Attention deficit hyperactivity disorder (ADHD) is a common neurodevelopmental disorder associated with impaired regulation of attention, activity level, and impulsivity. In real-world and laboratory environments, children with ADHD exhibit deficits in both error recognition (i.e., self-monitoring) and adaptive control (i.e., corrective behavioral change) (Shiels & Hawk Jr, Reference Shiels and Hawk2010). Post-error slowing (PES), which is defined by slowed response times immediately following an error, is often reduced among children with ADHD. This finding has been interpreted as support for the theory that ADHD symptoms are related to neurocognitive deficits in error detection and adaptive control (Willcutt, Doyle, Nigg, Faraone, & Pennington, Reference Willcutt, Doyle, Nigg, Faraone and Pennington2005). However, a small body of research posits a contrasting theory that PES among typically developing controls derives from delayed attention orienting following an infrequent event, such as an error (Notebaert et al., Reference Notebaert, Houtman, Van Opstal, Gevers, Fias and Verguts2009). This would suggest that reduced PES associated with ADHD is instead indicative of performance enhancement following novelty, which is consistent with literature reporting that novel environmental cues facilitate improved task performance in this population (Mullane, Corkum, Klein, McLaughlin, & Lawrence, Reference Mullane, Corkum, Klein, McLaughlin and Lawrence2011). Given the vast neurocognitive and behavioral heterogeneity of ADHD (Clarke, Barry, McCarthy, & Selikowitz, Reference Clarke, Barry, McCarthy and Selikowitz2001; Fair, Bathula, Nikolas, & Nigg, Reference Fair, Bathula, Nikolas and Nigg2012; Loo, McGough, McCracken, & Smalley, Reference Loo, McGough, McCracken and Smalley2018), the etiology of PES differences may be multiplicative. In the current study, we investigate competing hypotheses about the cognitive-behavioral origin of reduced PES in children with ADHD.
PES Differences and ADHD
A meta-analysis by Balogh and Czobar (Reference Balogh and Czobor2016) reported reduced PES associated with a diagnosis of ADHD, with a medium effect size (d = .42). However, several studies have failed to find PES differences among children with ADHD compared to controls (Van De Voorde, Roeyers, & Wiersema, Reference Van De Voorde, Roeyers and Wiersema2010; van Meel, Heslenfeld, Oosterlaan, & Sergeant, Reference van Meel, Heslenfeld, Oosterlaan and Sergeant2007), possibly due to moderating variables such as interstimulus interval (Balogh & Czobor, Reference Balogh and Czobor2016), task difficulty (Regev & Meiran, Reference Regev and Meiran2014), and error type (i.e., omission vs. commission; Epstein et al., Reference Epstein, Hwang, Antonini, Langberg, Altaye and Arnold2010). Importantly, an adult ADHD study found that reduced PES was not simply a marker of ADHD, but was linearly associated with the severity of current symptoms (Mohamed, Börger, Geuze, & van der Meere, Reference Mohamed, Börger, Geuze and van der Meere2016). Moreover, Shiels & Hawk (Reference Shiels and Hawk2010) reported that deficient PES was only evident among children with the predominantly inattentive subtype of ADHD, underscoring the importance of considering ADHD from a dimensional, rather than categorical, perspective. Altogether, evidence for a PES reduction in ADHD is mixed and appears to be influenced by methodological as well as individual differences (Shiels & Hawk Jr, Reference Shiels and Hawk2010).
