ADHD in Girls and Boys
Demographic characteristics are presented in Table 1. Forty three males with ADHD (M = 11.2 years), 37 females with ADHD (M = 11.9 years), 32 healthy control (HC) males (M = 11.4 years) and 18 HC females (M = 11.9 years) between 8 and 17 years participated in the study. The ADHD participants were recruited as consecutive referrals from seven outpatient Child and Adolescent Mental Health Centres in Innlandet Hospital Trust (IHT) for assessment of ADHD. All participants underwent a comprehensive assessment according to common clinical practice. Semi-structured clinical interviews (Kiddie-Schedule for Affective Disorders and Schizophrenia - K-SADS) were conducted separately for children/adolescents and parents to assess psychopathology. The interviewers were experienced clinicians, and were trained to high levels of interrater reliability for the assessment of diagnosis. The diagnostic evaluation with K-SADS was supplemented with information from the ADHD Rating Scale IV (ARS-IV), and the Child Behavior Checklist/6-18, which covers the DSM-IV symptoms for ADHD. Teacher reports describing school functioning, both academic and socially, which is mandatory on referral, were incorporated into the diagnostic evaluation. Diagnoses were considered positive if, based on a comprehensive evaluation of K-SADS, teacher information and rating scales, DSM-IV criteria were met.
Based on diagnostic evaluation with K-SADS, co-existing diagnoses within the group of males with ADHD included depression (4.7%), anxiety (4.7%), conduct disorder (4.7%), and oppositional defiant disorder (11.6%). Co-existing diagnoses within the females with ADHD included anxiety (8.1%), and oppositional defiant disorder (10.8%). Despite a low prevalence of co-existing diagnoses, parent and self-report scales indicated elevated levels of externalizing and internalizing symptoms in both males and females with ADHD when compared with normal developing counterparts. Exclusion criteria for all participants included prematurity (< 36 weeks), IQ below 70, a history of stimulant treatment or any disease affecting the central nervous system. None of the participants used any type of psychopharmacological medication. One boy with ADHD was excluded due to estimated IQ below 70. None were excluded due to history of stimulant treatment or any disease affecting the central nervous system.
All participants in the HC groups were screened for mental disorders with the K-SADS in separate interviews for children/adolescents and parents. The HC were recruited from local schools and were given a small compensation for participating. The HC could not have been treated for a mental disorder, have a psychiatric diagnosis, have had a head injury (with loss of consciousness) or known dyslexia. The four groups (ADHD/females, ADHD/males, HC/females, HC/males) did not differ significantly with regard to age and gender distribution. The Wechsler Abbreviated Scale of Intelligence (WASI) was administered to estimate IQ in all participants. The groups differed significantly with regard to IQ, F (3,126) = 4.60, p = .004, Eta = .099, and Bonferroni post-hoc analysis showed that both ADHD gender groups scored below the females in the HC group. On average, mothers of children in the HC group had 1.7 years more education than mothers of children with ADHD, F (3,126) = 6.80, p < 0.001. All parents/caregivers and participants above 12 years gave written informed consent in accordance with the Research Ethics Committee in Eastern Norway. All children under the age of 12 years provided oral consent to participate. The study was approved by the Regional Committee for Medical Research Ethics in Eastern Norway (REK-Øst), and by the Privacy protection ombudsman for research at Innlandet Hospital Trust. It was conducted in accordance with the Helsinki Declaration of the World Medical Association Assembly.
