Introduction
Schizophrenia and bipolar disorder are severe and complex neurodevelopmental disorders (Murray & Lewis, Reference Murray and Lewis1987; Weinberger, Reference Weinberger1987) and among the most costly and debilitating disorders for the affected individuals, their relatives, and broader society. As early signs of schizophrenia and bipolar disorder are rare in the general population, studies of enriched populations, such as children with familial high risk of these disorders, provide insight into disease processes and disease development over time. Children with a familial high risk of schizophrenia (FHR-SZ) show several impaired cognitive functions, such as lower intelligence (Agnew-Blais & Seidman, Reference Agnew-Blais and Seidman2013; Hemager et al., Reference Hemager, Plessen, Thorup, Christiani, Ellersgaard, Spang and Jepsen2018; Sugranyes et al., Reference Sugranyes, De La Serna, Borras, Sanchez-Gistau, Pariente, Romero and Castro-Fornieles2017), poorer working memory (de la Serna et al., Reference de la Serna, Sugranyes, Sanchez-Gistau, Rodriguez-Toscano, Baeza, Vila and Castro-Fornieles2017; Hemager et al., Reference Hemager, Plessen, Thorup, Christiani, Ellersgaard, Spang and Jepsen2018), attention deficits (Agnew-Blais & Seidman, Reference Agnew-Blais and Seidman2013; Burton et al., Reference Burton, Vangkilde, Petersen, Skovgaard, Jepsen, Hemager and Plessen2018; Cornblatt, Obuchowski, Roberts, Pollack, & Erlenmeyer-Kimling, Reference Cornblatt, Obuchowski, Roberts, Pollack and Erlenmeyer-Kimling1999; Hemager et al., Reference Hemager, Plessen, Thorup, Christiani, Ellersgaard, Spang and Jepsen2018; Hemager et al., Reference Hemager, Vangkilde, Thorup, Christiani, Ellersgaard, Spang and Plessen2019) and subtle deficits in interference control (Burton et al., Reference Burton, Vangkilde, Petersen, Skovgaard, Jepsen, Hemager and Plessen2018). Studies among children with a familial high risk of bipolar disorder (FHR-BP), on the contrary, show inconsistent results in relation to intelligence (Bora & Özerdem, Reference Bora and Özerdem2017; de la Serna et al., Reference de la Serna, Sugranyes, Sanchez-Gistau, Rodriguez-Toscano, Baeza, Vila and Castro-Fornieles2017; Hemager et al., Reference Hemager, Plessen, Thorup, Christiani, Ellersgaard, Spang and Jepsen2018; Sugranyes et al., Reference Sugranyes, De La Serna, Borras, Sanchez-Gistau, Pariente, Romero and Castro-Fornieles2017) attention, visual and verbal memory, and processing speed, except for consistent deficits in working memory (Bora & Özerdem, Reference Bora and Özerdem2017; Hemager et al., Reference Hemager, Plessen, Thorup, Christiani, Ellersgaard, Spang and Jepsen2018; Hemager et al., Reference Hemager, Vangkilde, Thorup, Christiani, Ellersgaard, Spang and Plessen2019) and deficits in cognitive flexibility (Burton et al., Reference Burton, Vangkilde, Petersen, Skovgaard, Jepsen, Hemager and Plessen2018; Patino et al., Reference Patino, Adler, Mills, Strakowski, Fleck, Welge and Delbello2013). Children with a FHR-SZ also exhibit motor impairments (Burton et al., Reference Burton, Hjorthoj, Jepsen, Thorup, Nordentoft and Plessen2016) and a high prevalence of attention-deficit/hyperactivity disorder (ADHD) (Ellersgaard et al., Reference Ellersgaard, Jessica Plessen, Richardt Jepsen, Soeborg Spang, Hemager, Klee Burton and Elgaard Thorup2018), whereas this was not evident for children with FHR-BP (Burton et al., Reference Burton, Thorup, Jepsen, Poulsen, Ellersgaard, Spang and Plessen2017; Duffy, Reference Duffy2012; Ellersgaard et al., Reference Ellersgaard, Jessica Plessen, Richardt Jepsen, Soeborg Spang, Hemager, Klee Burton and Elgaard Thorup2018).
A central part of children's development, including extending their cognitive, motor, and emotional capacities, entails learning from their mistakes by adapting their behavior to prevent future mistakes. Behavioral adaptation to situational demands is fundamental to daily functioning across the life span (Diamond, Reference Diamond2013; Ullsperger, Reference Ullsperger2006; Ullsperger, Danielmeier, & Jocham, Reference Ullsperger, Danielmeier and Jocham2014) and reflects a central part of our cognitive control processes. Specifically, error adaptation is the ability to adapt behavior after an erroneous response by slowing the response speed in a subsequent trial in order to improve accuracy and avoid additional errors (Rabbitt, Reference Rabbitt1966; Ullsperger, Harsay, Wessel, & Ridderinkhof, Reference Ullsperger, Harsay, Wessel and Ridderinkhof2010). This change in response speed is called post-error slowing (PES). PES may reflect the ability to assert cognitive control and is viewed as a marker of adaptation (Rabbitt, Reference Rabbitt1968). This adaptation process may lead to improved accuracy in post-error trials compared with post-correct trials. The improved accuracy as a consequence of post-error adjustments is called post-error improvement of accuracy (PIA) (Danielmeier & Ullsperger, Reference Danielmeier and Ullsperger2011; Marco-Pallares, Camara, Munte, & Rodriguez-Fornells, Reference Marco-Pallares, Camara, Munte and Rodriguez-Fornells2008).
