Emotional Dysregulation in Emerging Adult ADHD: A Key Consideration in Explaining and Classifying Impairment and Co-Occurring Internalizing Problems

Abstract

Objective:

The current study sought to clarify and harness the incremental validity of emotional dysregulation and unawareness (EDU) in emerging adulthood, beyond ADHD symptoms and with respect to concurrent classification of impairment and co-occurring problems, using machine learning techniques.

Method:

Participants were 1,539 college students (M_age = 19.5, 69% female) with self-reported ADHD diagnoses from a multisite study who completed questionnaires assessing ADHD symptoms, EDU, and co-occurring problems.

Results:

Random forest analyses suggested EDU dimensions significantly improved model performance (ps < .001) in classifying participants with impairment and internalizing problems versus those without, with the resulting ADHD + EDU classification model demonstrating acceptable to excellent performance (except in classification of Work Impairment) in a distinct sample. Variable importance analyses suggested inattention sum scores and the Limited Access to Emotional Regulation Strategies EDU dimension as the most important features for facilitating model classification.

Conclusion:

Results provided support for EDU as a key deficit in those with ADHD that, when present, helps explain ADHD’s co-occurrence with impairment and internalizing problems. Continued application of machine learning techniques may facilitate actuarial classification of ADHD-related outcomes while also incorporating multiple measures.

Keywords

ADHD emotional dysregulation impairment machine learning emerging adults

ADHD is characterized by symptoms of inattention and hyperactivity/impulsivity (American Psychiatric Association [APA], 2022), with the disorder being diagnosed in up to 5% of emerging adults (Song et al., 2021). ADHD in emerging adulthood (aged 18–29 years; Arnett, 2000) has been linked to functional impairment across academic, occupational, and social domains, as well as increased risk for co-occurring internalizing problems including depression and anxiety (Anastopoulos et al., 2016). Despite trends showing striking increases in ADHD’s prevalence (Chung et al., 2019), standardized assessment approaches that accurately capture the disorder’s expression and impact across multiple domains remain lacking, particularly for emerging adults. Such approaches, while also helping to explain ADHD’s co-occurrence with impairment and internalizing problems, could have benefits both in terms of transdiagnostic assessment and tailored treatment; specifically, they may help shift the focus from symptom reduction alone to enhancing quality of life and addressing co-occurring issues. Ongoing research in this area has primarily focused on ADHD in childhood, with studies using contemporary computational tools such as adaptive testing and machine learning algorithms to improve prediction of ADHD and its key correlates (e.g., functional impairment; Caye et al., 2019; Goh, Elkins, et al., 2023; Goh, Martel, et al., 2023; Mooney et al., 2023). In contrast, little work has focused on understanding the nature and improving classification of impairment and co-occurring problems in emerging adults with ADHD, despite a growing body of research underscoring their vulnerability and unmet needs across a wide and heterogeneous range of functional domains (Kosheleff et al., 2023).

Limitations in ADHD’s Symptom List During Emerging Adulthood

Commonly used assessment tools for ADHD in adulthood rely primarily on the Diagnostic and Statistical Manual of Mental Disorders, 5th Edition, Text Revision (DSM-5-TR) ADHD symptom list (APA, 2022). Yet, though ADHD is considered to be a chronic, lifelong concern (Faraone et al., 2021), many adults may not meet the threshold for a diagnosis potentially because symptoms of ADHD, particularly those within the hyperactive/impulsive symptom domain, were not written with adults in mind (e.g., “leaves seat in situations when remaining seated is expected;” Gibbins et al., 2010). Further stressing this point are the results of studies suggesting that although only 20% to 35% of children with ADHD continue to meet diagnostic criteria as emerging adults, a majority—60% to 72%—continue to experience clinically significant levels of corresponding impairment (Biederman et al., 2010). These findings indicate that ADHD symptoms, in isolation and as they stand in DSM-5-TR, may not be sufficient to adequately capture ADHD’s expression and wide-ranging impact particularly in emerging adulthood. Although the core DSM ADHD symptom list will never fully incorporate all related functional impairment and co-occurring problems like anxiety and depression, consideration of other well-validated risk markers may be needed to facilitate a better understanding of the nature of these difficulties in emerging adults with ADHD. Such consideration could also inform more comprehensive assessment and intervention tools that accommodate a focus on co-occurring impairment and internalizing problems in addition to symptoms (Fedele et al., 2010).

To address the limitations of the ADHD symptom list, studies have begun examining the incremental validity of non-symptom measures, with findings suggesting that some of these measures demonstrate incremental validity beyond symptoms in diagnosing ADHD (Sawaya et al., 2024). Yet, almost no work has been conducted investigating the incremental validity of non-symptom measures for classification of those with versus without impairment and internalizing problems that commonly accompany the disorder. Such work could identify non-symptom risk markers of ADHD that, despite being excluded from current diagnostic criteria, may be crucial in explaining the disorder’s link with common impairment domains and internalizing problems. If so, these risk markers could potentially be incorporated into more comprehensive assessment protocols, as well as interventions, to provide broader benefits beyond symptom reduction in emerging adults with ADHD.

Emotional Dysregulation and Unawareness

One proposed aspect of ADHD in emerging adulthood that may help address this critical issue is emotional dysregulation. Emotional dysregulation has been characterized as the expression and experience of emotions which are excessive relative to social norms, context, and developmental stage; indeed, emotional dysregulation involves rapid and poorly controlled changes in emotion, along with an anomalous allocation of attention to emotional stimuli (Shaw et al., 2014). A large body of work has highlighted emotional dysregulation as a multidimensional construct, with one model by Gratz and Roemer (2004) defining six dimensions: (a) Non-Acceptance of Emotional Responses, (b) Difficulties Engaging in Goal-Directed Behavior, (c) Impulsive Control Difficulties, (d) Lack of Emotional Awareness, (e) Limited Access to Emotion Regulation Strategies, and (f) Lack of Emotional Clarity. Of note, work investigating this model has suggested that the Lack of Emotional Awareness dimension, although associated with ADHD (Factor et al., 2016), may be distinct from the broader construct of emotional dysregulation (i.e., whereas other dimensions assess reactions to emotions, Lack of Emotional Awareness assesses noticing of emotion; Hallion et al., 2018). As such, instead of grouping dimensions together into a single construct of emotional dysregulation, we henceforth refer to these two constructs as Emotional Dysregulation and Unawareness (EDU).

Clinic-based studies have suggested rates of EDU in up to 70% of adults with ADHD, with the combination of ADHD and EDU associated with high risk for impairment, emotional difficulties, and substance use (Bodalski et al., 2019; Factor et al., 2016; Shaw et al., 2014; Soler-Gutierrez et al., 2023). Conceptual studies have proposed several models for understanding the connection between ADHD and EDU. In particular, research with adolescents and adults has suggested that facets of EDU related to negative emotional responses and lability, while not entirely overlapping with ADHD symptoms, may represent important features of the disorder that uniquely help explain its relation with negative outcomes (Beheshti et al., 2020). Such findings are consistent with the idea that EDU, when present in those with ADHD, may be an important component in explaining commonly observed impairment and co-occurring internalizing problems. If so, measurement of EDU as a supplement to symptom-based ADHD assessment may facilitate improved insight into risk for specific types of impairment and inform transdiagnostic assessment of problems that commonly co-occur with the disorder but are not necessarily encompassed by ADHD’s symptoms.

Using Machine Learning to Clarify EDU Dimensions’ Incremental Validity

Despite evidence supporting the incremental validity of EDU beyond ADHD symptoms (Hirsch et al., 2018), there remains a lack of understanding in how to best incorporate measurement of EDU-related deficits into assessment protocols in a robust manner. This is particularly evident in emerging adulthood; despite recommendations by Marshall et al. (2021) encouraging the use of multiple measures in assessing ADHD in emerging adults, no work, to our knowledge, has proposed a standardized method of integrating scores from these measures.

