Abstract
Avoidant behavior is a defining feature of pediatric anxiety disorders. Although prior research has examined it from the perspective of early information processing events, there has been relatively less consideration of the processes by which anxious youth make avoidant decisions and how these choices are reinforced over time. Studies of risk taking are valuable in this regard because they consider how individuals identify the pros and cons of their choices, how they weigh potential gains and losses and estimate their respective probabilities, and how they tolerate the uncertainty intrinsic to any decision. In this review, we place risk taking within existing models of information processing in pediatric anxiety disorders and highlight the particular value of this construct for informing models of developmental psychopathology and individual differences in outcome over time. We review existing behavioral and neurobiological studies of risk taking in anxious youth and conclude by identifying directions for future research.
Introduction
Anxiety disorders are the most common mental health condition affecting children and adolescents, with prevalence rates ranging from 10% to 20% (1). They are characterized by impairment across multiple domains of functioning and carry risk for a host of poor long-term outcomes (2). The burden of disease associated with these conditions is recognized around the globe (3,4) and within the United States in particular, where they are among the biggest drivers of health care costs for individuals under age 18 (5). Much of the cost—both personal and economic—associated with anxiety disorders can be traced to the behavioral avoidance that defines them. Across the 10 disorders under the anxiety umbrella in DSM-5, avoidance of stimuli or situations perceived as dangerous or threatening is a cardinal feature. This avoidance is self-reinforcing, shaping further retreat over time (6); it is a primary intervention target.
Among numerous factors that drive avoidant behavior, study of risk-taking behavior may be particularly informative. At its core, avoidant behavior is fueled by a desire to avoid danger, a feature that makes anxious youth vigilant for threat and prone to exaggerate their interpretations of it. However, risk taking moves beyond attentional and interpretation biases to encompass how potential gains and losses are evaluated, how uncertainty is navigated, and how behavioral decisions to approach or avoid are made. It holds particular value for providing an integrated view of how approach and avoidance systems—typically studied in isolation in anxiety—interact to produce symptoms and related impairments, and, crucially, how they develop over time. In the context of pediatric anxiety, it has the potential to inform not only our understanding of behavioral avoidance—one extreme of the risk-taking continuum—but also the subset of anxious youth who gravitate toward high-risk behaviors (e.g., substance use). Here, we define risk taking as involving decisions with multiple possible outcomes that can be estimated in terms of probability (where a broader range of possible outcomes corresponds to higher risk). “Risk” itself is determined based on whether there is uncertainty in the outcome; the outcome need not involve a possible adverse effect to be classified as a “risk.”
In this review, we focus on risk taking as an understudied construct in the pediatric anxiety literature. We begin by reviewing information processing (IP) models of anxiety and linking them to the literature on risk taking in youth, noting the relatively limited work on risk taking among anxious youth. We highlight the particular value of this construct for understanding the developmental psychopathology of pediatric anxiety disorders and individual differences in outcome, and we identify directions for future research.
Information Processing Models of Anxiety
Risk-taking behavior occurs within the context of larger IP events that are well documented in the anxiety literature (7–9). Work in this area highlights perturbations in attention, emotion processing, and working memory that define the phenotype of anxiety and may result in behavioral avoidance (10–13). Moreover, this work articulates how genes and environment influence these cognitive processes at the neural level and how they interact to produce the features characteristic of anxiety (12). Anxious individuals have a propensity to selectively attend to threatening stimuli (14–20) and to make threatening appraisals or biased interpretations of neutral stimuli (21–25). In addition, they may exhibit altered fear conditioning—and faulty threat–safety cue discrimination in particular—that crystalize fears over time (26–29). There is also evidence—albeit more mixed—for related memory biases (30,31). Collectively, these processes set the stage for risk-avoidant behavior. Yet, until recently, study of how these features give way to risky decision making and behavior—or avoidance thereof—in anxious youth has been relatively sparse (32,33).
Risky Decision Making
Individual decisions all involve some degree of risk—some larger than others—and are often examined through the lens of behavioral economics. Research in this area focuses on parsing the component processes of decision making to understand how individuals arrive at the choices they make, and it considers features such as the valence, magnitude, and timing of outcomes along with whether aspects of risk are observed or experienced directly. Within these models, risky decisions are those with multiple possible outcomes that can be estimated in terms of probability (where a broader range of possible outcomes corresponds to higher risk).
