Attention-Deficit/Hyperactivity Disorder Developmental Trajectories related to Parental Expressed Emotion

Case Identification Procedures

All families underwent a multi-gate screening process to establish eligibility and ADHD diagnosis. After completing a brief telephone screen to determine interest and to identify major rule outs (below), families visited the University for a research-based diagnostic-evaluation. A KSAD-S-E (Puig-Antich & Ryan, 1996) was administered to a parent by a trained master's level clinician. Inter-interviewer reliability was kappa>0.70 for all diagnoses with base rate >5% in the sample. Parents and teachers completed the following widely used, well-normed, standardized rating forms (1) The ADHD Rating Scale for DSM-IV (DuPaul et al., 1998), (b) the Conners 3rd Edition (Conners, 2008), (c) the Strengths and Difficulties Questionnaire (Goodman, 2001). Each of these measures has established validity and reliability and all yielded satisfactory intra-scale reliabilities in the current sample. Impairment was scored using the impairment section of the parent and teacher SDQ and the clinician impairment rating from the KSADS.

Children completed a valid (r=.88) and reliable (a=.93; Sattler, 2008) short IQ screen comprising Information, Vocabulary, and Block Design subtests of the Wechsler (2003) Intelligence Scales for Children-4th Ed (WISC-IV) and Wechsler (2005) Individual Achievement Test 2nd Ed Word Reading, Numerical Operations, and Pseudoword Decoding subtests.

All of this information was presented to a clinical diagnostic team in order to establish a best estimate diagnosis (Roy et al., 1997). The team included a board-certified child psychiatrist and a licensed clinical psychologist, both blind to the parental expressed emotion data. They independently reviewed all information to arrive at diagnoses using DSM-IV criteria, taking into careful account age of onset, duration, impairment, cross-informant convergence, and likelihood that another diagnosis would better account for the ADHD symptoms. Their agreement rate was acceptable for ADHD and as well as ODD and for all disorders with base-rate >5% in the study (all k >.80). Disagreements were resolved by discussion. Youth in the current ADHD cohort also meet ADHD criteria according to DSM-5.

Exclusion Criteria

Exclusion criteria include an estimated Full Scale IQ < 75 by our IQ screen, use of long-acting psychotropic medication (e.g., anti-depressants) by parent report, as well as presence of current major depressive episode, lifetime mania or psychosis, pervasive developmental disorder (including autism), or major medical/neurological disorders or injuries as identified on the KSADS, rating scales, or diagnostic team review. Other psychiatric disorders were free to vary. Children were also excluded if they could not be diagnosed with ADHD or designated a qualified control case at baseline (e.g., parent and teacher gave very different ratings; or ADHD symptoms were subthreshold).

ADHD Symptom Change

ADHD symptoms were obtained at all data collection time-points using the parent- and teacher-report ADHD Rating Scale (ADHD-RS; DuPaul, Power, Anastopoulos & Reid, 1998). Parents and teachers rated each of the DSM ADHD symptoms on a 4-point scale ranging from 0 (Never/Rarely) to 3 (Very Often). For children taking stimulant medications, parents and teachers were asked to rate the child's behavior when not taking medication. Symptom counts were determined using standard scoring procedures in which a symptom is considered present if rated as a “2” or “3” (DuPaul et al., 1998).

ODD Symptom Change

Parent-rated ODD symptoms for the growth models were obtained at all data collection time-points using the symptom count from the K-SADS-E ODD module because a symptom checklist was not available. Each symptom was coded as absent (0), maybe (1/2) or present (1) and a total symptom score created in this way. Teacher-rated ODD symptoms were obtained using a DSM-based symptom checklist (using the same scale as the ADHD-RS) in which symptoms rated a “2” or a ‘3′ were counted as present and a total symptom count was created and analyzed.

