anxiety
Sofie ValkUniversity of Amsterdam Brain and Cognitive Sciences
Track: Cognitive Sciences Feb – July 2012 30 EC Supervisor: Dr. Sonia Bishop U.C. Berkeley Coassessor: Elise Seip, PhD University of Amsterdam
The aim of this study was to assess non-‐emotional attention training in order reduce anxiety in high anxious individuals. Our study had several sub-‐goals, (1) we aimed to verify the reliability of the tasks we used, (2) we aimed to implement an emotion-‐free training paradigm in order to improve sustained attention and reduce trait anxiety. We conducted a behavioral study with healthy volunteers. We measured the reliability of the dot-‐probe paradigm, a frequently used measure of attention bias, and a letter-‐search task of high and low perceptual load superimposed on fearful faces, ambiguous faces, negative faces, or houses. However, we found no test-‐retest reliability in the dot-‐probe task and the letter-‐search task. We implemented an emotion-‐free training paradigm by adjusting the regular sustained attention to response task (SART) by adding audio cues and performance feedback to the task. Performance in the regular SART and the “training” SART was found reliable across days. Participants performed the training task for one hour on three consecutive days. The effect of the training was assessed with the regular SART without feedback, the dot-‐probe task and level of trait anxiety. We found that our participants’ performance in the SART task decreased after pre-‐training assessment, and we found that the training with feedback had a negative effect on performance. Furthermore we found a significant increase in trait anxiety in the low anxious group, the high anxious group showed no difference in trait anxiety level. Also the dot-‐probe paradigm showed not consistent results, possibly due to its unreliability. These findings suggest that the current assessment tasks for attention bias are unreliable and underline the importance of verification of reliability of used tasks in general. Furthermore we found our training paradigm did not improve performance in the tasks and even had a negative effect on level of trait anxiety and performance in the SART task. Possible causes and improvements on the current paradigm are discussed.
Introduction ... 4
Methods ... 7
Overview of tasks: ... 8
Study part I ... 12
Procedure ... 12
Results ... 13
Study part II ... 14
Procedure ... 14
Results ... 14
Self-‐reported personality measures: results ... 15
Assessment and training of sustained attention ... 16
Assessment of attention bias in high anxious individuals ... 18
Discussion ... 20
Literature ... 24
Appendix ... 29
Introduction
Anxiety is the displeasing feeling of fear and concern (Cannistraro & Rauch, 2003; Davison, 2003; Seligman, Walker, & Rosenhan, 2000) and is a generalized mood that can occur without an identifiable triggering stimulus. Anxiety disorders affect around one in five individuals (Cannistraro & Rauch, 2003; Kessler, Chiu, Demler, & Walters, 2005) creating a significant clinical burden (Kessler et al., 2003). Many affected individuals do not receive treatment (Mojtabai, Olfson, & Mechanic, 2002), and among those who do, many continue to suffer (Ballenger, 2001; Doyle & Pollack, 2003; Pine, Helfinstein, Bar-‐Haim, Nelson, & Fox, 2009). It is necessary to continue refining existing treatments and actively pursue more efficacious ones. Recently attention bias modification (Bar-‐Haim, 2010; Matthews & MacLeod, 2002) has started to emerge as one such treatment. The aim of attention bias modification is to alter the way people focus their attention in order to relieve symptoms of anxiety. This theory-‐driven treatment is based upon established experimental data on attention biases in anxiety. It has been found that people with high anxiety have an attention bias toward negative information, or even a more general attention bias that possibly reinforces the anxiety mechanism in the brain (Bar-‐Haim, Lamy, Pergamin, Bakermans-‐ Kranenburg, & Van IJzendoorn, 2007; Bishop, Jenkins, & Lawrence, 2007). Following the landmark study of Mathews & MacLeod in 2002, where they described that induced processing biases have a causal effect on anxiety, processing bias has been further explored as a possible new therapy to reduce anxiety. Multiple studies report to have reduced anxiety level with attention bias modification (Amir, Beard, Burns, & Bomyea, 2009; Amir, Bomyea, & Beard, 2010; Browning et al., 2011; Eldar & Bar-‐Haim, 2010; Klumpp & Amir, 2009). Traditional attention bias modification is assessed using tasks with emotional content, such as images of fearful faces. However, modulatory effects of emotion-‐free attention training paradigms on anxiety level, where the attention deficit itself is also addressed, as well as some methodological concerns and improvements of commonly used experimental paradigms are yet to be addressed.
