Authors: Jacob DeRosa (1Center for the Developing Brain, Child Mind Institute, New York, NY, USA.; 2Department of Psychology and Neuroscience, University of Colorado Boulder, Boulder, CO, USA.), Keri S Rosch (3Center for Neurodevelopmental and Imaging Research, Kennedy Krieger Institute, Baltimore, MD, USA.; 4Department of Neuropsychology, Kennedy Krieger Institute, Baltimore, MD, USA.; 5Department of Psychiatry and Behavioral Sciences, Johns Hopkins University School of Medicine, Baltimore, MD, USA.), Stewart H Mostofsky (3Center for Neurodevelopmental and Imaging Research, Kennedy Krieger Institute, Baltimore, MD, USA.; 5Department of Psychiatry and Behavioral Sciences, Johns Hopkins University School of Medicine, Baltimore, MD, USA.; 6Department of Neurology, Johns Hopkins University School of Medicine, Baltimore, MD, USA.), Aki Nikolaidis (1Center for the Developing Brain, Child Mind Institute, New York, NY, USA.)
Categories: Article, Delay Discounting, ADHD, Transdiagnostic, Development, Household Income
Source: Journal of child psychology and psychiatry, and allied disciplines
Doi: 10.1111/jcpp.13870
Authors: Jacob DeRosa, Keri S Rosch, Stewart H Mostofsky, Aki Nikolaidis
The tendency to prefer smaller, immediate rewards over larger, delayed rewards is known as Delay Discounting (DD). Developmental deviations in DD may be key in characterizing psychiatric and neurodevelopmental disorders. Recent work empirically supported DD as a transdiagnostic process in various psychiatric disorders. Yet, there is a lack of research relating developmental changes in DD from mid-childhood to adolescence to psychiatric and neurodevelopmental disorders. Additionally, examining the interplay between socioeconomic status (SES)/total household income (THI) and psychiatric symptoms is vital for a more comprehensive understanding of pediatric pathology and its complex relationship with delay discounting (DD).
The current study addresses this gap in a robust psychiatric sample of 1843 children and adolescents aged 5-18 (M = 10.6, SD = 3.17; 1219 males, 624 females). General Additive Models (GAMs) characterized the shape of age-related changes in monetary and food reward discounting for nine psychiatric disorders compared to neurotypical youth (NT; n=123). Over 40% of our sample possessed a minimum of at least three psychiatric or neurodevelopmental disorders. We used bootstrap-enhanced Louvain community detection to map DD-related comorbidity patterns. We derived five subtypes based on diagnostic categories present in our sample. DD patterns were then compared across each of the subtypes. Further, we evaluated the effect of cognitive ability, emotional and behavioral problems, and THI in relation to DD across development.
Higher discounting was found in six of the nine disorders we examined relative to NT. DD was consistently elevated across development for most disorders, except for depressive disorders, with age-specific DD differences compared to NTs. Community detection analyses revealed that one comorbidity subtype consisting primarily of Attention-Deficit/Hyperactivity Disorder (ADHD) Combined Presentation and anxiety disorders displayed the highest overall emotional/behavioral problems and greater DD for the food reward. An additional subtype composed mainly of ADHD, predominantly Inattentive Presentation, learning, and developmental disorders, showed the greatest DD for food and monetary rewards compared to the other subtypes. This subtype had deficits in reasoning ability, evidenced by low cognitive and academic achievement performance. For this ADHD-I and developmental disorders subtype, THI was related to DD across the age span such that participants with high THI showed no differences in DD compared to NTs. In contrast, participants with low THI showed significantly worse DD trajectories than all others. Our results also support prior work showing that DD follows non-linear developmental patterns.
We demonstrate preliminary evidence for DD as a transdiagnostic marker of psychiatric and neurodevelopmental disorders in children and adolescents. Comorbidity subtypes illuminate DD heterogeneity, facilitating the identification of high-risk individuals. Importantly, our findings revealed a marked link between DD and intellectual reasoning, with children from lower-income households exhibiting lower reasoning skills and heightened DD. These observations underscore the potential consequences of compromised self-regulation in economically disadvantaged individuals with these disorders, emphasizing the need for tailored interventions and further research to support improved outcomes.
Delay discounting (DD) is a phenomenon in which individuals prefer smaller, immediate rewards over larger, delayed rewards (Green & Myerson, 2010). Recent work has identified DD as a key transdiagnostic marker across a range of psychiatric and neurodevelopmental disorders, including Attention-Deficit/Hyperactivity Disorder (ADHD), anxiety, and Autism Spectrum Disorder (ASD) (Amlung et al., 2019), such that individuals with these disorders tend to show higher discounting rates compared to neurotypical (NT) controls (Pinto et al., 2014; Sohn et al., 2014). However, we are currently uncertain if DD can be considered a transdiagnostic marker of mental illness during childhood and adolescence. Answering this question is important, considering that DD is associated with various adverse outcomes, including addictive behaviors, impulsivity, and poor decision-making (Odum et al., 2020). In addition, identifying the age of onset of DD might help inform current transdiagnostic treatment efforts for targeting core DD behavioral processes to prevent adverse life outcomes and provide markers for assessing response to interventions (Pasion & Barbosa, 2019).
