That Was Easy! Expectancy Violations During Exposure and Response Prevention for Childhood Obsessive-Compulsive Disorder

Abstract

Violating expectancies during exposure therapy is proposed to promote inhibitory learning and improved treatment outcomes. Because people tend to overestimate how distressing emotionally challenging situations will be, violating expectations of distress may be an intuitive way to promote treatment outcome during exposure-based cognitive-behavioral therapy (CBT). This study evaluated overpredictions of distress during exposure tasks in 33 youth with obsessive-compulsive disorder (OCD; ages 8-17) participating in CBT. Youth with more variable prediction accuracy and a higher proportion of overpredictions experienced more rapid symptom reduction, b = −0.29, p = .002. Underpredictions were less common toward the end of therapy as youth experienced less severe OCD, b = 0.12, p= .001. Findings suggest that although youth often accurately predict the intensity of exposure, overpredictions are common as well. The frequency of these overpredictions promoted treatment outcome, supporting expectancy violations as one indicator of inhibitory learning during exposure therapy.

Keywords

OCD exposure therapy ERP inhibitory learning affective forecasting youth

Introduction

The Inhibitory Learning Approach to Exposure Therapy

Exposure therapy is a highly effective behavioral technique for youth with anxiety and obsessive-compulsive disorders (OCD; Ale, McCarthy, Rothschild, & Whiteside, 2015; Kendall et al., 2005). Although exposure-based cognitive-behavioral therapy (CBT) has established a strong evidence base, it is estimated that approximately 30% of youth do not respond to this treatment, and almost half do not reach remission (Öst, Riise, Wergeland, Hansen, & Kvale, 2016). Studying the mechanisms that underlie successful exposure can clarify how to deliver this therapy more effectively and improve treatment outcomes. Although exposure is a critical component in CBT for children with anxiety disorders and OCD, exposure and response prevention is particularly emphasized in CBT for childhood OCD (Ale et al., 2015) and, thus, is an ideal treatment to study when evaluating exposure therapy mechanisms.

Craske and colleagues (2008) provided the most recent comprehensive review of the mechanisms of exposure, proposing an “inhibitory learning” approach rooted in basic extinction learning research. Their framework is grounded in a Pavlovian fear conditioning model of anxiety and related disorders, which suggests that fear extinction occurs when a conditioned stimulus is repeatedly presented without an unconditioned stimulus that evokes a fear response (Craske et al., 2008; Craske, Treanor, Conway, Zbozinek, & Vervliet, 2014). Central to this framework is the emphasis on expectancy violations in promoting inhibitory learning (Brown, LeBeau, Chat, & Craske, 2017; Craske et al., 2008; Craske et al., 2014). From an extinction learning perspective, expectancy violations occur when a conditioned feared stimulus that has been previously associated with an unconditioned stimulus is presented without the unconditioned stimulus (Craske et al., 2008; Craske et al., 2014). A recent study with college students found that the retention of fear extinction was significantly positively associated with expectancy violations during extinction learning trials (Brown et al., 2017).

An example of an expectancy violation during exposure therapy may be when a person with arachnophobia expects to be bitten by a spider if he or she gets close and then touches a spider without being harmed. In this case, a learning network linking a spider with a neutral outcome would be more likely to develop, and the spider–danger association would be more likely to be inhibited. The importance of expectancy violations in exposure therapy was demonstrated in a study of college students with elevated anxiety sensitivity, in which reduced expectations of catastrophic outcomes during exposure therapy were found to mediate improvement (Deacon et al., 2013). In contrast, a recent examination of prolonged exposure therapy for adults with posttraumatic stress disorder found that although harm expectancies decreased across treatment, this reduction did not correspond with treatment outcome (De Kleine, Hendriks, Becker, Broekman, & van Minnen, 2017).

Unfortunately, feared outcomes are not always refutable in OCD (Jacoby & Abramowitz, 2016). For example, many patients with OCD fear that moral wrongdoing may result in losing their salvation; a therapist will struggle to definitively violate this expectation. Many people with OCD also report “not just right” experiences, and a feared outcome for them may be confronting stimuli that are “not right” that may bring about an intolerable level of distress (Summerfeldt, Kloosterman, Antony, & Swinson, 2014). Still others engage in rituals without being able to articulate a concrete feared outcome, a phenomenon particularly prevalent in children (Geller et al., 2001). Furthermore, intolerance of uncertainty is a central cognitive distortion that maintains obsessive distress, and thus building acceptance of uncertainty is a central focus in the treatment of people with OCD (Obsessive-Compulsive Cognitions Working Group, 2005).

An alternative way to violate expectancies could be violating expectations of distress (Craske et al., 2014). If children expect that exposure exercises will cause them more distress than they actually do, they could learn that these exercises are more tolerable than anticipated, and in turn develop learning networks between a feared stimulus and more manageable emotional reactions, which may also enhance therapeutic engagement. Affective expectancy violations may also be an indicator of progress to therapists, as it may suggest improved emotion regulation, extinction learning (Craske et al., 2014), or self-efficacy (van Hout & Emmelkamp, 1994), and they may also lead to less avoidance or compulsive behavior in the future (Craske, Rapee, & Barlow, 1988).

Anxiety Forecasting and Expectancy Violations

People often overestimate the magnitude of their emotional reactions to various situations (Wilson & Gilbert, 2013). This finding has been replicated across different circumstances and emotional states but is particularly pronounced when anticipating negative events (Gilbert, Pinel, Wilson, Blumberg, & Wheatley, 1998). Wilson and Gilbert (2003) have coined the expression “affective forecasting” to describe the degree to which people predict their own emotional experiences.