Anxiety can be conceptualized in two ways: trait anxiety and state anxiety. State anxiety is a transitory emotional state, or arousal, to perceived dangerous stimuli, while trait anxiety is a relatively stable behavioral disposition to respond anxiously to a wide range of perceived threatening stimuli (Spielberger, Gorsuch, & Lushene, 1983). Initial evidence suggests there is a disruption of the amygdala-‐ prefrontal circuitry in high anxious individuals, with deficient recruitment of prefrontal attention mechanisms found related to trait anxiety level and amygdaloidal hyper-‐responsiveness to threat found related to state anxiety (Bishop, 2007). More specifically, individuals in general are reported to show an
increased amygdala response to attended threat-‐related stimuli (fearful faces) compared to non-‐threat-‐ related stimuli (neutral faces), but only high state anxious volunteers showed an increased amygdala response to unattended threat-‐related stimuli (Bishop, 2007; Bishop, Duncan, & Lawrence, 2004). Trait anxiety, on the other hand, seems to specifically relate to threat-‐related biases in selective attention expressed by a reduced prefrontal control signal (Mathews & Mackintosh, 1998; Bishop, 2009). Increased attention to threat normally serves an adaptive function by facilitating the detection of danger (LeDoux, 2000). Considerable research demonstrates, however, that the attention system of anxious individuals is impaired (Bar-‐Haim et al., 2007; Mogg & Bradley, 1998). High trait anxious individuals have an attention bias favoring the threatening location (Derryberry & Reed, 2002), regardless of the actual threat value of a given stimulus. Furthermore, it has been shown that trait anxiety can be linked to impoverished recruitment of prefrontal attention control mechanisms to inhibit distractor processing especially when attention demands are low (Bishop, 2009).
This impaired attention control system found in high trait anxious individuals might relate to their style of information processing, rather than a deficit per se. If so, then it may be possible to train high trait anxious individuals to maintain attention focus more effectively. Attention regulation can be successfully manipulated through repeated practice (Amir et al., 2008; Eldar et al., 2008; Klumpp & Amir, 2010; MacLeod et al., 2002; Wadlinger & Isaacowitz, 2011) and might aid in normalizing the prefrontal attention mechanisms that are found to be deficient in high trait anxious populations (Bishop, 2009). Attention bias modification is believed to specifically target early attention processes by way of implicit/unconscious intervention (Bar-‐Haim, 2010; Eldar & Bar-‐Haim, 2010; Monk et al., 2008)
At present, paradigms used in attention training research consist of assessments of anxiety level and attention bias followed by training on a specific attention related task. The task used most often is the dot-‐probe task (MacLeod, Mathews, & Tata, 1986). In the dot-‐probe task, stimuli with neutral and emotional valence are presented in pairs on a computer screen. After these stimuli disappear, a visual probe (usually one or two dots) appears on the screen and participants have to indicate the identity of the probe. Response time serves as an indicator of where attention has been allocated. In an integrative review of recent studies on cognitive training with attention bias modification protocols, Hakamata and colleagues found that these interventions are effective, especially in clinical samples (Hakamata et al., 2010). However, there are several theoretical considerations to be addressed concerning these protocols. Firstly, it remains unknown whether the anxiety-‐reducing effect of these commonly used protocols might reflect a more general attentional control process (Bar-‐Haim, 2010). For example,
training paradigms that bias attention away from threat were compared with those that bias attention toward threat (Klumpp & Amir, 2009) and it was found that, in both cases, participants exhibited a relative decrease in anxiety during a speech challenge compared to participants in a placebo control condition. This could indicate that it is not the threat-‐related component of the training that benefits high anxious individuals, but only the attention aspect of the training. Secondly, most studies use the same task (with slight modifications) for both attention training and assessment. Therefore the results should be interpreted with caution, as the pre-‐ to post-‐training assessment differences may only reflect a habituation effect on the specific task used. Furthermore, the commonly used dot-‐probe paradigm has been reported to have relatively low test-‐retest reliability (Browning et al., 2011; Lebel & Paunonen, 2011; Staugaard, 2009; Schmulke, 2005). This means that testing from one day to the next could result in different effect sizes. High test-‐retest reliability is desirable for assessment, since the measurements used should reflect an objective standard, such as attention bias, and not change over time (Lebel & Paunonen, 2011).
The aim of the current study is to develop a non-‐threat-‐related attention training protocol with feedback in order to reduce anxiety and negative attention biases, including the use of a different task to assess the efficacy of this training. At the same time, we aim to address the low test-‐retest reliability present in commonly used attention training paradigms by exploring the option of another assessment task for attention bias.