Expressed Emotion

Parental EE was measured via the Five Minute Speech Sample (FMSS; Magana et al., 1986), which asks parents to describe their child and their relationship for five minutes in their own words without interruption or guidance from the administrator. The FMSS was audio-recorded and blindly coded. Tapes were transcribed to aid in interpretability and accuracy. Two expert, independent raters each blind to ADHD symptom trajectories or parent-teacher ratings, as well as to one another's codes, independently coded the FMSS for EE. The two raters were trained in the UCLA Family Project Lab where the measure originated and have served as co-raters and/or reliability raters across numerous independent samples.

Parental EE was coded in two steps, following established procedures and an established coding manual (Magana et al., 1986). First, the two subscales (Criticism and Emotional Over-Involvement) are coded as low, borderline, or high. These subscale scores are derived from coding specific aspects of the speech sample including the initial statement, the description of the parent-child relationship, critical remarks, and evidence of extreme self-sacrificing behavior or a lack of objectivity. Coding considered the respondent's speech content (via a transcript of the session) and tone (via listening to the session).

Then, a global EE score of “low” or “high” is assigned. A “high” final EE score is obtained only when a parent receives a “high” score in either (or both) of the Criticism and/or Emotional Over-Involvement domains. For the present report, the Criticism and Emotional Over-Involvement subscale scores. 10% of tapes were coded blindly by both raters percentages of agreement were 92% (k=.81) for the overall final score, 87% (k=.78) for criticism, and 84% (k=.74) for emotional over-involvement.

Symptom Growth Models

Age-based growth models were estimated separately for parent- and teacher-reported inattention, H/I, and ODD symptoms in the sample of children with ADHD. Of the 388 ADHD children, 253 were enrolled in a prospective longitudinal study and had multiple time points of data. The remaining children completed a single time point of assessment. Thus for the ADHD youth, 38% (n=147) had one time-point of assessment; 19% had two time points (n=73), and 43% had three time points (n=168). Retention was satisfactory for those for whom a follow-up visit was planned (exceeding 95%).

Models were calculated using MPLUS v.7.2 (Muthen & Muthen, 2014). Where variability in symptom trajectories was reasonably suspected (operationalized as variability around the slope with p<.15 to avoid Type II error, which was the priority at that preliminary stage of analysis), an unconditional linear latent class growth model (LCGA; Muthen & Muthen, 2000) with no predictors was fit. LCGA defines three latent factors (class, intercept, and slope) from a structural equation modeling framework. The latent class variable was regressed onto the intercept and slope factors to examine differences between the latent trajectory classes in the sample. Intercepts and slopes within each class were held equal. Two-, three-, and four- and five-class LCGA models were fit to the data and the best-fitting model was selected based on comparison of standard fit indices (BIC, CFI, RMSEA) and the parametric bootstrapped Likelihood Ratio Test (LRT), which assesses whether the k-class model significantly improves on the k-1 class model (Asparouhov & Muthén, 2012). Missing data modeling was handled using Full Information Maximum Likelihood.

EE Domain Analyses

EE analyses focused on a randomly selected subset of the longitudinal sample, for whom EE data were collected and coded at two time points one year apart (n=208 with ADHD, n=127 controls), using logistic regression analyses. To code stability of EE, Criticism, and Emotional Over-Involvement were dummy coded separately as present/absent for each variable at each of the two assessments. Children rated at high in a particular domain (i.e., EE, Criticism, Emotional Over-Involvement) at both assessments were considered “stable high;” those with high scores at only one assessment were “ever high;” and the others were “never high” in that particular domain of interest. To examine EE domain effects on longitudinal symptom trajectories, the 208 ADHD participants with two time-points of EE data were subjected to conditional growth models using EE domain (i.e., criticism, emotional over-involvement) dummy codes as predictors of the above described symptom trajectory classes. For ease of presentation only results for parental criticism and emotional over-involvement are presented. Results for “overall” EE scores largely mirrored that of the criticism domain (available from the first author).