One recent study demonstrated that elevated trait-level anxiety in college students significantly predicted overpredictions of nervousness/jitteriness during an upcoming week, suggesting that clinical populations may be especially biased in their emotional predictions (Wenze, Gunthert, & German, 2012). Accordingly, affective forecasting errors have been recognized in those with clinical anxiety for some time; Rachman (1994) first systematically described this pattern, in which the majority of his participants with anxiety disorders overestimated the amount of fear they would experience during a behavioral approach task. These overpredictions likely contribute to the development and maintenance of clinical anxiety, as avoiding situations that provoke uncomfortable emotional states has been said to underlie all emotional disorders (Barlow et al., 2014; Hayes, 2016).

Despite clear ties between affective forecasting biases and exposure therapy, only two studies have examined overpredictions of distress during exposure therapy. Van Hout and Emmelkamp (1994) studied expected versus actual fear experiences at three time points during the course of exposure therapy for adults with agoraphobia. They found that, on average, expected fear was greater than actual fear during exposure tasks but that predictions were no more accurate at the last session than they were at the first. Beyond finding increased self-efficacy immediately after overpredictions, van Hout and Emmelkamp (1994) did not evaluate other outcomes like posttreatment functioning.

More recently, Kircanski and Peris (2015) tested whether overpredictions of distress during exposure and response prevention for youth with OCD predicted treatment outcome. In contrast to their hypotheses, they found that children with greater expected versus actual distress during exposure trended toward less clinician-rated improvement at mid-treatment, though this relationship became nonsignificant at posttreatment and follow-up. The present study seeks to build on this work by addressing several questions about affective expectancy violations during exposure therapy that were not addressed by Kircanski and Peris (2015), including how often they occur, whether there are developmental differences in affective forecasts among children and adolescents, whether symptom severity corresponds with forecasting accuracy, whether expectations change across the course of therapy, and whether the frequency of expectancy violations affects treatment outcome.

Kircanski and Peris’s (2015) research into exposure therapy processes in youth was much needed, as they are rarely studied in youth compared with adults. When considering overpredictions of distress in this developmental group, it is worth noting that children with OCD are less likely to have insight into the excessiveness of obsessions and compulsions (Geller et al., 2001). Their expectations about their own emotional reactions and ability to cope with those experiences are likely to be less accurate as well, and so this phenomenon should be more thoroughly studied in this population.

The Present Study

The first goal of this study was to provide a description of the frequency of affective forecasting errors among children with OCD during exposure and response prevention. Research suggests that people become better predictors of their own emotional reactions when repetitively presented with the same stimuli (Rachman, 1994), but this phenomenon has not been evaluated during CBT, during which patients frequently face thematically similar but experientially novel exposure exercises as they progress in treatment. Thus, it is unclear whether children become more accurate predictors after participating in exposure in varying contexts.

Another goal was to test whether symptom severity positively corresponded with overpredictions of distress. These relationships were examined both between and within participants, evaluating whether children with more severe symptoms made greater overpredictions, and whether children overpredicted more dramatically on days when they are experiencing more severe obsessions and compulsions. The final goal of this study was to test whether overpredictions are associated with improved treatment outcome. Expectancy violations in the prediction of treatment outcomes were tested two ways: (a) by assessing the average difference between expected and actual distress level for each child, and (b) by assessing variability in prediction accuracy, as was done in a recent study of expectancy violations in a fear extinction paradigm (Brown et al., 2017). Youth who are more variable in their prediction accuracy would be expected to have more opportunities to experience substantial discrepancies between their expected and actual distress during exposures (i.e., more expectancy violations), and thus were expected to experience improved treatment outcomes.

Method

Procedure

The present analysis used data from an observational CBT treatment outcome study for youth with OCD that was approved by the University of Florida (UF) Institutional Review Board. Parent–child dyads were recruited through clinic flow at the UF OCD Program and through advertisements on the International OCD Foundation website. Either prior to or immediately after their intake appointment, child–parent dyads were approached about participation in the study. Participants were followed for a maximum of 15 sessions.

Therapists were licensed psychologists or psychology trainees at the postdoctoral, predoctoral intern, or doctoral student level supervised by licensed psychologists experienced in the treatment of pediatric OCD. Therapy was guided by CBT with exposure and response prevention principles (e.g., see Storch, McGuire, & McKay, 2018). Patients were seen by treatment teams of two to four therapists, each of which had a “treatment team leader” designated by faculty consensus. Therapists attended weekly group and individual supervision to discuss cases. For more information about the training model, see Balkhi, Reid, Guzick, Geffken, and McNamara (2016).

Participants

Inclusion criteria were age between 7 and 17 years old, a primary diagnosis of OCD as determined by a licensed psychologist, and a score of at least 16 on the Children’s Yale-Brown Obsessive Compulsive Scale (CYBOCS). Patients were excluded if they had a diagnosis of autism spectrum disorder or a significant medical/learning impairment that hindered their ability to complete questionnaires or participate in treatment.

Fifty-one parent–child dyads provided written consent/assent. Seventeen parent–child dyads were excluded from analyses because the child did not meet diagnostic criteria for OCD (n = 6), children or parents preferred to participate in treatment without research activities (n = 4), OCD was a secondary diagnosis (n = 3), the family did not have time for therapy (n = 1), distance (n = 1), the child could not fill out questionnaires due to a reading disability (n = 1), the child had autism spectrum disorder (n = 1), and the family failed to initiate treatment following the intake (n = 1).

The final sample consisted of 33 youth. Of the final sample, 19 patients completed all 15 sessions, nine completed therapy early, five dropped out. Reasons for drop out included satisfaction with outcome (n = 2), a desire to continue treatment without filling out questionnaires (n = 1), distance (n = 1), and financial barriers (n = 1).

Parents were primarily married (n = 29), non-Hispanic White (n = 29) mothers (n = 30). Patients participated in therapy sessions on a once-weekly, semi-weekly (two to three sessions per week), or intensive (four to five sessions per week) basis based on collaborative treatment planning with the patient and family. Each frequency of CBT treatment has been shown to result in similar treatment outcomes (Storch et al., 2007). A summary of demographic variables is shown in Table 1.

