The Impact of Irritability and Callous Unemotional Traits on Reward Positivity in Youth with ADHD and Conduct Problems

Abstract

Children with attention-deficit/hyperactivity disorder (ADHD) and conduct problems exhibit significant variability in functioning and treatment response that cannot be fully accounted for by differences in symptom severity. Reward responsivity (RR) is a potential transdiagnostic means to account for this variability. Irritability and callous-unemotional (CU) traits moderate associations between both ADHD and conduct problems with multiple realms of functioning. Both are theorized to be associated with RR, but associations in clinical samples are unknown. In 48 youth ages 5-12 with ADHD referred for treatment of conduct problems, we examined RR using a guessing task where participants select a door icon to win and lose money. Analyses focused on the reward positivity (RewP) event-related potential in response to gain and loss feedback, which reliably peaks approximately 300ms after feedback. Frequentist and Bayesian approaches assessed main effects of ADHD, Conduct Disorder (CD) and non-irritable Oppositional Defiant Disorder (ODD) symptoms with RR, plus interactions between symptoms and affective dimensions (irritability, CU). CD and ODD were hypothesized to be associated with altered RR, with irritability and CU moderating these associations. Across models, a reliable CD x irritability interaction emerged, indicating enhanced RewP when irritability was elevated and CD symptoms were low. CU did not moderate any associations with RR, and little support was found for associations between RR and other symptom domains. As neural response to reward varied with levels of irritability and CD symptoms, RR may hold potential as a clinically-relevant biomarker in youth with ADHD and conduct problems.

Attention-deficit/hyperactivity disorder (ADHD) is a common and impairing mental health disorder in childhood, impacting over 7% of youth globally (Thomas et al., 2015). Conduct problems, such as aggression and defiance, commonly present with ADHD (Waschbusch, 2002). These symptoms of oppositional defiant disorder (ODD) and conduct disorder (CD) are associated with increased impairment and higher rates of future psychopathology in youth with ADHD (Connor et al., 2010). However, there is significant variability in impairment and outcomes amongst youth with ADHD (Connor et al., 2010; Nigg et al., 2020), which is not fully accounted for by symptom frequency and severity (Karalunas et al., 2019). Therefore, other dimensions are needed to fully account for the differential outcomes.

Integrating affective, cognitive, and temperamental dimensions into the study of ADHD has gained traction as a means of understanding this well-documented heterogeneity. Affective symptoms in particular have been found to account for clinically meaningful differences in phenotypic and prognostic profiles (Goh et al., 2020; Karalunas et al., 2019). Irritability (i.e., proneness to anger) and callous unemotional (CU) traits (i.e., reduced capacity for empathy, remorse or guilt) (Blair, 2018; Brotman et al., 2017; Waschbusch et al., 2020a) have garnered particular attention because they are associated with distinct profiles of impairment, comorbidities, pathophysiology and outcomes (Hawes et al., 2021; Wakschlag et al., 2018). For example, irritability moderates the association of inattention with the speed at which perceptual information is processed (Haller et al., 2021) and appears preferentially sensitive to medication treatments in youth with ADHD (Fernández de la Cruz et al., 2015). CU moderates the course and outcomes of children with conduct problems across multiple domains, including the response to behavioral therapies (Bansal et al., 2019; Frick et al., 2014; Kimonis et al., 2019). Inhibitory control abilities among youth with conduct problems differ as a funciton of CU traits (Frick & Marsee, 2018). While irritability has been established as a distinct dimension of ODD (Burke et al., 2014), elevated rates are also seen in youth with ADHD without ODD (Rappaport et al., 2020; Shaw et al., 2014; Waschbusch et al., 2020a). Similarly, while CU traits are a diagnostic modifier of CD, they commonly occur in children with ADHD and ODD not meeting CD criteria (Haas & Waschbusch, 2012; Willoughby et al., 2011). Therefore, examination of the impact of irritability and CU should occur across the externalizing spectrum.

A critical next step in understanding the intersection of diagnostic criteria and affective symptoms in children with ADHD is linking them on a mechanistic level by identifying transdiagnostic processes. Reward responsivity (RR), defined as the neural and behavioral response to reward cues or the receipt or omission of reward, has been proposed as one possibility (Nigg et al., 2020). Reward dysfunction has been recognized in ADHD dating back to the dual pathway model as youth with the disorder display increased approach towards rewards and a preference for smaller immediate rewards. While there is a sizable literature documenting various behavioral indicators of aberrant reward responses, neural response to rewards remains an understudied area (Luman et al., 2010) that is critical for contextualizing behavioral results.

A growing literature supports the construct validity of event-related potentials for measuring reward responses (Proudfit, 2015). Most of the reward-related research in ADHD has focused on the reward positivity (RewP) component, an enhanced positivity to reward compared to loss or neutral feedback about 300ms after feedback over frontocentral sites that is reliably elicited across development (Kujawa et al., 2018). RewP, also referred to as feedback negativity, is thought to consist of an overlapping positive-going component sensitive to reward and a negative-going component sensitive to loss feedback. It is thought to reflect consummatory reward responses and/or reward prediction signals involved in reinforcement learning (Holroyd et al., 2008b; Proudfit, 2015). Supporting this possibility, RewP is correlated with reward-seeking behavior, positive emotionality, and activation of reward-related brain regions including the ventral striatum and medial prefrontal cortex (Bress & Hajcak, 2013; Kujawa et al., 2015). In children, RewP is often measured using simple guessing tasks where participants select an icon on a screen. Outcomes are actually presented randomly and not dependent on task performance, with RewP recorded in response to feedback indicating monetary gains and losses (Holroyd et al., 2008a; Umemoto et al., 2014).