Higher discounting in psychiatric and neurodevelopmental disorders is well-documented (Lempert et al., 2019) and is proposed as a potential endophenotype for many problematic behaviors. DD also may be an efficient marker of individual differences relevant to treatment outcomes (Ahn et al., 2011). Yet, mixed findings have been reported on DD as a reliable transdiagnostic process (Bailey et al., 2021). A detailed understanding of the developmental variance of DD (Anandakumar et al., 2018) and how it relates to psychiatric and neurodevelopmental disorders across development is key to understanding the utility of DD as a transdiagnostic marker in developing psychiatric populations. Proper characterization of such developmental patterns would require evaluating a large and enriched transdiagnostic sample. Indeed, more DD studies from a developmental perspective are needed. For example, Dekkers et al. (2012) highlight the need to examine ADHD-control differences in developmental trajectories of impulsive and risky decision-making to identify critical developmental periods of impulsivity and risk-related opportunities. This notion can be held for all prevalent disorders diagnosed in children and adolescents.
Heterogeneity in cognitive profiles and behavioral presentations within disorders has been recognized as an intricate problem hindering progress in psychiatric and cognitive research (Allsopp et al., 2019). Directions to address this issue have called for hybrid approaches to identify subtypes capable of better-explaining outcomes and advancing treatment-based tools (Feczko et al., 2019). Approximately 50% of mental health problems are established by age 14 and 75% by age 24, and the lifetime prevalence of two or more disorders was found to be between 17-27.7% (Kessler et al., 2005). ADHD co-occurs with depression in about 20-30% of patients, anxiety in over 25% of patients, and learning disorders in approximately 45% of patients (Larson et al., 2011; Michielsen et al., 2013). Individuals with co-occurring disorders such as ADHD, anxiety, and depression have been shown to have higher symptom severity and lower overall quality of life (Yang et al., 2013). Despite the prevalence of comorbidities, most studies have not assessed DD in psychiatric comorbidity, limiting our understanding of how underlying comorbidities contribute to DD.
It is crucial to address if children and adolescents with certain comorbid disorders are at a higher risk for DD to aid our understanding of the proper development of treatment for comorbid youth. Our objective is to investigate comorbidity clusters to determine if specific groups of co-occurring mental health disorders show increased DD severity. We neither assume nor imply that comorbidities cause DD but aim to explore their potential role in intensifying atypical reward valuation. In addition, we aim to uncover the impact of comorbidity clusters on DD to understand if varying disorder compositions (e.g., ADHD Combined, Anxiety, Depression) reveal heightened DD compared to other combinations (e.g., ADHD Combined, Learning Disorders, Depression). This approach recognizes the intricate connections between mental health conditions and enables a comprehensive examination of overlapping symptoms, cognitive processes, and emotional/behavioral problems related to DD.
Total Household income (THI), an indicator of social-economic status (SES), is necessary to understand and assess psychiatric illness. Lower THI has been linked to heightened delay discounting rates (Hampton et al., 2018). This relationship indicates the influence of financial well-being on an individual's cognitive and decision-making processes. Moreover, lower THI has been consistently associated with a higher prevalence of mental health issues and psychiatric disorders, including depressive disorders and substance use disorders (Zimmerman & Katon, 2005). The underlying mechanisms connecting THI to these disorders may stem from increased stress, limited access to resources, and other psychosocial factors exacerbating psychiatric symptoms. In addition, previous findings demonstrated that economic poverty, apart from SES, has been linked to higher DD in psychiatric disorders (Lorant et al., 2007). The heightened DD in individuals facing financial strain can be attributed to the necessity for immediate access to rewards, such as money (Oshri et al., 2019). Individuals in economically precarious situations may prioritize short-term gains over long-term rewards, as the former might address pressing needs or alleviate immediate stress. This pattern of decision-making may exacerbate or prolong psychiatric symptoms, creating a self-perpetuating cycle that underscores the importance of understanding the intricate relationship between SES, DD, and psychiatric illness. Therefore, comprehensively examining the interactions between these factors is essential.
DD was assessed using five different monetary items and one food item that were combined with factor analysis to extract factors of DD behavior (Koffarnus & Bickel, 2014). These factors were analyzed with a highly enriched sample of children and adolescents from the Healthy Brain Network Biobank (Alexander et al., 2017) to investigate two core aims.
Our first aim assessed transdiagnostic age-specific deviations of the most common psychiatric and neurodevelopmental disorders compared to those of neurotypicals (NTs) in multiple types of DD through developmental pattern modeling with Generalized Additive Models (GAMs). Age-specific deviations, occurring from ages 5-18, denote the average increased or decreased developmental disorder (DD) prevalence in individuals compared to neurotypicals. In other words, where the difference in the DD trend between a given disorder and neurotypicals significantly differs (deviates) from the null hypothesis (Difference in DD = 0). To do so, we compared the non-linear developmental trajectory of the DD patterns of each disorder to that of neurotypicals. We model these trajectories across the ages of 5-18 to evaluate if there is a significant difference between these two trajectories and at which ages. The Methods section, 2.8, outlines how these trajectory comparisons between disorders and neurotypicals were modeled and how we determined the age range in which these trajectories differed. To our knowledge, previous studies have yet to use these models to investigate within-group comparisons of DD as a function of chronological age.
Our second aim generated transdiagnostic subtypes using community detection to determine how certain disorders cluster together and whether DD differed across transdiagnostic subtypes and development. Finally, we investigated if household income is related to DD across these subtypes.