Measures of symptomatology The Child Behavior Checklist/6-18 (CBCL) is a widely used scale containing 7 competence items and 113 specific problem items, each of which is rated on a 0–2 metric. The 120 items assess adaptive behavior as well as eight narrow band factors (Anxious/Depressed, Withdrawn/Depressed, Somatic Complaints, Social Problems, Thought Problems, Attention Problems, Rule-Breaking Behavior and Aggressive Behavior) and two broadband factors (Externalizing and Internalizing symptoms) of co-existing symptoms. The 2001 revision also includes seven DSM-oriented scales consistent with DSM diagnostic categories (Affective Problems, Anxiety Problems, Somatic Problems, ADHD, Oppositional Defiant Problems and Conduct Problems). On the parent-report CBCL, we used seven of the narrow band factors (excluding Attention Problems) and five of the DSM-oriented scales (excluding ADHD) to assess co-existing symptoms. Elevated T-scores indicate a higher degree of co-existing internalizing and externalizing symptoms. Cross-cultural studies have demonstrated good discriminant validity with mean factor loadings across societies at .62. Acceptable reliability and validity of the Norwegian version of the CBCL are reported by Nøvik.
The Revised Children's Manifest Anxiety Scale, second edition (RCMAS-2) is a 49-item self-report instrument designed to measure anxiety symptoms in children 6 to 19 years of age. Children respond either "Yes" or "No" to all 49-items. The instrument reveals three anxiety factors: Physiological Anxiety, Worry and Social Anxiety. The three anxiety factors are summed yielding a Total Anxiety score. Elevated raw-scores indicate a higher degree of anxiety symptoms. The RCMAS Total Anxiety Scale has been found to have satisfactory psychometric properties with high test–retest reliability and consistent construct validity. Satisfactory psychometric properties have been replicated among other cultures as well.
The State-Trait Anxiety Inventory for Children (STAIC) includes two 20-item self-report scales that measure both enduring tendencies (Trait) and situational variations (State) in levels of perceived anxiety. Children respond on a three-point scale indicating varying degree of worry, feelings of tension, and/or nervousness. Elevated raw-scores indicate a higher degree of situational and temporal anxiety. In a quantitative review by Seligman and colleagues, the authors argue that the STAIC possess satisfactory psychometric properties.
The Short Mood and Feelings Questionnaire (SMFQ) is a 13-item self-report instrument designed to measure depressive symptoms in children 8 to 18 years of age. The SMFQ is derived from the original 30-item Mood and Feelings Questionnaire (MFQ) where children respond on a three-point scale ("not true", "sometimes true" and "true"). A net score was generated based on the 13 items with elevated raw-scores indicating a higher degree of depression symptoms. The SMFQ have demonstrated high internal consistency (Crohnbach's alpha = .90), and test-retest stability in children for a two-week period yielded an intra class correlation of .66. Angold and colleagues found SMFQ to correlate strongly with Children's Depression Inventory (CDI) and Diagnostic Interview Schedule for Children (DISC-C) depression scores (r = .67 and .51, respectively).
The Letter-Number Sequencing Test The Letter-Number Sequencing Test (LN) was used as a measure of working memory. The test consists of ten items. Each item contains three trials with the same number of digits and letters. The test administrator reads aloud each trial and asks the child to recall the numbers in ascending order and the letters in alphabetical order. In the present study, total correct recalled trials were examined. Lower scaled scores indicated difficulties with the task.
The Colour - Word Interference Test, Condition 3 The Colour - Word Interference Test, Condition 3 (CW 3) was used as a measure of inhibition. The examinee needs to inhibit an overlearned verbal response when naming the dissonant ink colours in which the words are printed. For the present study, completion time in seconds was examined. Lower scaled scores indicated difficulties with the task.
The Colour - Word Interference Test, Condition 4 The Colour - Word Interference Test, Condition 4 (CW 4) was used as a measure of cognitive flexibility. The examinee is asked to switch back and forth between naming the dissonant ink colours and reading the words. For the present study, completion time in seconds was examined. Lower scaled scores indicated difficulties with the task.
The Trail Making Test, Condition 4 The Trail Making Test, condition 4 (TMT 4) was used as a measure of cognitive flexibility. The examinee is asked to draw a line interchangeably between numbers and letters in the right order. For the present study, time to complete task was examined. Lower scaled scores indicated difficulties with the task.
The Design Fluency Test, Condition 3 The Design Fluency Test, condition 3 (DF) was used as a measure of cognitive flexibility. The examinee is asked to draw as many different designs as possible using four straight lines connecting five filled and empty dots interchangeably. The examinee is given 60 seconds for the task. For the present study, total correct responses were examined. Lower scaled scores indicated difficulties with the task.
