Sex differences in Cingulo-Opercular activation during risky decision-making in youth with externalizing disorders

Abstract

Background:

Risky decision-making deficits predict unsafe behaviors, but sex differences in decision-making are underexplored in high-risk youth with externalizing disorders. While boys with externalizing pathology are more likely to make risky decisions, it remains unclear how these patterns manifest in girls, whose brains may process risks differently. Our study investigates sex differences in risky decision-making neurobiological activation among at-risk adolescents to identify sex-specific vulnerabilities for risky behaviors.

Method:

168 adolescents divided into four groups of 81 externalizing males, 39 externalizing females, 33 control males, and 15 control females completed a risky decision-making task, the Balloon Analog Risk Task, during functional magnetic resonance imaging.

Results:

Our primary finding was that externalizing males showed greater activation in the right dorsomedial prefrontal cortex/dorsal anterior cingulate cortex as the chance of a balloon explosion increased while making riskier choices over safer choices, compared to all other groups.

Conclusions:

These findings highlight key sex differences in the neurobiology of risky decision-making in youth with externalizing psychopathology within the cingulo-opercular network. With this network’s involvement in cognitive control and impulse inhibition—functions critical for managing risky behaviors—understanding its role in the interaction between sex and externalizing disorders is crucial for targeted, sex-specific interventions preventing risky behaviors.

1. Introduction

Deficits in adolescent risky decision-making have been linked to a variety of risky and dangerous behaviors later in life, such as substance abuse (Bechara, 2005, Bechara and Damasio, 2002, Bechara et al., 2001), motor vehicle accidents (Reyna and Farley, 2006), and unsafe sexual practices which can lead to sexually transmitted infections and unplanned pregnancies (Steinberg, 2004). Youth with externalizing disorders, such as ADHD, oppositional defiant disorder, or conduct disorder, are characterized by impulsivity, aggression, hyperactivity, and rule-breaking (Colder et al., 2018, Cox et al., 2021). These disorders represent an ideal population for examining adolescent risky decision-making and behaviors because deficits in risky decision-making are not only characteristic of these disorders, but are also often linked to their increased incidence of future risky behaviors compared to youth without externalizing pathology (Bechara and Martin, 2004, Bobova et al., 2009, Caspi et al., 1996, Finn, 2002, Iacono et al., 2008, Sartor et al., 2007, Siebenbruner et al., 2006, Sullivan et al., 2022, Tarter et al., 2003, Wetherill et al., 2013). However, research on sex differences in risky decision-making among youth with externalizing disorders remains underexplored, limiting our understanding of how these differences may contribute to risk profiles and the development of targeted interventions for males and females at high-risk for unsafe behaviors. A well-studied mechanism of altered decision-making among youth and adults with externalizing psychopathology is an increased likelihood to make disadvantageous decisions and favor low-probability, high-reward choices (d’Acremont and Van der Linden, 2006, Drechsler et al., 2008, Drechsler et al., 2010, Duarte et al., 2012, Fairchild et al., 2009, Fischer et al., 2005, Galván et al., 2013, Hobson et al., 2011, Luman et al., 2010, Matthies et al., 2012, Matthys et al., 1998, Matthys et al., 2004, Miranda et al., 2009). The propensity to make risky, disadvantageous decisions is key to the risk phenotype, yet several researchers have found sex differences in this aspect of the risk profile with males consistently exhibiting a greater tendency towards risky decision-making compared to females (Crowley et al., 2015, Korucuoglu et al., 2020, Sidlauskaite et al., 2018). While this suggests that sex differences in risk-taking behavior may play a critical role in shaping overall risk, with males showing a stronger propensity for impulsive, disadvantageous choices, sex differences in the risk profile have been underexplored in youth with externalizing disorders. This leaves a critical gap in understanding how sex-specific patterns of decision-making may shape risk-taking behaviors, particularly among high-risk, externalizing youth, necessitating further investigation into the unique, sex-specific vulnerabilities of this population. Specifically, it remains unclear whether females with externalizing pathology exhibit the same deficits in risky decision-making as males with externalizing pathology, which have been associated with an increased risk of unsafe and dangerous behaviors.

Beyond behavioral measures, neuroimaging studies provide valuable insights into the cognitive and emotional mechanisms underlying risk. Insights from brain imaging studies are particularly important when studying risky decision-making behaviors as these are difficult to assess through laboratory and self-report measures. The neuroimaging studies which do evaluate sex in the context of risky decision-making have indeed found sex-specific mechanisms. These risky decision-making tasks typically involve choices between options with varying levels of risk, followed by feedback on the outcomes of those decisions. Regarding the decision phase, sex differences seem to arise in the frontostriatal and salience networks, with males showing greater activation in the frontostriatal network during both risky and safe choices, while females show greater activation in this network during safe choices only, and males activating left-sided salience network and females activating right-sided salience network before risky decisions (Crowley et al., 2015, Korucuoglu et al., 2020, Dir et al., 2019). Regarding the outcome phase, females activate frontoparietal and salience networks more when processing negative outcomes, where males activate frontoparietal cognitive control networks when processing positive outcomes (Cazzell et al., 2012). Therefore, we hypothesize reduced activation in fronto-striatal-thalamic regions during risky choices compared to safe choices for females. Males, however, may show stronger activation in frontoparietal and reward-related areas during risky decisions. For the outcome phase, we hypothesize females will show increased activation in frontoparietal and salience regions during negative outcomes, while males may engage frontoparietal networks for positive outcomes, leading to less activation in these regions for males in negative-minus-positive outcome contrasts. Externalizing psychopathology may further influence these sex differences, with female externalizing showing different patterns of activation compared to controls, closer to male activation patterns (Crowley et al., 2015).