Selection of covariates

The following variables were considered for use as either invariant or time-varying covariates as appropriate: age at assessment, sex of child, ethnicity and race of child, child stimulant medication use, two-parent family vs single-parent family status, family income, child estimated full-scale IQ, comorbid diagnoses, and sex of respondent for FMSS. Over 80% of FMSS respondents were women (mothers or female caregivers) at both time points. Covariates were selected for inclusion if they either (a) differed by EE rating or (b) differed by latent-trajectory class using a criterion of p<.05. This resulted in inclusion of the following covariates in all diagnostic group comparisons: age at assessment, sex, two-parent family vs single-parent family, income, IQ, and comorbid diagnoses (anxiety, mood disorders, ODD, and CD). For trajectory class analysis, it resulted in inclusion of IQ, comorbid diagnoses, and income. Respondent sex was also treated as a covariate in all analyses of parental EE domain. ODD was treated as a covariate in ADHD symptom trajectory analyses, and ADHD was treated as a covariate in ODD trajectory analyses.

ADHD diagnosis and expressed emotion

High parental EE Criticism was more likely among children with an ADHD diagnosis at each time point Table 1. Logistic regression (with covariates described above), showed that this result held for at time 1 (Cox and Snell R2=0.088, beta=0.270, p<.001) and Time 2 (Cox and Snell R2=0.053, beta=0.128, p<.001). Parental emotional over-involvement was not associated with ADHD diagnosis at Time 1, but was more likely among children with ADHD at Time 2 (Cox and Snell R2=0.013, beta=0.061, p=.021). Stability in high parental criticism was more common in the ADHD than non-ADHD group (21.6% vs 1.6%) even after controlling for covariates and comorbid anxiety, mood, and disruptive behavior diagnoses (i.e., ODD and CD; Cox and Snell R2=0.072, beta=0.076, p=.003), suggesting a specific association with ADHD, despite the prior literature linking EE to externalizing symptomology. ADHD and non-ADHD children did not differ with respect to stable high emotional over-involvement (p=.51).

Parent-rated Hyperactivity-Impulsivity (H/I)

For parent-rated H/I the unconditional model yielded a best-fitting four trajectory solution (4-class model: Adjusted BIC= 3775.2, Parametric Bootstrapped LRT test for 3 versus 4 classes p<.001; 5-class model: Adjusted BIC= 3831.5, Parametric Bootstrapped LRT test for 4 versus 5 classes p> 0.05). These classes are depicted in Figure 1. They were labeled as (1) Low H/I (hereafter, “Low), (2) Moderate H/I decreasing (hereafter, “Moderate”), (3) Remitting (high symptoms but improving), and (4) Persistent (high symptoms and remained high). In order to present results concisely, data is provided by symptom trajectory class by parent-rated H/I, teacher-rated inattention, and teacher-rated ODD in Tables 2-4, respectively, along with significant demographic and comorbid diagnostic data, while demographic and comorbid diagnostics that did not differ significantly by trajectory group is available in online supplemental Tables S1-S3. Thus, symptom trajectories and significant descriptive and diagnostic group differences are in Table 2 with non-significant differences in Table S-1.

Teacher-rated inattention

Here, the unconditional model yielded a best-fitting three trajectory solution (3-class model: Adjusted BIC= 3535.6, Parametric Bootstrapped LRT test for 2 versus 3 classes p<.001; 4-class model: Adjusted BIC=3782.1, Parametric Bootstrapped LRT test for 3 versus 4 classes p> 0.05). This solution is depicted in Figure 2, and the classes were labeled as (1) Low inattention (hereafter, “Low”), (2) Remitting, and (3) Persistent. Symptom trajectories and significant descriptive and diagnostic group differences are in Table 3 with non-significant differences in Table S-2.

Teacher-rated ODD

Again, a best-fitting three trajectory solution emerged (3-class model: Adjusted BIC=2624.5, Parametric Bootstrapped LRT test for 2 versus 3 classes p<0.001; 4-class model: Adjusted BIC=3782.1, Parametric Bootstrapped LRT test for 3 versus 4 classes p>.05). These are depicted in Figure 3. They were accordingly labeled as (1) Low ODD, (2) Remitting, and (3) “Worsening” (high symptoms and worsened over time Symptom trajectories and significant descriptive and diagnostic group differences are in Table 4 with non-significant differences in Table S-3.