These decisions may also involve varying degrees of uncertainty, defined here as the psychological state in which a decision maker lacks knowledge about which outcome will follow from which choice (34). Following the neuroeconomic convention to distinguish uncertainty and ambiguity, uncertainty is present when there are multiple possible outcomes with well-defined or estimable probabilities (35). Ambiguity is present when there are multiple possible outcomes whose probabilities are unknown or ill defined (36). Unlike uncertainty, in which individuals can compute the expected utility of different options, ambiguity renders the expected utility of different options incalculable because the probabilities are unknown (35). Thus, probabilities are known under uncertainty but unknown under ambiguity. Research based on psychological experiments (37) and personality tests (38) shows that people typically prefer options with known probabilities to those with ambiguous probabilities [e.g., (37,38)].
From an IP perspective, the choices we make about whether to approach or avoid are shaped by the features of a given situation to which we attend, our interpretation of them, and associated learning and memory processes (39,40), which may exist outside of conscious awareness. Because risk taking requires active deliberation and because it draws on prior learning, memories of outcomes of prior decisions, and appraisals of the current situation, it can be viewed as a step in the IP chain that links basic automatic processes (e.g., attention) to those that can be conceptualized as more deliberative (behavioral decisions).Risk taking occurs along a continuum that captures a range of healthy, adaptive responses as well as potentially maladaptive behaviors that occur at the extremes. Often, study of these extremes has focused on high levels of risk taking and impulsivity that are concerning, given the consequences they can engender (e.g., unwanted pregnancy, accidental injury, death) (41).
However, the continuum can also be conceptualized as including avoidance of risk at the other extreme, a response pattern likely to bring its own adverse outcomes by diminishing valuable opportunities for learning, exploration, and mastery of new situations. Whether distinct neural substrates underlie the different aspects of the risk-taking continuum remains an open question. The marked individual differences in risk taking are thought to be a product of both person-specific (age, genetics, temperament) and environmental (social context) influences (33,42–44), and they are of particular interest in anxiety, in which some youths go on to exhibit high levels of risk-taking behavior.
Age is recognized widely as one of the largest determinants of risk-taking behavior. Both behavioral and neuroimaging studies (45–47) find age-dependent shifts in risk taking such that it peaks during adolescence and corresponds to increases in novelty and sensation seeking during this stage of development. Healthy risk-taking behavior is adaptive, preparing young people for greater autonomy and helping them to explore their environment and calibrate their behavior over time (47). Accordingly, increases in novelty seeking, exploration, and risky behavior are observed cross-species and cross-culturally and are thought to serve an evolutionary purpose that supports independence and reproduction (48,49). At the same time, they account for many of the risks that accrue to youth during this stage of development, including increases in sexually transmitted diseases (50), unwanted pregnancy, distracted and drunk driving, and death ascribed to risky behaviors (51,52).
Behavioral Studies of Risk Taking
Although adolescent decision making has been the subject of considerable research (53), recent reviews have highlighted the challenges that arise in the study of risky decision making in particular (54). Indeed, while developmental differences in risk taking are documented in both real-world settings and epidemiological studies, as assessed using self-report and survey data (55,56), recent meta analyses and systematic reviews paint a more complex picture of developmental trends, which may result from laboratory-based risk-taking scenarios in which it is difficult to simulate real-life risk taking.
Most experimental studies rely on computer tasks that present participants with two (or more) options with varying degrees of risk and reward. The Balloon Analogue Risk Task (BART) is one such task, and it presents participants with a balloon on a computer screen that they can inflate by pushing a button. With each button press, the balloon is inflated, and the participant earns coins. However, at any moment there is a chance the balloon will pop, so the participant must weigh the benefits of earning coins against the costs incurred if it breaks. One strand of experimental research suggests a peak in reward processing (57) and risk taking (58) among adolescents relative to either children or adults using a reward sensitivity task and the BART, respectively. Separate work has suggested that adolescents may be relatively similar to adults in their attitudes toward an aversion of risk (59–61) but more distinct in their attitudes toward ambiguity (60,61), whereby there are no age-related changes in risk attitude but there are age-related changes in ambiguity attitude, with younger adolescents more tolerant of ambiguity than adults (60). Specifically, these studies suggest that adolescents are just as averse to uncertainty but are more willing to accept ambiguous conditions—situations in which the likelihood of winning/losing is unknown—than adults (61).
