Breaking the vicious circle of fear and avoidance in children with abdominal pain: A mediation analysis

Objectives: Exposure-based cognitive behavioral therapy via internet (Internet-CBT) has been shown to reduce symptoms and increase quality of life for children with functional abdominal pain disorders (FAPDs), but the mechanisms of change are unknown. The objective was to examine whether a change in symptom-specific fear and avoidance, i.e., gastrointestinal-specific anxiety (GI-anxiety) and gastrointestinal-specific avoidance (GI-avoidance), mediated changes in parent-reported abdominal symptoms for children receiving Internet-CBT compared with children receiving treatment as usual. A further aim was to assess if baseline levels of the proposed mediators moderated the mediation. Methods: Weekly assessments of child-reported mediators and parent-reported outcome from 90 children aged 8 – 12 who were included in a randomized controlled trial were used in univariate and multivariate growth models to test the direct effect of treatment on outcome and the indirect effects via mediators and moderated mediation. Results: Treatment condition significantly predicted the slope of the mediators ( a -path), in favor of Internet-CBT, and mediators were correlated with the outcome ( b -path). The indirect effects of the mediators on the outcome (cross-product of the a and b -paths) were significantly different from zero for both GI-avoidance, ab = 1.43, 95% CI [0.42, 3.23]; and GI-anxiety ab = 1.58, 95%CI [0.43, 3.62]. Baseline levels of the proposed mediators moderated the size of the mediation. Conclusions: GI-anxiety and GI-avoidance were mediators of change in Internet-CBT and high levels of the me- diators at baseline were associated with larger mediated effects. Healthcare professionals should be aware of, and inform families about, the potential benefits of reducing symptom-specific fear and avoidance.


Introduction
Pediatric functional abdominal pain disorders (FAPDs) are characterized by medically unexplained abdominal pain and other abdominal symptoms [1]. FAPDs are associated with internalizing symptoms, e.g., anxiety [2] and low quality of life [3]. The disorders are prevalent (13.5%) [4] and for about 40% of the children the symptoms remain into adulthood [5]. This chronicity may partly be explained by the fear and avoidance model, in which an individual who perceives pain as a threat reacts with avoidance [6,7]. Avoidance of acute pain has an evolutionary advantage as damage to the individual is minimized, but when the pain becomes chronic, avoidance maintains functional disability and symptom severity [6].
While the empirical support for pharmacological and dietary interventions is weak for pediatric FAPDs [8,9], studies of cognitive behavioral therapy (CBT) have shown promising results [10][11][12]. Multiple components are typically used in CBT, such as cognitive restructuring of maladaptive thoughts, exposure exercises, relaxation, and parent management techniques. However, it is largely unknown which treatment components are effective, through which mechanisms they work, and for whom CBT is effective.
Our research group have shown that exposure-based CBT is effective in reducing abdominal symptoms in adults, adolescents and children with FAPDs [11][12][13]. In exposure-based CBT for FAPDs, the patients gradually expose themselves to symptom-provoking stimuli (such as eating pizza) and approach situations in which symptoms are perceived as intolerable (such as being in school). This approach is hypothesized to decrease fear and avoidance related to symptoms and thereby enables symptom reduction. Exposure-based CBT addresses the patient's maladaptive behavioral, cognitive and emotional reactions to the symptoms, as has been suggested to be the most important target in psychological treatments for FAPDs [14]. Exposure exercises represent an activating and patient-involving treatment, as has been recommended in a recent review of the management of functional somatic syndromes [15].
An important aim of mediation analyses is to uncover mechanisms of change and to use such knowledge to maximize treatment outcomes. Mediation analyses use statistical methods to assess whether all or some of the treatment effect works indirectly via a mediator, referred to as the mediated or indirect effect [16]. When a treatment is compared with a control, mediation can be confirmed if (1) the treatment changes the mediator and if (2) the change in the mediator is correlated with the outcome.
A moderator is a variable that affects the strength of the association between two variables. Moderation can be combined with mediation to test for whom, or under what conditions, the mediational processes operate. Moderated mediation can thus reveal for which patients the process of change is most pronounced, and who therefore might be particularly suitable for an intervention that target that specific mediator [17]. Also, moderated mediation provides a stringent test of treatment theory. That is, given that a treatment is most likely to benefit those who are functioning poorly on the mediator at baseline, positive findings from moderated mediation provide further evidence for the theoretical underpinnings of the treatment model [18].
In CBT for pediatric FAPDs, avoidance [19], maladaptive cognitions [20], pain impairment beliefs, and pain reactivity [21] have been shown to mediate symptom improvement. However, only one of these studies [19] used a design where both potential mediators and outcomes were assessed repeatedly during treatment, which is recommended to reduce the risk of biased results [22].
The present study used data collected in a randomized controlled trial where exposure-based CBT via internet (Internet-CBT) was found superior to treatment as usual, with sustained improvements at a 36week follow-up evaluation [12]. The aim of the present study was to investigate whether gastrointestinal-specific anxiety (GI-anxiety) and gastrointestinal-specific avoidance (GI-avoidance) mediated the symptom improvement for children with FAPDs receiving Internet-CBT compared with children receiving treatment as usual. A further aim was to assess if high levels of the mediators at baseline moderated the mediation. We hypothesized that GI-anxiety and GI-avoidance would be mediators of change and that children with high baseline levels of the proposed mediators would have a larger mediated effect. This is, to the best of our knowledge, the first study to assess whether a reduction in GIspecific fear and avoidance mediate symptom improvement in pediatric FAPDs.