Reward Response in ADHD

Overall, there is little evidence that RewP is altered in youth with ADHD. For example, children with and without ADHD have shown similar RewP amplitudes to both loss and win feedback when monetary incentives are offered (Holroyd et al., 2008a; Umemoto et al., 2014). More recently, no association was seen between ADHD symptoms in children and either biological or behavioral measures of RR on the balloon analogue risk task (Tenebaum et al. 2018). Kallen et al. (2020) also observed no significant correlation between current ADHD symptoms and RewP amplitude in 300 girls with ADHD, and neural response during reward anticipation and receipt were not found to be predictive of ADHD symptoms in the Adolescent Brain and Cognitive Development (ABCD) study (Owens et al., 2021). However, RewP amplitude was inversely correlated with change in hyperactive-impulsive symptoms over two years in the Kallen (2020) study. Bunford et al. (2021) similarly found that RewP at age 9 was not associated with current levels of ADHD but was positively correlated with future hyperactive/impulsive symptoms. The collective evidence suggests that ADHD symptoms are not associated with abnormalities in RewP although it may impact the trajectory of ADHD symptoms over time.

Reward Response in ODD and CD

Processing the response to reward is theorized to be a core deficit for ODD and CD with multiple studies reporting that youth with elevated levels of conduct problems have difficulties learning from consequences (Blair, 2018) and exhibit increased approach to reward (Byrd et al., 2014; Luman et al., 2010). These behavioral observations have been interpreted as evidence of both decreased and increased neural response to reward (Byrd et al., 2014). Sympathetic tone, viewed as a marker of peripheral reward responsiveness, is often reduced in youth with ADHD and conduct problems (Musser et al., 2013). The few studies that observed significant associations between ADHD diagnostic status and RewP amplitude did not assess conduct problems (Gong et al., 2014; Groen et al., 2013). In contrast, Tenenbaum et al. (2018) observed that in youth with ADHD, CD symptoms were significantly associated with increased reward seeking behavior on the balloon analogue risk task while ODD symptoms correlated with altered autonomic indices of RR. These joint results highlight the importance of considering CD and ODD when assessing links between RewP and ADHD.

Reward Response in Youth with Irritability or CU Traits

Reward response abnormalities are theorized to be a core impairment underlying irritability and CU traits (Blair, 2018; Brotman et al., 2017); however, the two have rarely been examined together and never in a study measuring neural response to reward. Both neural hypersensitivity and hyposensitivity to reward have been associated with irritability which often presents as high levels of impulsive anger (Nigg et al., 2020; Wakschlag et al., 2018). Impulsivity and anger are associated with increased motivation to approach and enhanced neural response to reward (Tsypes et al., 2019). Irritability, independent of other psychiatric symptoms, has been shown to moderate the risk for abnormalities in neural processing of emotions in youth with genetic risks for impairments in this domain (Kessel et al., 2017). Whether irritability moderates associations with neural processing of reward in those at increased risk for RR abnormalities has not been examined. RewP studies in adults have shown that higher trait irritability predicts reduced RewP to loss but not win feedback (Deveney, 2019; cf. Tsypes et al., 2019). Similarly, youth with severe mood dysregulation, marked by persistent irritability, have abnormal behavioral and physiological responses to negative but not positive feedback during frustration paradigms (Deveney et al., 2013). These results suggest that irritability may be most strongly associated with abnormalities in physiological response to loss or nonreward.

While impairments in affiliative reward with CU have been most well studied, CU has been found to moderate a range of other reward driven behaviors (Budhani & Blair, 2005; Waller & Wagner, 2019). For example, CU traits were associated with the greatest abnormalities in parasympathetic and sympathetic activity thought to index reward response during an emotion induction task amongst youth with ADHD (Musser et al., 2013). While CU has not been examined in pediatric RewP studies, associations between the related construct of psychopathy and RewP amplitude have been found in adults (Pfabigan et al., 2011; Schulreich et al., 2013). In 24 college students, psychopathy ratings were significantly associated with RewP amplitude on a time estimation task for reward and loss feedback (Schulreich et al., 2013). In a similar sample of college students, Pfabigan et al. (2011) observed that antisocial personality disorder traits correlated with RewP amplitude in the loss condition. Interestingly, studies in normative samples examining associations of RewP to the broader construct of externalizing symptoms were negative, (Bernat et al., 2011) suggesting these associations may be specific to CU/psychopathy.

