Childhood abuse history and attention bias in adults

Abstract

Attention biases toward unpleasant information are evident among children and adults with a history of abuse and have been identified as a potential pathway through which abused children develop psychopathology. Identifying whether a history of childhood abuse affects the time course of attention biases in adults is critical, as this may provide intervention targets. The present study examined the time course of attention bias during an emotion-word Stroop task using event-related potentials (ERPs) in a sample of adults with a range of child abuse histories using a categorical approach (comparing adults with or without a history of moderate-to-severe childhood abuse) and a dimensional approach (analyzing the range from no abuse to severe abuse in a continuous manner). Although behavioral performance did not vary as a function of abuse history, adults with a history of moderate-to-severe childhood abuse showed ERP evidence of early reduced processing of emotional stimuli (smaller N200) and later reduced processing of emotional and nonemotional stimuli (smaller P300), followed by later increased processing of unpleasant stimuli (larger slow wave [SW]). Results suggest that early disengagement from emotional stimuli may help individuals with moderate-to-severe abuse histories to achieve normal behavioral performance on the emotion-word Stroop task. Additionally, regardless of analytic approach, adults with elevated levels of childhood abuse exhibited prolonged engagement (larger SW) specifically with unpleasant stimuli. Present results demonstrate attention bias patterns in adults with a history of childhood abuse and clarify the time course of attention bias. Results are discussed in the context of potential treatment implications.

1 |. INTRODUCTION

Child abuse impacts a wide range of cognitive, social, and emotional functions during childhood (for review, see Pechtel & Pizzagalli, 2011; Pollak, Cicchetti, Hornung, & Reed, 2000; Pollack, Cicchetti, Klorman, & Brumaghim, 1997; Pollak, Klorman, Thatcher, & Cicchetti, 2001; Shackman, Shackman, & Pollak, 2007). Consistent with this literature, attention bias toward negative information, or evidence of preferential attention of unpleasantly valenced or threatening stimuli, has been found in children with an abuse history (e.g., Cicchetti & Curtis, 2005; Pollak & Tolley-Schell, 2003; Shackman et al., 2007). Children with a history of abuse show evidence of an attention bias toward angry faces (Pollak et al., 1997; Shackman et al., 2007) and voices (Shackman et al., 2007) compared to children without a history of abuse. Notably, among abused children, increased attention to unpleasant information predicts symptoms of psychopathology (e.g., anxiety symptoms in Shackman et al., 2007), indicating that attention biases toward negative information may have important implications for mental health.

It is possible that attention biases evident among children with a history of abuse also persist into adulthood. One way that a history of childhood abuse may impact adult attention biases is via aberrant cognitive development resulting from a chronically activated stress response during childhood. It has been suggested that chronically elevated levels of glucocorticoids alter the neural structure and function, especially within brain regions that support the regulation of attention (including regions of the prefrontal cortex and dorsal attention network; Mizoguchi, Ishige, Takeda, Aburada & Tabira, 2004; for a discussion, see Heim, Plotsky, & Nemeroff, 2004). Given the foundational neural organization that occurs during childhood, ongoing chronic stress experienced during childhood may disrupt typical neurodevelopment, including processes that are critical for the normative processing of emotional stimuli. In-line with this possibility, there is evidence of atypical neural architecture following early stress and maltreatment (De Bellis et al., 2015; Tyborowska et al., 2018), and some of the changes associated with early stress predict cognitive functioning in adulthood (e.g., working memory, Hanson et al., 2012).

Despite the plausibility that early childhood experiences may impact attention to emotion in adulthood, if and how early abuse impacts attention bias in adults is currently understudied (Gibb, Schofield, & Coles, 2009; Young & Widom, 2014). A small but emerging literature demonstrates the evidence of attention biases for unpleasant information in cross-sectional samples of adults who report a history of childhood abuse (Caldwell, Krug, Carter, & Minzenberg, 2014; Gibb et al., 2009; Johnson, Gibb, & McGeary, 2010). For example, one study of adults with a childhood abuse history showed evidence of an attention bias toward angry faces (Gibb et al., 2009), while another showed difficulty disengaging from fearful faces (Caldwell et al., 2014). In addition to attention biases toward unpleasant information, evidence of heightened amygdala activity prompted by emotional stimuli more broadly (i.e., regardless of emotion type, e.g., angry, sad, and happy faces; van Harmelen et al., 2012), reduced attention to emotional and neutral stimuli (Weber et al., 2009), and increased attention toward pleasant (happy faces), but not unpleasant (threatening faces), stimuli (Fani, Bradley-Davino, Ressler, & McClure-Tone, 2011) have emerged.

One possible reason for mixed findings in the adult literature is that studies of attention biases related to early maltreatment do not always require that participants have a history of abuse per se. As a result, these studies have included individuals with diverse types of adverse experiences, such as the death of a parent or neglect (e.g., van Harmelen et al., 2012; Weber et al., 2009). In children, abuse is typically associated with attention biases toward unpleasant and threatening information, but neglect is associated with difficulty discriminating between different emotions (Pollak et al., 2000). Thus, attention bias toward unpleasant information may not be evident in studies that do not recruit participants on the basis of an abuse history. Another possible explanation for mixed findings in the adult literature is methodological differences across studies. For example, the van Harmelen et al. (2012) study required active responses (identification of face gender) during picture viewing, whereas the Weber et al. (2009) study involved passive picture viewing. It is possible that the requirement of a response supported engagement with the stimuli that were presented, whereas not requiring a response may have allowed participants to disengage their attention, resulting in reduced attention to emotional information. Given that this aspect of methodology may impact task engagement, it is critical to consider when investigating attention biases in individuals with a history of abuse, since abuse, especially severe abuse, can be associated with cognitive and behavioral avoidance of potential threat (e.g., Batten, Follette, & Aban, 2002).

The present study characterized the time course of attention bias in adults with a childhood abuse history by employing event-related potentials (ERPs). ERPs are manifestations of continuous processing related to the time course of attention bias. Among children who have experienced abuse, ERPs have been used to track the time course of attention bias. For example, one study found that children with an abuse history showed evidence of initial heightened attention toward threatening information (e.g., Shackman et al., 2007), reflected in N200 amplitude, a negative-going component that is theorized to measure orienting and involuntary stimulus categorization processes (Näätänen, 1990; Ritter, Simson, Vaughan, & Macht, 1982). Another study found evidence that P300 amplitude, a positive-going component that is associated with increased cognitive resource allocation (Sass et al., 2010; Shackman et al., 2007; Yee & Miller, 1994) was larger for angry faces (Pollak et al., 1997; Shackman et al., 2007) and voices (Shackman et al., 2007) than children without a history of abuse.

The present study used an emotion-word Stroop task to examine attention bias in adults with a history of childhood abuse. This Stroop task was selected because it is a commonly used measure of attention bias (for review see Bar-Haim, Lamy, Pergamin, Bakermans-Kranenburg, & Van Ijzendoorn, 2007), is designed for compatibility with ERP measurement (see Sass et al., 2010), and requires active rather than passive responding, which addresses a notable limitation of previous literature discussed above. The version of the emotion-word Stroop task that was used in the present study is a well-validated version that has been used in several ERP studies (Fisher et al., 2010, 2014; Sass et al., 2010, 2014; Stewart et al., 2010). In this version of the emotion-word Stroop task, a pleasant, neutral, or unpleasant word is presented in color on a computer screen, and individuals are instructed to respond to the color of the word while ignoring the task-irrelevant word meaning. Pleasant and unpleasant stimuli were matched on emotional arousal level, which allowed for determination of whether attention bias in adults with an abuse history is specific to unpleasant valence or to emotionally arousing stimuli in general.