Design
This study was based on data from a randomized controlled trial [12] registered in Clinicaltrials.gov in August 2016 (NCT02873078) and approved by the Regional Ethical Board in Stockholm, Sweden in August 2016 (2016/1289-31).

Recruitment and inclusion criteria
Physicians, who had been informed about the study via emails, visits to clinics and lectures, referred all participants to the study between September 2016 and April 2017. All referred children were regular patients at the physicians' clinics. The study was conducted at the Child and Adolescent Psychiatry Research Centre in Stockholm. Inclusion and exclusion criteria were assessed and informed consent was obtained from parents and children during the initial clinical assessment at the research clinic. Ninety children and their parents were randomized to either Internet-CBT or treatment as usual. Inclusion criteria were: (a) age ≥ 8 and ≤ 12 years, (b) at least one of following functional abdominal pain disorders according to the ROME IV criteria: irritable bowel syndrome, functional dyspepsia, or functional abdominal painnot otherwise specified [1], (c) stable dose at least one month if using psychopharmacological medications, (d) internet access, and (e) basic reading and writing skills (child and one parent). Children were excluded if they had (f) another somatic disease that explained their abdominal symptoms, (g) psychiatric or social problems that needed immediate care, including school absenteeism >40%, or (h) another ongoing psychological treatment.

Internet-CBT
Internet-CBT consisted of 10 weekly modules for children and 10 weekly modules for parents. The children and parents completed a mean of 9.3 and 9.2 out of 10 weekly modules, respectively. Therapist support was provided via written text messages within the online treatment platform. Short texts, images and films used were used to illustrate principles of the treatment. The treatment included an explanatory model of how abdominal pain can be maintained by GI-anxiety and GIavoidance (Fig. 1).
The exposure-based approach was presented as a means to break the vicious circle of fear and avoidance, and the families were taught that exposure exercises could help the children take control of their lives. In the children's modules, children mapped their behaviors related to GIavoidance, such as avoiding foods or situations in which they feared having symptoms. The children and parents made a hierarchy of exposure exercises based on this mapping and set their own behavioral goals for the treatment. The children themselves chose from the hierarchy which exposure exercises to engage in during the coming week. Examples of exposure exercises were to eat symptom-provoking food or to play with friends when having abdominal symptoms. An example of a goal was to go to school every day despite abdominal symptoms. A short mindfulness exercise was used to help children engage in the challenging exposure exercises without distracting themselves from the abdominal symptoms. Children in Internet-CBT were allowed to participate in other treatments, but were encouraged to decrease abdominal medications, in order to regain natural control of bowel habits and increase exposure to symptoms.
The main focus of the parents' modules was to encourage and facilitate their child's work with the exposure exercises. Another important aspect was to help the parents decrease their attention to their child's abdominal symptoms, which is in accordance with the exposurebased approach. Parents were taught to first briefly validate the child's pain experience and then help their child shift focus. Parents were also instructed to plan for joyful activities with their child without focusing on abdominal symptoms. A comprehensive description of the Internet-CBT protocol is available elsewhere [23].