Advances in machine learning provide a potential means to address this issue; specifically, random forest regression can be a technique used to inform data-driven classification algorithms (Walsh et al., 2017). Praised for its robustness and accuracy, this technique involves the creation of a “forest” of decision trees generated from varying sets of indicators designed for classification (i.e., with versus without impairment or co-occurring internalizing problems; Scornet et al., 2015). It both complements and provides numerous benefits over methods applied in prior work (e.g., cut-scores, linear regression, logistic regression, SEM, network analysis). Such benefits include random forest regression being data-driven, aimed at classification/prediction, less sensitive to outliers, not bound to linear relations between variables, and able to parse variable heterogeneity even in the context of multicollinearity. Prior work mostly in children has shown some success in using machine learning to incorporate measures of IQ, depressive symptoms, oppositional defiant and conduct disorders, and cognitive function, among others, with ADHD symptoms to predict outcomes (Caye et al., 2019; Goh, Elkins, et al., 2023; Lorenzi et al., 2023; Mooney et al., 2023). However, to our knowledge, no work has sought to apply the machine learning data analytic approach to the classification of ADHD’s correlates in emerging adults, despite evidence that ADHD phenotypes and, correspondingly, elements that best link the disorder with its external correlates, change throughout development (Olson, 2002).

One additional benefit provided by machine learning concerns clarification of the relative importance of different indicators via Variable Importance Scores. In random forest models, such scores are derived by ranking indicators according to their relative contribution to accurate classification (Ishwaran, 2007). Such an analyses could be helpful for determining the relative importance of the different EDU dimensions in the context of ADHD symptoms, particularly in light of prior work suggesting heterogeneity in EDU dimensions’ unique relations with several types of psychopathology (Ruan et al., 2023). Further, in contrast to inattention and EDU which may persist throughout the lifespan, hyperactivity/impulsivity has been found to decrease in salience throughout development (Faraone et al., 2021). As a result of this decrease, EDU may grow to be more important in capturing the impact of ADHD during emerging adulthood. Ultimately, identification of EDU dimensions most important for classification of ADHD-related impairment and co-occurring problems could inform the development of improved assessment tools and interventions focused on these dimensions to produce the greatest downstream benefits.

The Current Study

The current study had three primary aims. The first was to clarify EDU’s incremental validity, beyond ADHD symptoms, in improving classification accuracy of functional impairment and internalizing problems using machine learning techniques. We hypothesized that the inclusion of EDU would significantly increase accuracy of classification beyond ADHD sum scores, thus suggesting that EDU, when present, may help to explain commonly observed impairment or co-occurring internalizing problems in emerging adults with ADHD. Second, we validated accuracy of the classification model incorporating ADHD symptoms and EDU dimensions using data from a “test sample” distinct from that used for Aim 1 (although drawn from the same larger study), hypothesizing that model performance would remain consistent with results of Aim 1. Third, we sought to clarify the importance of EDU dimensions compared to ADHD symptom domains, hypothesizing they would be more important than hyperactivity/impulsivity and just as important as inattention, which would still emerge as the most important classifier.

Methods

Participants

Participants were 1,539 college students aged 18 to 25 years (M_age = 19.47 years, SD = 1.54) who self-reported a prior diagnosis of ADHD. Data from these participants was collected across eight universities in the United States. In this sample, 30.2% participants identified as male, 69.2% as female, 0.1% as intersex, and 0.5% preferring not to answer. Further, 17.5% participants identified as Hispanic/LatinX and 66.8% identified as White. Detailed demographic information is available in Table 1.

Table 1.

Descriptive Data for Demographic and Study Variables.

Variable	N = 1,539
Age [M (SD)]	19.47 (1.54)
Sex [n (%)]
Male	465 (30.2)
Female	1,065 (69.2)
Intersex	1 (0.1)
Prefer not to answer	8 (0.5)
Race [n (%)]^a
White	1,260 (81.9)
Multiracial	81 (5.3)
Black	66 (4.3)
Asian	64 (4.2)
Middle Eastern/North African	12 (0.8)
Pacific Islander/Native Hawaiian	2 (0.1)
Ethnicity (% Latinx/Hispanic)	12.5
College GPA [M (SD)]	3.24 (0.59)
ADHD symptom count [M (SD)]^b
Inattention	4.13 (3.19)
Hyperactivity-impulsivity	2.73 (2.52)
ADHD symptom severity [M (SD)]
Inattention	13.01 (7.41)
Hyperactivity-impulsivity	9.38 (6.25)
DERS [M (SD)]
Non-acceptance of emotional responses	15.56 (6.76)
Difficulty engaging in goal-directed behavior	16.7 (5)
Impulse control	14.18 (5.62)
Lack of emotional awareness	17.36 (5.74)
Limited access to emotion regulation strategies	20.9 (8.02)
Lack of emotional clarity	13.76 (4.55)
WFIRS: functional impairment [M (SD)]		Met study criteria for impairment [n (%)] ^c
Life skills	1.18 (0.69)	1,146 (74.5)
Self-concept	1.66 (0.98)	945 (61.4)
Work	0.62 (0.59)	335 (21.8)
Family	0.72 (0.67)	546 (35.5)
Social	0.67 (0.62)	563 (36.6)
School	1.04 (0.69)	1,004 (65.2)
Risk-taking behavior	0.49 (0.48)	647 (42.0)
DASS-21: internalizing problems [M (SD)]		Met study criteria for depression/anxiety [n (%)] ^d
Depression	6.49 (5.53)	645 (41.9)
Anxiety	6.21 (4.89)	731 (47.5)

A total of 42 participants had missing racial data or specified an unlisted racial category

ADHD Symptom Count scores were determined by counting the number of symptoms endorsed as “Often” or “Very Often.” The finding that many participants, despite endorsing prior ADHD diagnoses, did not meet current symptom criteria for ADHD (i.e., ≥5 symptoms in either domain) is consistent with prior work suggesting that ADHD severity decreases with age (Faraone et al., 2006). However, given work suggesting that patterns of ADHD may fluctuate over time (Sibley et al., 2022), we included all participants that reported prior diagnoses of ADHD.

In line with test recommendations (Weiss et al., 2018), any domain that had one item rated 3 or two items rated 2 was be considered to be impaired.

Cut-off sum scores greater than 6 and 5 met study criteria for depression and anxiety, respectively (Lovibond & Lovibond, 1995).

Procedures

Ethics approval was obtained from the Institutional Review Board at Syracuse University for this multisite study of ADHD symptoms in college students (i.e., Undergraduate Learning, Emotion, and Attention Research Network). Undergraduate students across the universities were recruited primarily via online participant management sites (i.e., Sona Systems, https://www.sona-systems.com), as well as classroom announcements and flyers. The survey was hosted on a secure online survey platform (i.e., Qualtrics, https://www.qualtrics.com) through which participants gave their informed consent to participate and then completed a series of measures. Data was collected from 12,501 college students, although only those endorsing a prior ADHD diagnosis (12.3%, consistent with prior ADHD research on online surveys; Wymbs & Dawson, 2019) were included in the current study. Of these 1539 participants, 1200 provided data on medication use with 579 (48%) endorsing current use of stimulants for ADHD. Four attention check questions were included, and data from participants who answered more than one attention check incorrectly were removed (n = 34). Participants were compensated with research credits for study participation. Some sites also offered entry into a raffle for a $100 gift card to incentivize participation.

Measures

ADHD Symptoms

To test the incremental validity of EDU over and above DSM’s ADHD symptoms, a self-report checklist was created based on the exact wording of ADHD symptoms in the DSM-5 (APA, 2013). DSM-5 symptoms used in the current study were identical to those in the newer DSM-5-TR (APA, 2022), with examples being “Fail to give close attention to details or make careless mistakes in schoolwork, at work, or during other activities (overlook or miss details or work is inaccurate)” and “Fidget with or tap hands or feet or squirms in seat.” Participants rated their current symptoms using the following scale: 0 (Never or rarely), 1 (Sometimes), 2 (Often), or 3 (Very often). Item scores were then added to create sum scores representing symptom severity on inattention (IA; α = .92) and hyperactivity/impulsivity (HI; α = .87) symptom dimensions.