In order to investigate the test-‐retest reliability in of our proposed paradigm, the reliability of the dot-‐ probe (MacLeod, Matthews, & Tata, 1986), the letter search task (Bishop, Jenkins, & Lawrence, 2007; Jenkins, Lavie, & Driver, 2005; Lavie, 2001) and the SART (Sustained Attention to Response Task, Robertson, Manly, Andrade, Baddeley, & Yiend, 1997) are explored and compared. The letter search task is an adaptation of a task commonly used to study perceptual load (Bishop, Jenkins, & Lawrence, 2007; Jenkins, Lavie, & Driver, 2005; Lavie, 2001). Bishop and colleagues found significant interactions between trait anxiety and performance in this task that we hoped to replicate. The SART is a non-‐ emotional go/no-‐go task, and has been successfully used to identify sustained attention deficits in stroke patients (Roberson, Manly, Andrade, Baddeley, & Yiend, 1997). In order to investigate attention training, here we used a variant of the SART with feedback for three days of attention training (“training SART”), and we assessed the efficacy of this non-‐emotional training using both the dot-‐probe task and a non-‐feedback variant of the SART (“regular SART”). The feedback we chose to implement was based on a previous study that showed auditory feedback at random intervals is beneficial in the context of the
SART (Manly et al., 2004). They found that periodic audio cues had a positive effect on accuracy without slowing down reaction time. Thus, we expected feedback during the training days would help participants to improve their training SART performance (Manly et al., 2004) as well as improve their performance in the regular SART and reduce attention biases in the dot-‐probe task. Furthermore we will provide our participants with visual feedback during the task breaks, to indicate their increase or decrease in reaction time and error rate (Hattie & Timperley, 2007). In order to assess the effect of the training we compared performance on the regular SART and dot probe, and self-‐reported trait anxiety, on the last day to the first day. Our expectations were threefold: (1) to find improvement in the regular SART, expressed by fewer mistakes without slowing of reaction time, (2) to find improvement in the dot-‐ probe task, signified by a smaller attention bias to threat, and (3), to find a decrease in trait anxiety.
Methods
ParticipantsAll participants were students at the local university. They were right handed with normal or corrected vision, and learned English as their first language. They were not taking any psychotropic medication and were unfamiliar with the tasks presented. Twenty-‐three participants (four male) took part in the study part 1 (mean age ± SD: 20.9 ± 1.6; range: 18-‐24 years). Twenty-‐three participants (seven male) took part in the study part 2 (mean age ± SD: 22.1 ± 2.9; range: 19-‐29 years). One participant (female) was excluded, since she was unable to participate in all five sessions, resulting in 22 remaining participants. Questionnaires
All participants completed several self-‐report questionnaires before beginning the experiments. They included the State Trait Anxiety Inventory (STAI: Spielberger et al., 1983), which provides a measure of anxiety in the present moment (state) and in general (trait), as well as the Attentional Control Scale (ACS: Derryberry & Reed, 2002), which provides a measure of ability to voluntarily control attention. The range of scores on both questionnaires is 20-‐80, with higher scores indicating greater anxiety or attentional control, respectively.
Overview of tasks:
All tasks were presented via custom scripts using the VisionEgg module (Straw, 2008) for Python 2.6 (van Rossum & Drake, 2001).
Letter Search Task
The letter search task was adapted from Jenkins and others (2005, experiment 2, and similar to ones used in previous perceptual load studies (Bishop et al., 2007; Lavie, 2001). On each trial, a string of six letters superimposed on a task-‐irrelevant unfamiliar face or house was presented. The face stimuli comprised four different individuals with fearful and neutral expressions taken from the Pictures of Facial Affect Series (Ekman & Friesen, 1976) as well as morphed versions of these faces (50% fear, 50% surprise; using FantaMorph Software; www.fantamorph.com, Abrosoft, 2012). Face stimuli were cropped with an oval to remove extraneous background information, and the background was filled with gray. Fearful facial expressions of conspecifics act as cues to potential danger, which have been shown to share some of the functional properties of “prepared” (intrinsically threat-‐related) fear stimuli (Lanzetta & Orr, 1986). They have also been used in the majority of studies arguing for the pre-‐attentive and automatic processing of threat-‐related stimuli (Bishop et al., 2007; Vuilleumier, Armony, Driver, & Dolan, 2001; Whalen et al., 1998; Whalen et al., 2004). Morphed faces were used because previous unpublished work from our lab showed that morphs specifically attracted more attention of the high anxious individuals due to their ambiguity. Furthermore, in addition to neutral faces, we used four images of houses as a non-‐facial neutral baseline, since neutral faces could appear threatening to high anxious individuals (Yoon & Zinbarg, 2008). The house stimuli were matched in size with the faces. In the high perceptual-‐load condition, the letter string comprised a single target letter (X or N) and five non-‐ target letters (H, K, M, W and Z) arranged in a pseudo-‐random order. In the low perceptual-‐load condition, the letter string comprised six Xs or six Ns, reducing attention search requirements. The high and low perceptual-‐load conditions were mixed within each block. The task was to indicate as quickly and accurately as possible whether there was an X or N in the letter string.