The amplitudes of the P100 and N200 ERP components were used as evidence of early attention bias. P100 is a positive-going component that is sensitive to early visual attention processes (e.g., Luck, Woodman, & Vogel, 2000) and is larger for emotionally arousing than neutral stimuli in individuals with high levels of anxious arousal in the context of an emotion-word Stroop task (Sass et al., 2010). In a sample of children raised in an adverse environment (those who were institutionalized at a young age) compared to children raised with their families, early and later occipital ERPs (P100, N170, P400) were smaller in the institutionalized children during a caregiver–stranger face recognition task (Moulson, Fox, Zeanah, & Nelson, 2009). The authors suggested that these small early and later ERPs were consistent with a pattern of widespread cortical hypoarousal.

N200 has been shown to be larger for unpleasant stimuli in children with an abuse history (Shackman et al., 2007) and to be modulated by emotional valence, emotional arousal, anxiety, and depression in adults in the context of an emotion-word Stroop task (e.g., Sass et al., 2010, 2014). In adults, an ERP component similar to N200, called early posterior negativity or EPN, is generally left-lateralized, evoked in the context of word stimuli, and larger for emotionally arousing than neutral words (e.g., Kissler, Herbert, Peyk, & Junghofer, 2007). In an MEG study of adults who experienced early life stress (ELS; measured using the Early Trauma Inventory, Bremner, Vermetten, & Mazure, 2000), a component analogous to the EPN (160–210 ms) was larger for emotionally arousing stimuli than neutral stimuli, but individuals with high levels of ELS exhibited smaller EPN amplitude overall than those with low levels of ELS, particularly in the right hemisphere (Weber et al., 2009).

P300 and slow wave (SW) amplitudes were used as measures of later attentional engagement. P300 amplitude has been found to be larger for angry faces and voices in children with an abuse history (Pollak et al., 1997; Shackman et al., 2007) and larger for emotionally arousing than neutral stimuli in an emotion-word Stroop task in adults (Sass et al., 2010, 2014). Broadly, frontal SW amplitude is larger for emotional than for neutral stimuli (Cuthbert, Schupp, Bradley, Birbaumer, & Lang, 2000; Diedrich, Naumann, Maier, Becker, & Bartussek, 1997) including in the emotion-word Stroop task in children (Perez-Edgar & Fox, 2007). In the context of the emotion-word Stroop task in adults, frontal SW is argued to reflect activity associated with conflict-related control adjustments, as it is larger following incongruent than congruent trials in Stroop tasks, including emotion- or alcohol-word Stroop (e.g., Bailey & Bartholow, 2016; van Hooff, Dietz, Sharma, & Bowman, 2008; West, Bailey, Tiernan, Boonsuk, & Gilbert, 2012). It has alternatively been argued to index extended processing of emotional stimuli in the context of the emotion-word Stroop task, with larger frontal SW amplitude associated with delayed disengagement from emotional stimuli in individuals scoring high on a measure of anxious arousal (Fisher et al., 2010).

Given that a history of child abuse has been associated with early and later ERP evidence of attention bias specific to unpleasant information in children, it was anticipated that early and later biases could be evident among adults with a child abuse history. Based on previous studies, it was hypothesized that adults with a history of childhood abuse would exhibit either larger N200 amplitude to unpleasant stimuli in particular (Shackman et al., 2007) or reduced amplitude to stimuli in general (Weber et al., 2009). It was further hypothesized that adults with a history of child abuse would exhibit altered P300 amplitude for unpleasant stimuli relative to adults without a history of child abuse, indicative of later attention bias effects (consistent with effects in children, e.g., Pollak et al., 1997; Shackman et al., 2007). P100 and SW amplitude were included for exploratory purposes, given their elicitation by the emotion-word Stroop task and lack of ERP studies investigating attention bias effects in adults reporting a history of childhood abuse.

Because hemispheric asymmetries in P100, N200, P300, and SW are present within the context of the emotion-word Stroop among children and adults (e.g., Fisher et al., 2010; Perez-Edgar & Fox, 2007; Sass et al., 2010; Thomas, Johnstone, & Gonsalvez, 2007) and among children and adults with a history of abuse or ELS (e.g., Moulson et al., 2009; Weber et al., 2009), ERP asymmetries were explored in the present study.

In the present study, both a categorical approach and dimensional approach were used. A categorical classification strategy (abuse vs. no abuse) is typically used due to the high skew of childhood trauma scores in samples unselected for trauma/abuse, and research typically fails to identify a significant impact of abuse on cognitive and emotional functioning unless individuals are selected for moderate-to-severe levels of one or more types of abuse (Bernet & Stein, 1999; Bevilacqua et al., 2012; Bradley et al., 2008; Choi, Jeong, Rohan, Polcari, & Teicher, 2009; Gibb et al., 2009; Gould et al., 2012; Pederson et al., 2004; Philip et al., 2014; Raine et al., 2001; Tomoda et al., 2009, 2011). However, a dimensional approach was also implemented, since prior studies have revealed linear relationships between the number of adverse childhood events and adult physical and mental health outcomes (Anda et al., 2006; Arata, Langhinrichsen-Rohling, Bowers, & O’Farrill-Swails, 2005; Chapman et al., 2004; Dube et al., 2009; Wells, Vanderlind, Selby, & Beevers, 2014), including attention bias (Fani et al., 2011). Furthermore, a dimensional approach has the potential to clarify the role of childhood trauma as a risk factor that may operate on a spectrum (McLaughlin & Sheridan, 2016). Such an approach also allows for an investigation of the relationships among the factors of interest (experience of abuse, psychopathology, attention abnormality) in a larger sample size (Fisher, Guha, Heller, & Miller, 2020).

2 |. METHOD

2.1 |. Participants

A local community sample was recruited through advertisements placed in local newspapers and a community psychology clinic run by the University of Illinois at Urbana-Champaign’s Department of Psychology to complete color- and emotion-word Stroop tasks (N = 144; data from three of these participants were also included in a study published by Silton et al., 2011). Primary inclusion criteria were right-handedness (based on the Edinburgh Handedness Inventory; Oldfield, 1971) and being a native English speaker. Participants were screened for and excluded on the basis of abnormal color vision, loss of consciousness greater than 10 min, claustrophobia, recent drug or alcohol use, excessive caffeine intake, lack of sleep, and current active substance abuse or dependence.

Of the 144 community participants who completed the Stroop tasks, seven participants had missing CTQ data and thus were excluded, leaving a final sample of 137 participants for the secondary analyses using a dimensional analysis strategy. About 64% of the participants identified as female and were between the ages of 19 and 51 years old (M = 35.30, SD = 9.27). Most of the sample identified as European American (n = 119, 86%), followed by Asian/Pacific Islander (n = 6), African American (n = 4), Native American (n = 3), Hispanic/Latino (n = 2), and Unknown/Prefer Not to Answer (n = 3).

Using a classification strategy that has been employed in several other studies (e.g., Bradley et al., 2008; Gibb et al., 2009; Heim et al., 2009), two groups were created from these 144 participants according to whether they reported having experienced a moderate-to-severe level of any type of abuse (emotional, physical, and/or sexual) or whether they did not report any history of abuse in childhood based on the short-form Childhood Trauma Questionnaire (CTQ; cutoff information provided in the Measures section). Twenty-eight individuals were classified as having an abuse history, and 25 individuals were classified as having experienced no abuse. Within the Abuse group, a history of moderate-to-severe emotional abuse was endorsed by 13 participants, physical abuse by nine participants, and sexual abuse by 16 participants, with some participants experiencing high levels of more than one type of abuse (n = 8). Across both groups, 60% of participants were female, between the ages of 19 and 50 years old (M = 34.80, SD = 9.21), and predominately identified as European American (n = 46, 87%), followed by Asian/Pacific Islander (n = 2), Native American (n = 2), African American (n = 1), Hispanic/Latino (n = 1), and Unknown/Prefer Not to Answer (n = 1).

