Attachment Representation Moderates the Effectiveness of Behavioral Parent Training Techniques for Children with ADHD: Evidence from a Randomized Controlled Microtrial

The current sample comprised 74 children (age 4–12) whose parents participated in a randomized controlled microtrial (Hornstra et al., 2021, N = 92) in which we investigated effects of antecedent-based and consequent-based techniques. Data from 18 children of the original sample were missing because of technical problems. Characteristics of the current subsample of children and their parents are presented in Table 1. The inclusion criteria for the study were (1) a Diagnostic and Statistical Manual of Mental Disorders-5 (DSM-5) based classification of ADHD, confirmed with the Diagnostic Interview Schedule for Children-IV, parent interview (DISC-IV; Shaffer et al., 2000), which we adapted to the DSM-5; (2) at least four parent-rated problem behaviors, as assessed with extensive list of 29 different inattentive, hyperactive, impulsive and oppositional defiant behaviors (Hornstra et al., 2021; Van den Hoofdakker et al., 2007) scored as three or higher on a five-point Likert scale, ranging from 1 (not severe) to 5 (extremely severe); (3) the child having an IQ > 70. This was derived from the patient file, or if not present measured with two subtests (‘Vocabulary’ and ‘Block design’; Sattler, 2008) of the Wechsler Intelligence Scale for Children (WISC-III-NL) or, if the child was younger than six years old, the Wechsler Preschool and Primary Scale of Intelligence (WPSSI-III-NL); (4) no current psychotropic medication use. If the child previously used psychotropic medication, he/she had to be off medication at least 4 weeks before the start of the study. Exclusion criteria were (1) the child having a diagnosis of autism spectrum disorder (as reported by the parent or derived from the patient file) or conduct disorder (derived from the adapted DISC-IV (Shaffer et al., 2000) or the patient file), because these children might have more intensive treatment needs than a two session experimental treatment, and to examine the effects of the separate techniques in a fairly homogeneous group, (2) the caregivers received BPT in the past year, and (3) the child was not living in the same house during the weekdays (because our primary outcome measure had to be reported by the same caregiver).

This study was registered at the Dutch Trial Register: https://​www.​trialregister.​nl/​trial/​6011. Medical ethical approval of the microtrial was waived by the Medical Ethical Committee of the University Medical Center Groningen (UMCG; METc 2016/197). After obtaining written informed consent, parents were randomized (using a random number generator, ratio of 1:1:1) to one of the two intervention conditions (i.e., antecedent- or consequent-based), or a waitlist control condition. Attachment representation was assessed prior to randomization. Outcome measures were assessed at three time points: prior to randomization (T0), during the week immediately after the intervention or the waiting period (T1), and three weeks after the intervention or waiting period (T2). After T2, all parents could receive care as usual. Parents randomly assigned to the waitlist control condition could receive one or both of the intervention conditions of choice of the current study at T2 as well. Since there were no guidelines for reporting on microtrials, we used the CONSORT-SPI 2018 extension (an extension of the main Consort statement; Montgomery et al., 2018) for reporting on randomized controlled trials of social and psychological interventions. The CONSORT-SPI 2018 differs from the standard CONSORT method as it extends 9 of the 25 items including background and objectives, trial design, participants, interventions, statistical methods, participant flow, baseline data, outcomes and estimation, and funding. Parents received a small compensation (€10) for participating in the study. For more details concerning the design of the study, see Hornstra et al. (2021).

There were two intervention conditions, one in which parents were trained in antecedent-based techniques (antecedent-based condition), and one in which parents were trained in consequent-based techniques (consequent-based condition). Both interventions were delivered individually and both caregivers (when applicable) were encouraged to engage in the sessions. The primary caregiver (i.e., the caregiver that spends most time with the child) always participated in the sessions. Each intervention condition consisted of two sessions (two hours each), and was provided if possible, in two consecutive weeks, by experienced psychologists who had comprehensive experience with ADHD and parent training and completed postgraduate training in behavior therapy. Both interventions were targeted to preselected problem behaviors from a list of 29 ADHD-related behaviors (including inattentive, hyperactive, and impulsive symptoms, and oppositional defiant behaviors; Hornstra et al., 2021; Van den Hoofdakker et al., 2007) in a specific situation (using the Home Situations Questionnaire; Breen & Altepeter, 1991) (see “Outcome Measure”).