The main reaction time contrasts that we planned were between conditions with emotional valence (fear and morph) and those with non-‐emotional valence (neutral faces and houses). Therefore, fearful and morphed faces were contrasted with neutral faces and houses in both low and high load conditions, for both day 1 and day 2.
Table 1: Planned Letter Search task reaction time contrasts
Day 1 Day 2
Low Load High Load Low Load High Load
Fear – Neutral Fear – Neutral Fear – Neutral Fear – Neutral Fear – House Fear – House Fear – House Fear – House Morph – Neutral Morph – Neutral Morph – Neutral Morph – Neutral Morph -‐ House Morph -‐ House Morph -‐ House Morph -‐ House
We expected to (1) find test-‐retest reliability across two days for all contrasts and (2) a correlation between anxiety level and RT difference (specifically under low load), where we expected high anxious individuals to be slower in the fearful and morphed face conditions in comparison to the house and neutral face conditions. Furthermore (3) we expected to find that the morphed face and the house RT difference would be largest because these images were the most different in emotional value. The level of accuracy was also measured, and we expected to find high anxious individuals to be less accurate in fearful and morphed face conditions in comparison to the house and neutral face conditions.
Sustained Attention to Response Task (SART) 1) Regular SART
The regular SART was adapted from a study that monitored sustained attention failures in patients with traumatic brain injury (Robertson et al., 1997). In each session, there were six runs consisting of 10 blocks each, divided evenly between letter and number blocks. Each block had 28 trials. Each run took about six minutes to complete. Participants were allowed a brief rest after each of the first five blocks. On each trial in the number blocks, a random digit from 1 to 9 was presented in the center of the screen for 250ms. The digits varied in size from trial to trial to prevent low-‐level visual influences on response. Following, a mask comprised of a zero superimposed on the letter X was presented during the Inter-‐ Stimulus-‐Interval (ISI) that varied randomly from pre-‐selected ISIs (0.75, 0.85, 0.95, or 1.05 s). A variable ISI was included to prevent participants falling into a regular temporal pattern of responding, succumbing to a speed-‐accuracy trade-‐off. The participants’ task was to press the spacebar upon presentation of each digit (go-‐trials) with the exception of the two or three occasions per block when the digit 3 (target) appeared, where they were required to withhold/inhibit their response (no-‐go trials). Letter blocks used letter stimuli rather than numbers in order to compare RTs between runs with only go trials and runs with both go and no-‐go trials.
We expected high anxious individuals would be faster in responding while making commission errors (failure to withhold a response on no-‐go trials). Furthermore, we expected high anxious individuals to make more commission errors. We also expected a negative relationship between attention control and number of commission errors, i.e., we predicted that participants with lower attention control would have more trouble inhibiting their response on no-‐go trials.
2) Training SART
The training SART was adapted from the SART, with the same set-‐up. However, instead of six runs there were only three. Each of the 10 blocks had double the number of trials (54 instead of 28). Instead of two to three presentations of the target digit 3, there were three to four presentations per block. During the training task, feedback was provided, by way of real-‐time audio feedback and performance feedback during the breaks. The audio feedback was made with the Audiere module for Python (audiere.sourceforge.net). It consisted of a short beep that sounded for 500ms every time the participants’ reaction time difference from one trial to the next deviated more than 0.75 SDs for four out of six “test” trials from the average reaction time difference of the previous six “baseline” trials.
Figure 1: A screenshot of the performance feedback in the training SART.
The performance feedback was presented during the breaks every five blocks (see: Figure 1). The reaction time feedback was based on the average reaction time during go trials in these five blocks in comparison to the average go trial reaction time overall. If the reaction time was faster than the average
reaction time, the diamond would be colored green and would be placed right of the vertical gray line; if the reaction time was slower than the average reaction time, the diamond would be colored red and placed left of the middle line. The baseline reaction time we used during practice was 500ms. Then, the average reaction time found during practice was used as the baseline during the main training task. It was hoped that providing feedback on reaction time would help participants be more aware of the speed with which they responded, which in turn may have reduced commission error rate. This was based on the finding that drifting of go-‐trial reaction times is a useful marker of increases in error rate (Manly, Robertson, Galloway, & Hawkins, 1999; Robertson, Manly, Andrade, Baddeley, & Yiend, 1997). In addition, we provided feedback on the average commission error rate during the last five blocks. If the commission error rate was less than the average commission error rate, the diamond would be colored green and would be placed right of the vertical gray line that indicates the previous commission error rate; if the commission error rate was more, the diamond would be colored red and place left of the vertical gray line. Furthermore, the location of the vertical gray line would indicate the average commission error rate overall, where the axis goes from 100% errors to 0% errors.