Cognitive control develops relatively slowly and is not fully mature until early adulthood (Cragg, Reference Cragg2016; Tamnes, Walhovd, Torstveit, Sells, & Fjell, Reference Tamnes, Walhovd, Torstveit, Sells and Fjell2013). This development depends on the maturation of various structures, including the anterior cingulate cortex (ACC) and the dorsolateral prefrontal cortex (Niendam et al., Reference Niendam, Laird, Ray, Dean, Glahn and Carter2012; Tamnes et al., Reference Tamnes, Walhovd, Torstveit, Sells and Fjell2013). Electrophysiologically, erroneous responses elicit error-related negativity (ERN), which is a neurocognitive marker that reflects ACC activity and cognitive control processes (Taylor, Stern, & Gehring, Reference Taylor, Stern and Gehring2007; van Veen & Carter, Reference van Veen and Carter2006). These processes are believed to lead to compensatory behavioral changes, such as PES (Ladouceur, Dahl, & Carter, Reference Ladouceur, Dahl and Carter2007). The literature indicates that the ERN amplitude becomes larger with age from childhood to adulthood, suggesting continued maturation of the neural system for cognitive control (Ladouceur et al., Reference Ladouceur, Dahl and Carter2007; Tamnes et al., Reference Tamnes, Walhovd, Torstveit, Sells and Fjell2013; van Meel, Heslenfeld, Rommelse, Oosterlaan, & Sergeant, Reference van Meel, Heslenfeld, Rommelse, Oosterlaan and Sergeant2012).
Brain maturation follows slightly different trajectories in girls and boys, and is influenced by chromosomes, gonadal steroid hormones, and cultural and environmental factors (Giedd, Castellanos, Rajapakse, Vaituzis, & Rapoport, Reference Giedd, Castellanos, Rajapakse, Vaituzis and Rapoport1997; Gogos, Ney, Seymour, Van Rheenen, & Felmingham, Reference Gogos, Ney, Seymour, Van Rheenen and Felmingham2019; McCarthy, Nugent, & Lenz, Reference McCarthy, Nugent and Lenz2017; Ruigrok et al., Reference Ruigrok, Salimi-Khorshidi, Lai, Baron-Cohen, Lombardo, Tait and Suckling2014). Age-related sex differences during development have also been documented in relation to the ERN amplitude from childhood to adulthood, with girls having a distinct development pattern from boys (including an earlier increase in ERN amplitude than boys) (Davies, Segalowitz, & Gavin, Reference Davies, Segalowitz and Gavin2004). Furthermore, among adults, males showed increased ERN amplitude relative to females (Larson, South, & Clayson, Reference Larson, South and Clayson2011) and females showed more pronounced PES at the behavioral level than males (Fischer, Danielmeier, Villringer, Klein, & Ullsperger, Reference Fischer, Danielmeier, Villringer, Klein and Ullsperger2016). To our knowledge, this issue has not been examined in a large sample of children before puberty.
Cognitive control is crucial for educational achievement, social and psychological development, and mental and physical health (Diamond, Reference Diamond2013). Individuals with schizophrenia and their unaffected first-degree relatives display deficits of cognitive control and altered ACC and dorsolateral prefrontal cortex connectivity during response inhibition tasks (Sambataro et al., Reference Sambataro, Mattay, Thurin, Safrin, Rasetti, Blasi and Weinberger2013). Children of parents with either schizophrenia or bipolar disorder have an increased risk for developing psychiatric disorders in adulthood (Rasic, Hajek, Alda, & Uher, Reference Rasic, Hajek, Alda and Uher2014) and associated cognitive problems. Therefore, investigating cognitive control is important to establish and identify both cognitive with-in person resilience factors or vulnerabilities, which may contribute to understand cognitive functioning in at-risk individuals. To our knowledge, no studies have assessed PES or PIA in children with FHR-SZ or FHR-BP.
The overarching purpose of this study was to assess behavioral markers of error adaptation in a large sample of children with a FHR-SZ or FHR-BP within a narrow age range. We hypothesized that children with FHR-SZ or FHR-BP would make more errors and exhibit less PES and PIA compared with controls during a task requiring cognitive control. We also hypothesized that boys would exhibit less error adaptation than girls, irrespective of familial high-risk status.
Method
The Danish Data Protection Agency approved the study protocol and we received permission to draw data from registers from the Danish Ministry of Health. The Danish National Committee on Health Research Ethics evaluated the protocol but, because of the absence of any intervention, ethical approval was not regarded necessary by the authority. Participating parents and legal guardians gave written informed consent.