Table 1.

Demographics.

Characteristic
Female sex n (%)	19 (58%)
Age M (SD)	12.5 (3.1)
Baseline CYBOCS score M (SD)	25.4 (5.4)
Race n (%)
White	29 (88%)
Asian	2 (6%)
Other	2 (6%)
Comorbid diagnosis n (%)
Depressive disorder	11 (33%)
Anxiety disorder	5 (15%)
Attention deficit/hyperactivity disorders	4 (12%)
Tic disorders	2 (6%)
Oppositional defiant disorder	2 (6%)
Comorbid eating disorder	1 (3%)
Psychotropic medications n (%)
Selective serotonin reuptake inhibitors	17 (52%)
Atypical antipsychotic medication	6 (18%)
Psychostimulants	4 (12%)
Serotonin antagonist and reuptake inhibitors	4 (12%)
Anxiolytics	2 (6%)
Benzodiazepines	1 (3%)
Alpha2-adrengergic agonist	1 (3%)
Treatment frequency n (%)
Weekly	14 (42%)
Semiweekly	6 (18%)
Intensive	13 (39%)

Note. CYBOCS = Children’s Yale-Brown Obsessive-Compulsive Scale.

Measures

OCD severity

The CYBOCS (Scahill et al., 1997) was used to assess OCD symptom severity. Both the interviewer-rated (CYBOCS-IR) and parent-rated (CYBOCS-PR; Storch et al., 2006) versions were used. Both have demonstrated strong psychometric properties (Scahill et al., 1997; Storch et al., 2006; Storch et al., 2004) and were highly correlated in this sample, r = .69. The internal consistency of the CYBOCS-IR and CYBOCS-PR were good, α = .89 and α = .90, respectively. The CYBOCS-PR was administered at every session and, thus, was used as the dependent variable to more reliably assess youth functioning across treatment. The CYBOCS-IR was used to control for baseline severity.

Expected and actual distress caused by exposure

A Subjective Units of Distress Scale (SUDS) was used to assess children’s expected and actual perceived distress on a 0 (no distress) to 10 (maximum distress) scale. This metric was first described by Wolpe and Lazarus (1966) in their seminal book on behavior therapy for anxiety. The SUDS scale continues to be a widely used, simple measure of anxiety in both clinical (e.g., Storch et al., 2018) and experimental (e.g., Kircanski et al., 2012; Rachman, 1994) settings. Before the first exposure of each session, children were asked to report how distressing they expected the exposure to be on a SUDS scale. After the session was over, they were asked to report their perceived actual difficulty of the task on the same 0 to 10 SUDS scale. We chose to measure “actual distress” this way, rather than the peak distress during exposure, because it mirrored the way the anticipated distress was assessed; as an overall account of the distress caused by the exposure, much in the same way youth were asked to rate exposures when creating a fear hierarchy.

Exposure therapy prediction accuracy

For descriptive analyses, prediction accuracy was measured by subtracting the actual score from the expected score. Predictions were considered “accurate” if the expected SUDS was within one standard error of measurement of the actual SUDS, “overpredictions” if the expected SUDS was more than one standard error of measurement of the actual SUDS, and “underpredictions” if the expected SUDS was more than one standard error of measurement less than the actual SUDS. The standard error of measurement is a commonly used metric to describe the reliability of a test score (e.g., on educational testing; Harvill, 1991), and was chosen as a simple empirical way to assess whether actual SUDS scores were reliably different than expected SUDS scores. The standard error of measurement is calculated as follows: SD × √(1 − r), where r is the reliability of the measure expressed as the Pearson correlation between items, and SD is the standard deviation of the measure (Harvill, 1991). Based on this calculation, an “accurate” prediction occurred when the actual SUDS were within 1.46 SUDS points of the expected SUDS. Because expected and actual SUDS were assessed using whole numbers, this translated to predictions being considered “accurate” if the expected SUDS was within one point of the actual SUDS. This empirically derived difference also holds clinical utility, as exposures that are expected to be two SUDS points different than the actual SUDS are likely perceived as experientially different, while an exposure in which expected and actual SUDS are within one unit of each other may be a less clinically noticeable difference.

Analysis

The first aim was to present a description of the pattern of over-, under-, and accurate predictions over the course of CBT using both categorical and continuous descriptors. Categorical descriptions included the proportion of youth who were classified as over-, under-, and accurate predictors based on their average discrepancy score, and the proportion of exposure sessions in which over-, under-, and accurate predictions occurred. Continuous descriptors included the average discrepancy score between expected and actual SUDS and the standard deviation of the discrepancy score.

Hypotheses were tested using multilevel modeling (MLM) with the procedures recommended by Singer and Willett (2003) due to this analysis’ ability to evaluate trajectories of change, to account for unequal time points in data, to analyze both fixed and random effects, and to assess both within-person and between-person characteristics (Tasca & Gallop, 2009). We followed a “rule of thumb” in MLM that suggests including five repeated measurements for a sample size of 30 participants (Maas & Hox, 2005; Snijders & Bosker, 1999). The present study included 33 youth who participated in an average of 13 CBT sessions and were assessed at each time point, and thus was adequately powered using these guidelines. Moderators of treatment outcome can be tested by evaluating the residualized interaction term between session number and an independent variable in predicting symptom severity, to determine whether an independent variable moderates the degree to which symptoms decrease across treatment.

Per the Singer and Willett (2003) recommendations, model fit in MLM is evaluated by comparing fit statistics between subsequent “nested models,” or models with additional independent variables added. The first model is the “unconditional means model,” which does not contain any independent variables. Significant improvement in −2 log likelihood (−2LL) values are determined by using chi-square analyses.