Data were obtained from the ongoing Child Mind Institute HBN Biobank (release 9.0), leveraging a community-based self-referral recruitment model to construct a transdiagnostic sample of 10,000 participants. The sample in the following analyses comprises 1843 (430 female) participants aged 5-18 (mean 10.16 ± 3.17). Participants included in the present work were based on complete data available for the ADT-5.
The ADT-5 obtained individual discount rates by measuring k for six unique items. This task directly measures the k value for a given item. Higher k values indicate greater DD (Odum, 2011). The ADT-5 presents a series of questions between some amount of a delayed item and half that amount available immediately. These amounts remain stagnant while the delay to the larger amount is adjusted to determine the k value. The first-choice trial is always between the amount of the item delayed 3 weeks and the amount of the same item available immediately. Depending on the choice made by the participant, the delay either adjusts up (delayed choice) or down (immediate choice) by 8 delays (index 8 or 24) for the next choice trial. This continues for five-choice trials, with the delay index adjusting by an amount half that of the previous adjustment. This results in 32 potential k values (2^5^) nearly evenly spaced (on a logarithmic scale) between 1 hour and 25 years, the same number of possible indifference points at each delay of the adjusting amount procedure above. Participants completed this task six times for different commodities and delayed amounts. These included three versions where the delayed amount was 5 now), 500 now), and 500,000 now). Fourth, participants completed a version presenting choices between 500 delivered 1 hour ago. Fifth, an “explicit zero” version was also included, which presented choices between 500 now, but with the options presented differently. In this version, delayed options were presented as “0 now” vs. “0 in [delay].” Sixth, a version of the task was completed presenting choices between 10 servings of the participant’s preferred snack food delivered after a delay vs. 5 servings now. Additional information on the ADT-5 can be found in the supporting information (Appendix S3.1).
Participants and their parents or legal guardian met with a licensed clinician who administered the Kiddie Schedule for Affective Disorders and Schizophrenia (K-SADS; (J. Kaufman et al., 1997); a semi-structured DSM-V-based psychiatric interview to derive a clinical diagnosis (if applicable; (Appendix S1.1).
Participants were administered the Wechsler Individual Achievement Test, 3^rd^ edition (WIAT-III; (D. Wechsler, 2005), the Wechsler Intelligence Scale for Children, 5^th^ edition (WISC-V; for participants ages 6-17 years old; (David Wechsler, 2012) and The Kaufman Brief Intelligence Test, Second Edition (KBIT-2; for participants ages 5-6 years old; (A. S. Kaufman & Kaufman, n.d.). The age-adjusted standardized index and subtest scores were examined as indicators of cognitive and academic abilities (WISC: Visual Spatial, Verbal Comprehension Fluid Reasoning, Working Memory, Processing Speed; WIAT: Numeracy, Pseudoword decoding, Spelling, Oral Expression, Listening Comprehension, Word Reading, Reading Comprehension, Math Problem Solving). (Appendix S3.3-5).
Questionnaires relating to behavior, financial status, and demographics were completed by participants and their parents or legal guardians over the course of their visits. Subtest T-scores (adjusted for age and sex) from the Child Behavior Checklist (CBCL; (Achenbach, 2001) were examined as indicators of emotional and behavioral problems (EBP).
Factor analysis was used to investigate latent choice DD constructs on the six commodities’ log-transformed k-values through an Exploratory Factor Analysis (EFA). Confirmatory Factor Analysis (CFA) was used to verify our factor structure using the R package “lavaan” (Rosseel, 2012) (Appendix S2.5-S.2.6).
Bagging enhanced Louvain community detection (LCD; (Blondel et al., 2008; Nikolaidis et al., 2022, 2021) was used to obtain data-driven diagnostic subtypes to better address the high comorbidity found within our DXs of interest. Our previous work has demonstrated the ability of bagging to enhance the signal-to-noise ratio of clustering and improve within and between sample reproducibility of clusters (Nikolaidis et al., 2020) (Appendix S2.4). All remaining DXs not within our primary DXs were coded under a created binary category (1 for any additional diagnoses, 0 for no additional diagnoses), dubbed “other”, which included communication, elimination, and motor developmental disorders.
GAMs were used to characterize age-related effects, along with diagnostic group and other demographic measures on the three DD factors using the “mgcv” package (Wood, 2018) in R. The first derivatives of the smooth function of age from the GAM model were calculated using finite differences to test for windows of significant change between groups across age. They then generated a simultaneous 95% confidence interval of the derivative (Simpson, 2018). Intervals of significant differences were identified as areas where the simultaneous confidence interval of the derivative does not include zero. Study site and sex were included as covariates in all GAMs. Follow-up GAMs covaried for Total household income (THI) and IQ, and missing data were handled via model-wise deletion. Multiple corrections were handled using the false discovery rate method (Benjamini & Hochberg, 1995). Specifically, we adjusted the significance threshold based on the number of diagnoses (n=10) and subtype (n=5) comparisons separately for each DD factor.
We first compared NTs to individual diagnostic (DX) groups of interest and based on their proportion relative to the rest of the sample, including the ADHD Combined Presentation (ADHD-C), ADHD Inattentive Presentation (ADHD-I), anxiety disorders, depressive disorders, learning disorders, Obsessive-Compulsive Disorder (OCD), Oppositional Defiant Disorder (ODD), and ASD. GAMs with ADHD-C and ADHD-I did not include individuals with ASD. Individuals who had a comorbid diagnosis of ASD and ADHD-C or ADHD-I were assigned to their own groups, respectively. We then computed a series of GAMs that investigated our DX transdiagnostic subtypes compared to NTs.
