A sample of 347 participants with ADHD-combined type (ADHD/C) aged 5–19 years participated in this study. This sample was based on the Dutch part of the International Multicenter ADHD Genetics (IMAGE) study. Between 2003 and 2006, the IMAGE study recruited families with at least one child with clinically diagnosed ADHD/C and at least one additional sibling regardless of possible ADHD status. Clinical diagnosis of each participant was assessed by health-care professionals from clinical child care centers in The Netherlands. In addition, the diagnosis was confirmed, using an extensive assessment protocol described below. The original sample (N = 459) was contacted and invited for follow-up on average 6.0 years (SD = 0.7) after enrolment; 75.6 % (N = 347) was retained successfully.
Selection procedures have been detailed previously . Briefly, inclusion criteria for the IMAGE study were an age of 5–19 years, Caucasian descent, IQ ≥70, no diagnosis of autism, epilepsy, general learning difficulties, brain disorders and known genetic disorders. Parent and teacher questionnaires were used to screen participants: Conners’ long version  and Strengths and Difficulties Questionnaire (SDQ ). T scores ≥63 on the Conners DSM-IV ADHD subscales Inattention (L), Hyperactivity/impulsivity (M), and Total symptoms (N), and scores ≥90th percentile on the SDQ Hyperactivity subscale were considered clinical. Participants scoring clinically on any of these subscales were administered the Parental Account of Children’s Symptoms (PACS), a semi-structured, standardized, investigator-based interview with the parents as informants . See  for the algorithm used to derive each of the 18 ADHD symptoms as defined by Diagnostic and Statistical Manual of mental disorders: (DSM-IV-TR ). Only participants with a diagnosis of ADHD/C based on the algorithm at baseline were included in the current study. Participants came from 282 different families, 81.6 % was male. Mean age at baseline was 11.4 years (SD = 2.8) and mean age at follow-up was 17.4 (SD = 2.8).
Diagnostic, symptom severity and overall functioning assessment
At follow-up, participants were screened for ADHD using the Schedule for Affective Disorders and Schizophrenia for school-age children—present and lifetime version (K-SADS), a semi-structured, standardized, investigator-based interview with the parents as informants, and when children were 12 years or older, also with the child (separately) . Participants with elevated scores on any of the screen items were administered the full ADHD interview. Additionally, parents completed the Conners’ Parent Rating Scale-Revised: Long version (CPRS-R:L ) and the Conners’ Teacher Rating Scale-Revised: Long version (CTRS-R:L , applied for participants <18 years), or the Conners’ Adult ADHD Rating Scales-Self-Report: Long Version (CAARS-S:L , applied for participants ≥18 years). A diagnostic algorithm was used to establish ADHD status according to DSM-5 criteria, which was similar to the algorithm used at baseline (for full description of diagnostic procedures see ). ADHD subtypes (combined, inattentive, or hyperactive/impulsive subtype) were established following DSM-5 criteria . Comorbidities were assessed using the PACS at baseline and using the K-SADS at follow-up. Classifications in both interviews were established according DSM-IV criteria for oppositional defiant disorder (ODD) and conduct disorder (CD). Classifications of DSM-IV anxiety-, mood-, and tic disorders were established in the K-SADS at follow-up.
For both the K-SADS and the PACS, interviewers underwent comprehensive training by a team under the supervision of E. Taylor at the London Institute of Psychiatry (IoP; PACS) or JB at the Donders Institute for Brain, Cognition and Behavior, Radboud University Medical Centre, Nijmegen (K-SADS). If additional interviewers were used, each center was responsible for their training and supervision. Inter-rater agreement for the PACS was 0.88 (range 0.71–1.00) and for the K-SADS 0.94 (ADHD), 0.89 (ODD), and 0.95 (CD) [17, 25]. The interviewers were trained clinicians (child psychiatrists, psychologists) or trained researchers. Persistence of ADHD was defined as meeting full DSM-5 criteria of ADHD/C at baseline, and meeting full DSM-5 criteria of ADHD regardless of subtype, at follow-up. Subthreshold persistence of ADHD was defined as meeting full criteria of ADHD/C at baseline, and meeting criteria of subthreshold ADHD at follow-up: <6 symptoms of inattention and hyperactivity/impulsivity, but ≥4 symptoms of inattention and/or hyperactivity/impulsivity at follow-up for children <18 years. For participants ≥18 years, thresholds were five and three symptoms respectively. Remission of ADHD was defined as meeting full criteria of ADHD/C at baseline, and not meeting criteria of (subthreshold) ADHD, any subtype, at follow-up.
To assess dimensional persistence, symptom change scores were calculated by subtracting follow-up raw scores from baseline raw scores on the CPRS-R:L ADHD scales L (Inattention), M (Hyperactivity/impulsivity), and N (Total symptoms) . Current ADHD symptom severity, and inattentive and hyperactive/impulsive symptom severity were assessed with the follow-up raw scores on the CPRS-R:L scales N, L and M, respectively. Scores on the Conners’ ADHD subscales represent combined measures of the number and severity of symptoms.
To measure current overall functioning, the Global Assessment Scale-score of the K-SADS (K-GAS) was administered at follow-up. After finishing the K-SADS interview, the interviewer rated psychological, social and academic functioning, resulting in an overall measure of the current level of functioning ranging between 1 (worst possible level of functioning) and 9 (best possible level of functioning) .