Joint Assessment of Behavioral and Affective Symptoms on Reward Positivity

To our knowledge, only one prior study evaluated RewP in children while assessing ADHD and multiple dimensions of conduct problems incorporating either irritability or CU. In this community sample, irritability at age 3 predicted increased RewP to monetary rewards at age 9; however, lifetime symptoms of ADHD and CD were unrelated to RewP, while ODD, which included the irritability dimension, was associated with a reduced RewP (Kessel et al., 2016). Results were limited by the restricted range of behavioral symptoms and the failure to assess CU.

To address some of the discrepancies in the extant literature, the present study evaluated the intersection of ADHD and conduct problems, including the affective dimensions of irritability and CU, in relation to neurophysiological responses to reward as measured by RewP. While there is evidence that irritability is associated with RewP in community samples (Kessel et al., 2016), associations have not yet been studied in clinical samples of youth with ADHD and conduct problems. Relationships between CU traits and RewP has not been explored in clinical or community populations of youth. The present study addresses these limitations by examining the unique and moderating effects of irritability and CU on RewP in children with ADHD referred for treatment of conduct problems. First, we tested whether ADHD, ODD (excluding the irritable dimension) and CD symptom scores were significantly associated with RewP amplitude. Based on the reported correlations of irritability and CU with each of these symptom domains (Brotman et al., 2017; Frick et al., 2014), CD was expected to have the strongest associations with RewP, followed by ODD, with no significant associations expected for ADHD symptoms. We then tested irritability and CU traits as moderators of the effects of ADHD, ODD and CD on RewP. It was hypothesized that both would moderate any associations found between CD, ODD and ADHD with RewP, with stronger effects seen for loss versus win feedback.

Method

Participants

Participants were 48 youth (33 boys, 15 girls), ages 5 to 12 (M = 8.08, SD = 2.09) recruited from an outpatient child psychiatry clinic. There were 37 (77%) participants that identified their race as White, 2 (4%) as Black, 4 (8%) as Other, with 4 (8%) identifying their ethnicity as Hispanic and 70% of parents having a college degree. They were pooled from three studies examining the role of irritability in children presenting for ADHD care. All studies were approved by the Penn State College of Medicine IRB and performed in accordance with the Declaration of Helsinki. Parents provided consent and children ages 7+ gave assent. All children met full diagnostic criteria for ADHD and had to have at least mild levels of ODD symptoms and associated impairment. There were 38 (79%) participants that met full ODD criteria and 15 that (31%) met full CD criteria (all children with CD met ODD criteria). Exclusion criteria were the presence of prominent traits of autism spectrum disorder and use of medications that could adversely impact EEG data collection (e.g., antiepileptics). An additional 5 children completed the monetary reward task but data were lost for 2 participants due to technical error and 3 were excluded for noisy EEG data, leaving the final analyzed sample of 48. There were no significant differences in demographics or behavior ratings between participants with usable and unusable EEG data. Clinical characteristics of the final sample are summarized in Table 1.

Table 1.

Summary of Diagnostic Measures

Measure	N	Mean	SD	Min	Max	alpha
Parent Ratings
ADHD mean scorea	48	1.99	0.51	0.67	2.94	.85
ODDa	48	1.61	0.74	0.13	3.00	.90
Non-irritable symptoms mean score	48	1.63	0.74	0.00	3.00	.83
Irritable symptoms mean score	48	1.56	0.87	0.00	3.00	.88
CD mean scorea	48	0.30	0.30	0.00	1.13	.79
Irritability sum scoreb	48	5.73	3.47	0	12	.90
Callous-Unemotional sum scorec	47	29.06	11.29	13	57	.90
Teacher Ratings
ADHD mean scorea	28	1.43	0.69	0.22	2.50	.92
ODD mean scorea	28	0.95	0.71	0.00	2.38	.86
Non-irritable symptoms mean score	28	1.01	0.77	0	2.40	.82
Irritable symptoms mean score	28	0.85	0.81	0	3.00	.83
CD mean scorea	28	0.20	0.33	0.00	1.57	.42
Diagnostic Interview
ADHD count	48	13.27	3.11	6	18	--
ODD count	48	5.29	2.55	0	8	--
CD count	48	1.23	1.63	0	6	--

Notes: Teacher ratings were unavailable for 20 participants because data were collected during summer break from school. The diagnostic interview was either the Disruptive Behavior Disorders Interview (21 children) or the Children’s Diagnostic Interview Schedule (27 children).

^a= Average symptom score on the Disruptive Behavior Disorders Rating Scale;

^b= Sum score on the Affective Reactivity Index;

^c= Sum score on the Inventory of Callous-Unemotional Traits

Procedure

Diagnostic Procedure

ADHD, ODD and CD were diagnosed using parent report on either the NIMH Diagnostic Interview Schedule for Children IV (Shaffer et al., 2000) or the Disruptive Behavior Disorders Interview (Hartung et al., 2005), depending on the parent study. The Disruptive Behavior Disorder Rating scale was also collected from teachers for all intakes done during the school year. When available, teacher and parent report were integrated using the “OR” rule for final diagnostic determination by a child psychiatrist or clinical psychologist. Impairment was evaluated using parent and teacher ratings on the Impairment Rating Scale (Fabiano et al., 2006).