THI was divided into bins (high and low to moderate) to investigate how economic status shapes DD patterns across transdiagnostic subtypes. The first GAMs compared DX subtypes within the high THI brackets. The second GAMs compared DX subtypes within the low/moderate THI groups. Our final analyses compared THI groups within individual subtypes to assess the economic impact. In these models, subtypes were individually extracted from the sample, and the group comparisons were between the THI high and low/moderate groups.
No outliers were detected for the DD items, and missing data (n=210) were excluded before analyses. Distributions of the k-values obtained from the ADT-5 assessment were negatively skewed and given a natural log transformation. ADT-5 performance did not significantly differ between the three data collection sites. Correlations among the six DD items (range = .18 - .49) indicated that discounting at different magnitudes generally overlapped but at varying effect sizes (Figure S1).
Parallel Analysis “scree” plot estimation (Horn, 1965) from the R package “paran”, informed the three-factor solution (Figure S2), and EFA (Table S1) investigated the underlying factor structure of the six ADT-5 items. CFA was evaluated with the comparative fit index (CFI 0.974), the Tucker Lewis index (TLI = 0.952), and the root-mean-square error of approximation (RMSEA = 0.06) and Omega (ω = 0.84); all indicating the three-factor model to be an appropriate fit. Factor one (smaller sooner monetary reward; SSMR) revealed positive loadings from the k-values of 1,000, 1,000 Past (range = .56 to .73). Factor two (Snack) loaded selectively on the snack item and Factor 3 on the $1,000,00 item (larger later monetary reward; LLMR). This three-factor solution was consistent with previous research that used dimensionality reduction as an exploratory method to detect underlying decision-making constructs (Koffarnus & Bickel, 2014; Anounk Scheres et al., 2010). The remaining analyses presented in this paper use the three derived DD factors. Correlations between the DD factors and Age, IQ, THI, SES can be seen in Figure S1.
Our first set of GAMs controlled for sex and study site effects resulting in significant age-constant differences for the ADHD-C group compared to NTs on the SSMR (t=2.839, p=.005, age-window of significant differences [AWD]=7.44 −17.88) (Figure 1A). For the snack factor, depressive disorders by age interaction emerged (F=3.389, p=.033, AWD=9.57-14.16) (Figure 1D) whereas constant effects across age were observed for youth with learning disorders (t=2.146, p=.032, AWD=5.04-17.88) (Figure 1C), anxiety disorders (t=1.973, p=.049, AWD=5.1-17.81; Figure 1B), and ADHD-I (t=1.995, p=.046, AWD=5.1-17.88) compared to NTs. Age-constant significant differences were observed for the LLMR factor among the ADHD-C (t=2.301, p=.022, AWD=5.06-17.88) (Figure 1E) and ASD without ADHD groups (t=2.035, p=.043, AWD=5.1-17.81; Figure 1F), such that both diagnostic groups showed higher DD compared to NTs. Statistical comparisons for all individual DX group GAM models can be found in Tables S2-S5. Table 2 includes a generalized summary of the findings for the Diagnosis-Neurotypical differences in DD, and on the Developmental Trajectory of DD.
Bagging enhanced LCD on the binary DX categories revealed five distinct DX transdiagnostic subtypes (Figure 2). Subtype names were determined based on their primary diagnostic composition and remarkable cognitive/academic (abbreviated “Cog”) and emotional/behavioral (abbreviated internalizing [Int], externalizing [Ext], and high overall emotional/behavior problems [EBP]) profiles.
Subtype ADHD-I/LD/AvgCog, was proportionally high in ADHD-I and learning disorders, with average cognitive/academic scores and relatively low to moderate emotional/behavioral problems.
Subtype ANX/ADHD-I/HighInt was proportionally high in ADHD-I and anxiety disorders, with average cognitive/academic scores and high internalizing problems.
Subtype ADHD-I/LD/LowCog was proportionally high in “other” disorders, learning disorders, and ADHD-I with low cognitive/academic scores and low to moderate emotional/behavioral problems.
Subtype ADHD-C/HighExt was proportionally high in ADHD-C, with generally average cognitive/academic scores and high externalizing behavior, thought, and social problems.
Subtype ADHD-C/ANX/HighEBP was proportionally high in ADHD-C and anxiety disorders, with average cognitive/academic scores and high emotional/behavioral problems.
One-way ANOVA tests revealed significant differences between subtypes on the CBCL, WISC, and WIAT subscales (Tables S6-S7). On average, we observe that NTs revealed the highest scores across the WISC and WIAT subscales and the lowest scores across CBCL subscales. Significant differences in sex (χ^2^ = 92.05, p < .001), Age (χ^2^ = 67, p < .001), THI (χ^2^ = 22.04, p < .001), and Race (χ^2^ = 35.16, p < .05) were observed across transdiagnostic subtypes (Table 1).