In sum, the current study, incorporating neuroimaging and behavioral assessments, aims to explore sex-specific mechanisms that underlie risky decision-making in male and female adolescents in a sample of 11–12-year-olds with and without externalizing disorders (e.g. ADHD). We hypothesize that nuances of the interaction between sex and externalizing pathology will result in externalizing females having activation closer to externalizing males, which will be greater frontoparietal/frontotemporal activation before risky choices and less activation before negative outcomes than control females. These sex-specific findings could guide new interventions and prevention strategies specifically aimed at defaults in risky decision-making processes among the highest-risk youth.

2. Methods

2.1. Sample

223 English-speaking, right-handed, 11–12-year-old participants and their caregivers completed psychiatric and behavioral assessments and an fMRI scan as part of an ongoing longitudinal study (Dir et al., 2019, Aloi et al., 2023, Dir et al., 2020, Hulvershorn et al., 2015, Kwon et al., 2021). For the externalizing (EXT) group, individuals met criteria for DSM-5-TR diagnoses of ADHD and a disruptive behavior disorder: oppositional defiant disorder, conduct disorder, or unspecified disruptive behavior disorder. The percentages of EXT pathology diagnoses and other comorbid DSM-5 diagnoses given for each group are presented in Table 1. Exclusion criteria at baseline were: (Bechara, 2005) lifetime history of bipolar disorder, psychotic symptoms, autism spectrum disorders or SUDs, (Bechara and Damasio, 2002) DSM 5-defined current major depressive disorder, (Bechara et al., 2001) history of neurological problems (e.g., epilepsy, traumatic brain injury, brain tumors), (Reyna and Farley, 2006) estimated Full Scale IQ <80 (to ensure participants understood the self-report forms and tasks) (Steinberg, 2004) active or debilitating medical conditions, (Colder et al., 2018) maternal substance use disorder during pregnancy, (Cox et al., 2021) MRI-related contradictions, (Bechara and Martin, 2004) reporting recreational drug use (other than caffeine), (Bobova et al., 2009) left-handedness, (Caspi et al., 1996) individuals with siblings already enrolled in the study. Patients taking psychostimulant medications were instructed to hold medication on study visit days and a urine drug screen was administered on study days to help confirm substance-naive status. All procedures were conducted according to the Indiana University Institutional Review Board.

Table 1.

Demographics.

	FC (n=15)	MC (n=33)	FEXT (n=39)	MEXT (n=81)	p-values
Age / M (SD)	11.93 (0.62)	12.01 (0.56)	12.02 (0.54)	11.83 (0.54)	0.24
Race / N (%)					0.90
White	9 (60.00)	22 (66.67)	22 (56.41)	52 (64.20)
Black	5 (33.33)	8 (24.24)	13 (33.33)	19 (23.46)
Multiracial	1 (6.67)	3 (9.09)	4 (10.26)	10 (12.35)
Hispanic Ethnicity / N (%)	4 (26.67)	3 (9.09)	3 (7.69)	7 (8.64)	0.17
Parental Education Max / N (%)					0.36
High School	1 (6.67)	0 (0.00)	5 (12.50)	9 (10.71)
Some college or college	7 (46.67)	18 (54.55)	23 (60.00)	42 (50.00)
Some graduate or graduate school	7 (46.67)	15 (45.45)	11 (27.50)	27 (33.33)
Unknown	0 (0.00)	0 (0.00)	0 (0.00)	3 (5.95)
Tanner Stage / N (%)					<0.01
1	2 (13.33)	14 (42.42)	8 (20.51)	33 (40.74)
2	4 (26.67)	10 (30.30)	5 (12.82)	34 (41.98)
3	5 (33.33)	4 (12.12)	6 (15.38)	10 (12.35)
4	4 (26.67)	3 (9.09)	12 (30.77)	3 (3.70)
5	0 (0.00)	2 (6.06)	8 (20.51)	1 (1.23)
Verbal IQ / M (SD)	114.33 (16.07)	114.79 (11.57)	106.82 (13.52)	107.94 (14.35)	0.03
ADHD Subtype / N (%)					<0.01 *
Hyperactive/Impulsive			4 (10.26)	8 (9.88)
Inattentive			28 (71.79)	27 (33.33)
Combined			7 (17.95)	46 (56.79)
Oppositional Defiant Disorder / N (%)			29 (74.36)	63 (77.78)	0.85*
Conduct Disorder / N (%)			3 (7.69)	6 (7.41)	1.00*
Unspecified DBD / N (%)			9 (23.08)	14 (17.28)	0.61*
Any Mood Disorder / N (%)	0.00 (0.00)	0.00 (0.00)	5 (12.82)	5 (6.17)	0.09
Any Anxiety Disorder / N (%)	0.00 (0.00)	0.00 (0.00)	11 (28.21)	12 (14.81)	<0.01

FC = female controls, MC = male controls, FEXT= externalizing females, and MEXT = externalizing males; DBD = disruptive behavior disorder; Any Mood Disorders included dysthymia and disruptive mood dysregulation disorder; Any Anxiety Disorder included separation anxiety, social anxiety, selective mutism, and generalized anxiety disorder

^*p-values for ADHD subtypes and disruptive behavior disorders are solely comparing externalizing males and females

2.1.1. Psychiatric disorders

Externalizing diagnoses were determined by a semi-structured clinical interview, the Kiddie Schedule for Affective Disorders and Schizophrenia-Present and Lifetime Version (KSADS-PL). The KSADS-PL, modified from DSM-5, determines current and lifetime psychiatric diagnoses (Kaufman et al., 1997) and was conducted with each participant and guardian(s) to assess psychiatric and SUD status at baseline.

