Evaluation of Attention-Deficit/Hyperactivity Disorder Medications, Externalizing Symptoms, and Suicidality in Children

Key Points Question Are there associations among attention-deficit/hyperactivity disorder (ADHD) pharmacotherapy, externalizing symptoms, and childhood suicidality (ie, ideation or attempts)? Findings In this cohort study of 11 878 children aged 9 to 10 years at baseline, externalizing symptoms showed expected associations with suicidality. ADHD pharmacotherapy was associated with less suicidality in children with externalizing symptoms, specifically among those with substantial symptoms burden, both at the baseline and 1-year follow-up assessment. Meaning These findings suggest that in children with substantial externalizing symptoms, ADHD medication use may be associated with less suicidality, in addition to its known clinical benefits in treating externalizing symptoms.


Sensitivity analyses
We conducted several sensitivity analyses for the main model. To evaluate robustness of results, we included established risk and protective factors for suicidality previously described in ABCD. 1 To probe the effects of our externalizing symptoms exposure variable, given established heterogeneity in the clinical phenotype, 2 we ran the main model using combinations of different symptom dimensions (ADHD including/excluding inattention, ODD and CD). To address effect of our choice of the dependent variable (child report suicidality), we ran two models using two different suicide measures: child report suicidal ideation and parent report suicidality. To control for depression/anxiety, we included in the main model KSADS-based depression/anxiety diagnoses. As an alternative to our handling of missing data using listwise deletion in the main analyses, we conducted sensitivity analyses running all models using an imputed dataset (created using the R library Amelia). To account for potential family-relatedness effects, we analyzed data including only one participant from a family, or excluding any related children, similar to previous work on ABCD suicidality. 1 Lastly, in attempt to better estimate clinical presentations where ADHD medication is especially associated with less suicidality, we matched participants receiving ADHD medications with high symptom loads at 3 thresholds (>1/2/3 SDs) with untreated controls and compared suicidality rates.

Exploratory analyses
To evaluate the moderating effect of sex on the main study question (interaction effect of symptoms and ADHD medication), we ran the main model adding the 3-way © 2021 Shoval G et al. JAMA Network Open.
interaction term of externalizing symptoms x ADHD medications x sex (including all of the 2-way interactions: symptoms x ADHD medications; externalizing x sex; ADHD medications x sex). A significant 3-way interaction would indicate that the moderating effect of ADHD medication on the relationship between externalizing symptoms and suicidality is different between the sexes. In addition, we ran sex-stratified analyses of the main model. This approach was also employed to examine effects in the children receiving any AD/AP medication.
To explore potential differences in the moderating effects of different ADHD medication classes on the association between externalizing symptoms and suicidality, we also ran the main analysis as four separate models (i.e., one for each medication class Past The externalizing symptoms variable was defined as the sum of externalizing hyperactivity symptoms from ADHD, and all symptoms of ODD and conduct disorder items in the K-SADS. Present and past were combined as one symptom; combination of 0 and NA was defined as 0. The variable was scored on a discrete scale from 0 -34 based on presence of symptoms in the Attempt current 29 0.26% Prevalence of suicidal ideation and attempts are shown at baseline assessment and at 1-year follow-up for the total sample. Total baseline sample included N=11 878. Missing data for suicidality measures was 0.6% at baseline. Longitudinal data on suicidality was available for n= 11 077 (missing for 6.7% of the original cohort). Current refers to past 2 weeks. We considered positive history of depression for every participant who was given any of the above diagnoses (coded as 1 in any of the ABCD elements above). History of anxiety was determined similarly using the above anxiety diagnoses.  197 0.228 a Model included age, sex, race, ethnicity, parents' education and marital status. The 3-way interaction was derived from a separate model including main effects and 2-way interactions of all the variables included in the 3-way interaction (i.e., symptoms x ADHD medications; symptoms x AD/AP medications; ADHD medications x AD/AP medications) b To improve interpretability of externalizing symptoms main effect, ADHD medication variable was regressed out of the sum of externalizing symptoms (resulting in a z-score), hence the obtained OR reflect a change in odds for a change in 1SD of externalizing symptoms.