2.2 |. Measures

The CTQ is a self-report questionnaire that assesses experiences of abuse and neglect during childhood and adolescence across five domains: emotional abuse, physical abuse, sexual abuse, emotional neglect, and physical neglect. Each CTQ subscale is comprised of five questions that are scored on a 5-point Likert scale (1 = never true, 5 = very often true), for a possible score range of five to 25 for each subscale. Individuals were included in the Abuse group based on established moderate-to-severe cutoff scores for any of the three abuse subscales of the CTQ: emotional (≥13), physical (≥10), and/or sexual abuse (≥8; Bernstein & Fink, 1998; Bernstein et al., 2003). Individuals were included in the No Abuse group if they endorsed no abuse (i.e., a score of 5) on each of the emotional, physical, and sexual abuse subscales. The CTQ has excellent psychometric properties in both clinical and nonclinical samples (Bernstein & Fink, 1998; Bernstein et al., 2003), and the observed internal consistency for the CTQ was in the acceptable range (observed Cronbach α’s for total severity: α = .87, emotional abuse: α = .86, physical abuse: α = .70, sexual abuse: α = .89). The CTQ can be used to classify individuals’ scores in each domain into qualitative categories (e.g., low-to-moderate abuse, moderate-to-severe abuse) or as a dimensional measure, and higher scores in each domain indicate more severe maltreatment experiences.

Participants completed questionnaire measures of depression and anxiety symptoms (see Tables 1 and 4). Questionnaires included an eight-item subset of the Anhedonic Depression subscale (Nitschke, Heller, Imig, McDonald, & Miller, 2001) and the 17-item Anxious Arousal subscale of the Mood and Anxiety Symptom Questionnaire (MASQ-AD8 and MASQ-AA, respectively; Watson, Clark, et al., 1995; Watson, Weber, et al., 1995) and the 16-item Penn State Worry Questionnaire (PSWQ; Meyer, Miller, Metzger, & Borkovec, 1990; Molina & Borkovec, 1994). Two anxiety measures were used to capture two separable, distinct dimensions of anxiety (Engels et al., 2007; Heller, Nitschke, Etienne, & Miller, 1997; Nitschke et al., 2001; Nitschke, Heller, Palmieri, & Miller, 1999; Sass et al., 2010; Sharp, Miller, & Heller, 2015): anxious arousal, which is associated with physiological arousal and panic symptoms (MASQ-AA), and anxious apprehension, which is associated with worry and somatic tension (PSWQ). Observed internal consistencies for the MASQ-AA, MASQ-AD8, and PSWQ were in the acceptable range (observed Cronbach’s α = .75, α = .88, and α = .95, respectively),

TABLE 1.

Descriptive statistics for the categorical analyses (moderate-to-severe abuse vs. no abuse groups, n = 52)

Mean (SD)	Abuse group (n = 27)	No abuse group (n = 25)
Age	36.3 (9.1)	32.8 (9.0)
Gender	59% female (n = 16)	60% female (n = 15)
Ethnicity	European American: 81%	European American: 92%
	Asian/Pacific Islander: 7%	Asian/Pacific Islander: 0%
	African American: 3%	African American: 0%
	Native American: 3%	Native American: 4%
	Hispanic/Latino: 3%	Hispanic/Latino: 4%
	Unknown: 3%	Unknown: 0%
% Moderate-to-severe emotional abuse	44% (n = 12)	0%
% Moderate-to-severe physical abuse	33% (n = 9)	0%
% Moderate-to-severe sexual abuse	59% (n = 16)	0%
MASQ-AD8*	16.89 (4.00)	13.12 (3.99)
MASQ-AA*	25.00 (5.11)	20.96 (3.16)
PSWQ	53.89 (15.89)	47.68 (13.63)

Abbreviations: MASQ-AA, mood and anxiety symptom questionnaire, anxious arousal; MASQ-AD8, mood and anxiety symptom questionnaire, anhedonic depression—8-item subscale; PSWQ, Penn state worry questionnaire.

^*Significant group difference, p < .01.

TABLE 4.

Correlations between total abuse severity, psychopathology symptoms, and reaction times on correct trials for the dimensional analyses (full sample, N = 136)

	MASQ-AD8	MASQ-AA	PSWQ	Pleasant RT	Neutral RT	Unpleasant RT
(log) CTQ total abuse	0.34**	0.26*	0.23*	0.02	0.04	−0.01
MASQ-AD8	–	0.42**	0.42**	−0.07	−0.03	−0.05
MASQ-AA	–	–	0.25*	−0.03	0.01	0.01
PSWQ	–	–	–	−0.03	0.01	0.01

Abbreviations: MASQ-AA, mood and anxiety symptom questionnaire, anxious arousal; MASQ-AD8, mood and anxiety symptom questionnaire, anhedonic depression—8-item subscale; PSWQ, Penn state worry questionnaire.

^*p < .01;

^**p < .001.

Participants were assessed for current and lifetime history of Axis I psychopathology using the Structured Clinical Interview for DSM-IV-TR Disorders (SCID-NP; First, Spitzer, Gibbon, & Williams, 2002). The presence of diagnosed psychopathology did not lead to exclusion in the present study and is included to provide information about sample characteristics. Diagnostic interviews were conducted by advanced doctoral students in clinical psychology and were reviewed by a consensus team consisting of a second interviewer and a clinical faculty supervisor (GAM), who has supervised over 2000 inpatient SCID cases and over 600 nonpatient research SCID cases (for method details see Bredemeier et al., 2010; Silton et al., 2011). Individuals were given a code of 1 in the present study if they met diagnostic criteria for an Axis I disorder and 0 if they did not.

2.3 |. Experimental design

Participants completed both color-word and emotion-word versions of the Stroop task in a counterbalanced order. Due to specific hypotheses regarding emotion processing, only the results of the emotion-word Stroop task are included in the present study. The emotion-word Stroop task consisted of blocks of pleasant and unpleasant words with alternating blocks of neutral words during EEG data collection. There were 256 trials across 16 emotion-word blocks (four pleasant, eight neutral, four unpleasant), with each block containing 16 trials. The 256 stimuli included in the emotion-word task were selected from the Affective Norms for English Words (ANEW) set (Bradley & Lang, 1999). A set of 64 pleasant words (e.g., birthday, ecstasy, laughter), two sets of 64 neutral words (e.g., hydrant, moment, carpet), and a set of 64 unpleasant words (e.g., suicide, war, victim) were included in the task. The words were carefully selected on the basis of established norms for valence, arousal, word length, and frequency of usage in the English language (Bradley & Lang, 1999; Toglia & Battig, 1978). Pleasant words were selected to be equivalent to unpleasant words in arousal rating (mean 6.53 for pleasant, 3.81 for neutral, 6.56 for unpleasant words) and to differ in valence rating (mean 7.83 for pleasant, 5.22 for neutral, and 2.47 for unpleasant words).

On a trial, a word was presented for 1,500 ms and was followed by a fixation cross for 275 to 725 ms (onset to onset ITI 2,000 ± 225 ms). Each trial consisted of one word presented in one of four ink colors (red, yellow, green, blue) on a black background. Participants’ responses were made with the middle and index fingers of each hand using a four-button response box. For the present analyses, half of the neutral words were randomly selected for inclusion.¹ For a full discussion of the experimental design methods, see Fisher et al. (2010), Sass et al. (2010), or Stewart et al. (2010).