EDU¹

The 36-item Difficulties in Emotion Regulation Scale (DERS; Gratz & Roemer, 2004) measures how frequently participants experience difficulties related to EDU. Items included statements such as “I am confused about how I feel” and “When I’m upset, I lose control over my behavior.” Participants responded to items using the following scale: 1 (Almost never), 2 (Sometimes), 3 (About half the time), 4 (Most of the time), or 5 (Almost always). Response scores were averaged into six subscales corresponding with Non-Acceptance of Emotional Responses, Difficulties Engaging in Goal-Directed Behavior, Impulse Control Difficulties, Lack of Emotional Awareness, Limited Access to Emotional Regulation Strategies, and Lack of Emotional Clarity dimensions (αs = .82–.93).

Functional Impairment

The Weiss Functional Impairment Rating Scale (WFIRS; Weiss, 2000) is a 70-item self-report measure of impairment that has shown robust psychometric properties for use in college students (Canu et al., 2020). Example items include “having problems with family” (Family domain), “problems taking notes” (School domain), and “problems keeping an acceptable appearance” (Life Skills domain). Participants responded to items using the following scale: 0 (Never or not at all), 1 (Sometimes or somewhat), 2 (Often or much), or 3 (Very often or very much). Items fall within Family, Work, School, Life Skills, Self-Concept, Social, and Risk-Taking Behavior domains (αs = .87–.93). For the current study, and in line with test recommendations (Weiss et al., 2018), any domain that had one item rated 3 or two items rated 2 was be considered to be impaired.

Internalizing Problems

The Depression, Anxiety, and Stress Scale (DASS-21) contains 21 items that assess depression, anxiety, and stress (Lovibond & Lovibond, 1995). Examples of items included “I felt scared without any good reason” and “I felt down-hearted and blue.” Participants rated items using the following scale: 0 (Did not apply to me at all), 1 (Applied to me some of the time), 2 (Applied to me a good part of the time), 3 (Applied to me very much, or most of the time). Item responses were summed to create depression (α = .91) and anxiety (α = .84) scores. Past studies have supported the validity of these dimensions (Zanon et al., 2021). Cut-off sum scores of 6 and 5, corresponding with moderate levels of depression and anxiety, respectively, were used to determine prevalence or absence of these internalizing problems (Lovibond & Lovibond, 1995).

Statistical Analyses

Transparency and Openness

We report how we determined our sample size, data exclusions (if any), manipulations, and measures in the study, and we follow JARS (Kazak, 2018). Data and research materials are available upon request. Pre-registered analyses (https://osf.io/2upw7/?view_only=6ff4ed43afc54d398c56700deef2e01e) and R-code (https://osf.io/t7x48/?view_only=9e377479232c45b4aa57425d436b8ea1) are available on Open Science Framework.

Random Forest Development

Random forests were created using the caret package (Kuhn, 2008) in R (version 4.3.2). Based on procedures recommended by Nguyen et al. (2021), participants were first randomly split into training (70% of the data) and testing samples (30%). Participants in the training sample who met study criteria for impairment were up-sampled (i.e., randomly duplicated with replacement) to match the size of the non-impaired sample to avoid bias in resulting algorithms, in line with recent studies suggesting random oversampling as the best means to address moderately imbalanced data (Wongvorachan et al., 2023). No adjustments were made to the testing sample. The training sample was then used to develop random forest models and tune them via the mtry hyperparameter (i.e., the number of random indicators considered at each split within trees), using fivefold cross-validation repeated 5 times. All other hyperparameters were set to the default criteria in the caret package (e.g., 500 trees per forest, one observation for terminal nodes). Optimal models were determined based on the receiver operator characteristic (AUC) curve metric, which provides an indication of the algorithm’s overall discriminative ability (Kumar & Indrayan, 2011).

Integrating EDU with ADHD Symptoms and Testing the Final Model

To test Hypothesis 1 (i.e., the inclusion of EDU would significantly improve model performance), we compared the performance of two random forest classification models across each impairment domain and internalizing problem using data from the training sample. First, we created an ADHD Algorithm by including IA and HI as indicators to generate the random forest model, with the mtry parameter set to 2. Second, we created an ADHD + EDU Algorithm with IA and HI along with EDU dimensions as indicators, while also testing all mtry values from 2 to 8 to determine the value that led to the best-performing model while maximizing the incremental validity of EDU dimensions. We set the seed prior to random sampling, so the same resampling datasets were used to create ADHD and ADHD + EDU Algorithms. This resulted in paired estimates of performance across training resamples. Hence, we could test whether the AUCs of model pairs significantly differed using paired t-tests, with any increase in AUC attributable to the inclusion of EDU dimensions (Kuhn & Johnson, 2018). For these comparisons, a cut-off of p < .05 was used to determine statistical significance. To test Hypothesis 2 (i.e., consistent performance in a distinct testing sample), we evaluated the algorithm in the distinct testing sample. Model performance was evaluated using AUC, with 0.5 suggesting no discrimination, 0.7 to 0.8 considered acceptable, 0.8 to 0.9 excellent, and greater than 0.9 considered outstanding (Hosmer et al., 2013).

To test Hypotheses 3 (i.e., EDU dimensions would be more important for machine learning models than HI, and just as important as IA) we calculated variable importance scores with respect to classification in the training sample for each impairment domain and internalizing problem. Scores were created by comparing classification error (i.e., mean-squared error) before and after each variable’s values were randomly shuffled (Ishwaran, 2007). These scores were scaled to fall between 0 and 100, with higher scores indicating greater importance (i.e., increased discrimination power between classes) within machine learning algorithms. To our knowledge, there is no existing statistical test to determine the significance of differences in variable importance scores; consequently, we evaluated the differences qualitatively.

Results

Testing Hypotheses 1 and 2: Examining the Incremental Validity of EDU²

See Tables 2 and 3 for detailed results of machine learning analyses. Initial analyses of the ADHD + EDU algorithm suggested that mtry = 2 yielded the best performing model for all outcomes. Our first hypothesis was that the inclusion of EDU would significantly improve classification performance with respect to impairment and co-occurring internalizing problems above and beyond ADHD sum scores. This hypothesis was supported. Comparison of AUC values for models without versus with EDU dimensions suggested that the inclusion of EDU significantly improved model classification (average increase in AUC = 0.14; range of AUC increase = 0.10–0.18; ps < .001). Further, average sensitivity across outcomes (i.e., the model’s ability to correctly identify those meeting impairment or internalizing problem criteria, ranging from 0 to 1 with higher values indicating greater accuracy) increased from 0.64 to 0.80, and specificity (the model’s ability to correctly identify those not meeting criteria on the same scale as sensitivity) increased from 0.71 to 0.80, after adding EDU to the classification model. Our second hypothesis was that model classification of the ADHD + EDU algorithm in a distinct testing sample would remain consistent with the training sample across outcomes. This hypothesis was partially supported. Consistent with those in the training sample, testing sample results suggested that the model demonstrated excellent performance in discriminating those meeting study criteria for impairment in Life Skills (AUC = 0.83) and Self-Concept (AUC = 0.85) domains, as well as Depression (AUC = 0.83) and Anxiety (AUC = 0.82). The model demonstrated acceptable performance with respect to Social (AUC = 0.73), School (AUC = 0.76), Family (AUC = 0.77), and Risk-Taking Behavior (AUC = 0.70) domains, and poor performance with respect to the Work domain (AUC = 0.66). However, compared to the training sample, examination of model accuracy in the testing sample suggested sensitivity decreased by 0.07 to 0.73, and specificity by 0.13 to 0.67, on average. Overall, results suggested that the inclusion of EDU significantly improved classification accuracy beyond IA and HI sum scores, with the resulting ADHD + EDU model also demonstrating acceptable to excellent performance in a distinct sample except with regard to classification of Work impairment.

Table 2.

Results of Machine Learning Analyses Classifying of ADHD-Related Impairment and Co-Occurring Problems.