Error Recognition versus Adaptive Control Theories of PES
An assumption inherent to the construct of PES is that individuals consciously or subconsciously recognize when an error has been committed, engendering subsequent behavioral change. Therefore, the leading explanations for reduced PES in ADHD are deficient error detection, impaired adaptive control, or both. The extant literature indicates mixed support for each hypothesis. In support of the error-detection theory, a study of school-aged children with ADHD found that the association between inattention and reduced PES was only evident on a task where there were no cues to facilitate error recognition (Shiels & Hawk Jr, Reference Shiels and Hawk2010). Likewise, a meta-analysis reported that PES is longer following errors that are more easily brought to conscious awareness (i.e., inhibitory commission errors) as opposed to choice errors, which could more easily go undetected (Balogh & Czobor, Reference Balogh and Czobor2016). In contrast with the adaptive control theory, Schachar et al. (Reference Schachar, Chen, Logan, Ornstein, Crosbie, Ickowicz and Pakulak2004) reported that PES was not associated with a behavioral measure of inhibitory control, and children with ADHD inconsistently showed PES after each error. However, this study also found that on trials for which PES was evident, the ADHD group showed shorter PES than controls, which would support a role of reduced adaptive control in PES. Evidence from event-related potential (ERP) studies is largely indicative of a deficit in early, automatic error detection associated with ADHD. Across multiple task modalities, individuals with ADHD tend to show attenuated amplitude of the error-related negativity (ERN) component (Balogh & Czobor, Reference Balogh and Czobor2016; Geburek, Rist, Gediga, Stroux, & Pedersen, Reference Geburek, Rist, Gediga, Stroux and Pedersen2013; Michelini et al., Reference Michelini, Kitsune, Cheung, Brandeis, Banaschewski, Asherson and Kuntsi2016). On the other hand, some studies have documented intact ERN amplitude among children with ADHD (Groom et al., Reference Groom, Cahill, Bates, Jackson, Calton, Liddle and Hollis2010; Wiersema, Van der Meere, & Roeyers, Reference Wiersema, Van der Meere and Roeyers2005), despite reduced PES and/or attenuated error positivity, the latter of which is thought to relate to conscious error processing and adaptive control.
Post-error accuracy increase (PAI) is another measure of adaptive control that has been documented among healthy individuals. However, PAI was uncorrelated with PES in a study of healthy adults (Danielmeier & Ullsperger, Reference Danielmeier and Ullsperger2011). Moreover, PES and PAI appear to have distinct neurobiological origins. A study of adult males reported that while increased PES was associated with suppression of response-related sensorimotor cortical activation following an error, performance improvement was associated with enhanced activation of a stimulus-specific sensory processing region (King, Korb, von Cramon, & Ullsperger, Reference King, Korb, von Cramon and Ullsperger2010). The finding that PES and PAI are at least partially non-overlapping among healthy individuals is consistent with an etiology of PES that is not explicitly tied to adaptive control.
Attention orienting theory of PES
Emerging research suggests PES may be at least partially driven by a neurocognitive response to low-frequency events (i.e., novelty) rather than specifically to error (Dutilh, Vandekerckhove, et al., Reference Dutilh, Vandekerckhove, Forstmann, Keuleers, Brysbaert and Wagenmakers2012; Notebaert et al., Reference Notebaert, Houtman, Van Opstal, Gevers, Fias and Verguts2009; Wessel, Danielmeier, Morton, & Ullsperger, Reference Wessel, Danielmeier, Morton and Ullsperger2012). In a small study of adults, Notebeart et al. (Reference Notebaert, Houtman, Van Opstal, Gevers, Fias and Verguts2009) found that they could elicit post-correct response slowing when correct responses were less frequent than errors. Similarly, a study of healthy adults demonstrated that response times were slower following novel auditory stimuli during an oddball task (Parmentier, Vasilev, & Andrés, Reference Parmentier, Vasilev and Andrés2019). Another healthy adult study (Núňez Castellar, Kühn, Fias, & Notebaert, Reference Núňez Castellar, Kühn, Fias and Notebaert2010) found that although PES was not correlated with neurophysiological evidence of error recognition (i.e., ERN amplitude), it was associated with another ERP component, the P3, which is elicited by novel stimuli and typically reduced in children with ADHD (Banaschewski et al., Reference Banaschewski, Brandeis, Heinrich, Albrecht, Brunner and Rothenberger2003; Gow et al., Reference Gow, Rubia, Taylor, Vallée-Tourangeau, Matsudaira, Ibrahimovic and Sumich2012; Keage et al., Reference Keage, Clark, Hermens, Kohn, Clarke, Williams and Gordon2006). This body of research strongly implies that PES is at least partially driven by attention orienting away from the task following infrequent, or novel, events.