The first session consisted of the following steps: (1) psycho-education about ADHD with specific information about a) how stimuli elicit behaviors, how executive functioning problems in children with ADHD may lead to problems in adapting their behavior to stimuli, and how antecedent-based techniques can be used to elicit desired behavior and prevent undesired behavior (i.e., antecedent-based condition), or b) how consequences influence behavior, how altered reward sensitivity in children with ADHD may cause problems in how their behavior is influenced by the environment, and how consequent-based techniques can support them by changing the consequences of behavior (i.e., consequent-based condition) (Van der Oord & Tripp, 2020); (2) the selection of one of the problem behaviors in a specific situation, based on frequency, severity, changeability, and burden to parents; (3) a detailed behavioral analysis of the selected behavior, and defining a desired target behavior, together with the parents; (4) a tailored intervention plan, consisting of a) antecedent-based techniques such as defining rules, giving clear instructions, anticipating misbehaviors, and providing routines, or b) consequent-based techniques such as planned ignoring, praise, rewards, and punishment, based on the function of the behavior (Virués-Ortega & Haynes, 2005); (5) practicing with the intervention plan, either through roleplay or visualization; (6) implementing the intervention plan by the parents at home for one week, after which the second session took place. The second session started with an evaluation of the previous week and, if necessary, adapting the intervention plan from the first session. After that, steps two to five were repeated for a different behavior in a specific situation. A more detailed description of the interventions can be found elsewhere (Hornstra et al., 2021).

Data was analyzed on an intention-to-treat basis. ANOVA’s (continuous variables) and chi-squared tests (categorical variables) were used to assess differences between the three conditions in demographic and baseline characteristics, using the Statistical Package for the Social Science (SPSS, version 26). We conducted multilevel analyses (mixed models), which takes missing data into account (Twisk et al., 2013), using Stata, version 16 (StataCorp, 2019). We distinguished three levels: outcomes (level 1) nested in participants (level 2), nested in therapists (level 3). Random intercepts at therapist level were included if the Likelihood Ratio Test showed a significant improvement of the model fit. Power-analysis of an ANCOVA with baseline score as covariate and three groups revealed that we needed a total sample of 90 children, and thus 30 children per group, to detect intervention effects on our primary, proximal outcome (alpha = 0.05, power = 0.80, df = 1). To assess main effects of the condition, Condition (AC, CC, control) was included as between subjects factor and time (T1, T2) as within subjects variable. T0 scores were added as a fixed factor to control for baseline differences in our outcome measure. Second, moderator effects were assessed by examining interactions between the effect of condition and attachment representations (continuous). For each variable (secure, ambivalent, avoidant, disorganized), the condition by variable interaction was added to the model to assess whether the change over time differed between levels of the potential moderator. Effects were averaged over T1 and T2, provided that problem behaviors remained stable between T1 and T2. To address the first aim of our study, we assessed whether attachment representation moderated the effect on the primary outcome when comparing both active conditions together to the control condition (i.e., brief BPT versus waitlist). To address the second aim, we tested if attachment representation moderated the effect on the primary outcome when comparing the active conditions with each other (i.e., antecedent-based condition versus consequent-based condition). Since these moderator analyses were hypotheses-generating, correction for multiple testing was not applied (Bender & Lange, 2001; Rothman, 1990). We conducted additional analyses with IQ and age as covariates. In case of a moderating effect, we assessed if conditions were differentially effective for different levels of the moderator based on median‐split. The significance level for all analyses was set at α = 0.05, two-tailed. Effect sizes were computed by dividing the regression coefficient by the (pooled) standard deviation.

In Supplementary Information, Table B, the effects of the intervention and the results of the moderator analyses are depicted. First, in line with previous findings in our larger sample Hornstra et al., 2021), we found a significant decrease in daily rated problem behaviors for both active conditions, compared to the control condition, with medium effect sizes (antecedent-based condition: d = 0.73, consequent-based condition: d = 0.56). The two active conditions did not differ from each other with regard to reduction in daily reported problem behaviors. Random intercepts at the ‘therapists’ level did not improve the model, therefore all models were reduced to two levels (observations clustered in children). Problem behaviors remained stable from T1 (i.e., post-intervention) to T2 (i.e., three week follow-up) within all conditions.

None of the attachment representations (secure, avoidant, ambivalent, disorganized) moderated treatment outcomes when training (both antecedent- and consequent based techniques) was compared to the waitlist control condition (see Supplementary Information, Table B for an overview of all analyses).