2.3.1 |. Electrophysiological recordings

Participants were seated in a comfortable chair in a quiet room while EEG data were recorded with a custom-designed Falk Minow 64-channel cap with Ag/AgCl electrodes spaced equidistantly. Procedures followed those recommended by Keil et al. (2014). Vertical and horizontal EOG were recorded with electrodes above and below each eye and near the outer canthus of each eye. The left mastoid served as the online reference during recording. Electrode impedances were maintained below 20 Kohms. Amplifier band pass was 0.1 to 100 Hz, with digitization at 250 Hz.

2.3.2 |. Data reduction

Reaction time (RT) was analyzed for correct trial responses between 350 and 1,400 ms poststimulus onset for pleasant, neutral, and unpleasant word trials. There were no RT outliers more than 3 SD from any emotion condition mean. For the main set of categorical analyses, one participant from the abuse group had an error rate 3 SD above the mean. This participant was excluded from all analyses, yielding a final sample size of 52 participants (abuse group: 27, no abuse group: 25). For the dimensional analyses, there were no outlier RTs greater than 3 SD from any emotion condition mean, but one participant was excluded for an error rate of 3 SD above an emotion condition mean, leaving a final sample of 136 participants.

Brain Electrical Source Analysis (BESA) software (Berg & Scherg, 1994) was used for artifact rejection and eye blink correction within ERP data. Spherical spline interpolation was used to transform the 64 electrodes to a standard set of 81 virtual scalp electrodes reflecting the 10–10 system (Perrin, Pernier, Bertrand, & Echallier, 1989). An average reference was created using the mean voltage over these 81 electrodes for each time point. Channels were baseline-adjusted by subtracting the average amplitude for 200 ms prior to the stimulus onset using MATLAB. To smooth the waveform averages, a 101-weight, 0.1–20 Hz digital filter was used for the P100 and N200 components, and a 0.1–8 Hz digital filter was used for the P300 and SW components (Cook & Miller, 1992; Edgar, Stewart, & Miller, 2005; Nitschke, Miller, & Cook, 1998). For each participant, a combination peak/area measure was calculated for correct trials by averaging voltage 48 ms around P100 and N200 peak amplitude, and 96 ms for P300 and SW amplitude, similar to Sass et al. (2010). This approach (Clayson & Miller, 2017) is less sensitive to trial-by-trial amplitude latency variability and spurious peak amplitudes, which can significantly impact the amplitude measurement (Clayson, Baldwin, & Larson, 2013; Luck, 2005).

Amplitudes were extracted within the following latency windows: P100: 60–128 ms, N200: 148–272 ms, P300: 412–616 ms, and SW: 840–1,400 ms. Following Sass et al. (2010), P100 and N200 electrode sites were chosen based on current source density and previous literature (e.g., Fisher et al., 2010; Sass et al., 2010). P300 and SW were chosen based on previous literature (e.g., Fisher et al., 2010; Sass et al., 2010) and where grand-average waveform effects were maximal. Amplitude values were averaged by hemisphere for each emotion type for P100 (left: P7, PO7, P9, PO9, right: P8, PO8, P10, PO10), N200 (left: P7, PO7, P9, PO9, right: P8, PO8, P10, PO10), P300 (left: P1, P3, right: P2, P4), and frontal SW (left: F1, F3, FC1, right: F2, F4, FC2).

Following outlier criteria used by previous ERP studies that used the emotion-word Stroop task (e.g., Sass et al., 2010, 2014), participants whose averaged amplitude values for any emotion condition that were more than 3 SD from the mean for a given component were excluded from analyses that included that component (categorical: P100: n = 1, no abuse group; N200: n = 0; P300: n = 1, no abuse group; frontal SW: n = 0; dimensional: P100: n = 7, N200: n = 13; P300: n = 4, frontal SW: n = 4). For RT, there were no outliers more than 3 SD from each emotion condition mean for any word type for categorical or dimensional analyses.

2.4 |. Data analysis

Two different forms of the general linear model (GLM) were used to address (a) whether adults with a history of no or severe-to-moderate levels of childhood abuse exhibit differences in ERP evidence of the time course of attention to emotional stimuli (i.e., categorical analysis) and (b) whether there are linear relationships between childhood abuse severity and ERP evidence of the time course of attention to emotional stimuli after accounting for current symptoms of depression and anxiety (i.e., dimensional analysis). For the categorical analysis approach, effects of RT were assessed using a Group (abuse, no abuse) × Emotion (pleasant, neutral, unpleasant word) repeated-measures ANOVA exploring linear (valence: pleasant compared to unpleasant) and quadratic (arousal: the average of pleasant and unpleasant compared to neutral) orthogonal polynomial trends. Linear and quadratic components of the Emotion factor tested hypotheses regarding whether unpleasant information specifically or emotional arousal generally was responsible for an Emotion effect. For each ERP component, a Group (abuse, no abuse) × Emotion (pleasant, neutral, unpleasant) × Hemisphere (left, right) repeated-measures ANOVA was conducted exploring linear and quadratic trends (described above). Whereas hemisphere was included to assess for possible laterality effects in relation to childhood trauma, main effects of laterality and/or interactions with emotion were not of primary interest. Tests were two-tailed, alpha level was set to .05, and Huynh-Feldt corrections were used when appropriate.

For the dimensional analysis approach, multivariate multiple linear regressions were conducted to examine the impact of childhood abuse severity (as a continuous measure) on RT for correct trials and on ERP amplitude scores for each component. This approach, similar to that of Fisher et al. (2014), allows multiple dependent variables to be entered into a common inferential model, accounting for covariance among the predictors across ERP components (see Figure 1). Predictors included in each of the ERP component models were total abuse severity (composite of emotional, physical, and sexual abuse CTQ scores, log-transformed due to skew-ness) and psychopathology symptoms (anxious arousal, anxious apprehension, and anhedonic depression scores). This combination allowed identification of the impact of abuse severity on the dependent variables (either RT or ERPs) beyond psychopathology symptoms. For the RT model, three dependent variables were entered (RT for pleasant, neutral, and unpleasant emotion conditions). For each ERP multivariate multiple regression model (predicting P100, N200, or P300), six dependent variables were entered. Specifically, amplitude scores for emotion (pleasant, neutral, unpleasant) over the left and right hemisphere (left, right) were the dependent variables. For the SW model, given that hemisphere did not interact with Group in the categorical analyses, hemisphere scores were averaged, with averages for emotion (pleasant, neutral, unpleasant) entered as dependent variables.

FIGURE 1.

The multivariate multiple regression model implemented for each ERP component. The four questionnaire scores on the left were predictors in a separate simultaneous regression for each ERP component score on the right. The figure illustrates the analysis of P100 scores. The same analysis was done for N200 and P300. For slow wave, amplitude for each emotion was first averaged across hemispheres, so three rather than six dependent variables were predicted

3 |. RESULTS

3.1 |. Demographics and psychopathology

Tables 1 and 2 provide descriptive statistics for the abuse and no abuse groups. The groups were similar in age (t (50) = 1.40, p = .166), gender composition (χ² (1, n = 52) = 0.003, p = .957), and anxious apprehension score (PSWQ, t (50) = 1.51, p = .138), but the abuse group had higher anxious arousal and anhedonic depression scores (MASQ-AA, t (50) = 3.40, p = .001 and MASQ-AD8, t (50) = 3.40, p = .001, respectively). Additionally, a higher proportion of the abuse group had a lifetime history of and currently met criteria for an Axis I disorder than the no abuse group (χ² (1, n = 52) = 10.19, p = .001 and χ² (1, n = 52) = 10.88, p = .001, respectively).

TABLE 2.