Interestingly, novel, extraneous stimuli may instead enhance behavioral performance among children with ADHD (Balogh & Czobor, Reference Balogh and Czobor2016; Tegelbeckers et al., Reference Tegelbeckers, Schares, Lederer, Bonath, Flechtner and Krauel2016; van Mourik, Oosterlaan, Heslenfeld, Konig, & Sergeant, Reference van Mourik, Oosterlaan, Heslenfeld, Konig and Sergeant2007). For example, van Mourik et al. (Reference van Mourik, Oosterlaan, Heslenfeld, Konig and Sergeant2007) reported that novel sounds reduced errors of commission among children with ADHD to a greater extent than among control children during a simple forced-choice ERP task. Similarly, Beike & Zentall (Reference Beike and Zentall2012) found that children with ADHD had improved reading comprehension for high compared to low novelty passages.
The current study
In the current study, we tested three hypotheses about the origin of PES differences among children with ADHD, which were not mutually exclusive (Table 1). Specifically, we evaluated whether PES differences among children with ADHD could be explained by reduced error monitoring, impaired adaptive control, and/or atypical orientation to novelty. We utilized two sets of tasks that varied on either task difficulty or error type to isolate each of these hypotheses and investigate associations with continuous measures of inattentive and hyperactive/impulsive symptom severity. Additionally, we examined moderating effects of both ADHD diagnosis and ADHD symptom severity on PES, in order to address previous methodological differences that might account for inconsistent findings in the literature.
Table 1. Competing hypotheses
Methods
Participants
One hundred and forty-one children, aged 7–11 years, were enrolled for participation in the research study. Recruitment was done via outreach to local mental health clinics, pediatric medical centers, schools, community interest groups, and community research study pools. The ADHD group was recruited based on parent report of a previous diagnosis of Diagnostic and Statistical Manual of Mental Disorders, 5th Edition (DSM-5) ADHD by a licensed psychologist, psychiatrist, or pediatrician. Collection of diagnostic data is an ongoing part of this study; for the current analyses, only ADHD participants whose diagnosis has been confirmed through standardized clinical interview using the Kiddie Schedule for Affective Disorders (Townsend et al., Reference Townsend, Kobak, Kearney, Milham, Andreotti, Escalera and Kaufman2019) or verified through record review by the supervising psychologist were included. Because prior research indicates continuous negative associations between ADHD symptom severity and PES (Mohamed et al., Reference Mohamed, Börger, Geuze and van der Meere2016), we included children with a broad range of inattentive and hyperactive/impulsive symptom severities in our ADHD group, rather than limiting the group to children with a particular clinical subtype. Exclusion criteria were autism spectrum disorder, IQ < 80, history of significant perinatal trauma or birth before 32 weeks, prenatal exposure to alcohol or drugs, or a history of seizures or abnormal EEG. Participants recruited as controls (n = 32) did not have a diagnosis of or concern for ADHD, nor did they have any immediate family members with a diagnosis of ADHD. A subset of participants were excluded following testing due to low IQ (n = 2), suspicion of autism spectrum disorder by the supervising psychologist on the study (n = 3), failure to abstain from medications prior to the visit (n = 2), suspicion of ADHD in a control subject based on parent ratings and observations by the supervising psychologist (n = 2), or identification of epileptiform waves during EEG (n = 1).
The final sample included 74 children with ADHD and 30 controls. The mean age was 9.12 years (SD = 1.37). Thirty-nine percent of the participants identified as non-White, and the proportion of females was 41%. ADHD and control groups did not differ on age, sex or ethnicity distribution (see Table 2). Psychiatric symptoms were estimated with parent report on the child behavior checklist (CBCL 6-18; Achenbach, Reference Achenbach and Rescorla2014). Within the sample recruited for elevated ADHD symptoms, 42% had T-scores in the clinically elevated range on the DSM-5 Affective, Anxiety, Somaticizing, Oppositional Defiant, or Conduct Disorder scales. None of the control subjects had elevated psychiatric symptoms by parent report. Participants also completed brief cognitive and academic testing using the Wechsler Abbreviated Scales of Intelligence, 2nd Edition, and the Wechsler Individual Achievement Test, 3rd Edition, respectively. ADHD participants had lower abbreviated full-scale IQ than control subjects (ADHD mean = 107.39, SD = 12.13, range = 85–143; control mean = 117.50, SD = 10.45, range = 91–135). Fifty-one percent of ADHD participants and 3% of control participants scored at least one standard deviation below average (standard score < 85) on WIAT-III single-word reading, pseudoword decoding, or numerical operations subtests, which is consistent with previously reported rates of coexisting learning disorders in ADHD (Reale et al., Reference Reale, Bartoli, Cartabia, Zanetti, Costantino, Canevini and Bonati2017).