eTable 5. Antidepressant-and Antipsychotic-Treated Children: Stratified Analysis and 3-Way Interaction Model
Abbreviations: ADHD = Attention Deficit Hyperactivity Disorder; AD = antidepressants; AP = antipsychotics; ext. symp. = externalizing symptoms; SE= standard error; OR= odds ratio; LL= lower limit; UL= upper limit; CI = confidence interval. Missing data was 3.9% for the interaction model, 3.7% in stratified model analyzing children without AD/PD and 5.1% in children receiving AD/AP. , race, ethnicity, parents' education and marital status, and whether child was receiving antidepressants or antipsychotics. The 3-way interactions are derived from a separate model including main effects and 2-way interactions of all the variables included in the 3-way interaction (i.e., symptoms x ADHD medications; symptoms x sex; ADHD medications x sex) b To improve interpretability of externalizing symptoms main effect, ADHD medication variable was regressed out of the sum of externalizing symptoms (resulting in a z-score), hence the obtained OR reflect a change in odds for a change in 1SD of externalizing symptoms. Abbreviations: ADHD = Attention Deficit Hyperactivity Disorder; AD = antidepressants; AP = antipsychotics; ext. symp. = externalizing symptoms; SE= standard error; OR= odds ratio; LL= lower limit; UL= upper limit; CI = confidence interval. CI = confidence interval. Missing data was 3.9% for the interaction model, 3.9% in stratified model analyzing boys and 3.8% in model analyzing girls. Participants who were lost to follow up did not differ in terms of age and sex and did not differ in rates of ADHD medication treatment and in suicidality rates. The participants lost to follow up differed in race/ethnicity, socioeconomic status, and had greater externalizing symptomatology. Abbreviations: ADHD = Attention Deficit Hyperactivity Disorder; MPH = methylphenidate derivative; meds = medications; SD = standard deviation. Weekend screen time 4.6 3.6 4.5 3.6 5.8 4.2 <.001 Because children that did and did not receive ADHD medication were significantly different in the total sample, we compared subgroups of children with high externalizing factors (>1, 2, and 3 standard deviations from the mean) and use of antidepressant or antipsychotic medications (yes/no). From each of these subgroups, equal amounts of children who did and did not receive ADHD medication were matched based on multiple demographic and clinical variables. a For all variables, missing data was smaller than 1.27% (151 participants out of the 11,878 participants at baseline ABCD assessment). b T-test and Chi-square comparisons for continuous and binary measures, respectively Abbreviations: ADHD = Attention Deficit Hyperactivity Disorder; AD/AP = antidepressant or antipsychotic medication; SD = standard deviation; NA = not available. Weekend screen time 7.2 5.7 6.4 3.6 0.498 b T-test and Chi-square comparisons for continuous and binary measures, respectively c Total participants in >3SD in the children receiving AD/AP included 32 participants of which only 7 participants did not take ADHD medication, therefore no matching was done on this group. Abbreviations: ADHD = Attention Deficit Hyperactivity Disorder; AD/AP = antidepressant or antipsychotic medication; SD = standard deviation; NA = not available. Binary logistic regression model with suicidality as the dependent variable testing main effect and interaction of externalizing symptom count (standardized) and ADHD medication (binary variable) and co-varying for age, parents' education, marital status, race (White, Black, Asian, other), Hispanic ethnicity, whether the child was receiving antidepressants or antipsychotics, and two binary variables indicating history of depression or anxiety diagnoses, as determined in the K-SADS interview. b Longitudinal model included current suicidality at the 1-year follow up assessment as the dependent variable, and co-varied for all variables as described above in addition to baseline suicidality and time between baseline and follow-up assessment. c For the interaction model we used one-tailed test for the interaction effect based on anticipated direction of medication protective effect as observed in cross-sectional model findings, and based on preregistered hypothesis for protective effect of ADHD medication in children with high externalizing symptoms. d The total count of externalizing symptoms was z-scored to allow easier interpretation of main effect, such that the obtained OR reflect a change in odds for a change in 1SD of externalizing symptoms. e Interaction term was introduced in a separate model. Abbreviations: ADHD = Attention Deficit Hyperactivity Disorder; externalizing = externalizing symptoms; meds = medications; SE = standard error; OR = odds ratio. Binary logistic regression model with suicidality as the dependent variable testing main effect and interaction of externalizing symptom count (standardized) and ADHD medication (binary variable) and co-varying for age, parents' education, marital