A meta-analysis by Defoe et al. (45) examined more than two dozen laboratory-based studies involving an array of widely used paradigms of risk-taking behavior and found evidence that adolescents take more risks than adults, a finding that was stronger among early versus older adolescents. However, in contrast to prior studies, these findings also suggested that adolescents made generally similar levels of risky choice when compared with children. Although this finding is complicated both by the small number of studies including child reference groups and by variability in the tasks that were employed, it underscores the need for further study.
In addition to accounting for variability in findings across extant developmental studies of risk taking, the measurement of risk-taking behavior is also noteworthy because it highlights the design trade-offs that come with various approaches as well as growing efforts to refine measurement. Questionnaire, self-report of life events, and event momentary assessment tap the propensity for risk taking in real life, but they are subject to the limitations of self-report and allow only broad understanding of the risky behaviors rather than the specific factors that motivate them. By contrast, in experimental settings, the basic structure of most risk-taking tasks involves the manipulation of different aspects of value-based decisions. Tasks vary from those that provide naturalistic risky decision-making probes (e.g., the Stoplight Driving Task wherein participants drive along a simulated track and yellow “stoplight” driving choices are tracked) to those that more carefully isolate specific components of decision making (e.g., expected reward value, level of uncertainty, or delay/immediacy of outcome). The former offer the advantage of ecological validity and may capture developmental differences—particularly between adolescents and adults—observed in real-world settings. Yet they often cannot isolate specific facets of decision making that would offer a mechanistic understanding of those developmental differences. As such, experimental paradigms borrowed from behavioral economics take a more agnostic approach that may not mirror real-world youth decision making as closely but are better situated to specify features of interest.
Behavioral economic frameworks have facilitated efforts to parse aspects of risky decision making that have been conflated in prior research (e.g., risk/ambiguity; risk aversion/loss aversion). For example, many commonly used risk-taking tasks involve gambles with known magnitude, valence, and probabilities (“Choose between an X% chance of winning $10 and or a sure win of $2”). However, more recent work using economic-based tasks that include conditions in which probabilities are known and conditions in which probabilities are unknown has explicitly delineated risk and ambiguity (60,61), revealing that adolescents may be more tolerant of ambiguity than adults—an important distinction given that these two aspects of decision making (risk and ambiguity) show only weak associations (62). Similarly, as decision making and risk taking are decomposed into their constituent parts, there has been further attention to distinctions between risk aversion and loss aversion. The former encompasses difficulty tolerating the uncertainty of the decision whereas the latter captures the weighing of negative outcomes more heavily than potential gains. As detailed later, these distinctions may be particularly important for the study of youth anxiety disorders.
Neuroimaging Studies of Risk Taking
Behavioral shifts in risk taking track closely to neurodevelopmental changes. For typically developing adolescents, risky behaviors are a product of heightened approach motivation and limited inhibition, factors that distinguish them from their adult counterparts in meta-analytical studies (63). However, these constructs are quite distinct and tap into different aspects of risk taking that may parallel those of reward-related response and cognitive control, respectively (64). Specifically, approach motivation has a curvilinear, inverted U-shaped relationship with age, similar to the pattern of reward-related response in the ventral striatum (VS), as shown in a longitudinal study in which 254 participants between the ages of 8 and 27 years were scanned twice when performing a gambling task (58). A separate longitudinal study provided a mechanistic account of this effect, whereby the VS exhibited greater coupling with the ventral tegmental area in early adolescence when the participants were in a motivational state (65); interestingly, this effect decreased by adulthood. Indeed, high levels of sensation seeking are linked to increased reward-related VS response (58). The data on inhibition suggest a linear developmental trajectory reflecting improvements from childhood to adolescence into adulthood (66–68), mirroring developmental refinements in prefrontal cortex (PFC) regions (69).While dual systems frameworks have focused on the interplay of cognitive control and reward systems, newer models include threat-related systems and their interaction with person-specific and environmental variables in contributing to motivated behavior (70). The triadic model of motivated behavior moves beyond the dual systems model by including not only motivation/approach and regulation nodes but also the avoidance system, centered on amygdala-related circuits; this model provides a basis for studying behavioral responses including risk seeking, emotional intensity, and social reorientation (70). In this model, avoidance is a passive construct that can be distinguished from inhibition, which is an active one. For example, one might “inhibit” the temptation to consume a cookie that is offered while one is on a diet whereas one would “avoid” the temptation by not buying cookies at a store. Neurobiologically, inhibition requires the active engagement of neural systems implicated in behavioral suppression (e.g., prefrontal regulatory systems), but recruiting these systems to achieve avoidance is likely unnecessary.