The Tower Test The Tower Test was used as a measure of planning. In this task the examinee is asked to construct several target towers by moving five disks, varying in size, across three pegs in the fewest number of moves possible. While doing this, the examinee is allowed to move only one disk at a time, and not to place a larger disk over a smaller disk. In the present study total achievement score was examined. Lower scaled scores indicated difficulties with the task.
The Letter Fluency Test The Letter Fluency Test (LF) was used as a measure of verbal fluency. This task includes three 60-seconds trials, where participants were asked to generate words fluently in an effortful, phonemic format with the letters F, A, and S. For the present study, total correct responses were examined. Lower scaled scores indicated difficulties with the task.
The BRIEF for children and adolescents aged 5 to 18 includes a parent form and a teacher form. In the current study, the Norwegian parent rating version was used. The BRIEF is composed of eight clinical scales (Inhibition, Shift, Emotional Control, Initiate, Working Memory, Plan/Organize, Organization of Materials and Monitor). Fallmyr & Egeland reported high internal consistency (Chronbachs α = .76-.92) on the Norwegian parent rating version of the BRIEF. These values are at the same level as Chronbachs α reported in the BRIEF manual (.80-.98). Elevated BRIEF T-scores indicate a higher degree of impairment.
Data analyses were conducted using the statistical package SPSS for Windows, version 15.0 (SPSS, Inc., Chicago, IL). Demographic characteristics were investigated using the Chi-square test for independence (nominal variables) and analysis of variance (ANOVA) (continuous variables) followed up by Bonferroni post-hoc tests for group comparisons when adequate. ANOVAs were carried out to investigate gender x diagnosis interactions in the three measurement domains (co-existing symptoms, neuropsychological EF tests, BRIEF).
In addition to tests of significance we also used an algorithmic modelling/data mining technique to explore gender differences in co-existing symptoms and EF ratings and measures. Classical statistical techniques are designed to test and reject the hypothesis that observed differences between groups have occurred by chance. Algorithmic modelling techniques have been developed to address a somewhat different question. Briefly, these techniques can identify from a sample of potential predictor variables the most important subset for categorizing subjects or predicting outcomes. Hence, we used this approach to delineate within each gender the subset of symptom ratings, EF measures and EF ratings that appear to be most important in discriminating children with ADHD from HC. Specifically, we used random forest classification and cross-validation (R packages randomForest 4.5-34 and caret 5.02-011) to identify and rank order different symptom ratings and EF measures for their degree of importance in differentiating ADHD from HC within each gender. Although importance and statistical significance often go hand-in-hand, the two are not necessarily the same. The approach has many advantages. In particular, it can provide meaningful results with smaller sample sizes than stochastic models. Further they are less susceptible to overfitting and multicollinearity, provide more accurate predictions, and do not make the unlikely assumption that the multivariate data being analyzed are multivariate normal.
Briefly, this is a form of "ensemble learning" in which a large number of unpruned decision trees are generated and their results aggregated. The random part comes in as each tree is constructed using a different bootstrap sample of the data, and each node is split using the best among a subset of predictors randomly chosen at that node. As Liaw and Wiener indicate this strategy performs very well compared to many other classifiers, including discriminant analysis, logistic regression, support vector machines and neural networks. It is primarily used in data mining and in genomic analysis, such as microarray studies.
Each decision tree was generated using results from 75% of the participants and then tested on the remaining 25% (validation set). This process was performed 5000 times on different random splits of the data to provide a cross-validated estimate of the predictive discriminant ability of the measures (accuracy, kappa) that would likely generalize to new cases. The importance of each variable in the cluster was assessed by calculating the decrease in predictive accuracy following the sequential permutation (effective randomization and elimination) of each variable in the cluster on the validation set. The most important variables were the ones whose effective elimination from the forest produced the greatest degradation in accuracy.