MLM uses a full-information maximum likelihood approach to impute missing data points. Missingness was evaluated by Little’s test for whether data are missing completely at random (Little, 1988). If they were found to not be missing completely at random, the following variables were tested as predictors of missingness: family income, baseline symptom severity, and comorbidity. Any significant predictors of missingness were included as covariates.

The second aim analyzed whether symptom severity corresponded with prediction accuracy. This was tested using expected distress as the dependent variable with actual distress as a covariate in the first nested model. This MLM included the following control models: missingness covariates, session, age, and treatment variables (treatment frequency, medication status). Age was included in this model to test whether younger children make more substantial overpredictions. Each nonsignificant predictor was dropped from subsequent models to optimize statistical power (with the exception of covariates that were included to control for missingness). The final nested model comprised mean-centered parent-report CYBOCS (to determine whether within-child symptom severity corresponded with prediction accuracy) and group-centered parent-report CYBOCS (to determine whether between-child symptom severity corresponded with prediction accuracy).

The third aim sought to determine whether prediction accuracy corresponded with treatment outcome, and was tested with an MLM using CYBOCS-PR as the dependent variable and the same control models described earlier. The final nested model tested the interaction between session number and child predictor status (i.e., whether children were categorized as over-, under-, or accurate prediction) and the interaction between session number and the standard deviation of each child’s prediction accuracy (i.e., whether variability in prediction accuracy corresponds with treatment outcome).

Results

Missingness

Data were not missing completely at random on the variables of interest (SUDS predictions, CYBOCS-PR), χ² = 45.5, p = .007. Baseline CYBOCS-IR scores were the only variable found to be negatively related to missingness on the CYBOCS-PR, r = −.14, p = .001, and on child prediction accuracy, r = −.10, p = .025, so it was used as a covariate in all models.

Characterizing Affective Forecasting Errors During CBT

Affective forecasting biases during exposure are displayed in Table 2. These trends did not appear to vary substantially between younger children (ages 8-12) and adolescents (ages 13-17).

Table 2.

Prediction Accuracy for Whole Sample Across All Sessions (322 Predictions).

	Everyone	Children (ages 8-12)	Adolescents (ages 13-17)
Underprediction %	11	11	10
Accurate prediction %	53	57	50
Overprediction %	36	31	40

Figure 1 shows that the proportion of correct and overpredictions for each session remained relatively stable throughout treatment, but that underpredictions generally only happened toward the beginning of therapy.

Figure 1.

Proportion of overpredictions, accurate predictions, and underpredictions across sessions.

Table 3 shows the number of children who were classified as over-, under-, and accurate predictors based on the average discrepancy between their expected and actual SUDS values. Although the mean discrepancy between actual and expected distress was less than one SUDS point, variability in discrepancy scores was relatively higher, with an SD of 2.2, indicating that there was substantial variability in the accuracy of children’s expectations. Again, clear differences between children and adolescents were not present.

Table 3.

Proportion of Children Who Were Under-, Over-, and Accurate Predictors Across All Sessions.

	Total N	UnderpredictorsN (%)	Accurate predictorsN (%)	OverpredictorsN (%)	Prediction accuracy^a M (SD)
Children (ages 8-12)	15	0 (0)	12 (80)	3 (20)	0.80 (2.1)
Adolescents (ages 13-17)	18	1 (6)	12 (67)	5 (28)	0.89 (2.3)
Total sample	33	1 (3)	24 (72)	8 (24)	0.85 (2.2)

Note. SUDS = Subjective Units of Distress Scale.

This figure was calculated by subtracting the expected SUDS value from the actual SUDS value for each exposure and response prevention exercise.

Relationship Between Prediction Accuracy and Symptom Severity

Nested models that included actual distress, baseline CYBOCS-IR, and session significantly improved the fit of the model predicting expected distress, while the models that added age, treatment frequency, and medication status did not. Thus, none of these latter variables were significant predictors of forecasting accuracy.

In the final model, adding the random and fixed effects of person-centered and group-centered CYBOCS-PR scores significantly improved the fit of the previously best fitting model, χ²(3) = 141.28, p < .001, though both person-centered and group-centered CYBOCS-PR scores were nonsignificant fixed predictors and showed nonsignificant random effects as well, ps > .1.

It was expected that child-centered CYBOCS-PR would show multicollinearity with the session variable (i.e., symptoms reduce as treatment progresses). Thus, we included another model post hoc that evaluated the independent impact of child-centered OCD severity on prediction accuracy without controlling for session. In this model, child-centered CYBOCS-PR scores had a significant negative effect, such that when children had less severe symptoms, they tended to overpredict the distress caused by exposure, b = −0.11, p = .009. This model also had improved fit statistic compared to the previously best-fitting model, χ²(2) = 129.78, p < .001. See Table 4 for a summary of nested model parameters.

Table 4.

Multilevel Model Investigating Child Prediction Accuracy, Using Expected Distress as the Dependent Variable.

	Unconditional means model	Model B	Model C	Model D	Model E^a	Model F^a	Model G	Model H
Fixed effects
Actual distress		0.53***	0.53***	0.51***	0.53***	0.52***	0.45***	0.49***
Baseline CYBOCS			0.054*	0.042	0.056	0.042	0.029	0.034
Session				0.097**	0.095***	0.096***	0.11**
Child (age 8-12) vs. adolescent (age 13-17)					−0.30
Taking psychotropic medication vs. not taking medication						0.19
Intensive vs. weekly^b						0.32
Semi-intensive vs. weekly^b						0.28
Group-centered CYBOCS-PR							0.038	0.047
Child-centered CYBOCS-PR							−0.034	−0.11**
Random effects
Session				0.0055
Group-centered CYBOCS-PR							<0.001	<0.001
Child-centered CYBOCS-PR							0.0076	0.0077
Fit statistics
–2LL	1,495.20	1,309.98	1,305.81	1,258.70	1,290.26	1,290.35	1,144.42	1,155.92

Note. CYBOCS = Children’s Yale-Brown Obsessive-Compulsive Scale; CYBOCS-PR = Children’s Yale-Brown Obsessive-Compulsive Scale, Parent Report; LL = log likelihood.