2.1.2. Sex

Each child participant’s sex was force choice reported as male or female by the adult guardian at baseline. Youth were asked at baseline “Which of the following best describes you?” with response options of “boy” or “girl.” Parent-reported sex was used to determine group membership. To account for potential gender diversity not fully captured by these questions, we created a dichotomous variable reflecting discordant gender identity. This variable was defined by participants whose self-reported gender identity (boy or girl) did not align with their guardian-reported sex (male or female).

The four groups were defined by presence or absence of externalizing disorders and sex (male or female) as reported at baseline. Reasons for subject exclusions are identified in Fig. 1. The final sample consisted of 168 subjects divided into four groups of externalizing males (n=81), externalizing females (n=39), control males (n=33), and control females (n=15).

Fig. 1.

Flowchart of Participant Exclusion Criteria.

2.2. Measures

2.2.1. Balloon analog risk task (BART)

In the fMRI scanner, participants completed an fMRI-compatible BART (Fukunaga et al., 2012, Rao et al., 2008) to explore brain activation during risky decision-making. The BART models real-world risky choices by balancing reward and negative outcome probabilities. BART performance correlates with impulsivity, adolescent risk-taking, and SUDs (Hopko et al., 2006, Hunt et al., 2005, Lejuez et al., 2007, Lejuez et al., 2003, Lejuez et al., 2003), making it a valuable tool for examining sex influences on risky decision-making in externalizing youth. During the BART, participants decide whether to risk cash rewards that increase with each balloon inflation or bank the amount and start inflating a new balloon (Fig. 2). Participants are instructed to “inflate the balloon as much as you can without popping it” to earn money for each unexploded balloon. Participants are told that they will win more money (paid in real cash immediately after scanning) for larger balloons and they will attempt this task three times in the scanner. The BART incorporates parametric increases in explosion probability over successive responses (Fukunaga et al., 2012). The parametric modulation employed the following probabilities of balloon explosions at each pump: 0% for $0.0; 2.1% for $0.05; 4.2% for $0.15; 6.3% for $0.25; 14.6% for $0.55; 23.9% for $0.95; 31.3% for $1.45; 43.8% for $2.05; 56.3% for $2.75; 68.8% for $3.45; 79.2% for $4.25; 89.6% for $5.15. A jitter function is applied to the timing of the stimulus presentations between decision and outcome phases of each trial to differentiate decision-making and feedback-related processes (Fukunaga et al., 2012). Participants were repeatedly presented a virtual balloon and had to press one of two buttons to either inflate the balloon (Choose Inflate), risking cash rewards that increased with the balloon’s size, or to stop inflating and bank the accumulated money for that balloon (Choose Win). If they choose to inflate, the balloon either expands, increasing the reward amount for that balloon (Outcome Inflate), or it bursts, resulting in the loss of the accumulated money for that balloon (Outcome Explode). Except for the initial inflation, explosions could occur at any size, with the risk increasing as the balloon grows larger. When participants opt to win (Choose Win), the accumulated rewards are banked, and a new balloon appears. During the three 8-minute sessions, participants could complete as many balloons as possible, with a maximum of 12 inflations per balloon. Before performing the BART in the fMRI scanner, participants practiced the task on a desktop computer. For measures of BART performance, we included variables indicative of risk-taking or tempo within the task. Risk-taking variables included average adjusted pumps (the average number of inflations on unexploded balloons), average money bet (the mean amount of money at stake when participants risked an inflation), number of choose inflate choices, choose win choices, outcome explosions, and total balloons completed. The tempo-related variable was reaction time (average time to press the response button for all choices, inflate or win, in milliseconds). Performance behavioral variables were averaged over a participant’s three runs.

Fig. 2.

BART Schematic.

2.2.2. MRI data acquisition

Participants completed brain scanning in <90-minute sessions on a research-dedicated 3.0-Tesla Siemens Prisma MRI scanner with a 32-channel head coil. A high-resolution 3D magnetization prepared rapid gradient echo (MPRAGE) scan consisting of 160 sagittal slices and 1.05 × 1.05 × 1.2 mm³ voxels was completed. For BART runs, a T2*-weighted gradient echo-planar imaging (EPI) sequence was used (54 axial slices; voxel size 2.5 × 2.5 × 2.5 mm³; TR/TE 1200/29 ms; flip angle 65°; field-of-view 220 × 220 mm²; matrix 88 × 88), using a multiband sequence with a multiband factor of 3. In the same order for each participant, and as the BART was the overall study priority, three 8-minute BART sessions were administered consecutively as the first task following an anatomical and resting-state scan and proceeding a delay-discounting task and white matter-focused diffusion scan.