Predictors and covariates
All predictor variables were assessed at baseline. Five different classes (italic) were investigated. Demographic variables: age, sex, and socio-economic status (SES) were measured. SES was calculated from the average educational levels of the parent(s), with educational levels ranging between 0 (no education) and 11 (university), according to an adapted Hollingshead scale  fitting the Dutch educational system. ADHD familiality: ADHD familiality was investigated by measuring the percentage of siblings with ADHD according to the PACS interview , and by establishing current parental ADHD status, based on the K-SADS interview (none versus one/both parent(s) with ADHD). ADHD characteristics: ADHD characteristics included ADHD symptom severity, impairment and age of ADHD onset. Symptom severity was measured by the raw score on scale N (range 0–54) of the CPRS-R:L . Impairment was measured using both the parent SDQ  and teacher SDQ (range 0–15) (parent and teacher ratings correlated r = 0.18 and were not combined). Age of onset of ADHD was assessed using the PACS interview . Comorbidities: Comorbid DSM-IV defined ODD (yes/no), CD (yes/no) and a screening of the presence of mood/anxiety symptoms (yes/no) were assessed with the PACS . Pharmacological treatment: cumulative intake of psychostimulants from age of onset until our baseline measurement and from age of onset until follow-up were calculated. Lifetime medication transcripts from pharmacies were available for 87 % and covered the lifespan for 31 % of participants. In addition, a questionnaire was administered to all participants and parents, which assessed lifetime history of psychostimulant medication. When pharmacy transcripts did not fully cover the self-reported treatment period, medication parameters of the missing period(s) were calculated from the questionnaire data and were added to the measures derived from the pharmacy. To optimally take into account daily dose and duration of pharmacological treatment, cumulative intake was calculated by multiplying the mean daily dose (average dose in milligrams for all exposed days; in line with prescription guidelines  and given larger direct effects of dexamphetamine on dopaminergic neurotransmission , dexamphetamine dose was multiplied by 2) with treatment duration corrected for age (treatment duration in months divided by [age minus the minimum start-age within the sample, i.e., 28 months]; see for further details ).
Two potential covariates were investigated. If the univariate relationship between follow-up interval and outcome measures was significant, follow-up interval was entered as a covariate in all subsequent analyses. In addition, study site was entered as a covariate in the final prediction models.
At baseline, families were recruited from clinics and via advertisements. Testing took place at the VU University Amsterdam or at the Donders Institute in Nijmegen. Participants were 48 h off medication during both baseline and follow-up assessments. All ratings of behavioral functioning pertained the participant’s functioning off medication. Families were financially compensated for participating in the study. Informed consent was signed by all participants at both measurements, and parents signed for all children in their family as well. The study was approved by the national ethics committee.
The percentage of missing data was <5 % for ADHD diagnoses, current ADHD symptom severity and overall functioning measures, 19 % for parental ADHD status, and between 0 and 9 % for the other predictors. Missing value analysis (expectation maximization) was performed for participants with one or two missing items on CPRS-R:L subscales, using all data reported in this study (scale L: 9 participants, scale M: 18 participants). K-GAS-scores were normalised by applying a Van der Waerden transformation.
For our first research question, ADHD persistence rates, percentage of comorbidities, mean symptom change and overall functioning scores were calculated. It was tested whether symptom severity decreased significantly over time and whether changes in hyperactivity/impulsivity symptom severity over time differed from changes in inattention symptom severity. To optimally correct for the familial dependency in our data, Generalized Estimating Equation analyses (GEE) were used, with an exchangeable correlation structure. Additionally, interaction-effects of symptom change between baseline and follow-up with age/age2 were tested. For our second research question, an optimal set of predictors for current ADHD symptom severity and overall functioning in participants with ADHD/C was derived in three successive analysis steps. In step 1, GEE analyses were ran on each of the five classes of predictors separately (see Supplemental Table S1 and Supplemental Table S2, available online), with current ADHD symptom severity or overall functioning as outcome measure respectively. The mean correlation between all predictors was 0.09 (0.001 < r < 0.45), indicating no collinearity. For all predictors, we tested linear effects with outcome measures, except for age. Literature indicated a possible non-linear (quadratic) relationship for the relation between age and our outcome measures . In step 2, predictors with a p value <0.15 in step 1 were entered into the final GEE model. Finally, in step 3, a backward selection (variables deleted when p >0.05) procedure was performed for model optimization. Additionally, to investigate possible moderating effects of age on the models for ADHD symptom severity and overall functioning, interactions between both age (assessed at baseline) and age2 and significant predictors of outcome were added to the final model. When an interaction-effect with age or age2 was significant, the finding was further explored by testing the final model in subsamples subdivided based on age at baseline (<12 years and ≥12 years). Final models were further tested separately for symptom severity of inattention and separately for hyperactivity/impulsivity as outcome measures, to explore whether the model was applicable to both symptom dimensions. Further, as the reliability and validity of the CPRS-R:L is only established for children under 18 years of age, we tested whether results of the final model for current ADHD symptom severity replicated in a subsample of children younger than 18 years. Second, the effect of missing data on the final models for both outcome measures was investigated, testing the final model using only participants with complete data. For our third research question, the first three steps of our GEE model were repeated, except that pharmacological treatment until baseline was replaced by pharmacological treatment until follow-up.