Study design and participants
The Danish High Risk and Resilience Study – VIA 7 is a prospective and population-based cohort established in Denmark between January 1, 2013 and January 31, 2016 (Thorup et al., Reference Thorup, Jepsen, Ellersgaard, Burton, Christiani, Hemager and Nordentoft2015) and consists of 522 Danish children aged 7 years with either no, one, or both parents diagnosed with schizophrenia spectrum psychosis or bipolar disorder. The cohort was constructed using the Danish Civil Registration System (Pedersen, Gotzsche, Moller, & Mortensen, Reference Pedersen, Gotzsche, Moller and Mortensen2006) to identify children born and living in Denmark who turned 7 years during the study period in combination with the Danish Psychiatric Central Research Register (Mors, Perto, & Mortensen, Reference Mors, Perto and Mortensen2011). This register, which consists of diagnoses of inpatient and outpatient contact, allowed identification of biological parents diagnosed with either schizophrenia spectrum psychosis defined as schizophrenia, delusional disorder, or schizoaffective disorder (ICD 10-codes F20, F22, F25 or ICD 8-codes 295, 297, 298.29, 298.39, 298.89, 298.99) or bipolar disorder (defined as ICD 10-codes F30, F31 or ICD 8-codes 296.19, 296.39). Control children (population-based children of parents with no diagnoses of schizophrenia spectrum disorder or bipolar disorder) were matched to children with FHR-SZ based on sex, age, and municipality. Children with FHR-BP were included as a nonmatched group.
Procedure
All the raters were blinded to group affiliation throughout the examination and analyses. We assessed error adaptation using a modified Eriksen flanker task. Furthermore, we used the total standard scores from the Movement Assessment Battery for Children – second edition (ABC-2) (Henderson, Sugden, & Barnett, Reference Henderson, Sugden and Barnett2007) as an estimate of motor function (published in detail elsewhere; see Burton et al., Reference Burton, Thorup, Jepsen, Poulsen, Ellersgaard, Spang and Plessen2017) to assess the impact of motor function in error adaptation. Moreover, we used a modified version of the ADHD-Rating Scale (Dupaul, Power, & Anastopoulos, Reference Dupaul, Power and Anastopoulos1998; Makransky & Bilenberg, Reference Makransky and Bilenberg2014) to assess symptoms of ADHD rated by primary caregivers (described in detail elsewhere; see Ellersgaard et al., Reference Ellersgaard, Jessica Plessen, Richardt Jepsen, Soeborg Spang, Hemager, Klee Burton and Elgaard Thorup2018) and investigated their relation to error adaptation. The current level of functioning of the child was evaluated using the Children's Global Assessment Scale (CGAS) (Shaffer et al., Reference Shaffer, Gould, Brasic, Ambrosini, Fisher, Bird and Aluwahlia1983).
Eriksen flanker task
The Eriksen flanker task (Eriksen & Eriksen, Reference Eriksen and Eriksen1974) is a visuospatial task involving an interference effect, which we used in a modified version implemented in E-prime 2. A short training session preceded the actual flanker task. The child was instructed to fixate on a dot presented in the center of a computer screen for 800 ms. Trials started with six horizontal flanker arrows appearing below the fixation. After 100 ms, a central target arrow appeared pointing either in the same direction as the flanker arrows in congruent trials (⋘<⋘, ⋙>⋙) or in the opposite direction in incongruent trials (⋘>⋘, ⋙<⋙) (Figure 1). The child was instructed to respond to the direction of the central arrow as quickly and accurately as possible with either a left or a right mouse button click. The target arrows and flanker arrows remained on the screen until a response was registered. Trials were terminated by the response and were immediately followed by 800-ms fixation screen of the next trial (the response–stimulus interval, RSI), leading to a response–target interval (RTI) of 800 ms + 100 ms = 900 ms. The intertrial interval (ITI) is the time from the end of one trial to the beginning of the next trial (Compton, Heaton, & Ozer, Reference Compton, Heaton and Ozer2017). Since trials in this flanker task finished with the response, the RSI was equal to ITI = 800 ms.
Figure 1. The Eriksen flanker task design and trial sequence. The child was instructed to fixate on a dot presented in the center of a computer screen for 800 ms. Trials started with six horizontal flanker arrows. After 100 ms, a central target arrow appeared either as congruent trials or as incongruent trials. The target arrows and flanker arrows remained on the screen until a response was registered. Trials were terminated by the response and were immediately followed by 800-ms fixation screen of the next trial. The response–stimulus interval (RSI) was equal to the intertrial interval (ITI) since trials in this flanker task design finished with the response.
The flanker task consisted of 400 trials divided into two blocks of 200 trials separated by a short break. The overall probability of congruent and incongruent trials and left and right responses, was held at 0.5, respectively. The occurrence of each trial type was pseudo-randomized separately for each child. An exclamation mark represented performance feedback and appeared when responses were erroneous or slower than an adaptive individual threshold value (i.e., the continuously updated mean response time plus 1.5 standard deviations (SDs)) (Burton et al., Reference Burton, Vangkilde, Petersen, Skovgaard, Jepsen, Hemager and Plessen2018; Eichele et al., Reference Eichele, Eichele, Marquardt, Adolfsdottir, Hugdahl, Sorensen and Plessen2017). After a correct response, a fixation dot was shown on the screen until the beginning of the next trial. The RSI after correct and noncorrect responses was the same.