Experimental Procedures

The experimental task was administered during a one-on-one session with a trained research assistant in a room free from distractions. The assessment session typically lasted 60 minutes, including “warm up” and task administration. All children taking CNS stimulants (n=19) had them held for 24 hours prior to testing. Fourteen children (29%) were taking other psychiatric medications at the time of testing, most commonly nonstimulants for ADHD.

The experimental task used to elicit event-related potentials to reward and loss feedback was a simple guessing reward task shown in previous research to reliably elicit the RewP in children and adolescents (Kujawa et al., 2018; see Figure 1). At the beginning of each trial, participants were presented with two doors and instructed to select the door that might have a prize behind it. The doors remained on screen until participants made a response. Next, the doors were replaced by a fixation mark (+), followed by feedback. Participants were told they could win $0.50, indicated by a green “↑,” or lose $0.25, indicated by a red “↓”, on each trial. Differences in value between reward and loss feedback allowed for equal numbers of each type of feedback (to control for probability effects on ERPs) but also allowed participants to accumulate money across the task. Next, a fixation mark appeared and was followed by the message “Click for the next round”, which remained on screen until the participant responded and the next trial began. The task included 30 win and 30 loss trials in a random order. Participants completed two practice trials to familiarize them with the feedback cues (i.e., upward and downward arrows). Participants were informed they could win up to $5, and all participants received the full $5 following completion of the task.

Fig. 1.

Example Trial on the Guessing Reward Task

Measures

Descriptive statistics and Cronbach’s alpha are reported in Table 1. The Disruptive Behavior Disorder Rating Scale (DBDRS) consists of 45 items measuring the DSM symptoms of ADHD, ODD and CD as rated by parents and teachers. Items are rated using a Likert ranging from 0 (Not At All) to 3 (Very Much). ADHD and CD scores were computed by averaging relevant items. Typical alphas range from .75 to .96 (Pelham et al., 1992) and validity is supported by significant correlations with diagnostic interviews and observed behaviors (Wright et al., 2007). To avoid overlap with irritability, a score representing the behavioral dimension of ODD (ODD-Non-irritable) was computed by averaging symptoms not loading into the irritable dimension: often argues with adults, often spiteful or vindictive, often blames others for his or her own mistakes or misbehavior, often actively defies or refuses to comply with adults’ requests or rules, and often deliberately annoys people (Burke et al., 2014; Waschbusch et al., 2020a).

The Affective Reactivity Index (ARI) consists of 7 items measuring irritability in youth on a Likert ranging from 0 (Not True) to 2 (Certainly True). The first six items measure severity which are summed to compute an overall irritability score. Internal consistency estimates (Cronbach’s alpha) from previous studies using this measure ranged from .88 to .92. The total score is correlated with impairment in clinical samples and differentiates between disorders with irritability as a core criteria versus those not explicitly requiring it (Stringaris et al., 2012).

The Inventory of Callous-Unemotional Traits (ICU) consists of 24 items measuring CU traits in youth using a 0 (Not At All True) to 3 (Definitely True) Likert. Items were reverse scored as needed and then summed to compute a total CU score. The total score has an average model-based reliability estimate (Omega) of .85 (Ray & Frick, 2020) and is significantly associated with impairment and other callousness measures (e.g., Bansal et al., 2020; Ray & Frick, 2020).

EEG Data Collection and Processing

Continuous EEG was recorded using a 32-electrode BrainProducts ActiCHamp system (Munich, Germany). Facial electrodes to measure electrooculogram (EOG) were attached above and below one eye and on either side of each eye with a ground on the back of the neck. Per the BrainProducts design, online data acquisition was referenced to a scalp electrode and then re-referenced to the linked mastoids (TP9/TP10) offline. Impedances were lowered below 30 kΩ. Data were sampled at 1000 Hz. Data processing was performed using BrainVision Analyzer (Brain Products, Munich, Germany), following established procedures for measuring the RewP in children (e.g., Bress et al., 2012) and adults (e.g., Ait Oumeziane & Foti, 2016). Specifically, data were band-pass filtered with cut-offs of 0.1 and 30 Hz, and segmented from 200 ms before to 800 ms after feedback. Ocular correction was conducted using Gratton’s algorithm (Gratton et al., 1983). Semiautomatic artifact rejection criteria were adjusted from prior work to apply more stringent criteria and optimize automatic detection of artifacts for consistency. A voltage step greater than 50 μV between sample points, maximum voltage difference of 175 μV within trials, a minimal allowed amplitude of −200 μV and maximal allowed amplitude of 200 μV, and lowest allowed activity of 0.5 μV within 100 ms intervals were automatically removed. Data were visually inspected and additional artifacts were removed as needed. Interpolation was used to resolve faulty recordings at single electrodes using the signal from surrounding electrodes. Included participants had a minimum of 12 artifact-free trials per condition at Cz. Segments were averaged separately for each condition and baseline corrected to the −200 ms window.