2.3. Covariates

2.3.1. IQ

IQ screening using the Wechsler Abbreviated Scale of Intelligence (WASI) (Axelrod, 2002) was used to ensure that all participants met a minimum threshold of general intelligence for understanding task instructions and as a possible covariate because prior work has found IQ to correlate with several aspects of risk calculation and decision-making (Brand and Schiebener, 2013, Deakin et al., 2004, Defoe et al., 2015, Flouri et al., 2019, Romer, 2010, Schutter et al., 2011, Toplak et al., 2010). Due to COVID-19 restrictions and the necessity of conducting study visits online, only Verbal IQ scores from the WASI are reported, as virtual visits did not permit completion of Performance IQ tasks. Two (n=2) individuals were unable to complete the WASI and two study clinicians estimated their IQ was highly likely to be >80 based on the overall clinical interview.

2.3.2. Pubertal development

Pubertal development was reported via parental identification of Tanner scale pictures, which range from 1 (prepubertal) to 5 (complete pubertal development) (Marshall and Tanner, 1968). Three (n=3) participants had unknown pubertal stages. Tanner staging was used as a covariate to account for distinct levels of physical development between our groups, knowing our group has shown effects of pubertal staging on BART before (Dir et al., 2019).

Data missingness for IQ and pubertal development was handled with multiple imputations utilizing the MICE package in R (Sv and Groothuis-Oudshoorn, 2023). Mean imputation per group (i.e., control females, control males, externalizing females, externalizing males) was used to account for IQ missing values. Proportional odds logistic regression (POLR) model (accounting for age and sex) was used to impute pubertal development.

2.4. BART behavior analysis

We first assessed whether the behavioral variables met the assumptions required for an ANCOVA, specifically testing for the normality of residuals and homogeneity of variance. Using diagnostic plots and statistical tests (e.g., Shapiro-Wilk test for normality and Levene’s test for homogeneity of variance), several variables were identified that deviated from these assumptions, which could compromise the validity of parametric analysis. To address these violations, the Box-Cox transformation method was implemented in R (Venables and Ripley, 2002) to identify an optimal transformation for each variable (e.g. square, square root, inverse square root, and logarithmic transformations). This approach systematically evaluates a range of power transformations (λ) to achieve a more normal distribution of residuals (Venables and Ripley, 2002, Box and Cox, 1964). Behavioral variables that showed significant deviations from normality or homoscedasticity were transformed accordingly, improving the fit to the assumptions of ANCOVA. Those variables which were not normally distributed or homoscedastic were Choose Inflate Count, which exhibited a negatively skewed distribution, and Choose Win Count, Balloons Completed, and Reaction Time, which all were positively skewed, before transformation. The applied Box-Cox transformations to better normalize those four variables were: squared, square root, inverse square root, and logarithmic, respectively. After applying the appropriate transformation, we analyzed the differences in behavioral performance across groups with ANCOVAs. Group membership was the primary independent variable of interest with two covariates: IQ and Tanner staging. Post-hoc analyses used estimated marginal means pairwise comparisons to explore specific group differences. These were conducted using the emmeans() package in R (Lenth, 2024) to compute adjusted means for each group, accounting for the effects of the covariates. To control the inflated risk of type I error associated with multiple comparisons, we applied Bonferroni correction during the pairwise tests.

2.5. MRI data preprocessing

Initially, echo planar image (EPI) scans underwent spin-echo unwarping to correct distortions using fMRIB Software Library (FSL)’s “topup” tool. Anatomical scans were registered to Montreal Neurological Institute-152 (MNI152) standard MRI template brain volume space, and EPI data were aligned with each participant’s anatomical scan using afni_proc.py (Cox, 1996, Cox and Hyde, 1997), a comprehensive processing pipeline in AFNI (Analysis of Functional NeuroImages). Also, using afni_proc.py, functional images were motion-corrected and spatially smoothed with a 6mm Gaussian kernel, and time-series normalization to a T1-weighted image was conducted, followed by an adjustment to a 100-scale per voxel. Lastly, unsupervised independent components analysis (ICA)-based denoising with ICA-AROMA (Pruim et al., 2015, Pruim et al., 2015) from FSL’s MELODIC tool was used for robust data cleaning by identifying and removing noise-related components.

In the first step of imaging quality assessment, subjects with motion artifacts during the anatomical scan, which led to failure in the volume registration step, were excluded from the analysis (n = 11 subjects). Then, further visual inspection of fMRI brain activity was conducted to ensure the presence of expected activation patterns corresponding to the task. This step involved examining the spatial distribution of activation maps to confirm typical task-related regions were engaged (e.g. visual cortex). Where expected activation in areas like the visual cortex were missing, those subjects were excluded (n = 5), and where there was widespread negative activation, which also could indicate global signal regression artifacts, those subjects were excluded (n = 7). Finally, if activation distribution within these maps were excessively dispersed, non-localized, or lacked a discernible pattern, this was attributed to motion artifacts and subjects exhibiting confirmed high motion were excluded (n = 13). Therefore, a total of 36 participants were excluded based on unsatisfactory functional imaging. Of note, these excluded participants did not differ significantly from the final sample (n = 168) in terms of age, race, ethnicity, parental education, Tanner stage, verbal IQ, or in proportions of ADHD subtypes, disruptive behavior disorders, mood, or anxiety disorders.