Model did not improve upon previous model.

Treatment types were dummy-coded and compared to weekly treatment.

p < .05. **p < .01. ***p < .001.

Prediction Accuracy and Treatment Outcome

Nested models that included baseline severity, age, psychiatric medication status, and treatment frequency did not improve the fit of the model compared with the previous model. Thus, they were all dropped from the following models, except for baseline severity due to its significant relationship with missingness. The model including the fixed and random effects of session was found to have significantly better fit than the previous model, χ²(2) = 228.07, p < .001, with significant fixed and random effects.

The final model added the interaction between session and predictor status as well as the interaction between session and prediction variability. Because only one child was found to be an underpredictor, the accurate and underpredictor groups were combined to compare their treatment outcome with that of overpredictors. This model had significantly better fit than the previous model, χ²(2) = 10.81, p = .004. The interaction between predictor status and session was nonsignificant, though the interaction between prediction variability and session was negative and significant. Visual inspection of this interaction suggested that those with more variable prediction accuracy experienced quicker symptom reduction than those with less variable prediction accuracy, though they started with more severe symptoms (see Figure 2). See Table 5 for a summary of model parameters.

Figure 2.

Expectancy accuracy variability moderates the rate of symptom reduction.

Table 5.

Multilevel Model Predicting Symptom Severity.

	Unconditional means model	Model B^a	Model C	Model D^a	Model E^a	Model F
Fixed effects
Baseline CYBOCS		0.28	0.33	0.33	0.15	0.22
Session			−0.80**	−0.78**	−0.78***	−0.78***
Child (age 8-12) vs. adolescent (age 13-17)				−1.51
Taking psychotropic medication vs. not taking medication					−1.22
Intensive vs. weekly^b					4.18
Semi-intensive vs. weekly^b					3.29
Session × Predictor status^c						−0.13
Session × Prediction variability						−0.29**
Random effects
Session			.16**	.27**	.27**	.22*
Fit statistics
–2LL	2,265.54	2,263.332	2,035.267	2,034.787	2,031.021	2,024.453

Note. CYBOCS = Children’s Yale-Brown Obsessive-Compulsive Scale; LL = log likelihood.

Model did not improve upon previous model.

Treatment types were dummy-coded and compared to weekly treatment.

The predictor status analysis compared overpredictors to accurate and underpredictors.

p < .05. **p < .01. ***p < .001.

In light of the significant interaction between the SD of prediction accuracy and session, the hypothesized relationship between the SD of prediction accuracy and the proportion of overpredictions was empirically tested post hoc. A partial correlation was done between SD of prediction accuracy and number of overpredictions, while controlling for total number of predictions. The correlation was positive and significant, r = .38, p = .033, suggesting that children with more variable prediction accuracy also had a higher number of overpredictions even when controlling for total number of predictions, indicating a higher proportion of overpredictions.

Discussion

Summary of Findings

Findings suggest that youth with more frequent affective expectancy violations during exposure and response prevention experience more symptom reduction during CBT, supporting the inhibitory learning approach to exposure therapy. The frequency of overpredictions of distress, however, was relatively less common than might have been expected, as children and adolescents accurately forecasted their SUDS over half of the time within one SUDS unit. Underpredictions of distress were more common toward the beginning of therapy.

Affective Forecasting Among Youth During Exposure and Response Prevention

Contrary to previous research (Rachman, 1994; van Hout & Emmelkamp, 1994; Wenze et al., 2012), this study found that youth often made accurate predictions about the distress exposure would cause, with their average expected distress falling less than one SUDS unit away from their actual reported distress. Children accurately predicted the difficulty of exposures within one SUDS unit 53% of the time, while they overpredicted the distress caused by exposures 36% of the time. This one unit threshold not only was empirically derived but also may serve as a clinically intuitive threshold for a surprise in the perceived difficulty of an exposure (i.e., an exposure that is at least two SUDS points different than expected). These results persisted despite studying a population that would be theoretically likely to overpredict distress, as individuals with more anxiety and emotional distress tend to make more substantial affective forecasting errors (Rachman, 1994; Wenze et al., 2012). Furthermore, this study investigated children and adolescents, a population that is less likely to have insight into the excessiveness of obsessions and compulsions, and thus may be especially susceptible to these biases (Geller et al., 2001).

This discrepancy with previous research likely arises from methodological differences between this and previous studies. The strongest empirical support for affective forecasting biases comes from studies using between-person designs by comparing participants who anticipate an affective experience with those who actually experience it (e.g., Gilbert et al., 1998). Interestingly, however, it seems that affective forecasting research has less frequently used within-person designs to test how often individuals make affective forecasting errors. The present study evaluated over-, under-, and accurate predictions over the course of exposure therapy using a standardized measurement of reliability that resulted in a clinically meaningful threshold of one SUDS unit, and thus was able to describe how often these over- or underpredictions occurred for an individual.

Predictors of Affective Expectancy Violations

This study found that being in a later stage of treatment was the only significant predictor of affective forecasting, such that youth tended to report greater expected than actual distress as therapy progressed. Visual inspection of prediction accuracy over the course of treatment suggests that this association likely arose from the almost disappearance of underpredictions toward the end of therapy (rather than an increased frequency of overpredictions). Underpredictions may be rare later on in therapy when youth had less severe symptoms because they likely made forecasts based on a long learning history that would suggest substantial distress during exposure but may not have recognized relatively recent clinical improvements they had made in treatment.

Contrary to hypotheses, those with more severe OCD were not found to overpredict distress to a greater degree. This may have been due to a restricted range of severity; including children with subclinical anxiety or obsessions as well may better illuminate whether children with internalizing disorders are less accurate at predicting their emotional experiences during challenging tasks. Younger children may also be expected to have less accurate expectancies, as younger people have been found to have poorer insight (Geller et al., 2001). Regardless, we found that there was no relationship between age and affective forecasting accuracy, suggesting that these biases persist across stages of development.