Post-hoc z-tests indicated that relative to NTs, ADHD-I/LD/LowCog and ADHD-C/ANX/HighEBP were proportionally higher in the <50k Total Household income bracket. NTs were also proportionally higher in the $100k+ bracket compared to ADHD-I/LD/LowCog and ADHD-C/HighExt and were proportionally lower in Non-Hispanic Black populations compared to those subtypes. Relative to the other transdiagnostic subtypes, ANX/ADHD-I/HighInt was characterized by a significantly higher proportion of Females, whereas the ADHD-C/HighExt and ADHD-C/ANX/HighEBP subtypes were characterized by a significantly higher proportion of males. See Table 1 for all significant demographic post-hoc comparisons. Table 2 includes a generalized summary of the findings for the Comorbidity Subtype-Neurotypical differences in DD and on the Developmental Trajectory of DD.
GAMs tested if diagnostic subtypes show differential DD compared to NTs across development (Tables S8-S11). ADHD-I/LD/LowCog showed significantly higher DD for the SSMR (t=−1.99, p=.47, AWD=8.68-13.59; Figure 3A) and the Snack (t=−2.33, p=.02, AWD=5.1-17.88; Figure 3B), across development, whereas a group-by-age interaction was revealed for the LLMR (F=2.876, p=.017, AWD=10.74-13.75; Figure 3D). ADHD-C/ANX/HighEBP showed higher discounting across development than NTs on the snack (t=−2.529, p=.012, AWD=5.1-17.59) (Figure 3C), which remained present across all covariate models.
THI measures the total income obtained each year by both primary caregivers or one primary caregiver if only one is present in the household. Notably, THI was a significant covariate in all the monetary factor models that compared the DD pattern of ADHD-I/LD/LowCog to the NTs and the other transdiagnostic subtypes. Post-hoc analyses examined how THI differences impacted DD across subtypes.
The first pair of GAMs compared ADHD-I/LD/LowCog to the rest of the sample for all individuals with low-moderate THI. The low-moderate THI sample comparisons (SSMR: t=3.003, p=.003, AWD=5.1-17.88; LLMR: t=3.095, p=.002, AWD=8.59-17.88) (Figure 4C-D) revealed differences between ADHD-I/LD/LowCog and the rest of the sample. However, no significant differences were observed in the high THI sample subtype comparisons (Figure 4A-B). Based on the significant differences observed across our THI analyses above, we probed to evaluate if THI impacted individuals within the ADHD-I/LD/LowCog subtype. These analyses revealed that the low-moderate THI group had significantly higher DD across development compared to the high THI group (SSMR: t=−3.992, p<.001, AWD=5.22-17.88; LLMR: t=−4.27, p<.001, AWD=6.93-17.88).
We first assessed the degree to which DD is a developmentally sensitive transdiagnostic process among children between the ages of 5-18 years. Compared to NTs, individuals with ADHD-C, ADHD-I, ASD without ADHD, depressive, anxiety, and learning disorders displayed higher DD for monetary or snack DD rewards. Our transdiagnostic subtypes offered the ability to address underlying heterogeneity across our diagnostic sample, revealing problematic transdiagnostic DD patterns in two distinct subtypes that included individuals with ADHD relative to NTs. Additionally, we found that household income is a critical risk factor to consider when examining monetary-based decision-making. Further, we find that lower THI exacerbates DD patterns in individuals with lower cognitive abilities. Overall, the modeling framework we present here allowed for a more precise characterization of DD as a transdiagnostic process across psychiatric and neurodevelopmental disorders and the important role of socioeconomic factors.
Our findings of elevated DD for monetary rewards in children with ADHD-C compared to NTs are consistent with prior work (Jackson & MacKillop, 2016; Marx et al., 2021; Patros et al., 2017) and extend this work to examine developmental changes and the impact of comorbidity. For individual diagnostic group comparisons, children with ADHD showed elevated DD for monetary rewards across development, suggesting that this may be a relatively stable trait within this population. In contrast, elevated DD was not observed in ADHD-I relative to NT for monetary rewards. This is consistent with limited prior work reporting that DD was specifically related to hyperactive/impulsive symptoms rather than inattentive symptoms (A. Scheres et al., 2008). However, children with ADHD-I did show elevated DD for food rewards that persisted across development. Given the lack of studies examining DD for non-monetary rewards across development, it will be important to replicate these findings. However, it may suggest that reward processing abnormalities are present across both the ADHD-C and ADHD-I presentations, but they may be commodity specific. Elucidating the task parameters that may lead to distinct patterns of DD in ADHD subgroups may be important for differential intervention approaches.
Elevated DD among individuals with anxiety disorders was specific to the food rewards. One issue relating this finding to previous work is that few studies exist on anxiety disorders and DD in developmental populations. One study investigating a sample of 44 adults with Social Anxiety Disorder (SAD) did not find elevated monetary DD compared to NTs (Steinglass et al., 2017). Examining the major types of anxiety disorders, such as SAD, was out of the scope of the present work. However, given the lack of research on DD within the major types of anxiety disorders, it is worth investigating how these types may differ in their developmental DD patterns across the different rewards.
Mixed findings have been reported regarding atypical DD in relation to depressive disorders (Pulcu et al., 2014). There is still a lack of research investigating developmental changes in DD within this population. Here we find that individuals with depressive disorders show increased DD for the food reward only specific to the developmental period beginning around late childhood and early adolescence. This difference compared to NTs peaked at 11.5 years old and began to level off after 14. Our findings for depressive disorders also align with previous work (Levitt et al., 2022) that found DD to be specifically associated with depression in a Structural Equation Modeling analysis. They note that DD may be linked to a specific aspect of depression, such as a symptom or cluster of symptoms or behaviors. In addition, some studies have shown an association between DD and hopelessness in depression, while others have found decreased DD in individuals with anhedonia (Lempert & Pizzagalli, 2010). Importantly, depressive disorders were the only disorder that showed non-constant differences in DD across development. This suggests that for individuals with depressive disorders, the transition from childhood to adolescence may be hallmarked by periods of elevated reward processing abnormalities.