2.6. Neuroimaging data analysis: subject level analysis

After preprocessing and noise reduction, runs were concatenated, and a general linear regression model (GLM) with random effects was used to estimate event-related responses in AFNI. Choice events, aligned to the repetition time (TR) that included the button press response, were modeled as Choose Inflate (choosing to continue inflating the balloon) or Choose Win (choosing to discontinue inflating and bank money) regressors. Outcome events were modeled as the TR that included balloon explosion (Outcome Explode), successful balloon inflation (Outcome Inflate), or the outcome of discontinuing inflations (Outcome Win). Balloon explosion probabilities were included as parametric modulators for each event-type regressor (e.g., Choose Inflate * P(explode), Outcome Inflate * P(explode)), except for Outcome Win, which has no uncertainty). Parametric modulators were incorporated to examine neural activation patterns that specifically track the escalating risk of both reward and loss as balloon size increases, providing insight into how brain regions dynamically respond to varying levels of decision-making uncertainty. To compare activation differences between conditions, individual subject activation maps were subjected to a voxel-wise subtraction using the 3dcalc tool in AFNI to compute the difference in activation between conditions, adjusting for individual differences in baseline activation. There were four contrast subtraction maps: Choose Inflate – Choose Win (modulated and unmodulated) and Outcome Explode – Outcome Inflate (modulated and unmodulated).

2.7. Neuroimaging data analysis: group level analyses

After unpacking the contrast subtraction maps for subject-level average activation to check for normality of imaging data to meet the requirements of ANCOVA analyses, we found that our imaging data activation violated the assumptions of normally distributed residuals across each subpopulation and homogeneity of variance. Therefore, we turned to a nonparametric analysis for our group analyses. For analyzing BART functional imaging, we conducted a whole-brain analysis using AFNI’s 3dKruskalWallis to identify brain regions that differed across groups. The statistical models tested whether there were significant differences between the rank of each subjects’ activation by sex/externalizing group across the brain and did not allow for inclusion of covariates. Multiple comparisons of this whole-brain voxel-wise analysis were addressed using cluster-wise thresholds (family-wise error rate of p < 0.05). Individual voxels were significant at p < 0.001, and a Monte Carlo simulation (AFNI’s 3dclustSim) determined the cluster size needed to correct for group-level significance, k=31 (p < 0.05). Significant clusters’ coordinate locations were mapped onto region names using standard Montreal Neurological Institute (MNI) brain atlases. Post-hoc analyses on extracted parameter estimates were conducted in R to further investigate the differences in brain activation identified by the Kruskal-Wallis test across the four sex/externalizing groups. The Kruskal-Wallis test was utilized over ANOVA with our imaging data due to its ability to account for unequal variances and sample sizes across groups (Hecke, 2012), providing a more robust analysis given the unbalanced nature of our data. As we could not include covariates in the Kruskal-Wallis whole brain analysis in AFNI, we controlled for covariates in the post-hoc analyses to determine whether the observed group differences in brain activation remained significant when adjusting for these factors. We performed ranked-based estimation regression with the rfit.default() function in R (Kloke and McKean, 2012). The reference group variable was releveled (first as female controls, second as male controls, and third as externalizing females), to 1) perform pairwise comparisons across all six group combinations; and 2) control for the pertained covariates (verbal IQ and Tanner stage due to reasoning described in Section 2.3 above).

2.8. Gender identity sensitivity analysis

To investigate whether gender identity might interact with the observed sex differences in brain activation patterns, we conducted a sensitivity analysis. In this analysis, we excluded imaging data from participants whose self-reported gender identity (boy or girl) did not match the guardian-reported sex (male or female; defined in our dichotomous variable as having a discordant gender identity). The whole-brain group level analyses described above were run without these participants to determine whether gender identity alignment with assigned sex influences the observed differences in brain activation patterns. The same parameters specified in the primary analysis were applied to this sensitivity analysis.

3. Results

3.1. Demographics and clinical characteristics

This cross-sectional analysis included 168 substance-naïve 11–12-year-olds who met criteria for usable imaging data, grouped as externalizing males (MEXT=81) and females (FEXT=39) and healthy control males (MC=33) and females (FC=15). Groups were similar on demographic, clinical and socioeconomic measures (Table 2) other than verbal IQ [F(3, 164) = 3.05, p = 0.03] and Tanner pubertal stage (χ² = 50.34, p < 0.01. Controls had higher IQ than externalizers, though all groups had above average IQ, and, as anticipated, males had significantly lower Tanner stages than the females. Due to significant differences in verbal IQ and Tanner pubertal stage, and reasoning described above, these variables were included as covariates in the behavioral analyses and imaging post-hoc analyses.

Table 2.

BART Behavior Performance Variables Mean (SD) by Group.

Behavior Outcome	FC	MC	FEXT	MEXT	P Value
Average Adjusted Pumps	4.46 (0.65)	4.83 (0.90)	4.73 (0.88)	4.87 (0.81)	0.32
Average Money Bet	0.23 (0.07)	0.26 (0.10)	0.27 (0.10)	0.27 (0.09)	0.41
Choose Inflate Count1	6772.30 (2017.71)	8587.70 (1601.27)	7835.67 (1496.84)	8377.30 (1558.40)	<0.01
Choose Win Count2	3.53 (0.45)	3.53 (0.69)	3.54 (0.55)	3.51 (0.52)	0.99
Outcome Explode Count	5.42 (2.35)	6.58 (2.37)	6.09 (2.05)	6.35 (2.08)	0.34
Balloons Completed3	0.24 (0.01)	0.23 (0.02)	0.23 (0.02)	0.23 (0.01)	0.41
Reaction Time (ms)4	7.17 (0.47)	6.77 (0.40)	6.91 (0.35)	6.79 (0.40)	<0.01

¹Squared (e.g. Choose Inflate Count)^2

²Square Root (e.g. sqrt(Choose Win Count))

³1/Square Root (e.g. 1/sqrt(Balloons Completed))

⁴Log (e.g. log(Reaction Time)); FC = female controls, MC = male controls, FEXT= externalizing females, and MEXT = externalizing males

3.2. BART performance

In our ANCOVA analyses, there were sex/externalizing group differences on squared Choose Inflate Count and log transformed Reaction Time (Table 2). Male controls (β = 1737.70, p < 0.01) and externalizing males (β = 1649.50, p < 0.01) had greater average squared Choose Inflate Counts than control females. Male controls (β = −0.41, p < 0.01) and externalizing males (β = −0.43, p < 0.01) had significantly faster reaction times (log transformed) than control females. Full results are presented in Table 3.