Outcome measures for the flanker task
PES was assessed from consecutive sequences of four trials (a CCEC trial sequence) consisting of an error trial (E), the two correct trials directly preceding the error trial (CE−1 and CE−2, respectively), and a correct post-error trial (CE+1) in order to avoid contamination by global fluctuations in performance during the experiment, such as time-on-task changes in motivation or response caution (Dutilh et al., Reference Dutilh, Ravenzwaaij, Nieuwenhuis, Van Der Mass, Forstmann and Wagenmarkers2012). Correspondingly, we calculated PES robust = mean reaction time (RT) post-error (CE+1) − mean RT pre-error, post-correct trials (CE−1), as suggested in a previous study (Dutilh et al., Reference Dutilh, Ravenzwaaij, Nieuwenhuis, Van Der Mass, Forstmann and Wagenmarkers2012). A minimum of five CCEC sequences per participant was required to calculate a meaningful PES (n = 3 of the controls were excluded). The PES for each subject was used in the overall mean PES of all the participants from the FHR-SZ, FHR-BP, and control groups. Outliers were removed before the assessment of PES. Trials with reaction times less than 200 ms and greater than 10,000 ms were removed. Thereafter, trials were discarded with a reaction time exceeding 3SDs from the mean. The reaction times are reported elsewhere (Burton et al., Reference Burton, Vangkilde, Petersen, Skovgaard, Jepsen, Hemager and Plessen2018).
Post-error accuracy was calculated as the average accuracy on all post-error trials (except trials 1 and 201 where no preceding trial was available). PIA was calculated by subtracting post-correct accuracy from post-error accuracy, which was expected to be positive if reflecting cognitive adaptation and adjusted behavior. Error rates were calculated as 1 − accuracy for both congruent trials and incongruent trials. Calculation of PES and PIA ignores whether errors occurred on congruent or incongruent trials. Therefore, PES and PIA were also assessed in follow-up analyses, where PES and PIA were calculated solely based on trials following incongruent trials. We refer to slowing after an error in an incongruent trial as PES after incongruent trials (PESincongruent) and the improvement in accuracy as PIA after incongruent trials (PIAincongruent). A minimum of five CCEC sequences per participant were required to calculate a meaningful PESincongruent, which resulted in n = 20 exclusions (n = 7 from the controls, n = 7 from the FHR-BP group, and n = 6 from the FHR-SZ group).
Statistical analysis
For the a priori hypothesis, the error adaptation measures (PES, PIA, PESincongruent, and PIAincongruent) were tested by assessing the effect of group in a mixed-model analysis with a random effect of matched set (including singleton cases). The model was adjusted for age and sex. In addition to these independent variables, we considered all three-way and two-way interactions of group, sex, and age. Statistically nonsignificant interaction terms were eliminated via backward stepwise regression, with the constraint that the model at each step had to be hierarchically well-formulated. All lower-order terms such as sex, age, and group were included in the model regardless of statistical significance.
We explored the following intermediate variables in the mixed model: error rate for incongruent trials, ADHD symptoms (rated by the caregiver ADHD-Rating Scale) (Ellersgaard et al., Reference Ellersgaard, Jessica Plessen, Richardt Jepsen, Soeborg Spang, Hemager, Klee Burton and Elgaard Thorup2018), and motor function (measured with the Movement ABC-2, total standard score) (Burton et al., Reference Burton, Thorup, Jepsen, Poulsen, Ellersgaard, Spang and Plessen2017) in which PES and PIA served as the dependent variables.
Variables with a skewed distribution were logarithmically transformed (log 2) before analysis and the results were backward transformed with antilogarithm. For the outcome variable error rate, we used the same mixed model and expanded the model with an unstructured covariance matrix, describing variance and correlation between the two outcomes for each child (congruent and incongruent). Covariates included age, sex, and condition (congruent or incongruent). We explored the intermediate variables ADHD symptoms and motor function in the repeated mixed model, in which error rate served as the dependent variable.
The relationship between PIA and PES (Danielmeier & Ullsperger, Reference Danielmeier and Ullsperger2011) was assessed by a Pearson correlation as was the relationship between PES and CGAS, measuring the child's current level of functioning. We considered p values <.05 as significant. All statistical analyses were conducted in SAS software (version 9.4).
Results
Our cohort consisted of 522 children aged 7 years. A total of 500 children completed the flanker task, of which 497 children had a minimum of five consecutive sequences of trials around errors, required to calculate a meaningful PES.
We assessed 192 (38.6%) children of parents with schizophrenia, 116 children (23.3%) of parents with bipolar disorder, and 189 (38.0%) children with parents without schizophrenia or bipolar disorder (Table 1). At group level, children exhibited PES and PESincongruent irrespective of familial high-risk groups (Table 2). The test for fixed effects of PES revealed a significant effect of sex (F (1, 492) = 5.17, p = .024), but no differences across groups (F (2, 492) = 0.30, p = .743) or across age (F (1, 492) = 1.12, p = .291). No interactions were significant (Table 2). The post-hoc test showed that boys displayed less prominent PES compared with girls (mean difference −56.4, 95% confidence interval (CI) [−105.2, −7.64], p = .024) (Figure 2). In the mixed model, a high error rate for incongruent trials led to a significantly smaller PES (estimate −65.49, 95% CI [−86.48, −44.50], p < .0001). The influence of ADHD symptoms was not significant (estimate −1.59, 95% CI [−4.40, 1.23], p = .27), nor was the effect of motor function (estimate 6.00, 95% CI [−1.76, 13.76], p = .13) on PES (Table 3). There was no relevant correlation between the child's current level of functioning (as evaluated using the CGAS) and PES in the overall sample (Pearson correlation r = −.03, p = .51), nor for the different high-risk groups (FHR-SZ: r = −.03, p = .66; FHR-BP: r = −.026, p = .78; controls: r = −.064, p = .39).
Figure 2. Results of the Eriksen flanker task: (a) means of post-error slowing after congruent and incongruent trials (PES) and (b) means of post-error slowing after incongruent trials only (PESincongruent) for boys (blue squares) and girls (red circles) between children with a familial high risk of schizophrenia (FHR-SZ), children with a familial high risk of bipolar disorder (FHR-BP), and control subjects. The error bars indicate 95% confidence intervals.