Our finding of elevated DD for monetary rewards among youth with ASD without comorbidity of ADHD is consistent with prior work that found individuals with only ASD discounted rewards more steeply than NTs and comorbid ASD with ADHD groups (Chantiluke et al., 2014). The elevated DD we observe for individuals with ASD may be due to their inability to effectively evaluate the magnitude of the LLMR due to certain cognitive deficiencies. The LLMR requires abstract prospective reasoning to consider the reward amount of $1,000,000 at delays of up to 25 years in the future. This notion aligns with previous findings on time-based and prospective processing deficiencies in individuals with ASD (Szelag et al., 2004; Williams et al., 2013). Our finding for individuals with ASD on the smaller soon monetary reward agrees with (Demurie et al., 2012), who also reported no difference in DD between ASD and their NTs peers. In addition, these findings are consistent with previous research (Antrop et al., 2006) that found that ADHD children, not children with Autism, show more heightened DD than controls. These delay-related features of small monetary reward choices can be considered a specific characteristic of ADHD and highlight the potential of delay-related motivational processes in differentiating ADHD and ASD.
A lack of research has also examined DD and learning disorders across development. Since learning disorders share a high prevalence of comorbidity with ADHD, individuals with these comorbidities may be driving the elevated DD. This notion seems to be confirmed by the elevated DD in one of our transdiagnostic subtypes, which consisted predominantly of ADHD-I and LD. We also did not observe differences in DD between individuals with OCD and NTs. Though mixed findings have been reported with OCD and DD, most studies show that individuals with OCD do not show elevated DD compared to NTs (Norman et al., 2017; Steinglass et al., 2017; Vloet et al., 2010). Finally, no differences in DD were discovered for individuals with ODD compared NTs. There is a lack of literature pertaining to ODD and DD across development; however, previous work did not find elevated DD in children with ODD and comorbid diagnosis and ADHD (Antonini et al., 2015).
Most diagnoses showed stable DD differences across development compared to NTs, except for depressive disorders, whose DD patterns showed higher variability across development. Our findings across these disorders support DD as a developmentally stable transdiagnostic process spanning internalizing and externalizing disorders. However, since these disorders are highly comorbid, alternative subtyping approaches are needed to clarify the most relevant clinical features/presentations across these disorders. The developmental stability and transdiagnostic nature of DD has important implications for understanding the underlying causes and mechanisms of DD in various disorders, and for early detection of psychiatric illness. Our findings support the idea that DD could be used as a potential endophenotype for psychiatric disorders (Amlung, 2023; Bickel, 2015; Ohmura et al., 2006).
The consistency in DD across developmental stages and diagnostic categories suggests that the underlying mechanisms and processes contributing to DD are consistent and could lead to the development of more targeted interventions and treatments (Kim-Spoon et al., 2015). The knowledge of DD’s stability could be used to predict future behavior more accurately, allowing for more effective interventions at an early age (Ohmura et al., 2006). Understanding the neurodevelopmental processes that underlie DD can help develop effective interventions and treatments for psychiatric disorders associated with impulsive decision-making and substance use, further emphasizing the importance of identifying DD as a developmentally stable transdiagnostic process (Kim-Spoon et al., 2015). Ultimately, the developmental stability and transdiagnostic nature of DD provide an avenue for improving our understanding of mental health and creating more effective interventions for psychiatric illness.
DD findings among our transdiagnostic subtypes have important implications for understanding transdiagnostic processes across these disorders. DD and self-regulation may not be specific to a particular disorder but rather a subset of individuals with particular comorbidities and more severe psychopathology. Nested heterogeneity across our disorders indicates the importance of considering the effect multiple diagnoses can have on influencing individual differences in cognitive and behavioral functioning within a given diagnostic category.
Our findings reveal two distinct transdiagnostic subtypes that display elevated DD relative to NTs. The subtypes with ADHD-I, learning disorders, and low cognitive/academic scores and with ADHD-C, anxiety disorders, and high emotional/behavioral problems both displayed elevated DD for food and monetary rewards (marginal effect for the latter group) compared to the NT group and did not significantly differ from each other in DD. Together, these subtypes provided key insights into how certain individuals will display elevated DD based on their multiple comorbidity profiles.
The subtypes revealed varying DD compared to neurotypicals during development, with ADHD-I/LD/LowCog showing higher DD in SSMR and Snack and ADHD-C/ANX/HighEBP exhibiting increased Snack discounting. This variation in elevated DD highlights the complexity of its origins, which may stem from factors like low critical thinking, immediate reward sensitivity, delay aversion, or poor impulse control. Our results thus encourage further investigation of distinct cognitive processes contributing to elevated DD in these specific subtypes. Perhaps deficits in cognitive reasoning abilities contribute to DD in the ADHD-I/LD/LowCog group. Reward valuation requires logic, critical thinking, and intuition to interpret and make sense of information, which are core cognitive reasoning processes (Hélie et al., 2017). These processes play an important role in DD by influencing an individual’s ability to decide the value of a reward that is available now versus a reward that will be available in the future. Individuals within the ADHD-I/LD/LowCog group likely possess deficits in these cognitive reasoning processes, making it difficult to understand the concept of future rewards and discount the value of these future rewards in favor of the more immediate rewards. Individuals within this group might benefit from training to improve their strategies for thinking about the long-term consequences of their actions and helping them to develop more effective problem-solving skills and impulse control strategies.