Importantly, the triadic model is based on the assumption that motivated behavior is the product of these three systems, which together exhibit functional but overlapping equilibrium that is contingent on both sustained and transient factors that vary across individuals. Broader still are conceptualizations that take into consideration the dynamic and interactive effects of multiple systems that, collectively, influence behavior and decision making (71,72). For example, functional networks have the capacity to support complex thought and action that any single element of the system would be unable to support alone. This is consonant with the principles of dynamic systems theory (73), which state that development can only be understood as the multiple, mutual, and continuous interaction of all levels of the developing system.
Risk Taking and Anxiety
Aversion to risk is well documented among adults (74–76) as assessed using the BART and the Iowa Gambling Task and, to a lesser extent, among children and adolescents with anxiety (33,59,77). Anxious individuals have a preference for certain outcomes over those perceived as risky, an inclination that can be motivated by both risk aversion and loss aversion. Studies of adults suggest that anxious individuals overestimate potential losses in gambling-based tasks in which participants choose between a “sure” and “risky” outcome (78) but that risk-avoidant behavior may be motivated more heavily by risk aversion than by sensitivity to potential loss (10). There is also some evidence that risk aversion is specific to anxiety versus other disorders (75).
Only a handful of studies have extended this work to anxious children and adolescents, and these studies have varied with respect to the segment of development and aspects of risk taking studied. In a sample of adolescents and young adults, the expected links emerged between social anxiety symptoms and self-reported risk aversion, but not risk perception (79). In the only study of loss aversion in youths to date, there were no differences between the anxious and comparison youths in performance on a mixed gambles task (80), which presented the participants with gambles in which they had a 50% chance of gaining or losing money. The participants decided whether or not they would accept the gamble. The amount of the potential gain and loss varied across these trials, allowing the calculation of an expected value and a determination of whether the participants were loss averse. Importantly, there have not yet been efforts to dissect risk versus loss aversion within the same task for anxious youth or to determine whether findings in the literature on adults replicate these results.
A separate study examined age and social context as determinants of risk taking among anxious children and adolescents and healthy controls (33), finding that anxious children became more risk avoidant when they were led to believe that peers were evaluating their performance. Surprisingly, this finding did not emerge in either the anxious or healthy control adolescents, underscoring the need for further inquiry. A final study used a community sample of youths to explore the intersection of anxiety level, social stress, and risk-taking behavior (81). Among youths with higher levels of anxiety there was evidence of enhanced sensitivity to the magnitude of reward such that highly anxious youths took more risks as the expected values shifted from negative to positive. Moreover, youths with higher levels of anxiety did not appear to differentiate their responses in risk-taking situations across different levels of stress, suggesting perhaps that anxious youths approach all risk taking with a certain degree of hypervigilance.
Neural Basis of Anxiety
Efforts to understand the neural underpinnings of anxiety have highlighted the fear circuitry that encompasses the amygdala, hippocampus, PFC (82–84). Consistent with animal models and adult human research (85), aberrant responses to threat are observed in these regions among behaviorally inhibited (86) or anxiety-disordered (82–84,87–89) youth. Previous studies (82,90) have suggested that the ventrolateral PFC regulates arousal through its effects on attentional control. Human lesion (91) and functional magnetic resonance imaging (fMRI) (92) work also implicate the ventromedial PFC in regulating amygdala activity in individuals with and without trait anxiety. Other research has focused on approach systems, finding evidence for heightened striatal sensitivity among adolescents with a history of behavioral inhibition (93–95) and/or current anxiety disorder (96).