Table 2. Sample characteristics
Note. IA = Inattention symptoms. HI = Hyperactivity/Impulsivity symptoms. PES = Post-Error Slowing. PAI = Post-error accuracy increase. PNS = Post-Novel Slowing. P-values are derived from Welch’s two-sample t-test, except for % female, which was tested with a chi-square analysis; ns = group difference not significant at p < .05.
Ethical considerations
Caregivers completed written informed consent and participating children completed written assent at the start of the in-person visit. All procedures were in compliance with the university institutional review board.
Procedures
Participants visited a university medical center for a single, 3-hr visit that included scalp electrophysiology, cognitive and academic testing, and parent report of behavioral, psychiatric, and medical histories. Children abstained from taking prescribed stimulants or other psychotropic medications for 48 hrs or longer prior to the visit, depending on the medication half-life and physician guidelines.
Measures
Experiments
PES was measured with four computerized tasks. Trials with omission errors were excluded from the analyses. An individual criterion for each task was greater than 50% accuracy on trials on which a response was made, and at least one post-error trial available after processing (see Table 2, for a number of individuals included in each task).
The first two tasks, “Easy ERP” and “Hard ERP,” were done during simultaneous scalp electrophysiology measurement. The electrophysiological results will be described in a separate manuscript. In both Easy and Hard ERP tasks, target (task-related) visual stimuli were presented alternately with irrelevant (non-task-related) visual stimuli, using a design adapted from experiments previously described by Jonkman and colleagues (Reference Jonkman, Kemner, Verbaten, Koelega, Camfferman, vd Gaag and van Engeland1997). Target stimuli included red, blue, green, and orange rectangles. The irrelevant stimuli included a white bracket presented 60% of the time; an identical bracket oriented in the opposite direction presented 20% of the time; and non-repeated white line drawings of animals and vehicles, presented 20% of the time. Each task lasted approximately 8 min and included up to three practice sets of 10 trials, followed by 140 target and 140 irrelevant stimuli, presented with a stimulus duration of 300 ms and interstimulus interval of 0.8–1.4 s. The Easy ERP task was a visual forced-choice discrimination task; participants were instructed to respond with a right-hand button press to blue rectangles (50%) and a left-hand button press to all other targets. The Hard ERP task was a 1-back task, in which participants were told to press the right button when two identical targets were presented consecutively (50%), and the left button for incongruent consecutive targets. In both Easy and Hard ERP tasks, participants were instructed not to respond to the irrelevant stimuli which were presented between each target stimulus. Thus, the irrelevant stimuli constituted a passive visual oddball paradigm that was integrated with the forced-choice and 1-back tasks. Participant behavior was monitored by the experimenter via camera and “bad trials” in which the child was not attending to the task or moving excessively were coded for exclusion from the analyses.
The next two tasks, “No-Beep” and “Beep,” were forced-choice and stop-signal tasks, respectively. These were completed in a quiet testing room, on a laptop using the EPrime 2.0 software (Schneider, Eschman, & Zuccolotto, Reference Schneider, Eschman and Zuccolotto2002). The No-Beep task consisted of 64 trials in which a continuous stream of randomly ordered X’s and O’s was presented one at a time, each followed by a fixation dot, with a stimulus duration of 500 ms and interstimulus interval varying from 1030 to 1050 ms. Participants used a keyboard to press one key for X and another for O. Errors were defined as responses in which the child pressed the incorrect key; that is, choice errors. The Beep task was identical, except that an auditory 1000-hz tone lasting 250 ms (the “stop signal”) was randomly presented prior to the onset of the X or O on 25% of trials. Children were instructed to inhibit their response when the stop signal was presented. The duration of time between the stop signal and visual stimulus began at 250 ms and subsequently decreased or increased by 50 ms depending on whether the previous response was a correct inhibition or incorrect commission, respectively (Logan, Cowan, & Davis, Reference Logan, Cowan and Davis1984). In the Beep task, errors were defined as trials with a stop signal on which the participant pressed the button when they should not have; that is, inhibitory control errors. The Beep task was preceded by a practice set of 10 trials; there were no practice trials for the No-Beep task.