Expectancy Violations in the Prediction of Treatment Outcome: Implications From an Inhibitory Learning Perspective

Consistent with the work of Kircanski and Peris (2015), overpredictions of distress during exposure failed to be a significant predictor of treatment outcome when evaluating each child’s average expected versus actual SUDS score. Variability in prediction accuracy, however, was found to moderate symptom reduction. This approach is consistent with how expectancy violations have been measured in previous human extinction learning research (e.g., Brown et al., 2017), as individuals with greater variability in the accuracy of their predictions have more opportunities to experience expectancy violations. Indeed, results from this study showed that children with more variability in their expectancy accuracy experienced a greater proportion of overpredictions, suggesting that a higher number of “pleasant surprises” during exposure contributed to more symptom reduction. Although youth with more variable prediction accuracy had more severe OCD toward the beginning of treatment, they experienced more rapid symptom reduction, ending therapy at a similar level of functioning as those with more consistent predictions, a finding that persisted even when controlling for baseline severity. The model suggested that the half of the sample with higher variability in their prediction accuracy experienced a 13-point CYBOCS reduction after 15 sessions, while the half of the sample with more stable predictions experienced a 9-point reduction.

These findings support the inhibitory learning approach to exposure therapy (Craske et al., 2008; Craske et al., 2014), which suggests that extinction learning can be optimized by promoting expectancy violations during exposure. From this perspective, variability in exposure contexts and frequent expectancy violations were expected to promote treatment outcome. More variability in exposure context should maximize the likelihood that extinction learning networks are retrieved in new situations, which includes internal contexts such as affect. From this framework, children who participate in exercises that are as challenging, less challenging, and more challenging than they expect may be more prepared to take on situations that arise in their everyday lives.

The finding that more variability in prediction accuracy moderated symptom reduction may have also indicated that “occasional reinforced extinction” occurred when children experienced underpredictions, or occasional pairing of the unconditioned stimulus with the conditioned stimulus during extinction learning (Woods & Bouton, 2007). It is thought this pairing followed by further presentations of the conditioned stimulus without the unconditioned stimulus protects against a “return of fear” (Krompinger, Van Kirk, Garner, Potluri, & Elias, 2018; Woods & Bouton, 2007). Thus, if children experience more distress than expected during an exposure exercise but also do several exposure exercises that are as challenging or easier than expected, they may learn that exposure rarely causes less distress than expected, and that perhaps they are able to manage them even when they do evoke a high level of distress. In contrast, if every exposure is as challenging or is easier than expected, children may confront a situation that is more anxiety-provoking than they anticipate in the real world and may experience a return of obsessive symptoms.

Clinical Implications

These results support clinical recommendations to maximize expectancy violations during exposure therapy (Craske et al., 2014; Jacoby & Abramowitz, 2016). When clinicians notice that children are inconsistent at predicting the distress caused by exposures, but often experience substantial mismatches, they may note this as one potential marker that these exercises are promoting inhibitory learning and thus leading to improved treatment outcomes. Clinicians may consider pointing out expectancy violations to their patients to ensure that they notice these mismatches. If exposure exercises are consistently as challenging as expected, clinicians may seek out new contexts in which to conduct exposure to more successfully promote more expectancy violations, such as doing exposure therapy outside the therapy office.

A clinician should not rely on exposure being easier than children anticipate, however, as this study suggests that children are likely to be accurate affective forecasters. Underpredictions are more likely to occur at the beginning of treatment, and thus counting on exposure therapy being easier than expected as a way to achieve buy-in is unlikely to be a successful approach.

The prevalence of accurate expectations of distress may suggest that clinicians should seek out other ways to violate expectations during exposure (e.g., specific feared outcomes, in their ability to cope with exposure exercises). Alternatively, the frequency of accurate affective forecasts may also indicate that it could be clinically wise to frame exposure therapy as a chance to practice willingness and acceptance to confront anxiety-provoking situations that have interfered with their lives in the past (e.g., doing homework a “not just right” way so they have time to spend with their family or pursuing their interests), as is articulated in an acceptance and commitment therapy approach to exposure therapy (Twohig et al., 2015). Indeed, it may be that children with more variable accuracy in their predictions are more accepting of different emotional experiences, another mechanism that may have driven more improvement in this group.

Limitations and Future Directions

The present study only assessed one kind of expectation, the distress exposure would evoke. Although the rationale for including this measure is grounded in a large body of literature, suggesting anticipation of emotional states is a robust predictor of behavior in clinical populations (Barlow et al., 2014; Hayes, 2016), other forms of expectancy violations occur, and investigators should continue to assess other “surprises” that happen during exposure therapy (e.g., about feared outcomes, their coping self-efficacy). Assessing fears so they can be disconfirmed in various ways can be challenging for a clinical researcher hoping to measure this phenomenon during therapy, especially because expectations may be more implicit than explicit. Future research interested in measuring expectancies during exposure may take an idiographic approach to capture the complexity of this construct.

A notable limitation of this study is that only the first exposure of each session was assessed. Although this allowed for the analysis of multiple expectations for each participant, obtaining ratings of expectations for each exposure would have led to more reliable results. Furthermore, this study was an observational analysis of children participating in therapy at a University training clinic that used multiple therapists with each child, a very particular environment in which to collect data that may not generalize to community settings. Future research should also use larger sample sizes with more diverse patient populations, both diagnostically and demographically.

This study evaluated expectations and predictions in a narrow time-frame, which has been argued to lead to weaker affective forecasting biases (Wilson & Gilbert, 2003). Future studies of distress predictions during exposure therapy may allow more time to elapse between assessments of expectations about exposure-related distress and the actual exposure tasks themselves (e.g., comparing distress rated on exposure hierarchies with actual reported distress at later sessions).