Diagnostic and statistical manual of mental disorders (DSM) diagnoses based on the SCID-IV-TR (structured clinical interview for the DSM, 4th edition, text revision) for the categorical analyses (moderate-to-severe abuse vs. no abuse groups)

	Current axis I disorder	Prior history of axis I disorder	Lifetime history
Abuse group (n = 27)	Total (≥1 dx): n = 15 (56%)	Total (≥1 dx): n = 23 (85%)	Total: n = 24 (89%)
	GAD, n = 4	MDD, n = 12	1. Disorder, n = 5
	Specific phobia, n = 4	Alcohol abuse, n = 6	2. n = 8
	Social phobia, n = 2	Alcohol dep, n = 7	3. n = 8
	Panic disorder, n = 1	Dysthymia, n = 2	4. n = 3
	PTSD, n = 1	Social phobia, n = 3
	MDD, n = 1	Specific phobia, n = 2
	Dysthymia, n = 1	Substance abuse (cannabis), n = 2
	Depressive disorder NOS, n = 1	Substance dep (cocaine), n = 2
	Anxiety disorder NOS, n = 1	OCD, n = 1
	Adjustment disorder, n = 1	Bipolar I, n = 1
		Bipolar II, n = 1
		Bulimia nervosa, n = 1
No abuse group (n = 25)	Total (≥1 dx): n = 3 (12%)	Total (≥1 dx): n = 10 (40%)	Total: n = 12 (48%)
	Social phobia, n = 2	MDD, n = 6	1. Disorder, n = 11
	Specific phobia, n = 1	Alcohol abuse, n = 2	2. n = 1
		GAD, n = 1
		Adjustment disorder, n = 1

Abbreviations: Dep, dependence; dx, diagnosis; GAD, generalized anxiety disorder; MDD, major depressive disorder; NOS, not otherwise specified; OCD, obsessive compulsive disorder; PTSD, posttraumatic stress disorder.

Table 3 provides descriptive statistics for the dimensional analysis (full) sample. Table 4 provides zero-order correlations between the CTQ total abuse score, PSWQ, MASQ-AD8, and MASQ-AA. Per Table 5, 63% of the sample had a lifetime history of an Axis I disorder, and 25% met criteria for a current Axis I disorder.

TABLE 3.

Descriptive statistics for the dimensional analyses (full sample, N = 136)

	Mean (SD)
Age	35.2 (9.3)
Gender	64% female
Ethnicity	European American: 86%
	Asian/Pacific Islander: 4%
	African American: 3%
	Native American: 2%
	Hispanic/Latino: 2%
	Unknown: 3%
CTQ total abuse	20.07 (6.10)
CTQ emotional abuse	7.79 (4.41)
CTQ physical abuse	6.41 (2.04)
CTQ sexual abuse	5.86 (2.47)
MASQ-AD8	14.97 (4.46)
MASQ-AA	22.26 (4.29)
PSWQ	47.04 (14.72)

Note: In this table, CTQ values are not log-transformed, as they were for analyses. The minimum value for each abuse subscale is 5 and for the abuse total is 15.

Abbreviations: CTQ, childhood trauma questionnaire; MASQ-AA, mood and anxiety symptom questionnaire, anxious arousal; MASQ-AD8, mood and anxiety symptom questionnaire, anhedonic depression—8-item subscale; PSWQ, Penn state worry questionnaire.

TABLE 5.

Diagnostic and statistical manual of mental disorders (DSM) diagnoses based on the SCID-IV-TR (structured clinical interview for the DSM, 4th edition, text revision) for the dimensional analyses (full sample)

	Current axis I disorder	Prior history of axis I disorder	Lifetime history
N = 136	Total (≥1 dx): n = 34 (25%)	Total (≥1 dx): n = 79 (58%)	Total: n = 86 (63%)
	Specific phobia, n = 8	MDD, n = 43	1. Disorder, n = 42
	Social phobia, n = 6	Alcohol abuse, n = 25	2. n = 20
	GAD, n = 5	Alcohol dep, n = 12	3. n = 18
	PTSD, n = 3	Social phobia, n = 7	4. n = 4
	Anxiety disorder NOS, n = 4	Dysthymia, n = 6	5. n = 2
	MDD, n = 3	Substance abuse (cannabis), n = 5
	Dysthymia, n = 3	Substance dep (cocaine), n = 4
	Depressive disorder NOS, n = 2	Substance dep (cannabis), n = 2
	Panic disorder, n = 1	Substance dep (polysubstance), n = 2
	Adjustment disorder, n = 1	GAD, n = 4
	Eating disorder NOS, n = 1	Specific phobia, n = 3
	Pain disorder, n = 1	OCD, n = 3
		Bipolar I, n = 3
		Bipolar II, n = 1
		Bipolar NOS, n = 1
		Bulimia nervosa, n = 2
		PTSD, n = 1
		Substance-induced mood disorder, n = 1
		Anorexia nervosa, n = 1
		Eating disorder NOS, n = 1
		Body dysmorphic disorder, n = 1
		Substance-induced psychosis, n = 1

Abbreviations: Dep, dependence; dx, diagnosis; GAD, generalized anxiety disorder; MDD, major depressive disorder; NOS, not otherwise specified; OCD, obsessive compulsive disorder; PTSD, posttraumatic stress disorder.

3.2 |. Reaction time (RT)

Comparisons between groups indicate that performance accuracy was high (mean error rate = 4.2%, SD = 2.3), and error rates were comparable (abuse group: M = 4.2%, SD = 2.2, No abuse group: M = 4.1%, SD = 2.4; t (50) = 0.16, p = .876). In the absence of a main effect of group, F (1, 50) = 0.64, p = .426, partial η² = 0.013, or a Group × Emotion interaction, F (2, 100) = 0.87, p = .422, partial η² = 0.017, there was a main effect of emotion, F (2, 100) = 4.31, p = .016, partial η² = 0.079. An arousal effect was evident across groups, quadratic emotion, F (1, 50) = 10.08, p = .003, partial η² = 0.168, with slower RT for emotionally arousing (pleasant: 701 ms, unpleasant: 706 ms) than neutral words (690 ms). The valence effect was ns, F (1, 50) = 0.61, p = .439, partial η² = 0.012, indicating that RT was similar for pleasant and unpleasant stimuli.

For the full sample, performance accuracy was high (mean error rate = 4.1%, SD = 2.2). Mean RT was slower for pleasant (M = 695 ms, SD = 93 ms) than neutral words (M = 686 ms, SD = 85 ms; t (135) = 2.68, p = .008), and for unpleasant (M = 698 ms, SD = 95 ms) than neutral words (t (135) = 3.71, p < .001). Pleasant and unpleasant RT did not differ. None of the predictors (anxious apprehension, anxious arousal, anhedonic depression, total childhood abuse) was significant in the multivariate model. Table 4 provides zero-order correlations between RT and PSWQ, MASQ-AA, MASQ-AD8, and CTQ total abuse score.

3.3 |. ERP components

Figure 2 presents the grand-average ERP waveforms, and Figure 3 shows current source density plots (P100, N200) and scalp voltage topographies (P300, SW). Current source density transforms the EEG to its second spatial derivative, which reduces the spread of focal brain activity on the scalp surface and enhances the contribution of the underlying cortical surface to the recorded electrode signal (i.e., it essentially acts a spatial high-pass filter; Hoechstetter et al., 2003; Nunez et al., 1999).

FIGURE 2.

Grand-average waveforms for each group. Rectangles indicate scoring interval for P100, N200, P300, and slow wave components. Time course is depicted on the x-axis beginning 200 ms before stimulus onset at 0 ms

FIGURE 3.