The Hard ERP task was characterized as more difficult than the Easy ERP task, based on greater cognitive and working memory load associated with the former. Error types on the Beep and No-Beep tasks differed in that No-Beep task errors were choice errors, while Beep task errors were inhibitory control errors.
Post-error slowing
Post-error slowing (PES) was calculated separately for each of the four tasks as the mean reaction time (RT) for responses that followed correct trials and preceded error trials, subtracted from the mean RT for responses that immediately followed an error trial (RTpost-error − RTpre-error). Following previous research, this maximized the proximity of post-correct and post-error trials and thus reduced bias introduced by RT variability (Dutilh, van Ravenzwaaij, et al., Reference Dutilh, van Ravenzwaaij, Nieuwenhuis, van der Maas, Forstmann and Wagenmakers2012). Positive PES values indicate that the participant slowed their RT, on average, following errors.
Post-error accuracy increase
Post-error accuracy increase (PAI) was calculated for each task as the difference in mean accuracy on trials following correct and preceding error trials, versus following error trials (ACCpost-error − ACCpre-error; range = 0–1). Higher PAI values indicated that the participant increased their accuracy, on average, following an error.
Post-Novel Slowing
Post-Novel Slowing (PNS) was derived from the Easy and Hard ERP tasks and calculated as the difference between mean RT on trials that followed a novel irrelevant stimulus and correct response trial, versus RT on trials that preceded a novel irrelevant stimulus, and followed both a standard irrelevant stimulus and a correct response trial (RTpost-novel − RTpre-novel). Thus, slower RT following as compared to immediately prior to a novel probe was indicated by a positive PNS value.
ADHD symptoms
ADHD symptom severity was measured with parent report on the Strengths and Weakness of ADHD Symptoms and Normative Behaviors (SWAN) questionnaire (Swanson et al., Reference Swanson, Schuck, Porter, Carlson, Hartman, Sergeant and Lakes2012), which uses a balanced 7-point Likert scale to measure the full spectrum of abilities on 18 ADHD symptoms from the DSM-5 (APA, 2013). Caregivers were instructed to rate their child’s behaviors (when off medications, if applicable) relative to same-age peers. Unlike a traditional DSM-5 ADHD checklist, SWAN scores show variance at both the adaptive and symptomatic tails of the behavior distribution (Arnett et al., Reference Arnett, Pennington, Friend, Willcutt, Byrne, Samuelsson and Olson2013). To calculate continuous ADHD symptom severity, ratings were coded numerically from 3 (far below) to −3 (far above) and averaged for the inattentive (items 1–9), hyperactive/impulsive (items 10–18), and total (items 1–18) symptom domains. Higher scores indicated more severe ADHD symptoms.
Analytic plan
Data preparation and analyses were conducted in R Studio. Data were inspected for normality and outliers (defined as > ±3 standard deviations) were removed (1 – 2 data points on two variables). Linear mixed-effects models with restricted maximum likelihood were estimated using R packages lme4 and lmerTest, with a random intercept specified for individual to account for repeated measures. Models were estimated separately for ERP tasks and Beep/No-Beep tasks. Baseline main effects models were first specified with PES as the dependent variable, and the following predictors: (1) number of error trials, (2) task, (3) ADHD diagnosis, and (4) total ADHD severity. To test each of our three hypotheses, we added relevant interactions and additional independent variables. For each model, reported p-values have been adjusted to account for false-discovery rates associated with multiple comparisons (Benjamini & Yekutieli, Reference Benjamini and Yekutieli2001).
Results
Preliminary analyses
T-tests were used to identify group differences on demographic and experimental variables. Compared to controls, the ADHD group had more post-error trials on the Easy and Hard ERP and No-Beep tasks, lower accuracy on the Hard ERP and No-Beep tasks, higher ADHD symptom severity, shorter PES on the Hard ERP task, and lower full-scale IQ (p’s < .04; see Table 2). Compared to females, males had lower accuracy on the Easy ERP task (t[65] = 2.38, p = .020) and fewer post-novel trials on the Easy ERP task after processing (t[90] = 2.25, p = .027). No other sex differences on demographic or experimental variables were statistically significant at p < .05. Pearson correlations between age, IQ and primary study variables (PES, PNS, PAI, task accuracy, number of post-error trials) revealed a linear association between age and Easy ERP PES (r = -.24, p = .017, adjusted p = .145) and age and Beep task accuracy (r = .27, p = .008, adjusted p = .097). Participant IQ was correlated with the number of Hard ERP post-error trials (r = -.22, p = .029, adjusted p = .177), No-Beep post-error trials (r = −.25, p = .031, adjusted p = .177), No-Beep accuracy (r = .15, p = .003, adjusted p = .095), and total ADHD severity (r = −.26, p = .009, adjusted p = .097). None of these correlations remained statistically significant at p < .05 after adjustment for false-discovery rate. There were no significant correlations between the remaining primary study variables and age or IQ (p’s > .210; adjusted p’s > .545).