Conclusion

The inhibitory learning approach to exposure therapy has garnered substantial clinical and empirical interest in recent years (e.g., Jacoby & Abramowitz, 2016; Kircanski & Peris, 2015). Although a growing body of experimental evidence has supported the tenets of inhibitory learning in promoting enhanced extinction learning in people (e.g., Brown et al., 2017; Kircanski et al., 2012), very few studies have tested whether indicators of inhibitory learning promote treatment outcome during the course of CBT. Even fewer have attempted to study youth despite the high prevalence of anxiety and OCD during childhood. The present study was the first to demonstrate the utility of expectancy violations in promoting symptom reduction during behavior therapy. Future research should continue to investigate other expectancy violations during exposure therapy for various patient populations with larger samples.

Footnotes

Acknowledgements

The authors would like to thank Michael Marsiske, PhD, David M Janicke, PhD, and Carol A Mathews, MD, for their thoughtful feedback on an earlier draft of this work. They would also like to thank the research assistants of the Division of Medical Psychology for contributing immense time and effort toward data collection for this study.

Declaration of Conflicting Interests

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

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was funded by a University of Florida Department of Psychiatry Seed Grant.

ORCID iD

Andrew G. Guzick

Author Biographies

Andrew G. Guzick is a doctoral candidate in clinical psychology at the University of Florida (UF). He is interested in studying the nature and treatment of pediatric obsessive-compulsive disorder (OCD) and anxiety disorders.

Adam M. Reid is a clinical psychologist and an assistant in psychology at McLean Hospital/Harvard Medical School. His research interests are in understanding the mechanisms of cognitive-behavioral therapy (CBT) for individuals with OCD and related disorders, and in the dissemination and implementation of evidence-based treatments.

Amanda M. Balkhi is a psychologist and acting director of the Balkhi Foundation. Her interests include strategies to improve training and dissemination of empirically validated interventions for pediatric populations.

Gary R. Geffken is a clinical psychologist at the Geffken Group in Gainesville, Florida. Prior to beginning his private practice, he was on the faculty at the UF for 29 years, where he developed a specialty in treating and studying individuals with OCD and OCD-related disorders.

Joseph P. H. McNamara is a clinical psychologist, and is an associate professor and chief of the Division of Medical Psychology in the Department of Psychiatry at UF. He conducts research that focuses on improving treatment outcomes for patients with OCD and related disorders.

References

Ale

C. M.

McCarthy

D. M.

Rothschild

L. M.

Whiteside

S. P.

(2015). Components of cognitive behavioral therapy related to outcome in childhood anxiety disorders. Clinical Child and Family Psychology Review, 18, 240-251.

Balkhi

A. M.

Reid

A. M.

Guzick

A. G.

Geffken

G. R.

McNamara

J. P.

(2016). The progress cascading model: A scalable model for teaching and mentoring graduate trainees in exposure therapy. Journal of Obsessive-Compulsive and Related Disorders, 9, 36-42.

Barlow

D. H.

Sauer-Zavala

Carl

J. R.

Bullis

J. R.

Ellard

K. K.

(2014). The nature, diagnosis, and treatment of neuroticism: Back to the future. Clinical Psychological Science, 2, 344-365.

Brown

L. A.

LeBeau

R. T.

Chat

K. Y.

Craske

M. G.

(2017). Associative learning versus fear habituation as predictors of long-term extinction retention. Cognition and Emotion, 31, 687-698.

Craske

M. G.

Kircanski

Zelikowsky

Mystkowski

Chowdhury

Baker

(2008). Optimizing inhibitory learning during exposure therapy. Behaviour Research and Therapy, 46, 5-27.

Craske

M. G.

Rapee

R. M.

Barlow

D. H.

(1988). The significance of panic-expectancy for individual patterns of avoidance. Behavior Therapy, 19, 577-592.

Craske

M. G.

Treanor

Conway

C. C.

Zbozinek

Vervliet

(2014). Maximizing exposure therapy: An inhibitory learning approach. Behaviour Research and Therapy, 58, 10-23.

Deacon

Kemp

J. J.

Dixon

L. J.

J. T.

Farrell

N. R.

Zhang

A. R.

(2013). Maximizing the efficacy of interoceptive exposure by optimizing inhibitory learning: A randomized controlled trial. Behaviour Research and Therapy, 51, 588-596.

De Kleine

R. A.

Hendriks

Becker

E. S.

Broekman

T. G.

van Minnen

. (2017). Harm expectancy violation during exposure therapy for posttraumatic stress disorder. Journal of Anxiety Disorders, 49, 48-52.

10.

Geller

D. A.

Biederman

Faraone

Agranat

Cradock

Hagermoser

. . . Coffey

B. J.

(2001). Developmental aspects of obsessive compulsive disorder: Findings in children, adolescents, and adults. The Journal of Nervous and Mental Disease, 189, 471-477.

11.

Gilbert

D. T.

Pinel

E. C.

Wilson

T. D.

Blumberg

S. J.

Wheatley

T. P.

(1998). Immune neglect: A source of durability bias in affective forecasting. Journal of Personality and Social Psychology, 75, 617-638.

12.

Harvill

L. M.

(1991). Standard error of measurement. Educational Measurement: Issues and Practice, 10(2), 33-41.

13.

Hayes

S. C.

(2016). Acceptance and commitment therapy, relational frame theory, and the third wave of behavioral and cognitive therapies–republished article. Behavior Therapy, 47, 869-885.

14.

Jacoby

R. J.

Abramowitz

J. S.

(2016). Inhibitory learning approaches to exposure therapy: A critical review and translation to obsessive-compulsive disorder. Clinical Psychology Review, 49, 28-40.

15.