Comparison of solutions

Teacher-rated inattention trajectories did not significantly overlap with either the parent-rated H/I (p=.645) or teacher-rated ODD (p=.211) trajectory classes, suggesting that these are distinct groups of children as rated by both parents and teachers, rather than the same groups being captured three ways. However, there was significant overlap between parent-rated H/I and teacher-rated ODD trajectory classes (X²(1)=19.99, p=.003). This overlap in turn was driven by a group of children (N=170, 43.8%) who both reporters identified as relatively low in externalizing symptoms. Thus, associations of EE may be distinct with ADHD versus ODD trajectories. We next examined that possibility.

Parent-rated H/I groups and parent EE

LCGA including covariates was used to test associations of the H/I trajectory groups with parental criticism coded as stable or not stable (see Methods). Class 1 (low), Class 2 (moderate), and Class 3 (remitting) were similar to each other (all p>.55), but each differed from Class 4 (persistent), which was more likely to be coded as stable high criticism; respective parameter estimate versus class 4 were 0.561, se=0.233, p=.016, parameter estimate=0.763, se=0.280, p=.009, and parameter estimate=0.631, se=0.220, p=.012, respectively, sample-size adjusted BIC=2504.13. In contrast, the dummy code for “Ever high” criticism (i.e., at time point 1 OR 2 instead of 1 AND 2) failed to detect group effects (all p>.12). Thus, it was not the mere presence of parental criticism at either time-point that accounted for differences in persistence versus decreasing symptoms of parent-rated H/I, but rather stability in criticism across time. With respect to emotional over-involvement, no differences were observed at any time point (all p>.43) or with respect to stability (all p>.24) for any of the H/I trajectory groups. This suggests that the overall EE effect is due to parental criticism rather than over-involvement.

While the preceding was interesting it could be due simply to negative halo effects by parents who experience negative affect and see their child as highly hyperactive also. To evaluate this we looked at teacher data two ways. First, parent H/I groups were compared on teacher rated hyperactivity-impulsivity (recall that these were not able to be divided into trajectories by teacher-rating). This analysis confirmed the parent-based results: parent-rated H/I trajectory classes significantly differed on teacher-rated H/I symptoms with changes in the anticipated directions based on trajectory classes (all p<.002) and ratings over time were also similar (details are presented in Table 2). The second check was we examined the parent EE measure in relation to teacher-rated dimensional symptoms of hyperactivity. The result was also confirmatory. Stable high parental criticism was associated with more teacher-rated H/I symptoms at all assessment points (all p<.05).

Teacher-rated inattention groups and parental EE

Next, we examined teacher-rated inattention trajectory classes. These did not differ in any respect with regard to parent high EE criticism at either time point or criticism stability, over-involvement at either time point, or over-involvement stability (all p>.3; see details in Table 3).

Teacher-rated ODD groups and parent EE

Lastly, we examined teacher-rated ODD trajectory classes. Teacher-rated ODD trajectory classes did not differ with respect to parent criticism stability (all p>.12), sample-size adjusted BIC= 1785.077. However, when looking at “Ever critical,” Class 1 (low) and Class 2 (remitting) were similar to each other (p=.52), but both differed from Class 3 (worsening), which was more likely to be coded as ever critical; parameter estimate=1.115, se=0.386, p=.004 and parameter estimate=1.352, se=0.5, p=.010, respectively, sample-size adjusted BIC= 1774.97. Specifically, teacher-rated worsening ODD was significantly associated with parental criticism at Time 1, but not at Time 2 (see Table 4). No teacher-reported ODD trajectory based group differences were observed with respect to emotional over-involvement at Time 1, Time 2, or stability over time (all p>.33).

Again, we checked results across rater

This confirmed that those in the teacher-rated ODD low, remitting, and worsening groups significantly different parent-rated ODD symptoms (all p<.001) with the worsening group having the highest numbers of symptoms and the low group having the lowest number of symptoms at all time-points (Table 4). Further, high parental criticism was associated with more parent-rated ODD symptoms at all assessment points (all p<.05). Thus, it appears that the effects of parental criticism held across raters as well.