Dot-‐probe task
The dot probe task is a visual-‐probe task (MacLeod, Mathews, & Tata, 1986; MacLeod et al., 2002) that is used to assess attentional bias to emotional information. The version we used has been adapted from Browning and colleagues (2011). On each trial, a pair of faces was briefly presented on a black background and followed by a probe (two horizontal or two vertical white dots), which appeared behind one of the faces. The faces were taken from the Pictures of Facial Affect Series (POFA; Ekman & Friesen, 1976), the NimStim face set (Tottenham et al., 2009), and the Karolinska Directed Emotional Faces (KDEF; Lundqvist, Flykt, & Ohman, 1998). They displayed positive (happy), neutral, or negative (fearful and angry) expressions, with each trial of the task presenting faces from two different valences and identities, but with the same gender. These combinations resulted in three possible face pair types: positive-‐neutral, positive-‐negative, and negative-‐neutral. During the task, the relative position of the faces (upper or lower) was counterbalanced throughout the experiment. Probe location was also counterbalanced between the relatively more negative and positive stimulus location, and stimulus duration (100ms, 500ms, 1000ms) was blocked. This resulted in six blocks of 48 trials, with the order of blocks counterbalanced (two blocks per stimulus duration). A unique set of faces, mixed between the POFA, NimStim and KDEF, was used for pre-‐ vs. post-‐assessment. During practice, simple colored geometric shapes were used in place of faces to avoid habituation while participants learned the button mapping for responding to the probes.
Importantly this task is able to assess whether the attention training produced, as expected, a reduction in high anxious individuals’ tendency to direct attention to negative information. We expected to find a correlation between level of anxiety and the mean reaction time difference between negative invalid and negative valid trials. That is, high anxious participants will be slower on average to respond on trials where the dot probe replaces the neutral face when a negative face is present (“neutral valid, negative invalid”) compared to trials when the dot probe replaces the negative face when a neutral face is present (“negative valid, neutral invalid”). We also expected to find a reduction in the mean RT difference between these conditions during post-‐training, indicating a reduction in attention bias. The same estimation of attention bias is done for the positive image conditions (“positive valid, neutral invalid” contrasted with “neutral valid, positive invalid”), however we did not hypothesize any relationship between anxiety level and RT to positive images.
Table 2: Dot Probe reaction time contrasts.
Day 1 Day 5
NegValidNeuInvalid -‐ NeuValidNegInvalid NegValidNeuInvalid -‐ NeuValidNegInvalid PosValidNeuInvalid -‐ NeuValidPosInvalid PosValidNeuInvalid -‐ NeuValidPosInvalid
Study part I
The goal of this study was to explore an alternative assessment task for attention bias in high anxious individuals. The test-‐retest reliability for both the letter search task and the regular SART were measured.
Procedure
The study took part in one hour on four consecutive days, at the same time of day. Participants were tested in a quiet room at the research unit. Upon first arriving, the participants filled out the STAI (Spielberger et al., 1983) and ACS (Derryberry & Reed, 2002). The experimenter read aloud the task instructions while the participants followed along. The participants then performed a short practice in order to familiarize themselves with the task with the instructor in the room. After the practice the instructor started the task and left the room. The letter search task was administered on the first two days and the SART was administered on the second two days (see: Table 3).
Table 3: The schedule for study part I.
Day 1 Day 2 Day 3 Day 4
Questionnaires
Letter-‐search task Letter-‐search task Regular SART Regular SART
Results
One participant (female) had to be excluded from the letter search task for not responding on more than 10% of the trials. Five participants (four female) had to be excluded from the SART for having greater than 10% omission rate (not responding on go trials).
Questionnaires
In our group, we found a median trait anxiety ± SD of 39 ± 9.9, range: 23-‐58 and a mean attentional control score ± SD of 55 ± 4.6, range: 45-‐62. These scores are typical of those published previously. Test-‐retest reliability
The letter search task: none of the eight reaction time contrasts reached significant test-‐retest reliability, neither in the case of reaction time nor accuracy.