In contrast, increased sensitivity to immediate reward or a heightened aversion to delay and related negative affect might contribute to DD in the ADHD-C/ANX/HighEBP group. A consequence of delay aversion in ADHD has been linked to heightened DD (Dekkers et al., 2022), impulsive behavior, and difficulty achieving long-term goals. This phenomenon is also closely related to sensitivity to reward immediacy, which is influenced by various factors, including cognitive and emotional processes, past experiences, and individual differences in personality and motivation (Chiew & Braver, 2011). Previous work has also shown that individuals more sensitive to reward immediacy tend to have a stronger preference for immediate rewards and are more likely to engage in DD. These factors likely contribute to the higher DD observed within the ADHD-C/ANX/HighEBP group. Individuals within this group may benefit from training on techniques to help them increase self-control, set and achieve long-term goals, and tolerate short-term discomfort for long-term gain.
Co-occurring affective and learning/cognitive problems may also contribute to DD among youth with ADHD. Individuals with ADHD have difficulty with impulse control and decision-making, which can lead to an increased risk of DD. In addition, individuals with a co-occurring diagnosis of depression may also have a reduced sensitivity to reward, which can lead to an even larger increase in DD. Our findings suggest that DD could potentially serve as an affordable screening tool to identify individuals at risk for these various disorders, offering an alternative to costly and complex neuropsychological evaluations. Consequently, treatment should prioritize addressing the specific disorders rather than solely focusing on DD. Individuals with ADHD and learning disorders may have increased difficulty understanding the long-term consequences of their actions, which can contribute to heightened DD. Addressing the heterogeneity within ADHD populations is important, as different subgroups may have different underlying causes of heightened DD and may require different treatment approaches. It is worth mentioning that individuals with ASD may have contributed to the lack of significant differences we observed in DD with other transdiagnostic subtypes. Further work will be needed to understand the influence of ASD comorbidities in relation to their interaction with DD patterns across development.
Clinicians and researchers should consider our findings when addressing the severity and heterogeneity within ADHD populations. Tailoring interventions to the specific needs of individuals with different comorbidity profiles may help reduce DD and its negative consequences. For example, individuals with ADHD-I/LD/LowCog may benefit from training to improve their cognitive reasoning abilities. In contrast, individuals with ADHD-C/ANX/HighEBP may benefit from training on self-control techniques to help them achieve long-term goals. Different subgroups may have different underlying causes of heightened DD and may require different treatment approaches.
DD could also be used as an early marker of later issues, identifying 'at-risk' youth who may benefit from early intervention efforts. Specifically, clinicians could use DD measures in conjunction with other cognitive and behavioral measures to assess the severity of ADHD and identify those at increased risk. By using DD measures in a personalized manner, interventions could be tailored to the specific needs of each individual, such as reducing DD in money and food and adjusted based on their progress. This approach could help address the heterogeneity within ADHD populations, provide more targeted and effective treatments, and help prevent or mitigate the adverse outcomes associated with DD.
We also found that transdiagnostic subtypes were significantly associated with sociodemographic, cognitive/academic, and behavioral indices that are important to understand in the context of the above findings. First, all transdiagnostic groups had lower cognitive and academic test scores (WISC and WIAT) compared to the neurotypical group. This is consistent with increasing evidence of cognitive deficiencies in major psychiatric disorders (Goodkind et al., 2015). Intellectual reasoning ability (IQ) was also negatively associated with DD, consistent with prior research (Shamosh & Gray, 2008), providing further support for the role of intellectual reasoning abilities in reward-based decision making. In addition, we captured a subtype (i.e., ADHD-I/LD/LowCog) with higher cognitive/academic impairments with no remarkable emotional/behavioral problems compared to the rest of the sample. Their DD patterns were consistent with prior work linking cognitive abilities and self-regulation (Hofmann et al., 2012). Examining the impact of THI within this subtype notably revealed critical insights into how socio-economic hardship might have a more detrimental effect on individuals with lower cognitive abilities.
Second, we found transdiagnostic group differences in racial composition and THI. In particular, the ADHD-I/LD/LowCog and ADHD-C/HighExt subtypes contained higher proportions of individuals from minority backgrounds and were also characterized by higher proportions of individuals from the lower THI households. This finding aligns with similar developmental large-scale studies investigating these populations (Lichenstein et al., 2022) and prior evidence of a positive association between household income and mental disorders and an increased risk of maladaptive outcomes (Ballenger, 2012) that may be the result of health disparities rooted in mental healthcare (McGuire & Miranda, 2008).
The higher DD we observed in lower THI households was specific to the monetary factors and to the ADHD-I/LD/LowCog subtype, supporting previous work that implicated economic constraint with lower executive functioning and self-regulation (Oshri et al., 2019). This indicates that self-regulation toward monetary-based decision-making begins at an early age, and children living under financial constraints tend to place greater value on smaller, immediate rewards and devalue larger, delayed rewards. It may be that this is ultimately adaptive for their financial situation, considering that delayed rewards are less certain, so perhaps opting for the sooner, guaranteed reward is adaptive in a low THI context.