Variable	ADHD algorithm				ADHD + EDU algorithm				Model comparison
Variable	AUC (SD)^a	Overall acc.	Sens.	Spec.	AUC (SD)^a	Overall Acc.	Sens.	Spec.	Δ AUC	t (dfs = 24)	p
Impairment domains
Life skills	0.88 (0.02)	0.82	0.72	0.91	0.98 (0.01)	0.92	0.87	0.97	0.10	21.1	<.001
Self-concept	0.77 (0.02)	0.71	0.64	0.77	0.92 (0.01)	0.84	0.80	0.87	0.15	33.0	<.001
Social	0.73 (0.02)	0.68	0.69	0.67	0.89 (0.03)	0.80	0.85	0.76	0.16	23.9	<.001
School	0.82 (0.03)	0.75	0.67	0.82	0.93 (0.01)	0.84	0.80	0.89	0.11	21.1	<.001
Work	0.67 (0.05)	0.61	0.61	0.60	0.83 (0.03)	0.74	0.77	0.72	0.16	12.4	<.001
Family	0.74 (0.04)	0.68	0.65	0.70	0.85 (0.02)	0.79	0.81	0.76	0.11	16.3	<.001
Risk-taking behavior	0.67 (0.03)	0.62	0.59	0.64	0.78 (0.03)	0.69	0.70	0.69	0.11	13.5	<.001
Internalizing problems
Depression	0.71 (0.03)	0.66	0.65	0.67	0.89 (0.02)	0.80	0.84	0.77	0.18	26.4	<.001
Anxiety	0.64 (0.03)	0.60	0.55	0.65	0.81 (0.03)	0.74	0.76	0.73	0.17	16.8	<.001

Mtry hyperparameter was set to 2 for all algorithms. All other hyperparameters were left at their defaults.

Table 3.

Detailed Statistics of Final Model Performance in the Testing Sample^a.

Variable	Threshold^b	AUC (95% CI)	Sensitivity	Specificity	PPV	NPV
Impairment domains
Life skills	0.65	0.83 [0.79, 0.87]	0.83	0.64	0.89	0.52
Self-concept	0.6	0.85 [0.82, 0.88]	0.73	0.83	0.88	0.64
Social	0.35	0.73 [0.68, 0.78]	0.75	0.58	0.52	0.79
School	0.65	0.76 [0.72, 0.80]	0.64	0.72	0.49	0.67
Work	0.5	0.66 [0.59, 0.73]	0.58	0.64	0.55	0.67
Family	0.45	0.77 [0.72, 0.82]	0.75	0.65	0.60	0.78
Risk-taking behavior	0.4	0.7 [0.65, 0.75]	0.71	0.55	0.57	0.7
Internalizing problems
Depression	0.4	0.83 [0.79, 0.87]	0.82	0.71	0.69	0.83
Anxiety	0.55	0.82 [0.78, 0.86]	0.74	0.74	0.74	0.74

Note. Sensitivity = the model’s ability to correctly identify those meeting impairment or internalizing problem criteria. Specificity = the model’s ability to correctly identify those who do not meet criteria. Positive predictive value (PPV) = the proportion of participants classified as meeting criteria by the model who truly meet criteria. Negative predictive value (NPV) = the proportion of participants classified as not meeting criteria by the model who truly do not meet criteria.

Mtry hyperparameter was set to 2 for all algorithms.

Youden’s index (sensitivity + specificity − 1; Youden, 1950) was used to select the threshold on the receiver operator curve (ROC) that optimized both sensitivity and specificity in determining accuracy metrics.

Testing Hypothesis 3: Importance of ADHD Symptoms and EDU Dimensions Across Algorithms

Detailed results are depicted in Table 4 and Figure 1. Our third hypothesis was that EDU dimensions would emerge as more important for classification than HI and just as important as IA. This hypothesis was supported. Examining rank order of ADHD sum scores and EDU dimensions across domains, IA emerged as most important for classification of four domains (Life Skills, Self-Concept, School, and Work), HI for two (Family and Risk-Taking Behavior), the Limited Access to Emotion Regulation Strategies dimension for two (Depression and Anxiety), and Non-Acceptance of Emotional Responses for one (Social). At least one EDU dimension emerged as more important than HI across six of nine impairment domains and internalizing problems. Similarly, at least one EDU dimension emerged as more important than IA across four impairment domains and internalizing problems.

Table 4.

Variable Importance in the Training Sample.

Variable	Life Skills	Self-concept	Social	School	Work	Family	Risk-taking behavior	Depression	Anxiety
Inattention	100	100	67.74	100	100	77.82	73.59	40.38	50.27
Hyperactivity/impulsivity	26.34	19.83	53.78	35.34	13.21	100	100	1.88	55.21
Non-acceptance of emotional responses	0	66.91	100	0	0	29.28	4.9	69.92	66.64
Difficulty engaging in goal-directed behavior	29.19	25.58	0	29.65	10.43	15.22	5.66	4.49	13.93
Impulse control difficulties	3.25	0	76.41	2.56	22.68	16.65	73.88	19.18	50.03
Lack of emotional awareness	24.42	14.78	37.36	8.57	11.37	0	0	0	0
Limited access to emotion regulation strategies	20.43	77.67	78.38	1.05	56.62	33.68	19.12	100	100
Lack of emotional clarity	37.5	23.56	85.76	9.23	38.44	56.69	28.62	43.72	51.87

Note. Scores were estimated by randomly shuffling each variable’s values and then comparing the resulting classification error of a forest to the error obtained before the shuffle. These values were averaged across all trees and then scaled to fall between 0 to 100. Higher scores indicate greater importance.

Figure 1.

Variable importance scores in machine learning analyses encompassing ADHD sum scores and EDU dimensions.

When comparing variable importance scores, results suggested that across impairment domains, IA was most important for classification (average variable importance = 78.87). The Limited Access to Emotion Regulation Strategies EDU dimension emerged as the next most important (54.11), followed by HI (45.07) and the Non-Acceptance of Emotional Responses (37.52) EDU dimension. Overall, results suggested IA as most important for facilitating model classification, with the Limited Access to Emotion Regulation Strategies dimension generally emerging as more important than HI.

Post Hoc Analysis: Replicating Findings in Those Without ADHD and Testing the Incremental Validity of Sex

It is possible that analyses described above might have suffered from a restriction of range problem, given that the sample used consisted only of those with prior ADHD diagnoses and thus likely elevated levels of ADHD symptom severity. To address this, we also explored the incremental validity of EDU beyond ADHD symptoms using data from all participants with and without prior ADHD diagnoses (n = 12,501). Consistent with primary analyses, comparison of AUC values suggested that the inclusion of EDU significantly improved model classification across all outcomes (increase in AUC with the inclusion of EDU: Life Skills = 0.07; Self-Concept = 0.11; Social = 0.20; School = 0.08; Work = 0.15; Family = 0.18; Risk-Taking Behavior = 0.19; Depression = 0.20; Anxiety = 0.18; ps < .001). Detailed information is available upon request from the corresponding author.

Past findings have suggested sex differences in ADHD’s expression during emerging adulthood (Fedele et al., 2010). As such, we tested whether the addition of sex into the ADHD + EDU algorithm significantly improved accuracy. These comparisons were conducted similarly to those described above (i.e., ADHD + EDU vs. ADHD + EDU + sex algorithms), although new bootstrapping procedures were conducted and accounted for any differences between primary and post-hoc results. Results suggested that sex did not significantly improve accuracy across any impairment domain or co-occurring problems (ps ranged from .30 to .92; average p = .56).