The Regular SART: Overall there was good test-‐retest reliability for go trial reaction time (r=0.81, p<0.00), commission error reaction time (r=0.56, p<0.02) and for commission error rate (r=0.84, p<0.00). There was poor test-‐retest reliability for omission error rate (r=0.17, p<0.51).
Correlation with trait anxiety
The letter search task: under low perceptual load, the day 1 morph-‐neutral RT difference (r=-‐0.49, p<0.02) and the morph-‐fear RT difference (r=-‐0.539, p<0.01) correlated negatively with trait anxiety. This indicates that high anxious individuals are significantly faster in the morph condition compared to the fear and neutral conditions. However, this was not the case on day 2. No other contrasts based on reaction time under low or high perceptual load were significant. Furthermore, under high perceptual load, the day 1 morph-‐neutral accuracy difference (r=0.430, p<0.05) and the morph-‐house accuracy difference (r=0.510, p<0.02) correlated significantly with anxiety. On day 2, the morph-‐house accuracy difference (r=462, p<0.03) also correlated significantly with anxiety. This indicates that high anxious individuals were significantly more accurate in the morph condition in comparison to in the neutral and fear conditions.
The Regular SART: The RTs and errors rates in the SART task did not significantly correlate with trait anxiety.
Study part II
The goal of the second part of the study was to investigate whether the SART training paradigm could be used to reduce attention bias in high anxious participants, reduce trait anxiety and improve sustained attention.
Procedure
The study took part in one hour on five consecutive days, at the same time of day. A subset (n=11) of the participants returned for a sixth day 19-‐42 days (mean ± SD 27.3 ± 8.4) after day 1. The participants were tested in a quiet room at the research unit. Upon first arriving, the participants filled out the STAI (Spielberger et al., 1983) and the ACS (Derryberry & Reed, 2002). The experimenter read aloud the task instructions while the participants followed along. The participants then performed a short practice in order to familiarize themselves with the task with the instructor in the room. After the practice the instructor started the task and left the room.
Table 4: The schedule for the study part II.
Results
Three participants had to be excluded from both the SART and the training SART due to a greater than 10% omission rate (two female participants) and failure to complete the study (one female) participant). One female participant had to be excluded from the dot probe due to failure to complete the study. Two participants (two female) were excluded from the follow-‐up due to a greater than 10% omission rate. After this, the number of participants in both the training and the regular SART is twenty and the number of participants in the dot probe is twenty-‐two. The follow-‐up study included nine participants. The two sub-‐groups of participants who participated in Study Part I and Study Part II were not significantly different from each other in trait anxiety (t(22)=1.33 p>0.19) nor attention control score (t(22)=0.50, p>0.62).
Day 1 Day 2 Day 3 Day 4 Day 5 Follow-‐Up
Questionnaires Dot-‐probe Regular SART
Training
SART Training SART Training SART Questionnaires Dot-‐probe Regular SART
Questionnaires Dot-‐probe Regular SART
Self-‐reported personality measures: results
Questionnaires
Trait anxiety scores were evenly distributed, with a median of 35 and n=11 people in the low anxious group, and n=11 people in the high anxious group. Furthermore, attentional control scores had a median of 54 and n=11 people in the low attentional control group and n=11 people in the high attentional control group. Anxiety and attentional control were not significantly correlated (r=-‐.230, p>0.30).
Effects of training on trait anxiety
Trait anxiety
There was a significant difference between the trait anxiety score on day 1 and day 5 in the low anxious group (m±sd day 1: 29.8±3.8, day 5 34.5±5.9, t(10)=-‐2.73, p<0.03). However there was no significant difference in the high anxious group (m±sd day 1: 43.4±5.3, day 2: 43.4±8.78). There was no significant correlation between the difference in trait anxiety pre-‐ to post-‐training and the trait anxiety level pre-‐ training (Figure 2).
Figure 2: Average trait anxiety scores in the low and high anxious groups for day 1 and day 2. * 0 10 20 30 40 50 60
Low Anxious High Anxious
Trait difference
Day 1 Day 5
Assessment and training of sustained attention
Test-‐retest reliability – SART Regular SART
The findings of reliability in the SART task from Study Part I are replicated in this study, with go trial reaction time (r=0.806, p<0.00) and commission error rate (r=0.724, p<0.00) consistent across days (day 1 and day 5).
Training SART
The SART task is also reliable in training, with go trial reaction time (r=0.944, p<0.00) and number of target commission errors (r=0.721, p<0.00) consistent across days (day 2 and day 3).