Table 1. Characteristics of children performing the Eriksen flanker task as participants in the Danish High Risk and Resilience Study – VIA 7
Note: FHR-SZ = familial high risk of schizophrenia; FHR-BP = familial high risk of bipolar disorder; Controls = population-based control children of parents with no diagnoses of schizophrenia spectrum disorders or bipolar disorder; NA = not applicable. Age at inclusion did not differ between groups, but because of differences in the time of testing, age for the flanker task differed between groups.
a Reported by the caregiver using the ADHD-Rating Scale (inattention, hyperactive, and impulsive symptoms)
b Total motor standard score of the child tested using the Movement Assessment Battery for Children – second edition (ABC-2)
c Children's Global Assessment Scale, measuring children's current level of functioning
d CBCL: Child Behavior Checklist.
Table 2. Error adaptation among 7-year-old children with either FHR-SZ or FHR-BP and control subjects measured by the Eriksen flanker task
Note: FHR-SZ = familial high risk of schizophrenia; FHR-BP = familial high risk of bipolar; Controls = population-based control children of parents with no diagnoses of schizophrenia spectrum disorders or bipolar disorder; PES = post-error slowing; PESincongruent = post-error slowing after incongruent trials; PIA = post-error improvement in accuracy; PIAincongruent = post-error improvement in accuracy after incongruent trials; RT =Reaction time.
a We tested PES, PIA, PESincongruent, and PIAincongruent by assessing the effect of group in a mixed-model analysis with a random effect of matched set (including singleton cases). The model was adjusted for age and sex
b We tested error rate, RT post-correct, and RT post-error by expanding the mixed model with an unstructured covariance matrix, describing variance and correlation between the two outcomes for each child (congruent and incongruent). Covariates included age, sex, and congruency.
c Post-hoc analysis showed that FHR-SZ had 8.22% slower post-correct RT compared with controls (p = .0002, 95% CI [1.086, 1.1276]), whereas no difference was detected between FHR-BP versus controls (p = .85)
d PESincongruent refers to PES calculated solely based on trials following incongruent trials
e PIAincongruent refers to PIA calculated solely based on trials following incongruent trials
Table 3. Error adaptation and the influence of predictors of ADHD symptoms, motor ability, or error rate.
Note: FHR-SZ = familial high risk of schizophrenia; FHR-BP = familial high risk of bipolar disorder; Controls = population-based control children of parents with no diagnoses of schizophrenia spectrum disorders or bipolar disorder. PES = post-error slowing; PIA = post-error improvement in accuracy; ADHD = attention-deficit/hyperactivity disorder.
a Estimate
b Estimated ratio
c Girls are reference level
d Incongruent are reference level
Some individual children exhibited PIA, but none of the three groups on average displayed PIA (Table 2). Accordingly, PIA did not differ across groups (F (2, 350) = 0.10, p = .902), with age (F (1, 464) = 0.18, p = .673), or with sex (F (1, 294) = 0.59, p = .445), and none of the two-way or three-way interactions were significant. A better motor function led to a larger PIA (estimate 0.003, 95% CI [0.0008, 0.0058], p = .01). ADHD symptoms had no effect on PIA (estimate −0.0005, 95% CI [−0.001, 0.0004], p = .29) (Table 3). The correlation between PES and PIA was significant for children in the control group (Pearson correlation r = .1998, p = .0058) (Figure 3), but not for children in the FHR-SZ (r = .066, p = .37) or FHR-BP (r = −.077, p = .41) groups.
Figure 3. Association between post-error slowing (PES) and post-error improvement of accuracy (PIA) for children in the control group (N = 189). PES on the X-axis is measured in milliseconds and PIA on the Y-axis is measured in percent.
Follow-up analyses showed that the test for fixed effects of PES after incongruent trials (PESincongruent) revealed no effect across groups (F (2, 473) = 0.25, p = .78), but a significant effect of sex (F (1, 473) = 4.00, p = .046) and age (F (1, 473) = 4.15, p = .042), and a significant age × group interaction (F (2, 473) = 4.75, p = .0091). Boys exhibited less PESincongruent than girls (mean difference −58.62, 95% CI [−116.2, −1.036], p = .046) (Figure 2b). The age × group interaction showed how PESincongruent increased with age for both the FHR-BP and FHR-SZ groups but not for the control group.
The test for fixed effects of PIA after incongruent trials (PIAincongruent) showed no significant differences across groups (F (2, 335) = 1.66, p = .19), sex (F (1, 269) = 0.53, p = .47), or age (F (1, 432) = 0.37, p = .55). None of the two-way or three-way interactions were significant.
Error rate differed between sexes, with boys making 25% more errors than girls (estimated ratio 1.25, 95% CI [1.093, 1.430], p = .0012). The error rate was clearly affected by congruency: children exhibited 59.5% less errors in the congruent trials compared with the incongruent trials (estimated ratio 0.405, 95% CI [0.382, 0.430], p < .0001). No effects of error rate across familial high-risk groups (F (2, 492) = 1.24, p = .29) or age (F (1, 492) = 2.11, p = .15) were present. A higher level of ADHD symptoms significantly increased the error rate by 1.3% (estimated ratio 1.013, 95% CI [1.005, 1.020], p = .0012), whereas a higher level of motor function significantly reduced the error rate by 6% (estimated ratio 0.94, 95% CI [0.921, 0.960], p < .0001) (Table 3).