Notably, the role of the striatum has—until recently—received limited attention in these models (97). Although often thought of in relation to its role in reward systems, it also links to several anxiety processes (e.g., attention bias, fear conditioning, motivation, intolerance of uncertainty) (97,98). The striatum is functionally linked to components of fear circuitry, particularly in the context of stress reactivity (99), and has strong interactive effects with the amygdala (100). Glutamatergic projections from the basolateral amygdala to the VS (101,102) facilitate the VS’s role as an interfacing region able to translate evaluative signals into action (100). The basolateral amygdala encodes the sensory and affective properties of stimuli and outcomes (i.e., value), which are then relayed to the VS to translate into value-based actions (e.g., pursue, avoid) (101–103); thus, amygdala–VS connections may be critical for learning and updating stimulus value. Preliminary evidence suggests that atypical engagement of the striatum is characteristic of youths with anxiety (104).
Neural Correlates of Risky Decision Making in Anxiety
Although research on the neural circuitry undergirding anxiety disorders has proliferated, study of the correlates of risky decision making in anxiety remains limited. This is unfortunate given that such work has the potential to inform our understanding of the developmental psychopathology of anxiety disorders by leveraging what is already known about risk taking among typically developing youth to understand divergent neurodevelopmental trajectories. Indeed, many anxiety disorders have their onset in adolescence during the very window when typically developing youth are exhibiting age-expected peaks in risk-taking behavior, underscoring the potential that risk-taking paradigms have to capture deviations in approach/avoidance behavior and how they develop over time.
Surprisingly, only one study to date has directly examined the neural correlates of risk taking in anxious youth. Using a sample of adolescents with anxiety disorder and healthy control adolescent, it found evidence for both behavioral and neurobiological differences during risky decision making on the Cups Task that varied as a function of both symptom severity and gain/loss condition (32). The Cups Task consists of decision trials that vary on explicit probabilities and reward value, and thus vary in expected value, under two contextual frames—a reward-motivated context (gain frame) and a nonrewarding context (lose frame). Participants decide between a certain monetary gain (or loss) of a small amount and an uncertain gain (or loss) of a larger monetary amount or no gain (or loss). On this task, the anxious youths made fewer risky choices during potential loss compared with the control youths, and these decisions were paralleled by divergent patterns of activation. Neural activation during risky choice was associated with individual differences in anxiety symptom severity, such that greater anxiety was associated with decreased recruitment of the VS in the gain condition and increasing recruitment of the amygdala, dorsolateral PFC, and medial frontal cortex in the loss condition (32). These data suggest that even when anxious youths make a risky choice, their neural circuitry is responsive to the potential negative outcomes of the risk rather than the potential earnings of the reward.
Related work has examined the neural correlates of risk avoidance in a community sample of young adults (mean age = 27.6 years), finding that risk aversion is associated with activation in both the striatum and precuneus activation during both risky and certain task conditions (105). More recently, research using event-related potentials methods has examined the intersection of risk and reward using the BART task among adults with a range of psychopathology, finding that dampened neural engagement in response to reward versus loss may characterize both extremes of aberrant risk-taking behavior (impulsivity and avoidance) (106). Further work with pediatric samples is needed in light of other research suggesting heightened striatal response to reward when the outcome was contingently associated with choice among youths with high levels of anxiety (104) as well as those with a history of behavioral inhibition (107).
Challenges and Future Directions
Taken together, this work highlights both the promise and challenges that accompany studies of risk taking. While the construct holds promise for linking early information processing to more deliberative decisions that may result in impairment, several methodological issues limit our understanding. First, across both behavioral and neurobiological investigations, definitions of risky behavior and its core components often differ, complicating efforts to link and compare findings. Measurement issues remain, and there is a clear need to probe the extent to which task-based avoidance maps to clinically relevant, real-world avoidance. Small sample sizes are common and understandable in this emerging line of research, yet they are ill-suited to capturing the heterogeneity and individual differences inherent in risk taking.