This was the first study to examine the influence of attachment representations on the effectiveness of behavioral techniques in a clinical sample of children with ADHD. A major strength of the study was the use of a story stem procedure to measure attachment representations, instead of other procedures such as self-reported questionnaires or observational methods. Representational measures such as the story stem procedure take possible biases such as social desirability, and the positive illusory bias of children with ADHD (i.e., the tendency to overestimate one’s own capacities; Volz-Sidiropoulou et al., 2016) into account. Given the broad age group of our study (4–12 years), the story stem procedure was an excellent option to measure attachment representations (Bosmans & Kerns, 2015), and seems to be more suitable than questionnaires for children with ADHD (Hornstra et al., 2019). The story stem procedure focused on the primary caregiver in the stories, and outcome was rated by the same primary caregiver. Therefore, the informant for the outcome measure and the person of focus with regard to the attachment representation match. Nonetheless, the current study should be interpreted with some limitations in mind. First, as mentioned above, we did not have a baseline measurement of parenting, including behaviors associated with child attachment (e.g., parental responsivity/sensitivity). Future studies could add this to gain more insight into the mechanism underlying the differential effects of BPT techniques. Second, parents received the training, but also rated our primary outcome. Therefore, an overestimation of the rated effects is a possible limitation (Daley et al., 2014). However, the use of EMA improves the ecological validity of the current study, as it reduces potential recall or memory bias (Russell & Gajos, 2020). Third, the follow up period was relatively brief (i.e., three weeks). Thus, it is not clear if moderative results extend to longer term effects of our intervention. Fourth, our sample was not explicitly selected on the base of insecure attachment representations, and approximately 50% of the children included in this study were securely attached. This could limit the generalizability of our findings to other, more vulnerable populations. Still, the percentage of children with an insecure or disorganized attachment representation was significantly and considerably higher than in typically developing children (Dekkers et al., 2021b), emphasizing the importance of attachment representations of children with ADHD as a potential moderator of treatment success. Fifth, the consequent-based techniques that were used for the individualized intervention plans could include a range of different techniques (i.e., planned ignoring, praise, rewards, and punishment). Potentially, certain consequent-based techniques, for example planned ignoring (a negative punishment technique) versus rewards, could be differentially effective for children (Yee & Shiota, 2015). Further research has to be conducted to examine exactly which techniques are most helpful and whether there is a differential effect of certain types of techniques for subgroups of children based on attachment representation, for example, punishment versus reward-based techniques. Sixth, because we assessed attachment representations of the children in the current sample at baseline, it remains unknown whether attachment changed following our brief BPT. Standard treatment strategies that target ADHD-symptoms seem to not in itself improve attachment security (Darling Rasmussen et al., 2021). The learning theory of attachment can help integrate attachment theory into clinical practice (Bosmans et al., 2020). Attachment based elements can strengthen regular antecedent- and consequent-based techniques in BPT (Bosmans et al., 2022). Evidence from parenting interventions in young children that specifically focus on parental sensitivity and the parent–child relation (e.g., Parent–Child Interaction Therapy [PCIT], Video-feedback Intervention to promote Positive Parenting and Sensitive Discipline [VIPP-SD]) show a beneficial effect of parenting programs on children’s attachment (Allen et al., 2014; Bakermans-Kranenburg et al., 2003; Juffer et al., 2017). This is in line with findings that children’s trust in caregivers can catch up due to improved parenting experiences (Bosmans et al., 2019; Verhees et al., 2020). However, in the current study, we did not include relationship building components to our training, which may have influenced our findings regarding the moderative effects of attachment representation. Relatedly, in the current study we assessed attachment representations towards the primary caregiver. In a minority of the cases, there is a discordance between attachment between both caregivers (Dagan & Sagi-Schwartz, 2018; Di Folco et al., 2020), and these discordant attachment representations can influence differential developmental outcomes (Dagan et al., 2022). Assessment of the attachment relationships towards both caregivers may provide a more complete picture of the influence of attachment on behavioral strategies. Seventh, in the current study we found an influence of attachment representations on the effectiveness behavioral techniques in a clinical sample of children with ADHD. Nevertheless, the question remains whether this is a unique effect within children with ADHD or a more general effect of attachment representations in broader samples of children with behavioral problems. Eighth, the current study was not a priori powered for moderation analyses. Our power-analysis was based on detection of differences between the intervention conditions and control condition on our outcome measure. Therefore, although generally microtrials can have greater power to detect moderation with smaller sample sizes (Howe et al., 2010; Howe & Ridenour, 2019), power to detect possible moderation effects in this study may have been too low, possibly leading to small effects remaining undetected. Lastly, it remains unknown which processes and mechanisms of change are involved in the differential effects of behavioral techniques, and future studies on the current topic, also in other samples, are therefore required (Kazdin, 2007).