Table 3.

BART Behavioral Variables from all Group Pairwise Comparisons.

Behavioral Outcome	FC vs MC		FC vs FEXT		FC vs MEXT		MC vs FEXT		MC vs MEXT		FEXT vs MEXT
	β	p	β	p	β	p	β	p	β	p	β	p
Average Adjusted Pumps	−0.33	1.00	−0.33	1.00	−0.39	0.70	0.00	1.00	−0.06	1.00	−0.06	1.00
Average Money Bet	−0.3	1.00	−0.05	0.64	−0.05	0.49	−0.01	1.00	−0.01	1.00	0.00	1.00
Choose Inflate Count1	−1737.7	<0.01	−967.6	0.35	−1649.5	<0.01	770.1	0.36	88.2	1.00	−681.9	0.35
Choose Win Count2	−0.01	1.00	0.06	1.00	−0.01	1.00	0.08	1.00	0.00	1.00	−0.07	1.00
Outcome Explode Count	−1.11	0.68	−0.82	1.00	−0.90	0.97	0.28	1.00	0.20	1.00	−0.08	1.00
Balloons Completed3	0.01	0.67	0.00	1.00	0.01	1.00	−0.01	0.92	0.00	1.00	0.00	1.00
Reaction Time4	0.41	<0.01	0.22	0.46	0.43	<0.01	−0.19	0.36	0.02	1.00	0.21	0.09

Results are the pairwise comparisons from the ANCOVA of behavior variables between the sex and externalizing groups, adjusted for verbal IQ and Tanner stage. The beta estimates (β) represent the magnitude and direction of the differences in the behavioral outcome between groups and “Group X vs Group Y” should be read as the first group, Group X, minus the second group, Group Y, so the first cell is read FC was associated with 0.33 less average adjusted pumps than MC. FC = female controls, MC = male controls, FEXT= externalizing females, and MEXT = externalizing males.

¹Squared (Choose Inflate Count)^2

²Square Root (e.g., sqrt(Choose Win Count))

³1/Square Root (e.g., 1/sqrt(Balloons Completed))

⁴Log (e.g., log(Reaction Time))

3.3. BART imaging results

Two clusters were identified with significantly different activity between groups using AFNI’s 3dKruskalWallis analyses. The two clusters demonstrated significant differences between the four sex/externalizing groups for Choose Inflate – Choose Win modulated activity and were located in the left inferior temporal gyrus (Fig. 3) and the right dorsal medial prefrontal cortex (dmPFC)/dorsal anterior cingulate cortex (dACC) (Fig. 4). Table 5 contains information on cluster size and coordinates.

Fig. 3.

Left inferior temporal gyrus cluster. Group differences on the parametrically modulated choice contrast in the left inferior temporal gyrus (Table 5). Group differences were driven by increased activation intensities as explosion probability increased in the male controls, male externalizing, and female externalizing groups compared to female controls. The violin plot displays the density of activation intensities during the modulated choose inflate—modulated choose win contrast (y-axis) for each group. *Indicates significant at p < 0.05.

Fig. 4.

Right dorsomedial prefrontal/dorsal anterior cingulate cortex cluster. Group differences on the parametrically modulated choice contrast in the right dorsomedial prefrontal/dorsal anterior cingulate cortex (dmPFC/dACC; Table 5). Group differences were driven by increased activation intensities as explosion probability increased in male externalizers. The violin plot displays the density of activation during the modulated choose inflate—modulated choose win contrast (y-axis) for each group. *Indicates significant at p < 0.05.

Table 5.

Regions with significant group differences during the BART in the modulated Choose Inflate vs Choose Win contrast. No clusters were significant in the outcome contrasts, nor unmodulated contrasts. Cluster size is defined by the number of voxels. Coordinates and Kruskal-Wallis statistics (K*) are provided for the peak voxel in each cluster.

	BA	Peak K* Value	Cluster Size	Talairach Coordinates
				X	Y	Z
Left Inferior Temporal Gyrus	37	21.25	41	−59	−56	−10
Right dmPFC/dACC	9/32	26.77	34	2	42	14

Abbreviations: dmPFC = dorsomedial prefrontal cortex, dACC= dorsal anteriorcingulate cortex, BA = Brodman Area

Regarding the unadjusted post-hoc analyses, in the left inferior temporal gyrus cluster, all groups had significantly greater modulated activation compared to control females; however, due to the small sample size of the reference group, this result should be interpreted with caution. In the right dmPFC/dACC, externalizing males had greater activation than the other groups. See Table 4 for reporting of all unadjusted and adjusted post-hoc results. Adjusting for covariates (“Adjusted” in Table 4) did not alter the direction nor significance.

Table 4.