Discussion

The current study sought to investigate whether EDU, when experienced by emerging adults with self-reported ADHD, may help to explain commonly observed impairment or co-occurring internalizing problems. To do so, we used machine learning to develop an algorithm encompassing EDU dimensions and ADHD sum scores that could discriminate between those reporting impairment and internalizing problems in a reliable and statistically robust manner. Consistent with our first hypothesis, results suggested that EDU significantly improved model performance, with AUC increasing by 11.4% with respect to classification of Life Skills impairment, 19.5% for Self-Concept impairment, 23.9% for Work impairment, 14.9% for Family impairment, 21.9% for Social impairment, 13.4% for School impairment, and 16.4% for Risk-Taking Behavior impairment beyond ADHD sum scores alone. Combining EDU with ADHD sum scores also increased AUC with respect to classification of Depression and Anxiety by 25.4% and 26.6%, respectively. In support of our second hypothesis, the ADHD + EDU classification algorithm achieved acceptable to excellent performance in a separate sample across all impairment domains and internalizing problems except for Work-related impairment, although some decreases in sensitivity and specificity were noted. Finally, and in support of our third hypothesis, results of variable importance analyses indicated IA as most important for classification across all impairment domains and internalizing problems. Further, the Limited Access to Emotional Regulation Strategies EDU dimension was more important for classification of impairment and co-occurring problems than HI, with the Non-Acceptance of Emotional Responses EDU dimension also demonstrating some importance for classification. Overall, results suggested that EDU, when present, may help to explain the presence of commonly observed impairment or internalizing problems in emerging adults with ADHD. Further, results provided support for the use of machine learning as a methodology to facilitate actuarial classification of co-occurring impairment and internalizing problems in emerging adults with ADHD.

EDU Demonstrated Incremental Validity Beyond ADHD Symptoms

In general, results supported EDU as playing a key role in explaining ADHD’s relation with impairment and co-occurring problems. Specifically in the training sample, average classification performance of only ADHD symptoms across outcomes was Acceptable, whereas performance of ADHD symptoms and EDU dimensions was Excellent. Additionally, overall classification accuracy increased by 0.11 on average, and appeared to be primarily driven by increases in sensitivity (0.16 on average, in comparison to specificity which increased by 0.08 on average). Unsurprisingly, the greatest benefits pertained to the classification of Depression and Anxiety, although classification of impairment domains, particularly Self-Concept, Social, and Work, also improved with the inclusion of EDU. Recent studies evaluating the accuracy of ADHD-related screening tools has set 0.8 as the cut-off for acceptable sensitivity and specificity (Mulraney et al., 2021). Based on these criteria, the ADHD algorithm would not have met criteria for any domain tested, whereas the ADHD + EDU algorithm met criteria for Life Skills, Self-Concept, School, and Family domains based on performance in the training sample. Although findings supporting EDU dimensions’ incremental validity are not necessarily surprising given existing research highlighting their role in contributing to ADHD-related phenotypes (Shaw et al., 2014), results are notable in supporting the idea that measures of ADHD symptoms and EDU dimensions can be incorporated into actuarial classification tools that significantly outperform those only including ADHD symptoms. In fact, results suggested that it may important to harness both ADHD symptoms and EDU dimensions’ unique validity within actuarial ADHD-related algorithms.

Application of the ADHD + EDU Algorithm to the testing sample also revealed that at the optimal threshold, the model classified those who met impairment and internalizing problem criteria more accurately than those who did not (i.e., sensitivity > specificity, average difference = 0.15) with respect to Life Skills, Social, Family, Risk-Taking Behavior, and Depression domains. Such findings suggested that EDU may be particularly helpful for identifying those experiencing difficulties in these areas, with further addition to the algorithm needed to improve its ability to correctly identify those who do not. The opposite may be true for Self-Concept, School, and Work domains (specificity > sensitivity, average difference = 0.08).

It should be noted that model performance in the testing sample decreased compared to the training sample with respect to AUC (average decrease = 0.1), sensitivity (0.07), and specificity (0.12). Although performance still ranged from Acceptable to Excellent, such decreases may have been due to upsampling procedures, which may have led to some overfitting of the model to the training sample. This is evidenced by follow-up findings that the greatest decreases in AUC from training to testing sample results occurred for impairment and internalizing problem domains that had the greatest class imbalances between those meeting versus not meeting criteria. Hence, although the addition of EDU appeared to provide meaningful benefits to performance beyond ADHD sum scores, further refinement of the model, including tuning in more balanced samples (without oversampling), may be needed before deployment into real-world settings.

Post-hoc analyses also provided support for EDU’s incremental validity even when we included participants without ADHD, which is notable in light of dimensional models conceptualizing ADHD as falling on a continuum (Haslam et al., 2006; Marcus & Barry, 2011). Individuals demonstrating subthreshold levels of ADHD have still been found to exhibit impairments and benefit from intervention (McLennan, 2016), and results suggested that EDU may be relevant for explaining ADHD’s relations with impairment and internalizing problems across the entire spectrum of ADHD symptom severity. Overall, findings provided additional support for the idea that EDU, when present in emerging adults with ADHD, may play a key role in explaining the presence of impairment or co-occurring internalizing problems and thus warrant attention in actuarial assessment and intervention protocols.

Limited Access to Emotion Regulation Strategies Was More Important for Classification Than HI

In general, results provided support for the idea that certain dimensions of EDU may be almost as important as IA, and more important than HI, for differentiating between emerging adults with ADHD who have co-occurring impairment and internalizing problems and those who do not. In particular, the Limited Access to Emotional Regulation Strategies dimension emerged as the most important for classification of Anxiety and Depression, and also demonstrated the second highest average variable importance value. IA emerged as most important for Life Skills, Self-Concept, School, and Work impairment classification, and also demonstrated the highest average variable importance value.

Conversely, although HI emerged as the most important for classification of two impairment domains (Family and Risk-Taking Behavior), its variable importance value fell below that of IA sum scores and the Limited Access to Emotion Regulation Strategies dimension, and was only slightly above the Non-Acceptance of Emotional Reactions EDU dimension. Such results provide support for prior work suggesting HI as demonstrating less salience during adulthood compared to IA (Biederman et al., 2010; Gibbins et al., 2010; Sibley et al., 2012). On the other hand, EDU has been found to remain stable or even worsen throughout adulthood (Deutz et al., 2020). As such, ADHD assessment during emerging adulthood may benefit from an increased focus on IA and EDU, particularly regarding acceptance of emotional reactions and access to emotional regulation strategies. Further, given the seeming importance of the Limited Access to Emotion Regulation Strategies and Non-Acceptance of Emotional Reactions dimensions, focusing intervention effects on these dimensions via emotional regulation strategies is likely to produce benefits beyond those focused only on symptom reduction.

Potential Benefits of Incorporating EDU Into ADHD-Related Clinical Practices

Despite elevated rates of impairment and co-occurring internalizing problems being identified as key features of ADHD in emerging adulthood (Anastopoulos et al., 2016; DuPaul, 2022), there has been very limited research providing guidance on how to incorporate different measures into assessment tools that can accurately provide insight into risk for these difficulties. The minimal work in this area has mostly proposed incorporating measures based on clinical judgment (Marshall et al., 2021), despite a large body of work supporting the superiority of actuarial methods for integrating different types of data (Sawyer, 1966). Results of the current study provided support for the idea that assessment of certain EDU domains, namely Limited Access to Emotion Regulation Strategies and Non-Acceptance of Emotional Reactions, may provide insight into ADHD’s connection with commonly co-occurring impairment domains and internalizing problems. As such, including measures of these domains in ADHD assessment protocols may help facilitate accurate classification of impairment and internalizing problem profiles in emerging adults with ADHD. Conversely, the Lack of Emotional Awareness domain consistently performed poorly across all models, so assessment of this domain may provide little benefit in classifying impairment and internalizing problems in those with ADHD. Results also provided support for the use of machine learning to incorporate different types of data and inform robust and replicable assessment practices. In contrast to many past studies that have created machine learning algorithms using time-consuming or costly measures (e.g., Das & Khanna, 2021), our findings suggested that accuracy of classification of ADHD-related impairment and co-occurring problems can be significantly improved through the addition of low-cost and relatively brief measures to the ADHD symptom list.

Limitations and Constraints on Generality

Despite EDU significantly improving the algorithm’s performance beyond ADHD sum scores, there was still room for improvement through refinement of the model or the addition of new relevant risk markers of ADHD (e.g., executive function, temperament/personality; Goh et al., 2020), particularly with respect to Work impairment classification which fell below the acceptable range in the testing sample. One important future direction concerns an examination of reasons for EDU dimensions’ importance in classification, due to the nuances of random forests methodology occurring in a “black box” that is relatively inaccessible. Hence, although random forest may be useful for identifying what variables are important for classification, the precise reasons why such variables are important are difficult to untangle. Some research has been conducted in this area, with findings suggesting EDU may stem from similar neural underpinnings as ADHD and mediate the disorder’s relation with common correlates (Shaw et al., 2014). However, additional work is needed to fully clarify reasons for EDU’s importance in explaining ADHD’s relation with impairment and internalizing problems. All measures, along with ADHD diagnosis, were based on self-report. Future studies should thus seek to validate findings using gold-standard diagnostic procedures and different reporters (i.e., informant report; clinical interview). Although correlations between predictors did appear to be strongly indicative of multicollinearity (>.8), some were moderate to large in strength. Thus, future work should replicate findings while determining whether these correlations impacted results. Since the data was collected at a single time point, an important future direction involves the use of longitudinal studies.