Methods
Procedure and Participants
Demographic characteristics are presented in Table 1. Forty three males with ADHD (M = 11.2 years), 37 females with ADHD (M = 11.9 years), 32 healthy control (HC) males (M = 11.4 years) and 18 HC females (M = 11.9 years) between 8 and 17 years participated in the study. The ADHD participants were recruited as consecutive referrals from seven outpatient Child and Adolescent Mental Health Centres in Innlandet Hospital Trust (IHT) for assessment of ADHD. All participants underwent a comprehensive assessment according to common clinical practice. Semi-structured clinical interviews (Kiddie-Schedule for Affective Disorders and Schizophrenia - K-SADS) were conducted separately for children/adolescents and parents to assess psychopathology. The interviewers were experienced clinicians, and were trained to high levels of interrater reliability for the assessment of diagnosis. The diagnostic evaluation with K-SADS was supplemented with information from the ADHD Rating Scale IV (ARS-IV), and the Child Behavior Checklist/6-18, which covers the DSM-IV symptoms for ADHD. Teacher reports describing school functioning, both academic and socially, which is mandatory on referral, were incorporated into the diagnostic evaluation. Diagnoses were considered positive if, based on a comprehensive evaluation of K-SADS, teacher information and rating scales, DSM-IV criteria were met.
Based on diagnostic evaluation with K-SADS, co-existing diagnoses within the group of males with ADHD included depression (4.7%), anxiety (4.7%), conduct disorder (4.7%), and oppositional defiant disorder (11.6%). Co-existing diagnoses within the females with ADHD included anxiety (8.1%), and oppositional defiant disorder (10.8%). Despite a low prevalence of co-existing diagnoses, parent and self-report scales indicated elevated levels of externalizing and internalizing symptoms in both males and females with ADHD when compared with normal developing counterparts. Exclusion criteria for all participants included prematurity (< 36 weeks), IQ below 70, a history of stimulant treatment or any disease affecting the central nervous system. None of the participants used any type of psychopharmacological medication. One boy with ADHD was excluded due to estimated IQ below 70. None were excluded due to history of stimulant treatment or any disease affecting the central nervous system.
All participants in the HC groups were screened for mental disorders with the K-SADS in separate interviews for children/adolescents and parents. The HC were recruited from local schools and were given a small compensation for participating. The HC could not have been treated for a mental disorder, have a psychiatric diagnosis, have had a head injury (with loss of consciousness) or known dyslexia. The four groups (ADHD/females, ADHD/males, HC/females, HC/males) did not differ significantly with regard to age and gender distribution. The Wechsler Abbreviated Scale of Intelligence (WASI) was administered to estimate IQ in all participants. The groups differed significantly with regard to IQ, F (3,126) = 4.60, p = .004, Eta = .099, and Bonferroni post-hoc analysis showed that both ADHD gender groups scored below the females in the HC group. On average, mothers of children in the HC group had 1.7 years more education than mothers of children with ADHD, F (3,126) = 6.80, p < 0.001. All parents/caregivers and participants above 12 years gave written informed consent in accordance with the Research Ethics Committee in Eastern Norway. All children under the age of 12 years provided oral consent to participate. The study was approved by the Regional Committee for Medical Research Ethics in Eastern Norway (REK-Øst), and by the Privacy protection ombudsman for research at Innlandet Hospital Trust. It was conducted in accordance with the Helsinki Declaration of the World Medical Association Assembly.
Measures
Measures of symptomatology The Child Behavior Checklist/6-18 (CBCL) is a widely used scale containing 7 competence items and 113 specific problem items, each of which is rated on a 0–2 metric. The 120 items assess adaptive behavior as well as eight narrow band factors (Anxious/Depressed, Withdrawn/Depressed, Somatic Complaints, Social Problems, Thought Problems, Attention Problems, Rule-Breaking Behavior and Aggressive Behavior) and two broadband factors (Externalizing and Internalizing symptoms) of co-existing symptoms. The 2001 revision also includes seven DSM-oriented scales consistent with DSM diagnostic categories (Affective Problems, Anxiety Problems, Somatic Problems, ADHD, Oppositional Defiant Problems and Conduct Problems). On the parent-report CBCL, we used seven of the narrow band factors (excluding Attention Problems) and five of the DSM-oriented scales (excluding ADHD) to assess co-existing symptoms. Elevated T-scores indicate a higher degree of co-existing internalizing and externalizing symptoms. Cross-cultural studies have demonstrated good discriminant validity with mean factor loadings across societies at .62. Acceptable reliability and validity of the Norwegian version of the CBCL are reported by Nøvik.