Discussion
This large population-based cohort study showed that, at age 7 years, children exhibited error adaptation in the form of PES. However, slowing of responses did not translate to improvements in accuracy (PIA). Familial high-risk status did not differentially affect PES or PESincongruent; however, the significant age × high-risk group interaction in relation to PESincongruent showed how PESincongruent increased with age for both the FHR-BP and FHR-SZ groups but not for the control group. Boys generally displayed less error adaptation than girls in relation to PES and PESincongruent. We detected a negative relationship between PES and error rate, with lower PES associated with a higher error rate. Motor function and ADHD symptoms did not influence PES. Although we detected no group or sex differences in relation to PIA, better motor function led to a larger PIA. Moreover, the two measures of error adaptation (PES and PIA) were significantly correlated for children in the control group, but not for children in the FHR-SZ or FHR-BP groups. No relationship between ADHD symptoms and PIA was identified. The error rate did not differ by familial high-risk status, but was influenced by congruency, sex, ADHD symptoms, and motor ability. Specifically, children exhibited more errors in incongruent trials and boys demonstrated more errors than girls across all groups. Furthermore, a higher level of ADHD symptoms increased error rates, whereas better motor function significantly reduced error rates.
Although error adaptation is not fully matured at age 7 years, our findings suggested that post-error adjustment behavior may be influenced by sex at this stage of development. Furthermore, our results indicated that familial high risk for severe mental disorders at this stage of cognitive control development did not significantly affect error adaptation, but the lack of correlation between PIA and PES for the familial high-risk groups was noteworthy.
Error adaptation has been measured in individuals with schizophrenia and bipolar disorder, but few studies have tested error adaptation in offspring (Patino et al., Reference Patino, Adler, Mills, Strakowski, Fleck, Welge and Delbello2013). Therefore, our study provides new knowledge regarding behavioral error adaptation in a large sample of children of primary school age with a familial high risk for severe mental disorders. We found that children displayed PES at this age, but there was no difference in PES between the groups. Related literature concerning adults diagnosed with schizophrenia is inconclusive (Abrahamse et al., Reference Abrahamse, Ruitenberg, Duthoo, Sabbe, Morrens and Van Dijck2016), with reports of absent or reduced PES (Alain, McNeely, He, Christensen, & West, Reference Alain, McNeely, He, Christensen and West2002; Carter, MacDonald, Ross, & Stenger, Reference Carter, MacDonald, Ross and Stenger2001; Kerns et al., Reference Kerns, Cohen, MacDonald, Johnson, Stenger, Aizenstein and Carter2005) as well as intact PES (Laurens, Ngan, Bates, Kiehl, & Liddle, Reference Laurens, Ngan, Bates, Kiehl and Liddle2003; Mathalon et al., Reference Mathalon, Fedor, Faustman, Gray, Askari and Ford2002; Perez et al., Reference Perez, Ford, Roach, Woods, McGlashan, Srihari and Mathalon2012; Polli et al., Reference Polli, Barton, Vangel, Goff, Iguchi and Manoach2006). Furthermore, to our knowledge, the only available study testing PES in individuals with bipolar disorder found no difference in PES from controls (Saunders, Goodwin, & Rogers, Reference Saunders, Goodwin and Rogers2016). Together with these studies, our study contributes to the discussion of error adaptation as an endophenotype reflecting underlying vulnerabilities in the neurodevelopmental process of developing severe mental illness. Thus, PES and PIA do not appear to fulfill the criteria for an endophenotype at this development stage. The immaturity was not specific to the FHR-SZ or FHR-BP groups as PES was present in all three groups (FHR-SZ, FHR-BP, and controls) and PIA in none. However, the relative immaturity of all three groups may have prevented us from detecting the high-risk-group-based differences that we expect to emerge at an older age (Reichenberg et al., Reference Reichenberg, Caspi, Harrington, Houts, Keefe and Caspi2010). As such, it is possible that different trajectories may be revealed because of later maturation. Follow-up studies into adolescence are necessary to provide knowledge of the development of error adaptation in the vulnerable group of children born with a familial high risk of severe mental disorders.
Brain maturation through childhood into adulthood may display sex differences (Kaczkurkin, Raznahan, & Satterthwaite, Reference Kaczkurkin, Raznahan and Satterthwaite2019), which may derive from different neurodevelopmental trajectories (Giedd, Raznahan, Mills, & Lenroot, Reference Giedd, Raznahan, Mills and Lenroot2012; Lenroot et al., Reference Lenroot, Gogtay, Greenstein, Wells, Wallace, Clasen and Giedd2007; Ruigrok et al., Reference Ruigrok, Salimi-Khorshidi, Lai, Baron-Cohen, Lombardo, Tait and Suckling2014) or possibly as a result of the more pronounced brain structure variability among males compared with females (Wierenga, Sexton, Laake, Giedd, & Tamnes, Reference Wierenga, Sexton, Laake, Giedd and Tamnes2018). These sex differences in the developing brain may contribute to differences in cognitive performance (Grabowska, Reference Grabowska2017; Gur et al., Reference Gur, Turetsky, Matsui, Yan, Bilker, Hughett and Gur1999). Our behavioral results showed there were sex-based differences in PES at age 7 years, whereby boys displayed less PES than girls, although there was no sex difference in relation to PIA. These findings are consistent with those from adult studies, which have reported that men displayed less PES than women (Fischer et al., Reference Fischer, Danielmeier, Villringer, Klein and Ullsperger2016; Thakkar et al., Reference Thakkar, Congdon, Poldrack, Sabb, London, Cannon and Bilder2014) but no sex effects on PIA (Fischer et al., Reference Fischer, Danielmeier, Villringer, Klein and Ullsperger2016). Furthermore, we detected a sex difference in relation to errors, with boys making more errors than girls. These findings may indicate that, at age 7 years, boys may not learn as much from their errors or adapt their behavior as appropriately as girls. Our behavioral finding of a sex difference may reflect different brain maturation at the age of 7 years between boys and girls in relation to errors and error adaptation.