(a) Current source density plots illustrating areas of maximal voltage for P100 (left: P7, P9, PO7, PO9, right: P8, P10, PO8, PO10) and N200 (left: P7, P9, PO7, PO9, right: P8, P10, PO8, PO10) after stimulus onset. Values range from red = +0.23 uV/cm² to blue = −0.23 uV/cm². (b) Scalp topographies illustrating areas of maximal voltage for P300 (left: P1, P3, right: P2, P4) and SW (left: F1, F3, FC1, right: F2, F4, FC2) after stimulus onset. For P300, values range from red = +3.92 uV to blue = −3.92 uV and for SW, values ranged from +0.49 uV to blue = −0.49 uV. BESA time-domain filter used for all graphs: 0.1–20 Hz

Tables 6 and 7 provide statistical results for the ERP group comparisons (early and late ERP components, respectively). Table 8 provides the results of the dimensional analysis approach and Table 9 provides the residual covariance matrices from the dimensional analyses.

TABLE 6.

Early ERP ANOVA results for the categorical analyses

P100
Hemisphere	Right > left	F (1, 49) = 4.42, p = .041, partial η2 = 0.083
N200
Hemisphere	Left > right	F (1, 50) = 7.44, p = .009, partial η2 = 0.129
Emotion		F (2, 100) = 4.10, p = .019, partial η2 = 0.076
	Pleasant < unpleasant	Linear, F (1, 50) = 5.25, p = .026, partial η2 = 0.095
Emot × Hemi		F (2, 100) = 7.17, p = .001, partial η2 = 0.125
	Left hemi: pleasant < unpleasant	Linear, F (1, 51) = 7.69, p = .008, partial η2 = 0.131
	Right hemi: neutral > pleasant & unpleasant	Quadratic, F (1, 51) = 6.91, p = .011, partial η2 = 0.119
	Pleasant: left > right	F (1, 51) = 6.24, p = .016, partial η2 = 0.109
	Neutral: left versus right	F (1, 51) = 3.51, p = .067, partial η2 = 0.064
	Unpleasant: left > right	F (1, 51) = 12.60, p = .001, partial η2 = 0.198
Group × Emot × Hemi		F (2, 100) = 2.57, p = .082, partial η2 = 0.049
		Quadratic, F (1, 50) = 4.29, p = .044, partial η2 = 0.079
	Abuse group: Hemisphere	F (1, 26) = 3.59, p = .069, partial η2 = 0.121,
	Abuse group: Emotion	F (2, 26) = 2.28, p = .112, partial η2 = 0.078,
	Abuse group: Emot × Hemi	F (2, 26) = 8.68, p = .001, partial η2 = 0.250
	Right hemi: neutral > pleasant & unpleasant	Quadratic, F (1, 26) = 13.23, p = .001, partial η2 = 0.337
	Right hemi: pleasant versus unpleasant	Linear, F (1, 26) = 1.27, p = .270, partial η2 = 0.047
	No abuse group: Hemisphere	F (1, 24) = 3.81, p = .063 partial η2 = 0.137
	No abuse group: Emotion	F (2, 24) = 2.54, p = .089, partial η2 = 0.096
	No abuse group: Emot × Hemi	F (2, 24) = 2.38, p = .103, partial η2 = 0.090

TABLE 7.

Later ERP ANOVA results for the categorical analyses

P300
Hemisphere		F (1, 49) = 3.73, p = .059, partial η2 = 0.071
Emotion		F (2, 98) = 0.22, p = .804, partial η2 = 0.004
Group	Abuse < no abuse group	F (1, 49) = 5.15, p = .028, partial η2 = 0.095
Group × Emot × Hemi		F (2, 98) = 4.52, p = .013, partial η2 = 0.084
	Abuse group: Emot × Hemi	F (2, 26) = 0.80, p = .454, partial η2 = 0.030
	No abuse group: Emot × Hemi	F (2, 23) = 4.48, p = .019, partial η2 = 0.163
	Right hemi: pleasant > unpleasant	Linear, F (1, 23) = 4.34, p = .048, partial η2 = 0.159
	Pleasant: left > right	F (1, 23) = 4.73, p = .040, partial η2 = 0.170
	Neutral: left versus right	F (1, 23) = 0.52, p = .480, partial η2 = 0.022
	Unpleasant: left versus right	F (1, 23) = 0.19, p = .670, partial η2 = 0.008
SW
Hemisphere	Right > left	F (1, 50) = 27.62, p < .001, partial η2 = 0.356
Emotion		F (2, 100) = 10.97, p < .001, partial η2 = 0.180
	Unpleasant > pleasant	Linear, F (1, 50) = 25.12, p < .001, partial η2 = 0.334
Emot × Hemi		F (2, 100) = 4.33, p = .016, partial η2 = 0.080
	Right hemi: unpleasant > pleasant	Linear, F (1, 51) = 38.94, p < .001, partial η2 = 0.433
	Left hemi: unpleasant > pleasant	Linear, F (1, 51) = 7.99, p = .007, partial η2 = 0.135
	Pleasant: right > left	F (1, 51) = 11.85, p = .001, partial η2 = 0.189
	Neutral: right > left	F (1, 51) = 13.85, p = .001, partial η2 = 0.210
	Unpleasant: right > left	F (1, 51) = 33.31, p < .001, partial η2 = 0.395
Group × Emot		F (2, 100) = 3.77, p = .027, partial η2 = 0.070
	Abuse group: Emotion
	Unpleasant > pleasant	F (1, 26) = 31.41, p < .001, partial η2 = 0.547
	No abuse group: Emotion	F (1, 24) = 1.84, p = .169, partial η2 = 0.071

TABLE 8.

Multivariate general linear model results for the dimensional analyses (full sample, N = 136)

P100
MASQ-AD	F (6, 122) = 1.40, p = .222, partial η2 = 0.064
MASQ-AA	F (6, 122) = 1.81, p = .103, partial η2 = 0.082
PSWQ	F (6, 122) = 1.31, p = .256, partial η2 = 0.061
CTQ total abuse	F (6, 122) = 0.90, p = .500, partial η2 = 0.042
N200
MASQ-AD	F (6, 124) = 0.88, p = .513, partial η2 = 0.041
MASQ-AA	F (6, 124) = 0.95, p = .465, partial η2 = 0.044
PSWQ	F (6, 124) = 1.02, p = .419, partial η2 = 0.047
CTQ total abuse	F (6, 124) = 1.99, p = .072, partial η2 = 0.088
P300
MASQ-AD	F (6, 122) = 1.11, p = .360, partial η2 = 0.052
MASQ-AA	F (6, 122) = 1.94, p = .079, partial η2 = 0.087
PSWQ	F (6, 122) = 0.89, p = .507, partial η2 = 0.042
CTQ total abuse	F (6, 122) = 1.49, p = .186, partial η2 = 0.068
SW
MASQ-AD	F (3, 125) = 0.75, p = .522, partial η2 = 0.018
MASQ-AA	F (3, 125) = 2.10, p = .103, partial η2 = 0.048
PSWQ	F (3, 125) = 1.43, p = .238, partial η2 = 0.033
CTQ total abuse	F (3, 125) = 2.91, p = .037, partial η2 = 0.065

Note: SW scores were averaged across hemisphere for each emotion type; see Results section.

Abbreviations: MASQ-AA, mood and anxiety symptom questionnaire, anxious arousal; MASQ-AD8, mood and anxiety symptom questionnaire, anhedonic depression—8-item subscale; PSWQ, Penn state worry questionnaire.

TABLE 9.