Unadjusted and Adjusted BART Activation for Choose Inflate-Choose Win modulated clusters.

Cluster Location	FC vs MC		FC vs FEXT		FC vs MEXT		MC vs FEXT		MC vs MEXT		FEXT vs MEXT
	β	p	β	p	β	p	β	p	β	p	β	p
Unadjusted
Left Inferior Temporal Gyrus	0.19	<0.01	0.16	<0.01	0.16	<0.01	−0.02	0.31	−0.03	0.20	0.00	0.90
dmPFC/dACC	−0.02	0.50	−0.02	0.40	0.04	0.04	0.00	0.85	0.06	<0.01	0.06	<0.01
Adjusted
Left Inferior Temporal Gyrus	0.19	<0.01	0.15	<0.01	0.16	<0.01	−0.04	0.10	−0.03	0.20	0.01	0.52
dmPFC/dACC	0.00	0.90	−0.02	0.52	0.06	<0.01	−0.01	0.52	0.06	<0.01	0.08	<0.01

Results are the post-hoc pairwise comparisons of cluster activation between the groups. The table includes the beta estimates and p-values for each post-hoc comparison of robust regression with the rfit.default() function in R. The beta estimates (β) represent the magnitude and direction of the differences in activation between groups and “Group X vs Group Y” should be read as the first group as Group X is the comparison group, so the first cell is read MC was associated with 0.19 greater activation units than FC in the left inferior temporal gyrus. Adjusted results includes covariates verbal IQ and Tanner stage in the robust regression. FC = female controls, MC = male controls, FEXT= externalizing females, and MEXT = externalizing males.

Abbreviations: dmPFC = dorsomedial prefrontal cortex, dACC= dorsal anterior cingulate cortex

3.4. BART imaging Results-Gender Sensitivity analysis

Eleven subjects were excluded for the gender sensitivity analysis due to the youth’s self-reported gender (boy or girl) not matching the guardian-reported sex (male or female). A whole brain analysis without the eleven subjects found that both clusters in the left inferior temporal gyrus and right dmPFC/dACC were again significant at p < 0.001 and exceeded the voxel size threshold of 31 voxels, as was observed when those participants were included.

4. Discussion

The goal of the current study was to examine sex differences in behavior and neural activation during risky decision-making among adolescents with and without externalizing disorders. In terms of behavioral performance on the BART, significant sex and externalizing group differences were found in both Choose Inflate Count and log-Reaction Time (both transformed). Specifically, male controls and externalizing males exhibited greater squared Choose Inflate Counts compared to control females, indicating more frequent risky choices for males. Additionally, these same groups had faster reaction times (log-transformed) than control females. Interestingly, although these behavioral findings align with established sex differences in risky decision-making—where males typically engage in riskier and more impulsive behavior (d’Acremont and Van der Linden, 2006, Drechsler et al., 2008, Drechsler et al., 2010, Duarte et al., 2012, Fairchild et al., 2009, Fischer et al., 2005, Galván et al., 2013, Hobson et al., 2011, Luman et al., 2010, Matthies et al., 2012, Matthys et al., 1998, Matthys et al., 2004, Miranda et al., 2009, Crowley et al., 2015, Korucuoglu et al., 2020, Sidlauskaite et al., 2018)—the results suggest that externalizing females did not display the same risky/impulsive patterns as their male counterparts. Instead, male controls exhibited more risky and impulsive behavior, suggesting that these differences may reflect broader sex-based tendencies in risk-taking behavior rather than being indicative of externalizing pathology or specific risk for unsafe behaviors, at least in the preadolescent period. This implies that sex differences in decision-making behavioral metrics on this task may not directly correlate with risky behaviors, particularly for externalizing females, but further study is needed to confirm.

Regarding brain imaging findings, the left inferior temporal gyrus and the dmPFC/dACC, nodes within ventral visual processing stream and cingulo-opercular circuits, respectively, were identified as differing between groups in our whole-brain analysis. The group differences were observed when contrasting a risky choice (Choose Inflate) with a safe choice (Choose Win), but only as those decisions became riskier (modulated) and not during any of the outcome phase contrasts. Findings in the left inferior temporal gyrus cluster revealed that all groups had significantly greater modulated activation compared to control females when making a risky choice (Choose Inflate) over a safe choice (Choose Win), as those choices became riskier (modulated). The left inferior temporal gyrus is frequently identified in risky decision-making studies likely because of its role in the ventral stream of visual processing (Sakagami and Pan, 2006)—playing a critical role in higher-order visual processing (Gross, 2008, Lafer-Sousa and Conway, 2013), visual working memory (Lafer-Sousa and Conway, 2013, Ranganath et al., 2004, Ranganath, 2006), and shows sustained activation during maintained visual representations of preferred objects (a particularly salient function in the current task where risk is represented visually as increasing balloon size) (Fuster and Jervey, 1981, Fuster and Jervey, 1982, Miller et al., 1993, Miyashita and Chang, 1988, Nakamura and Kubota, 1995). Therefore, our findings may be a preliminary indication that deficits in visual memory during risky decision-making could impair an individual’s ability to accurately evaluate risk by limiting their capacity to recall and integrate visual information about previous outcomes, leading to misjudgments in assessing potential rewards or consequences as the situation becomes riskier. However, because the observed group difference involves all groups compared to the smallest group, female controls, which has only fifteen participants and may be heavily influenced by outliers in this group, we view this finding as preliminary, warranting further investigation.