RewP was scored 275-375 ms after feedback at Cz, consistent with prior work in children of similar ages (Kujawa et al., 2018; Kujawa et al., 2014) and the timing and distribution where RewP was maximal in the sample (Figure 2). These measures demonstrated good split-half reliability at Cz (Spearman-Brown coefficients = .90 for RewP win and .86 for RewP loss). To isolate the variance in the ERP wave attributed to responses to reward and loss feedback, unstandardized residual scores were computed by partialing out the variance associated with each condition (Meyer et al., 2017). The resulting RewP win measure was a residual score adjusting for response to loss (M = 0.00, SD = 5.28, Min = −11.36, Max = 10.11). The resulting RewP loss measure was a residual score adjusting for response to wins (M = 0.00, SD = 5.87, Min = −14.33, Max = 11.34). Consistent with prior work (Kessel et al., 2016), we examined the different symptom dimensions in relation to the RewP residuals. More positive values for win (residual adjusting for responses to loss) and more negative values for loss (residual adjusting for responses to wins) are thought to reflect greater reward and loss responsiveness, respectively.

Fig. 2.

ERPs (Negative Up) and Scalp Distribution Depicting Neural Responses to Monetary Win vs. Loss Feedback in the Overall Sample

Data Analysis

We first computed bivariate correlations to examine whether RewP was associated with symptom scores (continuously defined using parent ratings from the Disruptive Behavior Disorder Rating Scale) or demographic measures. We used dimensional parent ratings rather than diagnoses to increase power and to better reflect the dimensional nature of the disorders (Barry et al., 2013). Next, regression analyses were computed to test whether irritability or CU traits moderate the association between symptom scores (ADHD, ODD-Non-irritable, CD) and RewP, with separate regressions computed for RewP to win and loss (both residual scores) and for each symptom score. Significant interactions were followed up by computing simple slopes tests at low (defined as the 16^th percentile of the sample) and high (defined as the 84^th percentile of the sample) values of irritability or CU. Age and sex were included as covariates in preliminary regressions but were dropped from final models as they were not significant predictors of RewP to wins or losses (p’s ≥ .1431). Analyses were computed using SPSS version 26, with regressions computed using version 3.4 of the PROCESS macro (Hayes, 2017).

Traditional frequentist statistics were supplemented with Bayesian multiple linear regressions conducted in JASP (JASP Team, 2018). The primary advantage of Bayesian statistics is that it quantifies the magnitude of support for both the alternative and null hypotheses, thus permitting inferences regarding evidence of absence, rather than merely absence of evidence (Rouder & Morey, 2012; Wagenmakers et al., 2016). Bayesian statistics are also generally more adept at handling small sample sizes (McNeish, 2016), which is particularly advantageous in the present study, given the relatively small sample and concern that null effects in the frequentist approach could be due to low power. JASP uses a model comparison approach in which all permutations of predictor variables (i.e., all possible combinations of main effects and specified interaction terms) are evaluated. In other words, the software indicates which variables should be included in the model, rather than the user. We therefore included all potential main effects (ADHD, ODD CD, irritability, and CU traits), and interactions of diagnostic symptoms with irritability and CU traits as predictors, resulting in 193 potential models. The best-fitting model, relative to an intercept-only null model, was selected. Importantly, this model comparison approach does not result in concerns about alpha inflation due to multiple comparisons because, as discussed below, Bayesian analyses do not rely on a p-value. Uniform priors were specified, which gives each model an equal probability before “seeing” the data.

Bayesian modeling provides a Bayes factor (BF), rather than a p-value, and is the relative probability of one hypothesis over the other, given the data. BF₁₀ is the probability of the alternative over the null hypothesis, and BF₀₁ is the probability of the null over the alternative hypothesis (Wagenmakers et al., 2016). Generally, a Bayes factor of 1-3 indicates weak evidence that should not be interpreted (i.e., there is no clear support in favor of either hypothesis), a Bayes factor of 3-10 indicates moderate evidence, and Bayes factor > 10 indicates strong evidence (Jeffreys, 1939; van Doorn et al., 2021).

Results

Bivariate Correlations

Bivariate correlations (Table 2) showed that RewP win and RewP loss were significantly and inversely correlated with each other, but neither was significantly correlated with any symptom measure. ODD-Non-IRR was significantly correlated with CD, IRR, and CU. CD was significantly correlated with irritability and CU.

Table 2.

Correlations between Study Variables

	1	2	3	4	5	6	7	8
1. Age	--
2. Gender (0 = female)	−.15	--
3. ADHDa	−.03	−.01	--
4. ODD-Non-irritablea	.03	.17	.28+	--
5. CDa	.03	.05	.14	.61*	--
6. Irritabilityb	−.11	.04	.12	.62*	.47*	--
7. Callous-Unemotionalc	.28+	−.08	.13	.43*	.56*	.22	--
8. RewP win residual	.07	−.11	−.22	.03	−.21	.16	−.04	--
9. RewP loss residual	−.02	.22	.21	.03	.14	−.09	−.01	−.86*

Notes: All ratings were completed by parents;

^a= Average symptom score on the Disruptive Behavior Disorders Rating Scale

^b= Sum score on the Affective Reactivity Index;

^c= Sum score on the Inventory of Callous-Unemotional Traits;