Multivariate multiple regression analyses: dependent variable residual covariance matrixes

	Hemisphere/Emotion	Left, Pleasant	Left, Neutral	Left, Unpleasant	Right, Pleasant	Right, Neutral	Right, Unpleasant
P100	Left, Pleasant	1.72	1.41	1.24	1.04	0.77	0.68
	Left, Neutral	1.41	2.24	1.42	0.65	1.02	0.68
	Left, Unpleasant	1.24	1.42	2.15	0.87	0.97	1.04
	Right, Pleasant	1.04	0.65	0.87	4.18	3.38	3.85
	Right, Neutral	0.77	1.02	0.97	3.38	3.48	3.45
	Right, Unpleasant	0.68	0.68	1.04	3.64	3.45	4.15
N200	Left, Pleasant	8.90	7.65	7.55	5.37	5.09	4.50
	Left, Neutral	7.65	7.80	7.46	5.03	5.89	4.80
	Left, Unpleasant	7.55	7.46	8.11	4.71	5.53	4.80
	Right, Pleasant	5.37	5.03	4.71	8.15	7.90	7.27
	Right, Neutral	5.09	5.89	5.53	7.90	9.78	8.22
	Right, Unpleasant	4.50	4.80	4.80	7.27	8.22	7.66
P300	Left, Pleasant	5.95	5.38	5.88	4.31	3.73	4.23
	Left, Neutral	5.38	5.66	5.60	4.09	4.02	4.12
	Left, Unpleasant	5.88	5.60	6.56	4.39	3.76	4.75
	Right, Pleasant	4.31	4.09	4.39	4.34	3.85	4.17
	Right, Neutral	3.73	4.02	3.76	3.85	4.20	3.89
	Right, Unpleasant	4.23	4.12	4.75	4.17	3.89	4.86
	Hemisphere/Emotion	Pleasant	Neutral	Unpleasant
SW	Pleasant	2.25	1.50	1.61
	Neutral	1.50	2.03	1.68
	Unpleasant	1.61	1.68	2.27

^aIndependent variables included in each model: mood and anxiety symptom questionnaire—anhedonic depression, mood and anxiety symptom questionnaire—anxious arousal, Penn state worry questionnaire, and childhood trauma questionnaire, abuse total score (log-transformed).

3.3.1 |. Early ERP components

P100

P100 was larger over the right than the left hemisphere, but no other effects emerged in the group or dimensional analyses.

N200

N200 amplitude was larger over left than right hemisphere and was smaller for pleasant than unpleasant stimuli. Examining the Emotion × Hemisphere interaction with simple-effects analyses, N200 was larger for unpleasant than pleasant words over the left hemisphere, and larger for neutral than emotionally arousing words over the right hemisphere. Additionally, N200 was larger over the left than right hemisphere for pleasant and unpleasant words, but not for neutral words.

Although the omnibus Group × Emotion × Hemisphere interaction was only marginally significant (p = .082), linear and quadratic trends were explored because of the a priori hypothesis that adults with a childhood abuse history would preferentially process unpleasant or emotionally arousing information early. Figure 4 shows that an Emotion × Hemisphere interaction was evident in the abuse group only. Specifically, smaller N200 amplitude was prompted by pleasant and unpleasant than by neutral words over the right hemisphere, whereas unpleasant or pleasant words did not differ in amplitude over the right hemisphere. In the dimensional analyses, none of the predictors was significant in the multivariate model (see Table 8).

FIGURE 4.

N200 amplitude arousal effect in the abuse group. The asterisk (p < .05) indicates a larger N200 for neutral than for emotional words over the right hemisphere ((pleasant + unpleasant)/2 < neutral). Error bars represent 1 SE

3.3.2 |. Later ERP components

P300

Figure 5 shows a smaller P300 amplitude overall for the abuse than no abuse group (also evident in the middle rows of the upper vs. lower portions of Figure 2). Following the a priori hypothesis that adults with a childhood abuse history would preferentially process unpleasant or emotionally arousing information at a later time point, the Group × Emotion × Hemisphere interaction was followed up with simple-effects Emotion × Hemisphere ANOVAs for each group. Following up the Emotion × Hemisphere interaction in the no abuse group, Emotion ANOVAs were conducted separately for each Hemisphere within that group. Figure 6 shows that P300 was larger for unpleasant than pleasant words over right hemisphere.

FIGURE 5.

P300 amplitude Group effect. The asterisk (p < .05) indicates a smaller overall P300 for the abuse than the no abuse group. Error bars represent 1 SE

FIGURE 6.

P300 amplitude Emotion × Hemisphere effects in the no abuse group. The asterisks (p < .05) indicate a larger P300 for pleasant words over the left than right hemisphere and a larger P300 for unpleasant than pleasant words over right hemisphere. Error bars represent 1 SE

In the dimensional analyses, although ns in the multivariate model (p = .186), total abuse severity predicted smaller P300 amplitude to unpleasant words in the right hemisphere (p = .015, partial η² = 0.046). None of the other predictors was significant in the multivariate model (see Table 8).

Slow wave

SW was larger over right than left hemisphere and was larger for unpleasant than pleasant words. Examining the Emotion × Hemisphere interaction in simple- effects analyses, larger SW amplitude emerged for unpleasant than pleasant words over both left and right hemispheres. Additionally, SW amplitude was larger over right than left hemisphere for pleasant, neutral, and unpleasant words.

Following the a priori hypothesis that a history of childhood abuse would bias attention toward unpleasant or emotionally arousing information, the Group × Emotion interaction was dissected with separate Emotion ANOVAs for each group. Figure 7 shows that larger SW amplitude was prompted by unpleasant than pleasant words for the abuse group only.

FIGURE 7.

Slow wave (SW) amplitude Hemisphere × Emotion effect for the abuse group. The asterisk (p < .05) indicates larger SW for unpleasant than pleasant words. Error bars represent 1 SE

In dimensional analyses, the CTQ total abuse score was the only significant predictor in the multivariate model (see Table 8). Three separate follow-up univariate hierarchical regressions were run, with PSWQ, MASQ-AD, and MASQ-AA entered in the first step and total abuse severity entered in the second step as predictors of the pleasant, neutral, and unpleasant SW scores. Even with variance that total abuse severity shared with the three psychopathology scores thus removed, higher residualized total abuse predicted larger SW amplitude for unpleasant (ΔR² = .03, B = 0.18, p = .050), but not pleasant (ΔR² < .01, B = 0.009, p = .923) or neutral (ΔR² < .01, B = 0.003, p = .977), words.

4 |. DISCUSSION

The present study sought to identify the time course of attention bias in adults with and without self-reported childhood abuse experiences. The question of whether early or late attention biases are evident among adults with a history of childhood abuse was investigated using ERPs and RT during an emotion-word Stroop task. A categorical (extreme-groups) analysis approach (Fisher et al., 2020), commonly implemented in the abuse literature (Bradley et al., 2008; Gibb et al., 2009; Heim et al., 2009), revealed evidence of early reduced attention. Specifically, N200 amplitude prompted by emotionally arousing words was smaller than for neutral words over right hemisphere among adults with a history of moderate-to-severe childhood abuse. This result suggests that early stages of semantic analysis for emotional words (Kissler et al., 2007) are dampened in individuals with a history of moderate-to-severe childhood abuse, which is consistent with studies using other stimuli and tasks indicating dampened occipital-parietal ERP waveforms (P100, N170, P400) in children raised in an early adverse environment (Moulson et al., 2009). Reduced early attention is also broadly consistent with the findings of a study of ELS in adults, which found that EPN localized to the right hemisphere (similar to N200 in the present study) was generally suppressed for pleasant, neutral, and unpleasant stimuli in those with ELS (Weber et al., 2009). According to Weber et al. (2009), blunted EPN amplitude among those with ELS may reflect a dampening of cortical arousal resulting from a chronically hyperactive stress response during childhood, consistent with ERP evidence of blunted cortical activity in studies of maltreated children (e.g., Moulson et al., 2009).