status, race (White, Black, Asian, other), Hispanic ethnicity, and whether child was receiving antidepressants or antipsychotics. b Longitudinal model included current SI at the 1-year follow up assessment as the dependent variable, and co-varied for all variables as described above in addition to baseline SI and time between baseline and follow-up assessment. c For the interaction model we used one-tailed test for the interaction effect based on anticipated direction of medication protective effect as observed in cross-sectional model findings, and based on preregistered hypothesis for protective effect of ADHD medication in children with high externalizing symptoms. d To improve interpretability of externalizing symptoms main effect, ADHD medication variable was regressed out of the sum of externalizing symptoms (resulting in a z-score), hence that the obtained OR reflect a change in odds for a change in 1SD of externalizing symptoms. e Interaction term was introduced in a separate model. Abbreviations: ADHD = Attention Deficit Hyperactivity Disorder; meds = medications; externalizing = externalizing symptoms; SE = standard error; OR = odds ratio; SD = standard deviation. Externalizing X ADHD meds c -0.37 0.087 18.054 0.69 <.001 Note-Parent report of suicidality was not released by ABCD for the 1-year follow-up longitudinal assessment. a Binary logistic regression model with suicidality as the dependent variable testing main effect and interaction of externalizing symptom count (standardized) and ADHD medication (binary variable) and co-varying for age, parents' education, marital status, race (White, Black, Asian, other), Hispanic ethnicity, and whether child was receiving antidepressants or antipsychotics. b To improve interpretability of externalizing symptoms main effect, ADHD medication variable was regressed out of the sum of externalizing symptoms (resulting in a z-score), hence that the obtained OR reflect a change in odds for a change in 1SD of externalizing symptoms. c Interaction term was introduced in a separate model. Abbreviations: ADHD = Attention Deficit Hyperactivity Disorder; meds = medications; externalizing = externalizing symptoms; SE = standard error; OR = odds ratio; SD = standard deviation. a Binary logistic regression model with suicidality as the dependent variable testing main effect and interaction of externalizing symptom count (standardized) and ADHD medication (binary variable) and co-varying for age, parents' education, marital status, race (White, Black, Asian, other), Hispanic ethnicity, and whether child was receiving antidepressants or antipsychotics. b Model 2 is similar to model 1 in addition to the four risk and protective factors described previously: [Janiri et al. 2020]: family conflict, weekend screen time, parental supervision and positive school involvement. c To improve interpretability of externalizing symptoms main effect, ADHD medication variable was regressed out of the sum of externalizing symptoms (resulting in a z-score), hence that the obtained OR reflect a change in odds for a change in 1SD of externalizing symptoms. d Interaction term was introduced in a separate model. e Longitudinal model included current suicidality at the 1-year follow up assessment as the dependent variable, and co-varied for all variables as described above in addition to baseline suicidality and time between baseline and follow-up assessment. f For the interaction model we used one-tailed test for the interaction effect based on anticipated direction of medication protective effect as observed in cross-sectional model findings, and based on preregistered hypothesis for protective effect of ADHD medication in children with high externalizing symptoms. Abbreviations: ADHD = Attention Deficit Hyperactivity Disorder; meds = medications; externalizing = externalizing symptoms; SE = standard error; OR = odds ratio; SD = standard deviation. Binary logistic regression model with suicidality as the dependent variable testing main effect and interaction of externalizing symptom count (standardized) and ADHD medication (binary variable) and co-varying for age, parents' education, marital status, race (White, Black, Asian, other), Hispanic ethnicity, and whether child was receiving antidepressants or antipsychotics. b To improve interpretability of externalizing symptoms main effect, ADHD medication variable was regressed out of the sum of externalizing symptoms (resulting in a z-score), hence that the obtained OR reflect a change in odds for a change in 1SD of externalizing symptoms. c Interaction term was introduced in a separate model. The main model was repeated twice to account for the sets of related children in the total sample. In the first, all related children were excluded. In the second, one child from each family was randomly selected for inclusion. Abbreviations: ADHD = Attention Deficit Hyperactivity Disorder; meds=medications; externalizing = externalizing symptoms; SE = standard error; OR = odds ratio; SD = standard deviation.