DD can be an adaptive strategy in certain circumstances, such as when an individual faces financial or resource scarcity, uncertainty about the future, and rewards in general (Kraft & Kraft, 2021). In such cases, it may be beneficial to prioritize immediate needs over long-term goals. For instance, lower THI children may have grown up in environments where resources were scarce. Therefore, it may have been adaptive for them to prioritize immediate rewards over future rewards to ensure their survival (Frankenhuis & Nettle, 2020). Additionally, low THI children may experience more unpredictability and instability in their environment, which can increase the perceived value of immediate rewards (Frankenhuis & Nettle, 2020). Therefore, it may be adaptive for low THIS children to have higher levels of DD. However, DD can also be maladaptive when it interferes with an individual's ability to achieve important long-term goals, such as education or health outcomes. Finally, some individuals may have greater difficulty regulating DD due to genetic factors, neurological problems, or other underlying conditions, such as those implied with ADHD (Jackson & MacKillop, 2016).
Given the associations between this type of behavior and multiple maladaptive outcomes (MacKillop et al., 2011), intervention-based strategies (Amagir et al., 2018) might help train these at risk-populations. Training children and adolescents may assist them in improving DD. One approach is cognitive-behavioral therapy (CBT), which can help individuals develop cognitive and behavioral skills to delay gratification. Another approach, "preference reversal," involves training individuals to change their preferences for immediate versus delayed rewards. This can be done by providing immediate rewards for choosing a delayed reward and vice versa. Additionally, providing education about the benefits of delaying gratification and the costs of impulsive behavior can help promote improved DD.
Third, there was evidence of differential representation of males and females among the DX transdiagnostic subtypes. Specifically, a higher proportion of females was present in the ANX/ADHD-I/HighInt subtype consistent with prior work linking sex differences in anxiety disorders and inattention to be more common in females than impulsivity (Quinn & Madhoo, 2014). In contrast, the ADHD-C/HighExt subtype was characterized by proportionally higher males, consistent with previous work that found boys with ADHD had higher levels of hyperactivity symptoms and externalizing behaviors (Hasson & Fine, 2012).
Finally, we found that age was negatively associated with DD, supporting established studies showing that DD is age-sensitive (Bixter & Rogers, 2019) and is sensitive to developmental changes in self-regulation and impulsivity (Orgeta, 2009). Importantly, we generally did not see evidence of differential changes in DD across development as a function of psychopathology, with the exception of depressive disorders as discussed above.
The complex nature of reward-based decision-making requires innovative approaches to improve outcomes for individuals with neurodevelopmental conditions. CBT is a practical application of personalized medicine that can help individuals identify and change cognitive distortions that may contribute to their tendency to discount the value of future rewards. Individuals may believe they will never have the money to achieve their goals or will never be able to save enough money to make a difference in the long term. CBT can help these individuals challenge these negative thoughts and develop more positive and realistic perspectives on their financial situation. A trained clinician could also help individuals develop self-regulation strategies to resist the temptation to spend money on immediate rewards. This can include techniques such as mindfulness, goal setting, and self-monitoring. For example, individuals can be taught to focus on the long-term benefits of saving money, such as being able to pay off student loans or buy a car, rather than the short-term pleasure of spending money on unnecessary items.
While the current work possesses many strengths, the ADT-5 was based on hypothetical rewards, which may rely more heavily on abstract reasoning than if concrete, tangible rewards were used, particularly for younger children, and may not reflect real-world DD. However, prior work in adults has shown that hypothetical rewards produce DD behavior akin to real currency (Koffarnus & Bickel, 2014). In addition, we could not assess validity in decision-making during this task. Finally, this study is limited by our sample size's low racial and ethnic diversity. Given the previous findings on DD differences across racial and ethnic groups (Hampton et al., 2018), future studies should examine developmental differences across these populations with similar approaches applied in the current work.
This study leveraged unique modeling approaches and a large, highly heterogeneous transdiagnostic sample to accurately characterize developmental differences in DD behavior. Our findings suggest that DD is a multi-faceted indicator of psychiatric and neurodevelopmental illness and severity across children and adolescents. This work has important implications for behavioral economics, education, psychology, and neuroscience. Previous findings (van den Bos et al., 2014) suggest that distinct striatal pathways are related to DD, and structural/functional connectivity between the striatum and prefrontal cortex and subcortical areas play a role in self-regulation and impulsivity. Further work should investigate the neural mechanisms, such as functional brain network organization and cortico-striatal system maturation, to determine their role in reward-based decision-making.
Future studies should also investigate the effectiveness of tiered intervention strategies (Rung & Madden, 2018) to reduce DD using methods such as mindfulness (Scholten et al., 2019), financial literacy programs (Amagir et al., 2018), and behavioral training (Ashe & Wilson, 2020) in school and clinical settings. These interventions should specifically target individuals with lower cognitive abilities and from lower and moderate household income brackets. Extensive work is needed using the approaches outlined here to evaluate its reproducibility and to gain a more cohesive understanding of these findings and their clinical relevance. In sum, our study highlights DD as a transdiagnostic marker of psychiatric and neurodevelopmental disorders in children and adolescents, revealing preliminary evidence of its complex nature, heterogeneity across disorders, and potential for identifying high-risk individuals, while emphasizing the importance of investigating developmental deviations and the impact of socio-economic factors on self-regulation.