Baseline linear models
Results of all linear models are reported in Table 3. First, we tested for main effects of ADHD diagnosis, ADHD symptom severity, number of error trials and task on PES, separately within ERP and No-Beep/Beep experiment pairs. Across ERP tasks, shorter PES was associated with ADHD diagnosis (B = −95.44, SE = 29.79, p = .0064) and more error trials (B = −1.93, SE = 0.71, p = .0098). ADHD symptom severity showed an unexpected effect in that greater symptom severity was associated with longer PES (B = 34.69, SE = 12.15, p = .0095) once diagnosis was controlled. There was no main effect of task on PES (B = −6.63, SE = 16.64, p = .6908). Across Beep/No-Beep tasks, there were no statistically significant main effects of task, error trials, ADHD diagnosis or ADHD severity (p’s > .7618).
Table 3. Linear models
PES = post-error slowing; PAI = post-error accuracy increase; PNS = post-novelty slowing. Main and interaction effects significant at p < .05 after FDR adjustment are italicized.
Hypothesis 1: error recognition
To test the hypothesis that reduced error recognition explains PES differences in ADHD, we added interaction terms between ADHD diagnosis and task, and between ADHD severity and task. With these interactions added, the main effects of the ERP model remained consistent. A moderating effect of ERP task was found wherein there was a greater difference in PES between controls and ADHD participants during the Hard as compared to the Easy task, (B = −129.34, SE = 59.07, p = .0447; Figure 1). The interaction between ADHD severity and task was not significant (B = −34.30, SE = 23.97, p = .1540) (see Figure 1).
Fig. 1. Reduced PES among ADHD participants as compared to controls was significantly more pronounced during the Hard ERP task.
Next, we repeated these analyses with the Beep and No-Beep tasks. The main effect of task (i.e., error type) emerged, wherein PES was shorter during the No-Beep task, but this association did not remain statistically significant after false-discovery rate correction (B = −129.36, SE = 59.11, p = .0902). Likewise, an interaction between task and ADHD diagnosis – suggesting greater effect of task on individuals with ADHD – approached significance after correction (B = 230.62, SE = 92.03, p = .0792). There was no interaction between task and ADHD severity (p = .2623).
Hypothesis 2: adaptive control
To test whether variance in PES is associated with PAI, we added PAI as an independent variable in the main effects model, as well as interactions between PAI and ADHD diagnosis; between PAI and ADHD severity; and PAI and task. Using the ERP tasks, main effects reported in the baseline model remained consistent. However, PAI was not associated with PES; nor were any of the three interaction terms (p’s = .550). Using the Beep/No-Beep tasks, no main or interaction effects were statistically significant in this model (p’s ≥ .665).
Hypothesis 3: attention orienting
To evaluate the hypothesis that attention orienting explained PES differences in ADHD, we tested two separate models. First, we evaluated a model in which ADHD diagnosis and symptom severity interacted with the number of error trials. With the error trials interactions included in the model, main effects of ADHD diagnosis, symptom severity and error trials no longer survived FDR adjustment for the ERP tasks (p’s > 1.54) nor the No-Beep/Beep tasks (p’s = .505). Moreover, the error trials interaction terms were not statistically significant for either pair of experiments (ERP task p’s = .888; No-Beep/Beep task p’s = .505).
Next, we tested a model that included PNS as a predictor, as well as interactions between PNS and ADHD diagnosis; PNS and ADHD severity; and PNS and task. Only ERP data were used for these analyses as there were no novel stimuli during the No-Beep/Beep tasks. Main effects remained consistent with those reported in the baseline model. There were no significant main or interaction effects of PNS (p’s > .423).