The categorical approach also revealed smaller P300 amplitude in response to emotional and neutral stimuli for adults with a history of moderate-to-severe childhood abuse than those without a history of abuse. An overall blunting of P300 is consistent with Metzger, Orr, Lasko, McNally, and Pitman (1997), who found that adults with trauma histories and a diagnosis of PTSD exhibited smaller P300 than adults without PTSD during an emotion-word Stroop task (across positive, neutral, and traumatic words). Similarly, Bremner et al. (2004) found that women with a trauma history and PTSD exhibited less anterior cingulate (ACC) activity than women without PTSD during an emotion-word Stroop task. Araki et al., 2005) suggested that an overall reduction in P300 amplitude may reflect, in part, aberrant stimulus discrimination, impaired attention, and/or emotion dysregulation resulting from ACC dysfunction. In-line with this possibility, ACC volume has been found to be smaller in adults with versus without a history of ELS, and this is not accounted for by current mood symptoms (Cohen et al., 2006).

Despite the abuse group showing evidence of smaller N200 and P300 amplitude prior to RT, their behavioral performance did not differ as a function of abuse, using either categorical or dimensional analytic strategies. Thus, adults with a history of childhood abuse (at least as represented in the present sample) can perform the emotion-word Stroop task in a manner that is overtly indistinguishable from that of individuals without an abuse history, despite processing differences prior to (N200, P300) and following (SW) response execution. Present results suggest that earlier disengagement from emotional stimuli may help individuals with moderate-to-severe abuse to achieve normal behavioral performance on the emotion-word Stroop task.

As hypothesized, evidence of an attentional bias toward unpleasant stimuli emerged in the abuse group after RT (SW). This pattern of results emerged in the dimensional analyses as well, which included adults with a range of abuse history. Furthermore, the relationship between childhood abuse and prolonged attention to unpleasant words (SW) was not accounted for by current symptoms of anxiety and depression. Interpreted in the context of the emotion-word Stroop literature, larger SW amplitude prompted by unpleasant stimuli may indicate that individuals with a history of higher levels of childhood abuse experience more conflict between attentional capture (by word meaning) and task goals (to ignore word meaning) when unpleasant stimuli are present (e.g., Bailey & Bartholow, 2016; van Hooff et al., 2008). Alternatively, given that the emotion-word Stroop task did not elicit behavioral evidence of conflict specific to unpleasant stimuli (RT), it is possible that SW reflected selective extended engagement with unpleasant stimuli (Cuthbert et al., 2000).

Present results highlight the potential importance of targeting attention biases in interventions with individuals with a history of childhood abuse. Evidence of later bias toward unpleasant stimuli (SW amplitude) in the context of an emotion-word Stroop task is in-line with findings of attention bias toward unpleasant information in adult PTSD samples (for reviews see Bar-Haim et al., 2007; Buckley, Blanchard, & Neill, 2000). In a community inpatient study of adult active duty-military members receiving treatment for PTSD that included either Prolonged Exposure or Cognitive-Processing Therapy in addition to medication, participants who completed a form of attention training that trained attention away from threat and toward neutral stimuli showed greater reductions in PTSD and depression symptoms than participants who did not receive the attention training intervention (Kuckertz et al., 2014). Training attention toward neutral (away from unpleasant) information among adults with a history of childhood abuse similarly may be a useful adjunctive therapy approach, which may help to reduce later, prolonged engagement with unpleasant information. Additionally, for individuals with a history of moderate-to-severe levels of childhood abuse, attention training targeting early, semantic-level processing (e.g., EPN, N200) of emotional information may be helpful in normalizing “dampened cortical responses” to emotional stimuli observed in previous studies (e.g., Moulson et al., 2009; Weber et al., 2009). These intervention ideas are speculative, as to our knowledge no studies have specifically targeted attention biases among adults with a history of childhood abuse. Such work may be a fruitful direction to explore in future studies.

A notable limitation of the present study was reliance on retrospective, self-reported abuse. One benefit of using self-report questionnaires is that they can be completed quickly and easily, making them a relatively efficient research tool. However, self-report is associated with biases and other factors that can contribute to inaccurate responding (Furnham, 1986). For example, it is possible that individuals are unable to accurately recall how much abuse they experienced in childhood, or they may choose not to report accurately. Evidence is mixed regarding individuals’ ability to accurately provide accurate self-reports of past abuse (Brewin, Andrews, & Gotlib, 1993; Goodman et al., 2003). For example, whereas one prospective study by Goodman et al. (2003) found that the clear majority (81%) of adult participants who had documented cases of abuse during childhood accurately recalled having been abused, a recent meta-analysis found that only half of participants with a history of observed childhood maltreatment (e.g., child protective services records) retrospectively accurately reported it (Baldwin, Reuben, Newbury, & Danesem, 2019). When possible, additional methods of abuse reporting should be included.

The present study was not designed to unequivocally disentangle the impact of childhood abuse history and psychopathology on the time course of attention bias, the latter of which is associated with blunted attention (e.g., major depressive disorder; Weber et al., 2009) and attention biases in the emotion-word Stroop task (e.g., Sass et al., 2010, 2014). The presence of psychopathology is an inherent challenge for studies that examine the impact of childhood abuse in adulthood, however, as early adverse events, including childhood abuse, predict substantially increased rates of lifetime mental health disorders (Bernet & Stein, 1999; Chapman et al., 2004; Fergusson, McLeod, & Horwood, 2013; Green et al., 2010; Kessler et al., 2010). Indeed, categorizing individuals according to childhood abuse history in the present study revealed that almost all participants with a history of moderate-to-severe abuse (89%) had a lifetime history of at least one mental health disorder, which is similar to an estimate of 80% lifetime history of psychopathology among individuals who reported at least moderate childhood sexual abuse (Molnar, Buka, & Kessler, 2001), indicating that the study’s sample was representative of adults with a history of moderate-to-severe abuse. In the full sample, approximately half of participants endorsed a lifetime history of a mental health disorder, which is similar to estimated lifetime rates of mental health disorders among community samples (Kessler et al., 1994, 2005). Additionally, psychopathology scores (e.g., van Rijsoort, Emmelkamp, & Vervaeke, 1999) and CTQ abuse subscale scores were similar to those of other studies that included community participants (Scher, Stein, Asmundson, McCreary, & Forde, 2001), indicating that the full sample was representative of a community sample unselected for psychopathology and childhood abuse. That psychopathology symptoms did not account for the relationship between higher levels of childhood abuse and larger SW amplitude for unpleasant words suggests that depression and anxiety do not play a straightforward role in determining the impact of childhood abuse on later attention bias toward unpleasant information, at least in an unselected sample of community adults.

It may be noted that the large-N, dimensional analysis did not cross-validate some of the findings from the small-N, extreme-groups analysis. Including the psychopathology scale scores is interpretively problematic (Miller & Chapman, 2001; Verona & Miller, 2015), as they likely shared (and thus removed) some of the variance appropriately attributable to childhood abuse history. Furthermore, as reviewed above, the literature on adult effects of childhood abuse documents nonlinear effects of abuse severity, such that low-to-moderate scores often are not associated with low-to-moderate degrees of the impact found for higher abuse scores. Thus, the lack of cross-validation does not undermine the findings for P100, N200, and P300. Nevertheless, the dimensional analyses were informative in showing that the extreme-groups effects were not entirely attributable to internalizing psychopathology.

Present results add to a small but growing number of reports providing evidence of aberrant attention bias in adults with a history of childhood abuse. Present findings of early and later reduced attention to emotional information followed by normal behavioral performance, followed by later increased attention to unpleasant information, specifically clarify the time course of effects using ERPs in an emotion-word Stroop task. Although it is not known whether childhood abuse history per se was directly responsible for the ERP effects observed in the present study, childhood abuse may set into motion a series of cognitive and emotional changes that ultimately contribute to attention biases in adults and, therefore, is critical to consider. Identifying the development and impact of attention biases during child and adulthood is an important step for future studies.