Currently, no consensus has been reached as to the mechanisms underlying PES (Notebaert et al., Reference Notebaert, Houtman, Opstal, Gevers, Fias and Verguts2009; Purcell & Kiani, Reference Purcell and Kiani2016; Ullsperger & Danielmeier, Reference Ullsperger and Danielmeier2016), how to measure PES, or which ITI (Compton et al., Reference Compton, Heaton and Ozer2017) or RSI (Danielmeier & Ullsperger, Reference Danielmeier and Ullsperger2011; Jentzsch & Dudschig, Reference Jentzsch and Dudschig2009) conditions are optimal for ascertaining PES. Some researchers considered PES to result from an increased motor threshold implemented by a general motor inhibition to avoid future mistakes (Marco-Pallares et al., Reference Marco-Pallares, Camara, Munte and Rodriguez-Fornells2008), while others interpreted PES as related to cognitive control processes associated with committing errors (Botvinick, Braver, Barch, Carter, & Cohen, Reference Botvinick, Braver, Barch, Carter and Cohen2001; Plessen et al., Reference Plessen, Allen, Eichele, Van, Hovik, Sorensen and Eichele2015). Alternatively, PES may reflect an unspecific orientation response to uncommon or surprising events rather than the result of cognitive control (Notebaert et al., Reference Notebaert, Houtman, Opstal, Gevers, Fias and Verguts2009). Studies also differ in the way PES is measured. Some opted to subtract the average RT on correct trials following correct trials from the average RT on correct trials following errors (Hajcak, McDonald, & Simons, Reference Hajcak, McDonald and Simons2003; Kerns et al., Reference Kerns, Cohen, MacDonald, Cho, Stenger and Carter2004; Mathalon et al., Reference Mathalon, Fedor, Faustman, Gray, Askari and Ford2002), while others chose an arguably more robust measure (Dutilh et al., Reference Dutilh, Ravenzwaaij, Nieuwenhuis, Van Der Mass, Forstmann and Wagenmarkers2012) in which PES was calculated based on the sequence surrounding trials with erroneous responses to avoid fluctuations in task engagement confounding PES. In addition, the duration of ITI (Compton et al., Reference Compton, Heaton and Ozer2017) and RSI (Danielmeier & Ullsperger, Reference Danielmeier and Ullsperger2011; Jentzsch & Dudschig, Reference Jentzsch and Dudschig2009) potentially play a role in ascertaining PES and how errors affect accuracy. Among adults, long ITIs led to increased post-error accuracy, while shorter ITIs led to impaired post-error accuracy (Wessel, Reference Wessel2018). In short RSI situations, adults showed larger PES and lower post-error accuracy (Danielmeier & Ullsperger, Reference Danielmeier and Ullsperger2011; Jentzsch & Dudschig, Reference Jentzsch and Dudschig2009). Furthermore, PIA has been seen at long mean ITIs (e.g., 900–2250 ms) (Danielmeier & Ullsperger, Reference Danielmeier and Ullsperger2011; Marco-Pallares et al., Reference Marco-Pallares, Camara, Munte and Rodriguez-Fornells2008). We calculated PES with the robust measure suggested by a previous study (Dutilh et al., Reference Dutilh, Ravenzwaaij, Nieuwenhuis, Van Der Mass, Forstmann and Wagenmarkers2012). The ITI in our study was 800 ms, which is considered a short ITI (Compton et al., Reference Compton, Heaton and Ozer2017). The fact that we found that children showed PES but no PIA at age 7 years could be explained by the short ITI. However, our results could also reflect the general development of the cognitive control system, as increasing PES and PIA with age has been reported in children from age 8 to 19 years (Overbye et al., Reference Overbye, Walhovd, Paus, Fjell, Huster and Tamnes2019). This highlights the need for longitudinal developmental studies to track the emergence and changes in PES and PIA across development.
Even though PIA and PES both reflect adaptive processes, they may be caused by different underlying processes because they are not necessarily correlated and do not always co-occur (Danielmeier & Ullsperger, Reference Danielmeier and Ullsperger2011). This was consistent with our findings that 7-year-old children exhibited PES but not PIA at the group level, even though some children in all three groups exhibited PIA. However, we only found a significant positive correlation between PES and PIA for the control group, indicating higher PES was associated with higher PIA (although PIA remained negative for most children). For the children in the control group with a positive PIA, PES was reached between 0 and 500 ms (see the upper right quadrant of Figure 3), whereas for the majority of control group children with a negative PIA, PES was between −500 ms and + 500 ms (Figure 3). No correlation between PIA and PES was established for children in the FHR-SZ and FHR-BP groups, indicating different associations between PIA and PES in these familial high-risk groups. PES was calculated using a pre–post measure to remove the effects of global performance shifts, as suggested by Dutilh et al. (Reference Dutilh, Ravenzwaaij, Nieuwenhuis, Van Der Mass, Forstmann and Wagenmarkers2012). Because the pre–post method is not applicable to PIA, PIA was calculated using a standard procedure which allowed post-error adaptation to be confounded with changes in motivation, error rate, and so on across the experiment. However, this might be the reason why the correlation between PES and PIA was observed only for the control group. It could be that PIA in the high-risk groups is more strongly driven by global performance shifts (e.g., motivation or response caution decrease during the trial, as the child gets tired, which results in more errors and inaccurate responses), thus reducing the common variance with (control-related) PES.