Sustained attention training task
In order to compare the number of commission errors between the training SART and the regular SART is it necessary to compare the error ratio, which is the number of commission errors divided by the total number of opportunities. Comparing the reaction times and commission error rate of the training SART with the regular SART, we found performance to significantly worsen with feedback. The go trial reaction time speeded up between the first assessment day and the second training day (m±sd day 1: 0.458±0.09, day 2: 0.400±0.07, t(19)=3.39, p<0.003), and the commission error rate increased (m±sd day 1: 0.242±0.14, day 2: 0.398±0.16, t(19)=-‐4.43, p<0.00). In fact, in comparison to day 1, the average reaction times for go trials were faster in all the following days (see figures 3 and 4).
Improvement in sustained attention -‐ pre-‐ vs. post-‐training.
The average reaction time per block in the SART task speeded up significantly between the pre-‐training assessment and the post-‐training assessment (m±sd day 1: 0.46±0.10, day5: 0.38±0.05, t(19)=3.88, p<0.00). The number of commission errors increased significantly as well (m±sd: day 1: 0.77±0.50, day 5: 0.99±0.48, t(19)=-‐2.18, p<0.04). However, there were no significant correlations between error ratio change pre-‐ to post-‐training and day 1 trait anxiety level or day 1 attention control.
Improvement sustained attention -‐ follow-‐up
In an exploratory follow-‐up with a subset of our participant group of study 1b (n=9), trait anxiety m±sd 36±5.8, attention control m±sd 52 ±6.8), we found that the patterns of decreased reaction time and increased error ratio indeed persisted even after a few weeks. However, since only 9 of our participants returned, the results can only interpreted for exploratory purposes (see Appendix, figures 7 and 8).
Figure 3: Reaction time across days for go trials in the regular and training SART. Days 2-‐4 were the training days,
marked in pink.
Figure 4: Commission error ratio across days, indicating the number of errors divided by the number of
opportunities. Days 2-‐4 were the training days, marked in pink. 0.000 0.100 0.200 0.300 0.400 0.500 0.600 1 2 3 4 5
SART-‐ Go trial reacRon Rme
0.000 0.050 0.100 0.150 0.200 0.250 0.300 0.350 0.400 0.450 0.500 1 2 3 4 5
Assessment of attention bias in high anxious individuals
Test-‐retest reliability Dot-‐probe:
In order to measure attention bias for negative images, the difference between the reaction time for the two different neutral and negative contrasts is measured, as discussed in the Methods. It is expected that high anxious individuals respond more slowly when the probe replaces the neutral face, because they are attentive to the negative face. To measure test-‐retest reliability, the difference in bias between the first day of testing, day 1, and the last day of testing, day 5, is measured.
The dot-‐probe task is not reliable (table 5). The reaction time in the 500ms condition in the negative and neutral versus neutral and negative contrast seemed to be reliable across days at first. Further inspection showed us that the high correlation across days in the 500ms duration of the NegValidNeuInvalid -‐ NeuValidNegInvalid contrast stems from one data point. When this point was removed, the correlation between days was r=0.047, which is not significant.
Table 5: The test-‐retest reliability per stimulus duration.
Day 1 – Day 5
Duration NegValidNeuInvalid -‐ NeuValidNegInvalid PosValidNeuInvalid -‐ NeuValidPosInvalid
100 r=0.279 r=0.228
500 r=0.591 r=-‐0.071
1000 r=-‐0.195 r=0.316
Improvement dot-‐probe pre-‐ vs. post-‐training
We found no significant differences between the high and low anxiety groups for any of the stimulus durations in the negative valid, neutral invalid versus the neutral valid, negative invalid contrast, for day 1 or day 5 (figures 5 and 6). Furthermore there were no significant changes across days, either in the group as a whole, or in the low or high anxious group specifically. We hypothesized that the high anxious group would be slower in moving attention away from the negative images, so we expected the neutral valid, negative invalid reaction time to be slower than the negative valid, neutral invalid reaction time. However, we found this was only the case in the 100ms and 1000ms stimulus duration conditions on day 1, but this was not significantly different from the reaction time difference found in the low anxious group. In the 500ms condition, the high anxious group was faster in the neutral valid, negative invalid
condition. Post-‐training, this effect was reversed. However, there was no significant difference between groups.
The positive valid and neutral invalid versus the neutral valid and the positive invalid showed the same pattern of results, with no findings relevant to attention bias in anxiety.
Figure 5: Negative valid, neutral invalid vs. neutral valid, negative invalid on Day 1. A positive RT difference indicates that reaction time was faster when the dot replaced the neutral image.
Figure 6: Negative valid, neutral invalid vs. neutral valid, negative invalid on Day 5. A positive RT difference indicates that reaction time was faster when the dot replaced the neutral image.