Further, the current work used a convenience sample of college students who mostly identified as female, which contrasts with prior work estimating prevalence of ADHD in males versus females at a 1:1 to 2:1 ratio (Cortese et al., 2016). Such a focus represents an important clinical aim, given that ADHD symptoms were created primarily based on observations of males who tend to express higher levels of hyperactivity/impulsivity compared to females (reviewed in Babinski, 2024). Females with ADHD remain underdiagnosed despite some evidence suggesting that they may report levels of impairment higher than males (Fedele et al., 2012), so the development and validation of actuarial assessment protocols in females represents an important aim. However, further validation is needed to determine the incremental validity of EDU in samples with more male-identified participants as well. Relatedly, the racial makeup of this cohort, which was predominantly White, imposes limitations to the degree to which results may generalize to other racial and ethnic groups. Overall, further work is needed to confirm results in other populations outside of that used in the current study to provide further validation for the incremental validity of EDU and the potential of incorporating measures of it into algorithms with ADHD symptoms to facilitate actuarial classification.

Conclusion

The ability to explain ADHD’s link with impairment and internalizing problems is of great clinical importance, particularly in emerging adulthood. Results of the current study suggested that EDU dimensions, particularly Limited Access to Emotion Regulation Strategies and Non-Acceptance of Emotional Reactions, may provide unique insights in line with this aim. Further incorporation of EDU dimensions into actuarial ADHD assessment procedures via machine learning could meaningfully improve classification of impairment and co-occurring internalizing problems in emerging adults with ADHD. Additionally, continued establishment of the role of EDU in ADHD could facilitate the implementation of interventions aimed at addressing the most salient outcomes.

Footnotes

Acknowledgements

The authors acknowledge Jasmine Black and Abigail Cummings for their help in writing the manuscript. The authors thank all participants for making this work possible. The current study uses data from an ongoing multisite study of ADHD symptoms in college students (i.e., the Undergraduate Learning, Emotion, and Attention Research Network [U-LEARN] study). This study has been ongoing for several years with manuscripts published using data from prior academic years. Some of this data (24%) partially overlaps with those used in the current study. However, the current study is the first to use data collected between Fall 2021 and Spring 2023. Further, no prior manuscripts have examined relations between emotional dysregulation and functional impairment/internalizing problems as was done in the current study.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iDs

Patrick K. Goh

Ashlyn W. W. A. Wong

Da Eun Suh

Elizabeth A. Bodalski

Yvette Rother

Cynthia M. Hartung

Elizabeth K. Lefler

Notes

Author Biographies

Patrick K. Goh, PhD, is an Assistant Professor of clinical psychology at the University of Hawai‘i at Mānoa. His research centers on improving attention-deficit/hyperactivity disorder (ADHD) assessment through consideration of inter- and intrapersonal risk factors, particularly those within emotional and sociocultural domains.

Ashlyn W. W. A. Wong, BA, is a Graduate Student at the University of Hawaiʻi at Mānoa. Her research focuses on cultural factors affecting the perception, identification, and treatment of ADHD.

Da Eun Suh, MA, is a Doctoral Candidate in Clinical Psychology at the University of Hawai‘i at Mānoa. Her research focuses on investigating cultural factors influencing social anxiety and ADHD, with a particular interest on the Asian and Asian American population.

Elizabeth A. Bodalski, MPS, is a Doctoral Candidate in Clinical Community Psychology at the University of South Carolina. Her research focuses on ADHD in young adults and adults. She is especially interested in understanding difficulties with emotion dysregulation among individuals with ADHD.

Yvette Rother, MA, is a Doctoral Student of clinical-community psychology at the University of South Carolina. Her research focuses on various attention-deficit/hyperactivity disorder (ADHD) outcomes, such as protective and risk factors among youth and college students.

Cynthia M. Hartung, PhD, is a Professor of Psychology at the University of Wyoming. Her research focuses on the assessment and treatment of ADHD in adolescents and emerging adults and sex/gender differences in psychopathology.

Elizabeth K. Lefler, PhD is a Professor of Psychology at the University of Northern Iowa. Dr. Lefler’s research interests include the assessment of ADHD in understudied populations, as well as sex/gender differences in psychopathology.

References

American Psychiatric Association. (2013). Diagnostic and statistical manual of mental disorders (5th ed.).

American Psychiatric Association. (2022). Diagnostic and statistical manual of mental disorders (5th ed., text rev.). https://doi.org/10.1176/appi.books.9780890425787

Anastopoulos

A. D.

DuPaul

G. J.

Weyandt

L. L.

Morrissey-Kane

Sommer

J. L.

Rhoads

L. H.

Murphy

K. R.

Gormley

M. J.

Gudmundsdottir

B. G.

(2016). Rates and patterns of comorbidity among first-year college students with ADHD. Journal of Clinical Child and Adolescent Psychology, 47(2), 236–247. https://doi.org/10.1080/15374416.2015.1105137

Arnett

J. J.

(2000). Emerging adulthood: A theory of development from the late teens through the twenties. American Psychologist, 55(5), 469–480. https://doi.org/10.1037/0003-066X.55.5.469

Babinski

D. E.

(2024). Sex differences in ADHD: Review and priorities for future research. Current Psychiatry Reports, 26(4), 151–156. https://doi.org/10.1007/s11920-024-01492-6

Beheshti

Chavanon

M. L.

Christiansen

(2020). Emotion dysregulation in adults with attention deficit hyperactivity disorder: A meta-analysis. BMC Psychiatry, 20(1), 1–11. https://doi.org/10.1186/s12888-020-2442-7

Biederman

Petty

C. R.

Evans

Small

Faraone

S. V.

(2010). How persistent is ADHD? A controlled 10-year follow-up study of boys with ADHD. Psychiatry Research, 177(3), 299–304. https://doi.org/10.1016/j.psychres.2009.12.010

Bodalski

E. A.

Knouse

L. E.

Kovalev

(2019). Adult ADHD, emotion dysregulation, and functional outcomes: Examining the role of emotion regulation strategies. Journal of Psychopathology and Behavioral Assessment, 41, 81–92. https://doi.org/10.1007/s10862-018-9695-1

Canu

W. H.

Hartung

C. M.

Stevens

A. E.

Lefler

E. K.

(2020). Psychometric properties of the Weiss Functional Impairment Rating Scale: Evidence for utility in research, assessment, and treatment of ADHD in emerging Adults. Journal of Attention Disorders, 24(12), 1648–1660. https://doi.org/10.1177/1087054716661421

10.

Caye

Agnew-Blais

Arseneault

Gonçalves

Kieling

Langley

Menezes

A. M. B.

Moffitt

T. E.

Passos

I. C.

Rocha

T. B.

Sibley

M. H.

Swanson

J. M.

Thapar

Wehrmeister

Rohde

L. A.

(2019). A risk calculator to predict adult attention-deficit/hyperactivity disorder: Generation and external validation in three birth cohorts and one clinical sample. Epidemiology and Psychiatric Sciences, 29(e37), 1–9. https://doi.org/10.1017/S2045796019000283

11.

Chung

Jiang

S. F.

Paksarian

Nikolaidis

Castellanos

F. X.

Merikangas

K. R.

Milham

M. P.

(2019). Trends in the prevalence and incidence of attention-deficit/hyperactivity disorder among adults and children of different racial and ethnic groups. Journal of the American Medical Society Network, 2(11), e1914344. https://doi.org/10.1001/jamanetworkopen.2019.14344

12.