One study documented a positive correlation between PIA and PES in children and adolescents (Overbye et al., Reference Overbye, Walhovd, Paus, Fjell, Huster and Tamnes2019), whereas studies of adults either showed no correlation between PIA and PES (Carp & Compton, Reference Carp and Compton2009; Danielmeier & Ullsperger, Reference Danielmeier and Ullsperger2011) or confirmed a relationship (Hajcak et al., Reference Hajcak, McDonald and Simons2003). The prolonged maturation of the cognitive control system means it is unclear if our behavioral error adaptation data reflected immature cognitive control processes or an automatic orienting response. Further studies during development are needed to clarify this point.
A higher rate of errors is commonly reported among individuals with ADHD (Balogh & Czobor, Reference Balogh and Czobor2016; Van De Voorde, Roeyers, & Wiersema, Reference Van De Voorde, Roeyers and Wiersema2010), which was consistent with our finding that a higher level of ADHD symptoms increased error rates. A meta-analysis of 1,667 patients with ADHD with a broad age range, including adults with a mature cognitive control system (age range 6–41 years, mean age 12.2 ± 7.9 years), documented diminished PES in patients with ADHD as a group compared with controls (Balogh & Czobor, Reference Balogh and Czobor2016). In contrast to our expectations, we could not replicate that ADHD symptoms significantly influenced PES or PIA at age 7 years in our sample of 497 children. These results could reflect the immaturity of their cognitive control system.
Brain imaging studies have documented involvement of the cerebellum, posterior parietal cortex, and primary motor cortex in error adaptation in addition to the ACC (Desmurget et al., Reference Desmurget, Grea, Grethe, Prablanc, Alexander and Grafton2001). Our results showed that better motor function significantly reduced error rates. Moreover, our results suggest that better motor function improved the ability to display PIA, whereas motor function did not influence PES. These findings indicate the involvement of different brain areas and processes in error adaptation.
Our study had major strengths, including the novelty of assessing error adaptation in a large, same-age, pre-pubertal sample with a familial high risk of severe mental disorders. In addition, we assessed motor function and ADHD symptoms in the same children, which allowed us to investigate the impact of these parameters in relation to error adaptation. The Eriksen flanker task provided performance feedback after errors. This implies that post-error adaptation does not solely rely on internal error detection (as in typical ERN studies) but could also rely on feedback processing. This somewhat limits interpretation of the results and could include a potential explanation as to why no significant differences between groups were found. A further limitation of our study was that cognitive control was only assessed using a behavioral task. Future studies would benefit from assessing these abilities using brain imaging and electrophysiology to gain a more profound understanding of the neural underpinnings and development of cognitive control before puberty.
Conclusions
At age 7 years, children exhibited PES and PESincongruent, but the slowing of responses did not translate to improvements in accuracy. Although we found no differences in PES, PIA, or error rate in the FHR-SZ or FHR-BP group compared with the control group, PES and PESincongruent showed sex-related differences, with boys displaying less error adaptation than girls. We detected a correlation between PES and PIA for children in the control group, but not for children in the familial high-risk groups. We detected a negative relationship between PES and error rate, with lower PES associated with higher error rates. Error rates were found to be influenced by sex, congruency, ADHD symptoms, and motor ability. Our findings suggest that error adaptation behavior at age 7 years displays specific sex differences at this stage of development. Furthermore, our results indicate that familial disposition of severe mental disorders at this stage of cognitive control development does not influence error adaptation, except for the differential relationship between PIA and PES.
Acknowledgments
We thank: all the children and their families who participated in the Danish High Risk and Resilience Study – VIA 7; C. Tjott for training and supervision in the assessment of the ABC-2; M. Skjærbæk, A. Søndergaard, M. Gregersen, A. Ranning, H. Jensen, M. Melau, C. Gregersen, H. Stadsgaard, K. Kold Zahle, and M. Toft for contributing to data collection; C. Bøcker Pedersen and M. Giørtz Pedersen for retrieving the register extract; J. Ohland and M. Chaine for help with data management; and P. B. Mortensen, T. Werge, D. Hougaard, and A. Børglum for collaboration in iPSYCH.
Funding Statement
This work was supported by the Mental Health Services of the Capital Region of Denmark, the Independent Research Fund Denmark (#9039-00220B and #9037-00169B), the Lundbeck Foundation Initiative for Integrative Psychiatric Research (iPSYCH) (grant nos. R102-A9118 and R155-2014-1724), Aarhus University, the Tryg Foundation, and the Beatrice Surovell Haskell Fund for Child Mental Health Research of Copenhagen (grant no. J.NR 11531). The sources of funding had no involvement in this work.
Conflicts of Interest
None.