Discussion
The aim of this pilot study was to assess non-‐emotional attention training in order reduce anxiety in high anxious individuals. Our study had several sub-‐goals, (1) we aimed to verify the reliability of the tasks we used, (2) we aimed to implement an emotion-‐free training paradigm in order to improve sustained attention and reduce trait anxiety. We found no test-‐retest reliability in the dot-‐probe task and the letter-‐search task; we found that our participants’ performance in the SART task decreased after the day one assessment, and we found that the training with feedback had a negative effect on performance. In the next section, we will address the possible causes of our findings, the implications for interpretability of past results and provide suggestions for future improvements.
Participant groups. Our participants’ level of trait anxiety was comparable to the published norms for this age group (Spielberger, 1983). However, we only had two people with a trait anxiety score higher than 50. Previous research has suggested that the expected effects of anxiety on dot-‐probe reaction time only show when people are clinically or highly anxious (Hakamata, 2010), instead of moderately anxious, as was the case on our group. Furthermore, it has been found that high anxious individuals with low attention control specifically benefit from attention training to reduce anxiety (Derryberry & Reed, 2002; Bar-‐Haim, 2010). Within our participants however, only four high anxious individuals reported low attention control. This low number may have influenced our results. Lastly, taking into account the demographics of our subjects, who were all students in the final period of their semester, stress and sleep deprivation might have biased our results.
Reliability. We found that the dot probe task has a low reliability, as has been described earlier in the literature (Lebel & Paunonen, 2011; Staugaard, 2009; Schmulke, 2005). The task we wanted to assess alternatively, the letter-‐search task, also showed low reliability. Reliability is an important issue for many implicit tests (Lebel & Paunonen, 2011). Thus, it is imperative to develop measurements that are indeed reliable in order to accurately assess effects of training. The results also indicated there were no consistent significant correlations between anxiety and planned contrasts.
Interestingly, within the letter-‐search paradigm, the highest correlations we found were all contrasts with the morph face condition. The morph face was included in the study in order to include an additional ambiguous image that could help differentiate between high and low anxious participants. Negative biases in the interpretation of emotionally ambiguous stimuli have all been held to characterize anxiety (Lissek et al., 2005; MacLeod, Rutherford, Campbell, Ebsworthy, & Holker, 2002;
Mathews & MacLeod, 2002; Phelps, Delgado, Nearing, & LeDoux, 2004). Our results indicate there is a consistent difference between low and high anxious individuals in the processing of morph faces, however, not in the way we initially hypothesized. In the contrasts with other images, whether fearful faces, neutral faces or houses, we did not find this effect. High anxiety was related to a relatively faster reaction time and more accurate responses in trials with ambiguous images. This effect was not present in low anxious individuals. Although contrary to our expectations and unpublished work in the lab, it seems high anxious people recruited cognitive control mechanisms more strongly in the case of ambiguous faces. Also contrary to previous research (Bishop, Jenkins, & Lawrence, 2007), we did not find a higher correlation of anxiety and planned contrasts in the low perceptual load condition.
In the dot-‐probe paradigm, the reaction time is sometimes faster when the dot replaces a negative image and the negative image is invalid and sometimes faster when the dot replaces the negative image and the neutral image is invalid. This effect might be caused by different attention systems (Bar-‐Haim, 2010), however, our findings are highly inconsistent so concrete conclusions cannot be drawn.
The SART task was found to be reliable across days. However, there was no correlation with anxiety scores. Furthermore, we found our participants’ performance decreased in the task after the first day. An exploratory follow-‐up with a sub-‐group of participants a few weeks later informed us that this decrease in performance was consistent, even a month later. One explanation could be that the participants, when unfamiliar with the task, were more attentive. After the first experiment day, the participants might have been not motivated to pay attention and tried to rush through the task by just pressing the spacebar as soon as possible, even though this had no effect on length of the task. Indeed, during debriefing, participants mentioned their boredom with the task and their feeling of lack of challenge in the task, as well as the feeling that improvement was impossible.
Training. Feedback had no positive effect on performance of the SART task. There can be two explanations for this. It could be that during the training, they aimed at improving the reaction time only. In our current design, gradually speeding up reaction time is associated with positive reaction time feedback and possibly even no audio feedback. Furthermore, many participants indicated they did not find the audio feedback helpful and even mentioned it could be distracting. A previous study has shown feedback is beneficial in the context of the SART (Manly et al., 2004), however they implemented the feedback between different sub-‐blocks in the experiment and contrasted the reaction time and omission errors with blocks that had no feedback. This might have resulted in different effects. Although it seems intuitively correct to use feedback to aid people, this assumption not always correct (Hattie &