^*p < .05,

⁺p < .10

Regressions Testing Irritability and CU as Moderators

The overall ADHD regression model was not significant for either RewP win [F(5, 41) = 1.28, p = .29, R² = .07] or RewP loss [F(5, 41) = 0.40, p = .84, R² = .05], nor were any regression coefficients significant within the models. The overall ODD-Non-irritable regression model was not significant for either win [F(5, 41) = 0.65, p = .66, R² = .08] or loss [F(5, 41) = 0.52, p = .76, R² = .06] nor were any regression coefficients significant within the models. The overall CD regression model in table 3 was significant for both RewP win and RewP loss. The regression coefficients (see Table 3) showed a significant main effect of irritability for wins and a CD*irritability interaction for both win and loss (see Figure 3). Simple slopes follow up tests of RewP wins (see legend of Figure 3) indicated that higher levels of CD were associated with a less positive RewP for youth high in irritability (defined as the 16^th percentile of the sample) but not those low in irritability (defined as the 84^th percentile of the sample). In contrast, simple slopes tests of RewP loss (see legend of Figure 3) showed that higher CD levels were associated with a more positive RewP for youth high in irritability but not those low in irritability.

Table 3.

Regressions Examining IRRITABILITY and CU as a Moderators of the Association Between CD and RewP

Predictor	b(SE)	p-value	BF	Predictor	b (SE)	p-value	BF
WINS				LOSSES
Intercept	0.72 (0.83)	.39	---	Intercept	−0.66 (0.93)	.48	---
CD	−2.78 (3.92)	.48	3.63	CD	0.85 (4.39)	.85	3.50
IRR	0.51 (0.24)	.04	3.68	IRR	−0.38 (0.27)	.17	4.11
CU	−0.01 (0.08)	.88	---	CU	0.03 (0.09)	.74	---
CD*IRR	−2.15 (0.88)	.02	6.92	CD*IRR	3.14 (0.98)	.01	11.89
CD*CU	0.13 (0.27)	.64	---	CD*CU	−0.37 (0.30)	.23	---
Model: F(5, 41) = 2.69, p = .0341, R2 = .25				Model: F(5, 41) = 2.62, p = .0380, R2 = .24

Notes: CD = conduct disorder, CU = callous-unemotional, IRR= Irritability. BF = Bayes Factor for inclusion in the model. BF >3 is considered moderate evidence for inclusion. Please note that the Bayes Factors included in this table are derived from the best-fitting model for each outcome, which did not include CU traits.

Fig. 3.

Simple Slopes Depicting the CD*Irritability Interaction for RewP Wins (residual scores accounting for RewP Loss) and RewP Losses (residual scores accounting for RewP Wins)

Bayesian Results

Testing all combinations of main effects and interaction terms specified above, the best-fitting model for RewP win included the main effects of CD, ADHD, irritability, as well as the interactions between CD and irritability and between ADHD and irritability (BF₁₀ = 6.65 R² = .32). The Bayes factor indicates that the data were 6.65 times more likely to be observed under the alternative hypothesis than the null hypothesis. Although the best-fitting model included ADHD, the posterior summary of coefficients did not support the inclusion of the main effect of ADHD (BF_inclusion = 0.96) or the ADHD x irritability interaction (BF_inclusion = 1.73)¹. There was moderate support for the main effect of CD (BF_inclusion = 3.63), the main effect of irritability (BF_inclusion = 3.68), and the CD x irritability interaction (BF_inclusion = 6.92). Consistent with the results from the frequentist approach, a significant CD*Irritability interaction was observed. Greater levels of CD symptoms were associated with reduced (less positive) RewP to wins when irritability was high (b = −7.22, BF = 5.60). However, CD symptoms were also significantly associated with RewP when irritability was low (b = 1.91, BF = 4.98).

The best-fitting model for RewP loss also included main effects of CD, ADHD, irritability, and the CD x irritability and ADHD x irritability interactions [BF₁₀ = 11.31, R² = .34]. The posterior summary of coefficients provides support for the main effect of CD (BF_inclusion = 3.50), the main effect of irritability (BF_inclusion = 4.11), the CD x irritability interaction (BF_inclusion = 11.89) and the ADHD x irritability interaction (BF_inclusion = 5.07). Once again, the CD x irritability interaction was robust. Greater levels of CD symptoms were associated with a reduced (i.e. less negative) RewP to loss when irritability was high (b = 6.63, BF = 3.62.). However, CD was also significantly associated with RewP when irritability was low (b = −4.64, BF = 4.08). Despite the moderate evidence for including the ADHD x irritability interaction, there was no evidence that ADHD was associated with RewP to loss at either high (b = −0.83, BF = 1.85) or low (b = 2.75, BF = 1.82) levels of IRR. Finally, although non-irritable ODD and CU are not presented above because the best-fitting models did not include these variables, it is important to note that there was moderate-to-strong evidence against the inclusion of non-irritable ODD or CU in any model (BF₀₁ = 3.22 – 28.09). In other words, frequentist analyses showed null results for these variables, and Bayesian analyses provided evidence that that non-irritable ODD and CU should not be included in any model for RewP for win or loss.