Cortese

Faraone

S. V.

Bernardi

Wang

Blanco

(2016). Gender differences in adult attention-deficit/hyperactivity disorder: Results from the National Epidemiologic Survey on Alcohol and Related Conditions (NESARC). Journal of Clinical Psychiatry, 77(4), 7626. https://doi.org/10.4088/JCP.14m09630

13.

Das

Khanna

(2021). A robust machine learning based framework for the automated detection of ADHD using pupillometric biomarkers and time series analysis. Scientific Reports, 11(1), 16370. https://doi.org/10.1038/s41598-021-95673-5

14.

Demler

O. V.

Pencina

M. J.

D’Agostino

R.,

Sr . (2012), Misuse of DeLong test to compare AUCs for nested models. Statistics in Medicine, 31(23), 2577–2587. https://doi.org/10.1002/sim.5328

15.

Deutz

M. H. F.

Geeraerts

S. B.

Belsky

Deković

van Baar

A. L.

Prinzie

Patalay

(2020). General Psychopathology and dysregulation profile in a longitudinal community sample: Stability, antecedents and outcomes. Child Psychiatry and Human Development, 51(1), 114–126. https://doi.org/10.1007/s10578-019-00916-2

16.

DuPaul

G. J.

(2022). The conceptual and diagnostic importance of ADHD-related impairment: A Commentary on Arildskov et al. (2021). Journal of Child Psychology and Psychiatry, 63(2), 238–240. https://doi.org/10.1111/jcpp.13467

17.

Factor

P. I.

Rosen

P. J.

Reyes

R. A.

(2016). The relation of poor emotional awareness and externalizing behavior among children with ADHD. Journal of Attention Disorders, 20(2), 168–177. https://doi.org/10.1177/1087054713494005

18.

Faraone

S. V.

Banaschewski

Coghill

Zheng

Biederman

Bellgrove

M. A.

Newcorn

J. H.

Gignac

Al Saud

N. M.

Manor

Rohde

L. A.

Yang

Cortese

Almagor

Stein

M. A.

Albatti

T. H.

Aljoudi

H. F.

Alqahtani

M. M. J.

Asherson

Atwoli

, . . . Wang

(2021). The world federation of ADHD international consensus statement: 208 evidence-based conclusions about the disorder. Neuroscience and Biobehavioral Reviews, 128, 789–818. https://doi.org/10.1016/j.neubiorev.2021.01.022

19.

Faraone

S. V.

Biederman

Mick

(2006). The age-dependent decline of attention deficit hyperactivity disorder: A meta-analysis of follow-up studies. Psychological Medicine, 36(2), 159–165. https://doi.org/10.1017/S003329170500471X

20.

Fedele

D. A.

Hartung

C. M.

Canu

W. H.

Wilkowski

B. M.

(2010). Potential symptoms of ADHD for emerging adults. Journal of Psychopathology and Behavioral Assessment, 32, 385–396. https://doi.org/10.1007/s10862-009-9173-x

21.

Fedele

D. A.

Lefler

E. K.

Hartung

C. M.

Canu

W. H.

(2012). Sex differences in the manifestation of ADHD in emerging adults. Journal of Attention Disorders, 16(2), 109–117. https://doi.org/10.1177/1087054710374596

22.

Gibbins

Weiss

M. D.

Goodman

D. W.

Hodgkins

P. S.

Landgraf

J. M.

Faraone

S. V.

(2010). ADHD-hyperactive/impulsive subtype in adults. Mental Illness, 2(1), e9. https://doi.org/10.4081/mi.2010.e9

23.

Goh

P. K.

Elkins

A. R.

Bansal

P. S.

Eng

A. G.

Martel

M. M.

(2023). Data-driven methods for predicting ADHD diagnosis and related impairment: The potential of a machine learning approach. Research on Child and Adolescent Psychopathology, 51(5), 679–691. https://doi.org/10.1007/s10802-023-01022-7

24.

Goh

P. K.

Martel

M. M.

Barkley

R. A.

(2020). Clarifying ADHD and sluggish cognitive tempo item relations with impairment: A network analysis. Journal of Abnormal Child Psychology, 48(8), 1047–1061. https://doi.org/10.1007/s10802-020-00655-2

25.

Goh

P. K.

Martel

M. M.

Jones

P. J.

Bansal

P. S.

Eng

A. G.

Elkins

A. R.

Thaxton

M. H.

Barkley

R. A.

(2023). Clarifying relations between ADHD and functional impairment in adulthood: Utilization of network and machine learning approaches. Assessment, 30(2), 316–331. https://doi.org/10.1177/10731911211050921

26.

Gratz

K. L.

Roemer

(2004). Multidimensional assessment of emotion regulation and dysregulation: Development, factor structure, and initial validation of the difficulties in emotion regulation scale. Journal of Psychopathology and Behavioral Assessment, 26, 41–54. https://doi.org/10.1023/B:JOBA.0000007455.08539.94

27.

Hallion

L. S.

Steinman

S. A.

Tolin

D. F.

Diefenbach

G. J.

(2018). Psychometric properties of the Difficulties in Emotion Regulation Scale (DERS) and its short forms in adults with emotional disorders. Frontiers in Psychology, 9, 1–12. https://doi.org/10.3389/fpsyg.2018.00539

28.

Haslam

Williams

Prior

Haslam

Graetz

Sawyer

(2006). The latent structure of attention-deficit/hyperactivity disorder: A taxometric analysis. Australian and New Zealand Journal of Psychiatry, 40(8), 639–647. https://doi.org/10.1080/j.1440-1614.2006.01863.x

29.

Hirsch

Chavanon

Riechmann

Christiansen

(2018). Emotional dysregulation is a primary symptom in adult attention-deficit/hyperactivity disorder (ADHD). Journal of Affective Disorders, 232, 41–47. https://doi.org/10.1016/j.jad.2018.02.007

30.

Hosmer

D. W.

Lemeshow

Sturdivant

R. X.

(2013). Applied logistic regression (3rd ed.). John Wiley & Sons, Inc.

31.

Ishwaran

(2007). Variable importance in binary regression trees and forests. Electronic Journal of Statistics, 1, 519–537. https://doi.org/10.1214/07-EJS039

32.

Kazak

A. E.

(2018). Editorial: Journal article reporting standards. American Psychologist, 73(1), 1–2. https://doi.org/10.1037/amp0000263

33.

Kosheleff

A. R.

Mason

Jain

Koch

Rubin

(2023). Functional impairments associated with ADHD in adulthood and the impact of pharmacological treatment. Journal of Attention Disorders, 27(7), 669–697. https://doi.org/10.1177/10870547231158572

34.

Kuhn

(2008). Building predictive models in R using the caret package. Journal of Statistical Software, 28(5), 1–26. https://doi.org/10.18637/jss.v028.i05

35.

Kuhn

Johnson

(2018). Applied predictive modeling. Springer.

36.

Kumar

Indrayan

(2011). Receiver operating characteristic (ROC) curve for medical researchers. Indian Pediatrics, 48(4), 277–287. https://doi.org/10.1007/s13312-011-0055-4

37.

Lovibond

P. F.

Lovibond

S. H.

(1995). The structure of negative emotional states: Comparison of the Depression Anxiety Stress Scales (DASS) with the Beck Depression and Anxiety Inventories. Behaviour Research and Therapy, 33, 335–343. https://doi.org/10.1016/0005-7967(94)00075-U

38.

Lorenzi

C. H.

Teixeira Leffa

Bressan

Belangero

Gadelha

Santoro

M. L.

Salum

G. A.

Rohde

L. A.

Caye

(2023). Replication of a predictive model for youth ADHD in an independent sample from a developing country. Journal of Child Psychology and Psychiatry, 64(1), 167–174. https://doi.org/10.1111/jcpp.13682

39.

Marcus

D. K.

Barry

T. D.

(2011). Does attention-deficit/hyperactivity disorder have a dimensional latent structure? A taxometric analysis. Journal of Abnormal Psychology, 120(2), 427-442. https://doi.org/10.1037/a0021405

40.