Discussion
Summary
In the current study, we examined competing hypotheses about the cognitive origin of PES differences in children with ADHD. Our results are consistent with the first hypothesis, that PES differences in ADHD are driven by reduced error recognition. ADHD diagnosis was associated with shorter PES in the Hard ERP task, and to an extent, the No-Beep task, tasks that had high cognitive load and choice, as opposed to inhibitory, errors. We interpret this finding as indicating that PES differences in ADHD are more evident during tasks where errors are difficult to detect.
Error recognition may be divided into subconscious versus conscious processes. The former is conceptualized as an automatic, bottom-up process mediated by the anterior cingulate (Hester, Foxe, Molholm, Shpaner, & Garavan, Reference Hester, Foxe, Molholm, Shpaner and Garavan2005) and does not necessarily correlate with post-error behavioral change. In contrast, conscious error awareness, in which the individual is able to communicate the recognition of their error, has been linked to activation in the bilateral insular cortex, pre-supplementary motor area, and prefrontal-parietal circuitries (Hester et al., Reference Hester, Foxe, Molholm, Shpaner and Garavan2005; Klein et al., Reference Klein, Endrass, Kathmann, Neumann, von Cramon and Ullsperger2007). Similarly, electrophysiological research indicates a temporal sequence in which automatic error awareness, measured via the error negativity component around 50 ms after an incorrect response, is followed by conscious awareness of the error, reflected in an error positivity component around 300 ms (Ullsperger, Fischer, Nigbur, & Endrass, Reference Ullsperger, Fischer, Nigbur and Endrass2014). These electrophysiological indices may not be dependent on one another, underscoring the notion that conscious and unconscious error recognition constitute distinct neural processes (Di Gregorio, Maier, & Steinhauser, Reference Di Gregorio, Maier and Steinhauser2018; Endrass, Reuter, & Kathmann, Reference Endrass, Reuter and Kathmann2007). In the current study, because our measures were strictly behavioral, the differences seen among children with elevated ADHD symptoms were presumably due to conscious error awareness.
PAI was not correlated with PES, consistent with previous research (Danielmeier & Ullsperger, Reference Danielmeier and Ullsperger2011; van Meel et al., Reference van Meel, Heslenfeld, Oosterlaan and Sergeant2007). This challenges conclusions of prior studies that have interpreted reduced PES in ADHD as an indication of impaired behavioral self-regulation (Shiels & Hawk Jr, Reference Shiels and Hawk2010), to the extent that it can be measured by PAI. On the other hand, we did find that ADHD symptom severity was associated with increased PES but comparable PAI, once diagnosis was controlled. This implies a response time–accuracy trade-off associated with greater ADHD symptoms, such that high-severity ADHD children were slowing their post-error response rate to maximize performance. This could have implications for behavioral interventions for this clinical population. Specifically, it may be useful to focus behavioral interventions on increasing error awareness, rather than behavior modulation following errors. This would be consistent with evidence-based behavioral therapies, which aim to heighten awareness of errors and as well as correct behaviors through positive reinforcement (Kazdin, Reference Kazdin1997). Of course, the simplicity of our task limits the degree to which inferences can be made about the role of error recognition and behavioral adaptation in functional outcomes among children in real-world settings. We encourage future research to utilize more complex tasks, with multiple response modalities (e.g., motoric and verbal) as well as variable levels of subtlety in error cueing, to further evaluate adaptive control deficits in this population.
While error recognition and adaptive control are arguably performance-enhancing responses to error, attention orienting may be considered impairing, despite its similar effect on behavior (i.e., slowed RT). Neuroimaging studies have proposed novelty facilitates performance via frequent activation of dopaminergic reward pathways (Volkow et al., Reference Volkow, Wang, Kollins, Wigal, Newcorn, Telang and Ma2009). This theory is consistent with the cognitive-energetic model of ADHD (Sergeant, Reference Sergeant2000), which emphasizes reduced engagement and attenuated reward sensitivity as a neurocognitive etiology of clinical symptoms. We specifically tested the hypothesis that attention orienting to infrequent events explains PES variance, as suggested by Notebaert and colleagues (Notebaert et al., Reference Notebaert, Houtman, Van Opstal, Gevers, Fias and Verguts2009; Núňez Castellar et al., Reference Núňez Castellar, Kühn, Fias and Notebaert2010). The current study is unique in that we used tasks in which novel irrelevant stimuli were embedded, which allowed for a clear distinction between PES and PNS. As reported in prior literature, more frequent error trials were associated with reduced PES during the ERP tasks, suggesting that error-related novelty may contribute to PES differences. However, our study was unique in that we further examined the effect of novelty using the novel irrelevant stimuli presented prior to target stimuli during the ERP tasks. Contrary to the attention orienting theory of ADHD, we did not find a linear association between PES and PNS in these experiments, nor was there an interaction with ADHD variables. Thus, the effect of the number of error trials on PES is more likely explained by an association between higher PES and overall performance on the task, which may be a proxy for intelligence and which was not moderated by ADHD diagnosis.