Discussion

In a treatment-referred sample of youth with ADHD and conduct problems, we examined the contributions of symptoms of ADHD, non-irritable ODD and CD to reward responsivity (RR) as measured by reward positivity (RewP) and the moderating effects of CU and irritability on these associations. A robust interaction between CD symptoms and irritability was observed for responsiveness to wins and losses across both frequentist and Bayesian analyses. For children with greater irritability, CD symptoms were consistently associated with RewP amplitudes. Examination of the slopes in Figure 3 suggests that RewP amplitude is relatively enhanced (i.e., more positive for wins and more negative for loss) in children with low levels of CD symptoms. At low levels of irritability, this association was more tenuous, with Bayesian but not frequentist analyses supporting a link between CD and RR. As hypothesized, ADHD symptoms were not associated with RR in frequentist analyses and weakly and inconsistently associated in Bayesian analyses. The non-irritable dimension of ODD was consistently not associated with RR across models. Unexpectedly, CU did not moderate the association between CD and RR for either RewP win or loss across both the frequentist and Bayesian approaches. In fact, Bayesian analyses provided moderate-to-strong evidence against including CU in any model, indicating that the null results are not merely a result of low statistical power in frequentist regressions.

CD is typically associated with diminished neural activity when anticipating reward in MRI studies using the monetary incentive delay task (Hawes et al., 2021; Veroude et al., 2016). There has been much less examination of its association with the response to reward receipt. In the ABCD study, Hawes et al. (2021) observed increased striatal activation to reward receipt in youth with elevated conduct problems. Irritability was not assessed nor were the specific associations of ODD and CD with RR. The observed increased striatal activation may have been driven by irritability, and therefore our results may be consistent with what was observed in this large community sample. In the present study, the direct effect of CD symptoms on RR was only observed in Bayesian analyses, but the interaction between CD and irritability predicting RewP to both wins and losses was robustly present across both frequentist and Bayesian approaches. The significant interaction, even in models without significant main effects, emphasizes the importance of assessing the entire spectrum of conduct problems – including irritability and CU--when examining RR in samples of ADHD. Irritability and CD are established risk factors for depression (Erskine et al., 2016; Stringaris et al., 2009), with a reduced RewP increasing depression risk (Kujawa et al., 2019). In contrast, we observed an enhanced RewP when irritability was higher. Most work examining the links between RewP and psychopathology in children has not examined the change in RewP over time so it remains unclear if an acquired blunting of a previously elevated RewP amplitude predicts future depression.

We found little-to-no evidence in either analytic model to support that ADHD was associated with abnormalities in RR when accounting for conduct problems. The failure to account for conduct problems may partially explain the discrepant findings across studies in ADHD. Therefore, screening youth with ADHD for conduct problems is warranted given their high rates of comorbidity (Connor et al., 2010; Shaw et al., 2014). These results suggest that youth with ADHD without elevations in irritability or CD symptoms may evidence a normative behavioral response to rewards, as has been observed elsewhere (Fosco et al., 2015).

No significant associations with RR were observed for the behavioral dimension of ODD across models. The irritable and behavioral ODD dimensions have unique associations with future comorbidity and impairment (Burke et al., 2014), supporting potential distinctions in their relationships with RR. Kessel (2016) observed that lifetime ODD symptoms predicted reduced RewP in childhood while preschool DMDD symptoms, which are similar to irritable ODD symptoms, predicted increased RewP in childhood. These opposing associations with RewP are suprising given the strong correlations between ODD dimensions (Burke et al., 2014). Varying results across studies may be due to differences in sample and methodology. In contrast to this cross sectional assessment of children refered for treatment, the longitudinal community study by Kessel (2016) that examined the impact of early chidhood behavior problems on future reward responsivity did not assess the unique impact of the behvioral dimesion of ODD on RR.

Results suggest that irritability may be associated with increased neural sensitivity to reward feedback and are largely consistent with prior research of hyper-responsiveness to reward in children and adults with irritability (Kessel et al., 2016; Tsypes et al., 2019). No prior study in a treatment seeking sample has examined the joint effects of CD and irritability on RR. In participants with elevated irritability, greater CD symptoms were associated with a suppressed RewP, such that the RewP amplitudes in response to both wins and losses were close to 0 for children with high CD symptoms. Replication is needed to verify if elevated irritability in the presence of mild versus marked CD symptoms demarcates the subset of youth with ADHD with the greatest enhancements in neural response to reward.

In contrast to irritability, CU traits were not associated with RR abnormalities across analyses. Initial studies in community samples of adults found associations between psychopathy and RewP (Pfabigan et al., 2011; Schulreich et al., 2013). However, a more recent report in a larger community sample using a passive gambling task similar to the guessing task employed here found no relation between the two (Salim et al., 2015). In the earlier works, RewP amplitude was measured relative to the P2 amplitude, a nonstandard scoring approach (Pfabigan et al., 2011) or assessed during a time estimation task that is meaningfully different than a simple guessing task (Schulreich et al., 2013), potentially explaining differences. Hawes et al. (2021) did observe associations between CU and increased RR across multiple brain regions, but irritability was not assessed. Consistent with our results, Blair et al. (2020) found no evidence that CU traits or ADHD symptoms were associated with temporal discounting of rewards while irritability levels correlated with the degree of temporal discounting in adolescents even after accounting for levels of other conduct problems. These results add to the accumulating evidence that CU traits may not contribute to the reinforcement learning deficits seen in youth with CD.