Kendall

P. C.

Robin

J. A.

Hedtke

K. A.

Suveg

Flannery-Schroeder

Gosch

(2005). Considering CBT with anxious youth? Think exposures. Cognitive and Behavioral Practice, 12, 136-148.

16.

Kircanski

Mortazavi

Castriotta

Baker

A. S.

Mystkowski

J. L.

Craske

M. G.

(2012). Challenges to the traditional exposure paradigm: Variability in exposure therapy for contamination fears. Journal of Behavior Therapy and Experimental Psychiatry, 43, 745-751.

17.

Kircanski

Peris

T. S.

(2015). Exposure and response prevention process predicts treatment outcome in youth with OCD. Journal of Abnormal Child Psychology, 43, 543-552.

18.

Krompinger

J. W.

Van Kirk

N. P.

Garner

L. E.

Potluri

S. I.

Elias

J. A.

(2018). Hope for the worst: Occasional reinforced extinction and expectancy violation in the treatment of OCD. Cognitive and Behavioral Practice. Advance online publication. doi:10.1016/j.cbpra.2017.12.002

19.

Little

R. J.

(1988). A test of missing completely at random for multivariate data with missing values. Journal of the American Statistical Association, 83, 1198-1202.

20.

Maas

C. J.

Hox

J. J.

(2005). Sufficient sample sizes for multilevel modeling. Methodology, 1(3), 86-92.

21.

Obsessive Compulsive Cognitions Working Group. (2005). Psychometric validation of the obsessive belief questionnaire and interpretation of intrusions inventory—Part 2: Factor analyses and testing of a brief version. Behaviour Research and Therapy, 43, 1527-1542.

22.

Öst

L. G.

Riise

E. N.

Wergeland

G. J.

Hansen

Kvale

(2016). Cognitive behavioral and pharmacological treatments of OCD in children: A systematic review and meta-analysis. Journal of Anxiety Disorders, 43, 58-69.

23.

Rachman

(1994). The overprediction of fear: A review. Behaviour Research and Therapy, 32, 683-690.

24.

Scahill

Riddle

M. A.

McSwiggin-Hardin

Ort

S. I.

King

R. A.

Goodman

W. K.

. . . Leckman

J. F.

(1997). Children’s Yale-Brown Obsessive Compulsive Scale: Reliability and validity. Journal of the American Academy of Child & Adolescent Psychiatry, 36, 844-852.

25.

Singer

J. D.

Willett

J. B.

(2003). Applied longitudinal data analysis: Modeling change and event occurrence. Oxford, UK: Oxford University Press.

26.

Snijders

Bosker

(1999). Multilevel analysis: An introduction to basic and applied multilevel analysis. London, England: SAGE.

27.

Storch

E. A.

Geffken

G. R.

Merlo

L. J.

Mann

Duke

Munson

. . . Goodman

W. K.

(2007). Family-based cognitive-behavioral therapy for pediatric obsessive-compulsive disorder: Comparison of intensive and weekly approaches. Journal of the American Academy of Child & Adolescent Psychiatry, 46, 469-478.

28.

Storch

E. A.

McGuire

J. F.

McKay

(2018). The clinician’s guide to cognitive-behavioral therapy for childhood obsessive-compulsive disorder. Cambridge, MA: Academic Press.

29.

Storch

E. A.

Murphy

T. K.

Adkins

J. W.

Lewin

A. B.

Geffken

G. R.

Johns

N. B.

. . . Goodman

W. K.

(2006). The Children’s Yale-Brown Obsessive Compulsive Scale: Psychometric properties of child-and parent-report formats. Journal of Anxiety Disorders, 20, 1055-1070.

30.

Storch

E. A.

Murphy

T. K.

Geffken

G. R.

Soto

Sajid

Allen

. . . Goodman

W. K.

(2004). Psychometric evaluation of the Children’s Yale-Brown Obsessive-Compulsive Scale. Psychiatry Research, 129, 91-98.

31.

Summerfeldt

L. J.

Kloosterman

P. H.

Antony

M. M.

Swinson

R. P.

(2014). Examining an obsessive-compulsive core dimensions model: Structural validity of harm avoidance and incompleteness. Journal of Obsessive-Compulsive and Related Disorders, 3(2), 83-94.

32.

Tasca

G. A.

Gallop

(2009). Multilevel modeling of longitudinal data for psychotherapy researchers: I. The basics. Psychotherapy Research, 19, 429-437.

33.

Twohig

M. P.

Abramowitz

J. S.

Bluett

E. J.

Fabricant

L. E.

Jacoby

R. J.

Morrison

K. L.

. . . Smith

B. M.

(2015). Exposure therapy for OCD from an acceptance and commitment therapy (ACT) framework. Journal of Obsessive-Compulsive and Related Disorders, 6, 167-173.

34.

van Hout

W. J.

Emmelkamp

P. M

. (1994). Overprediction of fear in panic disorder patients with agoraphobia: Does the (mis) match model generalize to exposure in vivo therapy? Behaviour Research and Therapy, 32, 723-734.

35.

Wenze

S. J.

Gunthert

K. C.

German

R. E.

(2012). Biases in affective forecasting and recall in individuals with depression and anxiety symptoms. Personality and Social Psychology Bulletin, 38, 895-906.

36.

Wilson

T. D.

Gilbert

D. T.

(2003). Affective forecasting. Advances in Experimental Social Psychology, 35, 345-411.

37.

Wilson

T. D.

Gilbert

D. T.

(2013). The impact bias is alive and well. Journal of Personality and Social Psychology, 105, 740-748.

38.

Wolpe

Lazarus

A. A.

(1966). Behavior therapy techniques: A guide to the treatment of neuroses. Elmsford, NY: Pergamon Press.

39.

Woods

A. M.

Bouton

M. E.

(2007). Occasional reinforced responses during extinction can slow the rate of reacquisition of an operant response. Learning and Motivation, 38, 56-74.