Marshall

Hoelzle

Nikolas

(2021). Diagnosing attention-deficit/hyperactivity disorder (ADHD) in young adults: A qualitative review of the utility of assessment measures and recommendations for improving the diagnostic process. The Clinical Neuropsychologist, 35(1), 165–198. https://doi.org/10.1080/13854046.2019.1696409

41.

McLennan

J. D.

(2016). Understanding attention deficit hyperactivity disorder as a continuum. Canadian Family Physician, 62(12), 979–982.

42.

Mooney

M. A.

Neighbor

Karalunas

Dieckmann

N. F.

Nikolas

Nousen

Tipsord

Song

Nigg

J. T.

(2023). Prediction of attention-deficit/hyperactivity disorder diagnosis using brief, low-cost clinical measures: A competitive model evaluation. Journal of the Association for Psychological Science, 11(3), 458–475. https://doi.org/10.1177/21677026221120236

43.

Mulraney

Arrondo

Musullulu

Iturmendi-Sabater

Cortese

Westwood

S. J.

Donno

Banaschewski

Simonoff

Zuddas

Döpfner

Hinshaw

S. P.

Coghill

(2022). Systematic review and meta-analysis: Screening tools for attention-deficit/hyperactivity disorder in children and adolescents. Journal of the American Academy of Child and Adolescent Psychiatry, 61(8), 982–996. https://doi.org/10.1016/j.jaac.2021.11.031

44.

Nguyen

Q. H.

H. B.

L. S.

Al-Ansari

H. V.

Tran

V. Q.

Prakash

Pham

B. T.

(2021). Influence of data splitting on performance of machine learning models in prediction of shear strength of soil. Mathematical Problems in Engineering, 2021, 4832864. https://doi.org/10.1155/2021/4832864

45.

Olson

S. L.

(2002). Developmental perspectives. In Sandberg

(Ed.), Hyperactivity and attention disorders of childhood (2 ed., pp. 242–289). Cambridge University Press.

46.

Ruan

Q. N.

Chen

Y. H.

Yan

W. J.

(2023). A network analysis of difficulties in emotion regulation, anxiety, and depression for adolescents in clinical settings. Child and Adolescent Psychiatry and Mental Health, 17(1), 1–11. https://doi.org/10.1186/s13034-023-00574-2

47.

Sawaya

Miller

J. C.

Raines

J. M.

(2024). Review of studies on incremental validity of assessment measures used in psychological assessment of attention-deficit hyperactivity disorder. Assessment, 31(2), 518–537. https://doi.org/10.1177/10731911231159933

48.

Sawyer

(1966). Measurement and prediction, clinical and statistical. Psychological Bulletin, 66(3), 178–200. https://doi.org/10.1037/h0023624

49.

Scornet

Biau

Vert

J. P.

(2015). Consistency of random forests. The Annals of Statistics, 43(4), 1716–1741. https://doi.org/10.1214/15-AOS1321

50.

Shaw

Stringaris

Nigg

Leibenluft

(2014). Emotion dysregulation in attention deficit hyperactivity disorder. American Journal of Psychiatry, 171(3), 276–293. https://doi.org/10.1176/appi.ajp.2013.13070966

51.

Sibley

M. H.

Arnold

L. E.

Swanson

J. M.

Hechtman

L. T.

Kennedy

T. M.

Owens

Molina

B. S. G.

Jensen

P. S.

Hinshaw

S. P.

Roy

Chronis-Tuscano

Newcorn

J. H.

Rohde

L. A.

MTA Cooperative Group (2022). Variable patterns of remission from ADHD in the Multimodal Treatment Study of ADHD. The American Journal of Psychiatry, 179(2), 142–151. https://doi.org/10.1176/appi.ajp.2021.21010032

52.

Sibley

M. H.

Pelham

W. E.

Molina

B. S. G.

Gnagy

E. M.

Waschbusch

D. A.

Garefino

A. C.

Kuriyan

A. B.

Babinski

D. E.

Karch

K. M.

(2012). Diagnosing ADHD in adolescence. Journal of Consulting and Clinical Psychology, 80(1), 139–150. https://doi.org/10.1037/a0026577

53.

Soler-Gutiérrez

A. M.

Pérez-González

J. C.

Mayas

(2023). Evidence of emotion dysregulation as a core symptom of adult ADHD: A systematic review. PloS One, 18(1), e0280131. https://doi.org/10.1371/journal.pone.0280131

54.

Song

Zha

Yang

Zhang

Rudan

(2021). The prevalence of adult attention-deficit hyperactivity disorder: A global systematic review and meta-analysis. Journal of Global Health, 11, 04009. https://doi.org/10.7189/jogh.11.04009

55.

Walsh

E. S.

Kreakie

B. J.

Cantwell

M. G.

Nacci

(2017). A random forest approach to predict the spatial distribution of sediment pollution in an estuarine system. PLoS One, 12(7), e0179473. https://doi.org/10.1371/journal.pone.0179473

56.

Weiss

M. D.

(2000). Weiss Functional Impairment Rating Scale (WFIRS) self-report. University of British Columbia. https://naceonline.com/AdultADHDtoolkit/assessmenttools/wfirs.pdf

57.

Weiss

M. D.

McBride

N. M.

Craig

Jensen

(2018). Conceptual review of measuring functional impairment: Findings from the Weiss Functional Impairment Rating Scale. Evidence-Based Mental Health, 21(4), 155–164. https://doi.org/10.1136/ebmental-2018-300025

58.

Wongvorachan

Bulut

(2023). A comparison of undersampling, oversampling, and SMOTE methods for dealing with imbalanced classification in educational data mining. Information, 14(1), 54–68. https://doi.org/10.3390/info14010054

59.

Wymbs

B. T.

Dawson

A. E.

(2019). Screening Amazon’s Mechanical Turk for adults with ADHD. Journal of Attention Disorders, 23(10), 1178–1187. https://doi.org/10.1177/1087054715597471

60.

Youden

W. J.

(1950). Index for rating diagnostic tests. Cancer, 3(1), 32–35. https://doi.org/10.1002/1097-0142(1950)3:1<32::AID-CNCR2820030106>3.0.CO;2-3

61.

Zanon

Brenner

R. E.

Baptista

M. N.

Vogel

D. L.

Rubin

Al-Darmaki

F. R.

Gonçalves

Heath

P. J.

Liao

H.-Y.

Mackenzie

C. S.

Topkaya

Wade

N. G.

Zlati

(2021). Examining the dimensionality, reliability, and invariance of the Depression, Anxiety, and Stress Scale–21 (DASS-21) across eight countries. Assessment, 28(6), 1531–1544. https://doi.org/10.1177/1073191119887449

Emotional Dysregulation in Emerging Adult ADHD: A Key Consideration in Explaining and Classifying Impairment and Co-Occurring Internalizing Problems

Abstract

Objective:

Method:

Results:

Conclusion:

Keywords

Limitations in ADHD’s Symptom List During Emerging Adulthood

Emotional Dysregulation and Unawareness

Using Machine Learning to Clarify EDU Dimensions’ Incremental Validity

The Current Study

Methods

Participants

Procedures

Measures

ADHD Symptoms

EDU 1

Functional Impairment

Internalizing Problems

Statistical Analyses

Transparency and Openness

Random Forest Development

Integrating EDU with ADHD Symptoms and Testing the Final Model

Results

Testing Hypotheses 1 and 2: Examining the Incremental Validity of EDU 2

Testing Hypothesis 3: Importance of ADHD Symptoms and EDU Dimensions Across Algorithms

Post Hoc Analysis: Replicating Findings in Those Without ADHD and Testing the Incremental Validity of Sex

Discussion

EDU Demonstrated Incremental Validity Beyond ADHD Symptoms

Limited Access to Emotion Regulation Strategies Was More Important for Classification Than HI

Potential Benefits of Incorporating EDU Into ADHD-Related Clinical Practices

Limitations and Constraints on Generality

Conclusion

Footnotes

Acknowledgements

Declaration of Conflicting Interests

Funding

ORCID iDs

Notes

Author Biographies

References

EDU¹

Testing Hypotheses 1 and 2: Examining the Incremental Validity of EDU²