, & W. W. McKinlay (Eds.), A handbook
*Schmitter-Edgecombe, M., & Rogers, W. A. (1997). Automatic process of neuropsychological assessment (pp. 241–266). Hove, England: Erl-
development following severe closed head injury. Neuropsychology, 11, baum.
296 –308. van Zomeren, A. H., & Brouwer, W. H. (1994). Clinical neuropsychology
Shallice, T. (1988). From neuropsychology to mental structure. Cam- of attention. New York: Oxford University Press.
bridge, England: Cambridge University Press. Williamson, D. J. G., Scott, J. G., & Adams, R. I. (1996). Traumatic brain
Sharpe, D. (1997). Of apples and oranges, file drawers and garbage: Why injury. In R. L. Adams, O. A. Parsons, J. L. Culbertson, & S. J. Nixon
validity issues in meta-analysis will not go away. Clinical Psychology (Eds.), Neuropsychology for clinical practice. Etiology, assessment and
Review, 17, 881–901. treatment of common neurological disorders (pp. 9 – 64). Washington,
*Shum, D. H. K., McFarland, K., & Bain, J. D. (1994). Effects of closed- DC: American Psychological Association.
head injury on attentional processes: Generality of Sternberg’s additive *Willison, J., & Tombaugh, T. N. (2006). Detecting simulation of attention
factor method. Journal of Clinical and Experimental Neuropsychol- deficits using reaction time tests. Archives of Clinical Neuropsychol-
ogy, 16, 547–555. ogy, 21, 41–52.
*Shum, D. H. K., McFarland, K., Bain, J. D., & Humphreys, M. S. (1990). *Wilson, B. A., Evans, J. J., Emslie, H., Balleny, H., Watson, P. C., &
Effects of closed-head injury on attentional processes: An information- Baddeley, A. D. (1999). Measuring recovery from post traumatic am-
processing stage analysis. Journal of Clinical and Experimental Neuro- nesia. Brain Injury, 13, 505–520.
psychology, 12, 247–264. *Withaar, F. K., & Brouwer, W. H. (2003). Divided attention after closed
*Simpson, A., & Schmitter-Edgecombe, M. (2000). Intactness of inhibi- head injury. Zeitschrift fur Neuropsychologie, 14, 203–211.
tory attentional mechanisms following severe closed-head injury. Neu- *Zahn, T. P., & Mirsky, A. F. (1999). Reaction time indicators of attention
ropsychology, 14, 310 –319. deficits in closed head injury. Journal of Clinical and Experimental
*Spikman, J. M., Deelman, B. G., & van Zomeren, A. H. (2000). Executive Neuropsychology, 21, 352–367.
functioning, attention and frontal lesions in patients with chronic CHI. Zakzanis, K. K. (2001). Statistics to tell the truth, the whole truth, and
Journal of Clinical and Experimental Neuropsychology, 22, 325–338. nothing but the truth: Formulae, illustrative numerical examples, and
Spikman, J. M., van Zomeren, A. H., & Deelman, B. G. (1996). Deficits of heuristic interpretation of effect size analyses for neuropsychological
attention after closed-head injury: Slowness only? Journal of Clinical researchers. Archives of Clinical Neuropsychology, 16, 653– 667.
and Experimental Neuropsychology, 18, 755–767. Zakzanis, K. K., Leach, L., & Kaplan, E. (1999). Neuropsychological
Spreen, O., & Strauss, E. (1998). A compendium of neuropsychological differential diagnosis. Lisse, the Netherlands: Swets & Zeitlinger.
tests: Administration, norms, and commentary. New York: Oxford Uni-
versity Press.
*Stablum, F., Leonardi, G., Mazzoldi, M., Umilta, C., & Morra, S. (1994). Received March 15, 2006
Attention and control deficits following closed head injury. Cortex, 30, Revision received November 6, 2006
603– 618. Accepted November 20, 2006 䡲

You might also like