Unexpectedly, comparable associations between RewP and symptom dimensions were observed to win and loss feedback. RewP amplitude during win and loss feedback were robustly correlated. Prior work in clinical samples of youth (Deveney et al., 2013) and normative samples of adults (Deveney, 2019) found RR abnormalities preferentially during loss feedback. Children are more likely to exhibit impaired cognitive processing and abnormal autonomic reactivity when frustrated (Deveney et al., 2013), and it has been suggested that the RewP may be similarly state dependent (Angus et al., 2015). All of these prior studies employed frustrative non-reward paradigms with participants intentionally exposed to recurrent loss feedback. In contrast, win and loss feedback on the guessing task is designed to ensure participants end with winnings. Therefore, guessing tasks may not induce anger to the same degree as frustrative nonreward tasks, potentially explaining the failure to observe greater RewP alterations with loss feedback.

This study has multiple strengths, including assessment of irritability and CU traits using established measures (Bansal et al., 2020; Stringaris et al., 2012), use of an objective measure of RR to avoid issues with common method variance and two different analytic methods that largely replicated results. Including Bayesian analyses strengthened the capacity to make null inferences from the evidence of absence. Additional strengths include the established reliability and validity of the RewP elicited by the guessing task (Bress & Hajcak, 2013; Kujawa et al., 2018) and the quality of the EEG data, with 90% of participants providing usable data despite their young age. As task complexity reduces reliability for event-related potentials (Meyer et al., 2013), we employed a simple guessing task over a frustrative nonreward design, which may have increased the rate of usable data. In addition, the employed task allows for rapid administration of multiple trials which may enhance reliability for measuring neural response to reward.

The study also has several limitations. It is cross-sectional, precluding the ability to examine the trajectory of RR. Although bigger than most past studies assessing RewP in clinical samples (e.g., Groen et al., 2013; Holroyd et al., 2008a; van Meel et al., 2011), the sample was relatively small, particularly compared to community samples (e.g., Kessel et al., 2016). Sample size concerns are somewhat lessened by the addition of the Bayesian analysis, which is more adept at handling small sample sizes and strengthens confidence in null findings. The study examined only monetary rewards, and ADHD has also been associated with deficits in social reward processing (Babinski et al., 2019; Foulkes et al., 2014). We did not include a non-ADHD control group, which may have limited the range for some symptom dimensions. The employed task measured response to receipt of reward feedback, but it cannot assess reward anticipation or other aspects of reinforcement learning; however, past work has extensively focused on the neural response to reward anticipation (Veroude et al., 2016). While there is appreciable support for the distinct dimensions of ODD (Burke et al., 2014), this is the first study to examine the specific associations of each dimension with any event-related potential. As there are not established measures of just the behavioral dimension of ODD, results regarding its association with RewP should be viewed with caution. Elevated irritability is associated with increased depressive symptoms in youth (Burke et al., 2014) which impact RewP amplitude (Kujawa et al., 2019). Unfortunately, depressive symptoms were not systematically assessed across our sample. While there is no evidence that nonstimulant medications impact RewP, these medications were not held during testing due to their pharmacokinetic profiles, which could have impacted results.

Clinical Implications

There is growing evidence that individual variances in RR may inform the response to pharmacological and psychosocial treatments for internalizing disorders (Burkhouse et al., 2018). In children with internalizing disorders, psychosocial interventions appear to enhance RewP amplitude, with changes in reward network connectivity mediating treatment response (Luby et al., 2020). It has been theorized that RR could predict the efficacy of behavioral treatments for ADHD (Li, 2018), but initial results using parent report of RR have not proved impactful (van Langen et al., 2020). Our results appear to suggest that in youth with ADHD and mildly elevated levels of CD symptoms, irritability is associated with enhanced neural response to reward, supporting reward programs as a core component for the psychosocial treatment of irritability in such youth. As both irritability and ADHD are associated with increased temporal discounting (Blair et al., 2020), it seems prudent to ensure that initial goals be quickly attainable to prevent rewards from serving as frustration triggers. CU traits were not associated with abnormalities in RR. Studies examining effects of reward-emphasized treatments in youth with CU traits provide mixed results (Kimonis et al., 2019; Waschbusch et al., 2020b). Presently, there appears to be insufficient evidence that youth with CU will experience enhanced outcomes by modifying existing evidence-based behavioral interventions to emphasize reward.

In youth with ADHD, we observed that neural response to reward receipt varies with irritability and CD symptom intensity. Other conduct problems and callous unemotional traits were unrelated to reward responsivity, and little evidence was seen to support that ADHD is associated with altered neural response to reward. Further exploration of the capacity of neural measures of reward responsivity to subtype the heterogeneous population of children with ADHD appears warranted as well as exploring if changes in RewP amplitude mediate response to behavioral therapies for conduct problems in children with ADHD and elevated irritability.

Availability of data and material:

Data will be made available upon request to the corresponding author.