Unlike prior studies, we included both ADHD diagnosis and symptom severity in our models to test for the possibility that ADHD symptom severity, rather than diagnosis, explained PES (Mohamed et al., Reference Mohamed, Börger, Geuze and van der Meere2016), and that methodological differences in ADHD characterization might account for conflicting effect sizes reported across studies in the extant literature. Surprisingly, not only did ADHD diagnosis and symptom severity explain independent variance in PES, the directionality of these associations was opposing. While participants with ADHD did indeed show reduced PES in the Hard ERP and No-Beep tasks, ADHD symptom severity was associated with longer PES. This suggests that the neurocognitive factors underlying PES in ADHD are not necessarily the same factors that influence parent ratings of ADHD symptom severity. This may be similar to the weak correlations reported between behavioral ratings and neuropsychological test performance in children with ADHD (Jonsdottir, Bouma, Sergeant, & Scherder, Reference Jonsdottir, Bouma, Sergeant and Scherder2006; Toplak, Bucciarelli, Jain, & Tannock, Reference Toplak, Bucciarelli, Jain and Tannock2008). A future direction should include including alternative tests of symptom severity, such as teacher or clinician ratings, or objective measures of inhibitory control or sustained attention.
A limitation of our study was that we lacked power to simultaneously test the effects of coexisting psychiatric diagnoses and symptoms on PES. ERP research suggests that anxiety increases error awareness while depression decreases error monitoring (Bress, Meyer, & Hajcak, Reference Bress, Meyer and Hajcak2015; Moser, Moran, Schroder, Donnellan, & Yeung, Reference Moser, Moran, Schroder, Donnellan and Yeung2013). In future work, we plan to examine whether coexisting psychopathology in our sample moderates the outcomes we report in this study. Another limitation is the low number of participants who had a sufficient number of errors during the Beep/No-Beep tasks to be included in analyses. Given that the direction of effects was similar across ERP and Beep/No-Beep models, the lack of statistically significant findings in the latter might be explained by reduced data points. Relatedly, by excluding individuals with extremely high or low accuracy on our tasks, we may have curtailed some meaningful variance, particularly in the ADHD sample. Future work could take advantage of adaptive task designs (similar to the approach used in our Beep task) to better control for individual differences in task performance. Finally, the substantial heterogeneity in ADHD suggests that individual differences in the etiology of PES may be more informative than group-level results. For example, response control among females with ADHD may only differ from same-sex controls on tasks with high cognitive load, while males with ADHD show impairment regardless of task difficulty (Seymour, Mostofsky, & Rosch, Reference Seymour, Mostofsky and Rosch2016). Although it was beyond the scope of the current study, investigation of demographic and cognitive moderators of error recognition and adaptive control will be critical to the translational impact of this work, specifically with respect to developing precision medicine care for children and families affected by ADHD.
Conclusions
Differences in PES among children with ADHD are driven by reduced error recognition during tasks with high cognitive demand. Our study did not find evidence to support reduced adaptive control or increased novelty orienting among children with ADHD during forced-choice, 1-back, or stop-signal tasks. The clinical implications of this study are that facilitation of error recognition may improve behavioral performance among children with ADHD.
Financial Support
This work was funded by grants from the National Institute of Mental Health (A.B.A., 5K99MH116064) and the Klingenstein Third Generation Foundation (A.B.A., 2020 ADHD Fellowship).
Conflicts of Interest
The authors have no conflicts of interest to declare.
Ethical Standards
Caregivers completed written informed consent and participating children completed written assent at the start of the in-person visit. All procedures were completed in accordance with the Helsinki Declaration and in compliance with the university institutional review board.