These challenges are particularly important for understanding anxious youths who will exhibit heightened risk taking (e.g., substance use, unprotected sex, impulsive decisions), which is often theorized to serve as a strategy for regulating affective distress (108–114). Whether the tendency toward heightened risk taking is specific to some anxiety disorders versus others and whether it is modulated by age and/or contextual factors also remains unclear. Moreover, our view of how development intersects with risk taking for anxious youths is limited by study designs that often do not include a child reference group or compare only adolescents and adults. Relatedly, the traditional reliance on case-control designs may hamper efforts to understand individual differences in risk taking by failing to consider the full continuum of the anxiety phenotype and its relationship with risk. Finally, the growing move toward more refined measurement of risk taking has yet to fully translate to studies of pediatric anxiety. Beyond the challenge of fMRI tasks that do not precisely mimic the risk taking (or avoidance) observed in real life, studies have yet to parse closely related constructs, including aversion to risk, loss, and ambiguity, that would deepen our understanding of the choices anxious youths make.
These methodological shortcomings are to be expected as the pediatric anxiety literature comingles with the adolescent risk-taking literature. As work in this area moves forward, we highlight four areas for advancement and innovation. First, there is a need for longitudinal designs that can shed light on relevant developmental processes and in doing so capture key transitions both into and out of adolescence. Second, future work will benefit from a dimensional perspective geared toward capturing individual differences in both anxiety and risk taking. Longitudinal studies of community youth reveal striking heterogeneity in anxiety outcomes during the transition to adolescence (115–117). Yet individual differences are seldom a focus of developmental fMRI studies. The broad brushstroke group comparisons often conducted among different age groups and/or between pathological and healthy groups reveal sweeping developmental trajectories, but such comparisons may rob these studies of the power to attain a nuanced understanding of individual differences in neural sensitivity and phenotypic variability. Third, we echo the calls of others in the field to carefully consider the component and interactive contributions of striatal activation to youth anxiety. Given the bidirectional projections between the striatum and other affective regions (e.g., amygdala) and its role in learning, zooming in on the striatum’s role in anxiety will expand our understanding of the network of regions associated with anxious phenotypes. Finally, we stress the importance of integrating other strands of developmental research (e.g., sleep, puberty, inflammation) that might provide a more complete view of both risk taking and anxiety. Each of these topics merits its own review, and we refer readers to the growing literature linking puberty and risk taking, a rich topic beyond the scope of the current review [see (58,118,119)].
However, to illustrate the value of this broader developmental view, we highlight sleep as a domain of interest that undergoes sweeping changes in adolescence (120), has clear links to risk-taking behavior, and also impacts emotional well-being. Sleep is a fundamental need; when sleep is insufficient, it impacts virtually every domain of functioning, including learning, memory, decision making, brain function, and mental health (120). A growing body of work has highlighted not only the sleep disturbances that adolescents face (121) but the link of these disturbances to cognitive control and reward sensitivity neural circuitry, crucial determinants of risk-taking behavior (122). Separate work has linked it to adolescent depression (123) and, to a lesser extent, anxiety (124). An fMRI comparison of anxious and nonanxious adolescents found that anxiety modulated the effects of sleep duration on neural activity in the anterior cingulate and hippocampus in response to negative faces (125). Prospective longitudinal data suggest that sleep disturbance commonly precedes anxiety during this developmental window (124,126), and growing daily diary evidence suggests developmental sensitivity of this effect in early adolescence (127). Yet there remains a need to determine whether poor sleep exacerbates anxiety symptoms or vice versa and how each of these processes influences behavioral decisions. Such knowledge will be important for designing effective interventions that improve both sleep and anxiety. Indeed, recent work suggests that a sleep intervention can improve sleep in anxious youth and that sleep improves over the course of anxiety treatment (128).In conclusion, the study of risk taking has the potential to elucidate a central impairment in pediatric anxiety disorders and to provide meaningful guidance about how we might intervene. Although risk aversion is an accepted feature of the anxiety phenotype, our understanding of risk taking in this population is still emerging. Future work, rooted in behavioral economics paradigms that deconstruct the decision-making process into its component pieces, will inform our understanding of how youth arrive at the decisions they make, how these choices are shaped by individual and contextual factors, and how they are moderated by other key developmental processes.