The Revised Children's Manifest Anxiety Scale, second edition (RCMAS-2) is a 49-item self-report instrument designed to measure anxiety symptoms in children 6 to 19 years of age. Children respond either "Yes" or "No" to all 49-items. The instrument reveals three anxiety factors: Physiological Anxiety, Worry and Social Anxiety. The three anxiety factors are summed yielding a Total Anxiety score. Elevated raw-scores indicate a higher degree of anxiety symptoms. The RCMAS Total Anxiety Scale has been found to have satisfactory psychometric properties with high test–retest reliability and consistent construct validity. Satisfactory psychometric properties have been replicated among other cultures as well.
The State-Trait Anxiety Inventory for Children (STAIC) includes two 20-item self-report scales that measure both enduring tendencies (Trait) and situational variations (State) in levels of perceived anxiety. Children respond on a three-point scale indicating varying degree of worry, feelings of tension, and/or nervousness. Elevated raw-scores indicate a higher degree of situational and temporal anxiety. In a quantitative review by Seligman and colleagues, the authors argue that the STAIC possess satisfactory psychometric properties.
The Short Mood and Feelings Questionnaire (SMFQ) is a 13-item self-report instrument designed to measure depressive symptoms in children 8 to 18 years of age. The SMFQ is derived from the original 30-item Mood and Feelings Questionnaire (MFQ) where children respond on a three-point scale ("not true", "sometimes true" and "true"). A net score was generated based on the 13 items with elevated raw-scores indicating a higher degree of depression symptoms. The SMFQ have demonstrated high internal consistency (Crohnbach's alpha = .90), and test-retest stability in children for a two-week period yielded an intra class correlation of .66. Angold and colleagues found SMFQ to correlate strongly with Children's Depression Inventory (CDI) and Diagnostic Interview Schedule for Children (DISC-C) depression scores (r = .67 and .51, respectively).
Neuropsychological EF Tests
The Letter-Number Sequencing Test The Letter-Number Sequencing Test (LN) was used as a measure of working memory. The test consists of ten items. Each item contains three trials with the same number of digits and letters. The test administrator reads aloud each trial and asks the child to recall the numbers in ascending order and the letters in alphabetical order. In the present study, total correct recalled trials were examined. Lower scaled scores indicated difficulties with the task.
The Colour - Word Interference Test, Condition 3 The Colour - Word Interference Test, Condition 3 (CW 3) was used as a measure of inhibition. The examinee needs to inhibit an overlearned verbal response when naming the dissonant ink colours in which the words are printed. For the present study, completion time in seconds was examined. Lower scaled scores indicated difficulties with the task.
The Colour - Word Interference Test, Condition 4 The Colour - Word Interference Test, Condition 4 (CW 4) was used as a measure of cognitive flexibility. The examinee is asked to switch back and forth between naming the dissonant ink colours and reading the words. For the present study, completion time in seconds was examined. Lower scaled scores indicated difficulties with the task.
The Trail Making Test, Condition 4 The Trail Making Test, condition 4 (TMT 4) was used as a measure of cognitive flexibility. The examinee is asked to draw a line interchangeably between numbers and letters in the right order. For the present study, time to complete task was examined. Lower scaled scores indicated difficulties with the task.
The Design Fluency Test, Condition 3 The Design Fluency Test, condition 3 (DF) was used as a measure of cognitive flexibility. The examinee is asked to draw as many different designs as possible using four straight lines connecting five filled and empty dots interchangeably. The examinee is given 60 seconds for the task. For the present study, total correct responses were examined. Lower scaled scores indicated difficulties with the task.