The second cluster that differed between groups was located in the dmPFC and dACC. In the dmPFC/dACC, male externalizers displayed increased activation in the choose inflate – choose win contrast as the probability of explosion increased. These regions are part of the cingulo-opercular (or salience) network, which supports cognitive control, task-set maintenance, sustained attention, and salience detection—functions often implicated in decision-making, externalizing behaviors, and addiction (Lerman-Sinkoff et al., 2017, Pierce et al., 2023). The dmPFC and dACC are essential for cognitive control and behavioral flexibility, particularly in maintaining internal goals, conflict monitoring, and adjusting cognitive processes across sensory, memory, and motor systems (Botvinick et al., 1999, Carter et al., 1998, Dosenbach et al., 2007, van Veen et al., 2001). The dmPFC also plays a significant role in set-shifting and response inhibition, aiding in modifying learned motor behaviors and inhibiting impulses (Bissonette and Roesch, 2015, Floresco et al., 2008, Hamel et al., 2022, Narayanan and Laubach, 2006, Modirrousta and Fellows, 2008). Regarding sex differences in normal function of the dmPFC, multiple studies have found that the PFC develops around two years earlier and matures more extensively in females than males, while males have larger PFC volume throughout development (Barkley-Levenson et al., 2013, Dennison et al., 2013, Lenroot et al., 2007, Li et al., 2014, Raznahan et al., 2010, Raznahan et al., 2014, Urošević et al., 2014, Zuo et al., 2010).

Given this cingulo-opercular network’s function in cognitive control, the dmPFC/dACC arises as a crucial area for risky decision-making. Studies consistently demonstrate that dmPFC activity is associated with more conservative, risk-averse behavior (Lv et al., 2021, Rodrigo et al., 2014, van Duijvenvoorde et al., 2015, van Leijenhorst et al., 2010, Xue et al., 2009, Xue et al., 2010), and similarly, the dACC has been implicated in monitoring ongoing decisions, with activation increasing as individuals engage in risky or ambiguous choices (Telzer et al., 2013, Krain et al., 2006). With these prior findings, if choosing to win is the risk-averse choice, and risk-aversion increases activation in this region normally, then risky choice activation minus risk-averse choice activation would be smaller, which it was in every group besides our highest risk male externalizers. Heightened dmPFC/dACC activation during risky choices (inflating), whereas typical patterns show increased activation during safer choices (winning) may reflect dysregulation in cognitive control, conflict monitoring, and impulse inhibition during decision-making among the high-risk males. Prior findings of sex differences in dmPFC/dACC activation during risky decision-making, align with prior work and our findings: Crowley et al. found that males exhibited heightened activity in the left medial and dorsolateral PFC regions prior to making cautious choices while females demonstrated greater engagement in the corresponding right-sided areas (Crowley et al., 2015). Therefore, aberrant activation in the cingulo-opercular network, in cognitive control, when evaluating a risky vs safe decision is likely a large part of the risk profile for unsafe behaviors carried by our male externalizers.

Our study has several limitations. First, the small female control group limits the generalizability of the findings, particularly the finding in the inferior temporal gyrus, and the unequal group sizes may affect the statistical power and overall interpretation of the BART behavioral results as they were analyzed with an ANCOVA. Next, while we aimed for our gender sensitivity analysis to account for possible gender nonconformity in our sample, we were ultimately underpowered to be able to detect differences between gender conforming and gender non-conforming youth. However, the results of our gender sensitivity analysis suggest that using sex assigned at birth was a reasonable approach given the small sample size of gender non-conforming youth. In future studies, we would aim to collect more comprehensive data on both gender and sex to better capture these variables’ effects. Second, we were unable to include covariates in our whole-brain analysis, limiting our ability to determine whether controlling for covariates would have altered the observed group differences in brain activation. While we adjusted for covariates after extracting imaging activation, and the results remained consistent, this represents only a partial check of the covariates’ potential influence. Finally, as this study employed a cross-sectional design, its findings may have limited generalizability to other populations or contexts.

In summary, this work provides new insights into the neural mechanisms underlying risky decision-making which are affected by sex and externalizing pathology. We found that decision-making during the choice phase relies on cingulo-opercular circuits, with specific sex and externalizing differences emerging in modulated brain activity as decisions become riskier. This suggests that risk-processing mechanisms in the cingulo-opercular circuits, specifically intact cognitive control, conflict monitoring, and impulse inhibition during decision-making, during the decision phase, may be critical in understanding sex differences in risky behaviors, warranting further investigation. Based on the findings of dysregulated cognitive control and heightened dmPFC/dACC activation during risky decision-making in male externalizers, a targeted behavioral therapy focusing on impulse control, conflict monitoring, and risk evaluation may be beneficial, particularly in males. This therapy could emphasize strategies to enhance self-regulation and improve choice-phase decision-making under risky conditions, helping these individuals better distinguish between risky and safe choices, which is critical in managing their heightened risk for unsafe behaviors. Future directions of this work are to continue evaluating whether females with externalizing disorders have distinct risk profiles for unsafe behaviors and further, to establish whether sex differences in externalizing decision-making relates to any differences in real-world risky behaviors. In summary, these findings emphasize that specific cingulo-opercular network brain regions, the dmPFC and dACC, during the choice phase of risky decision-making may contribute to heightened risk-taking behaviors observed in males with externalizing disorders. This work highlights an area to target with sex-specific interventions during the pre-adolescent/adolescent period.