The Tower Test The Tower Test was used as a measure of planning. In this task the examinee is asked to construct several target towers by moving five disks, varying in size, across three pegs in the fewest number of moves possible. While doing this, the examinee is allowed to move only one disk at a time, and not to place a larger disk over a smaller disk. In the present study total achievement score was examined. Lower scaled scores indicated difficulties with the task.
The Letter Fluency Test The Letter Fluency Test (LF) was used as a measure of verbal fluency. This task includes three 60-seconds trials, where participants were asked to generate words fluently in an effortful, phonemic format with the letters F, A, and S. For the present study, total correct responses were examined. Lower scaled scores indicated difficulties with the task.
Inventory Based Information of EF
The BRIEF for children and adolescents aged 5 to 18 includes a parent form and a teacher form. In the current study, the Norwegian parent rating version was used. The BRIEF is composed of eight clinical scales (Inhibition, Shift, Emotional Control, Initiate, Working Memory, Plan/Organize, Organization of Materials and Monitor). Fallmyr & Egeland reported high internal consistency (Chronbachs α = .76-.92) on the Norwegian parent rating version of the BRIEF. These values are at the same level as Chronbachs α reported in the BRIEF manual (.80-.98). Elevated BRIEF T-scores indicate a higher degree of impairment.
Data Analyses
Data analyses were conducted using the statistical package SPSS for Windows, version 15.0 (SPSS, Inc., Chicago, IL). Demographic characteristics were investigated using the Chi-square test for independence (nominal variables) and analysis of variance (ANOVA) (continuous variables) followed up by Bonferroni post-hoc tests for group comparisons when adequate. ANOVAs were carried out to investigate gender x diagnosis interactions in the three measurement domains (co-existing symptoms, neuropsychological EF tests, BRIEF).
Random Forest Classification
In addition to tests of significance we also used an algorithmic modelling/data mining technique to explore gender differences in co-existing symptoms and EF ratings and measures. Classical statistical techniques are designed to test and reject the hypothesis that observed differences between groups have occurred by chance. Algorithmic modelling techniques have been developed to address a somewhat different question. Briefly, these techniques can identify from a sample of potential predictor variables the most important subset for categorizing subjects or predicting outcomes. Hence, we used this approach to delineate within each gender the subset of symptom ratings, EF measures and EF ratings that appear to be most important in discriminating children with ADHD from HC. Specifically, we used random forest classification and cross-validation (R packages randomForest 4.5-34 and caret 5.02-011) to identify and rank order different symptom ratings and EF measures for their degree of importance in differentiating ADHD from HC within each gender. Although importance and statistical significance often go hand-in-hand, the two are not necessarily the same. The approach has many advantages. In particular, it can provide meaningful results with smaller sample sizes than stochastic models. Further they are less susceptible to overfitting and multicollinearity, provide more accurate predictions, and do not make the unlikely assumption that the multivariate data being analyzed are multivariate normal.
Briefly, this is a form of "ensemble learning" in which a large number of unpruned decision trees are generated and their results aggregated. The random part comes in as each tree is constructed using a different bootstrap sample of the data, and each node is split using the best among a subset of predictors randomly chosen at that node. As Liaw and Wiener indicate this strategy performs very well compared to many other classifiers, including discriminant analysis, logistic regression, support vector machines and neural networks. It is primarily used in data mining and in genomic analysis, such as microarray studies.
Each decision tree was generated using results from 75% of the participants and then tested on the remaining 25% (validation set). This process was performed 5000 times on different random splits of the data to provide a cross-validated estimate of the predictive discriminant ability of the measures (accuracy, kappa) that would likely generalize to new cases. The importance of each variable in the cluster was assessed by calculating the decrease in predictive accuracy following the sequential permutation (effective randomization and elimination) of each variable in the cluster on the validation set. The most important variables were the ones whose effective elimination from the forest produced the greatest degradation in accuracy.
SHARE