Treatment as usual
In treatment as usual, patients and parents were informed that they would be offered Internet-CBT after 10 weeks. During this time, they were allowed to continue with any current treatment and seek any further help that they deemed necessary. Treatments included visits to health care professionals, medications, and dietary interventions, reported in detail in the article of the randomized controlled trail [12].

Outcome variable
The outcome variable was the PedsQL Gastrointestinal symptoms scale (PedQL). The PedQL consists of 9 items assessing abdominal symptoms (abdominal pain, diarrhea, constipation, nausea, vomiting, abdominal discomfort, passing gas, not feeling hungry, and bloating) that are rated on a 5-point scale ranging from never (0) to almost always (4). It is specifically developed to fit children with FAPDs and has an acceptable internal consistency (Cronbach's alpha 0.77) [24]. The PedQL is transformed to a reversely scored 0-100 scale, with higher values indicating milder symptoms. In this study, we used parental assessments of the outcome variable. Parents' reports of the PedQL were in line with the children's reports, but showed larger effects and were therefore chosen to maximize power [12]. However, children's assessments of the PedQL were also analyzed as an outcome variable. These results are presented in Appendix C.

Proposed mediators
GI-avoidance (BRQ-C) was assessed with a child-adapted and shortened version of the Irritable Bowel Syndrome Behavioral Responses Questionnaire [25]. The BRQ-C comprises 11 items and the total score range from 11 to 77. It includes items like "I avoid exercise when I have stomach pains" and "I avoid certain foods when I have bowel problems". GI-anxiety (VSI-C) was assessed with a child-adapted and shortened version of the Visceral Sensitivity Index [26]. The VSI-C comprises 7 items and the total score ranges from 0 to 35 with items like "I often worry about problems in my belly" and "When I feel discomfort in my belly, it frightens me". Both the proposed mediators were child-rated. The adaptations of the original scales were made to adjust the scales to a child-population with different kinds of FAPDs, as they were originally developed for adults with irritable bowel syndrome. Further, there was a need to shorten the scales to enable repeated measurements without overloading the children with questions. The adaptations and validations of the BRQ-C and the VSI-C are described in detail in a separate article (in manuscript) and a brief description is provided in the Appendix A.

Time points of assessment
Abdominal symptoms were assessed weekly by the parents (and the children) during the treatment (week 1-10). At baseline and follow-up (weeks 0 and 11), another version of the PedQL that has a one-month recall period instead of one week was used. Because the mediation analysis was based on weekly assessments, the 0 and 11 week PedQL assessments were not included in the analysis. GI-anxiety and GIavoidance were assessed bi-weekly by the children (with a one-week recall period) using the planned missing design wave missing [27]. This design was chosen to decrease the workload for the patients. Half the patients were randomized to assess GI-avoidance on weeks 1, 3, 5, 7, and 9 and GI-anxiety on weeks 2, 4, 6, 8, and 10, and the other half were randomized assess GI-avoidance on weeks 2, 4, 6, 8, and 10 and GIanxiety on weeks 1, 3, 5, 7, and 9. Both GI-anxiety and GI-avoidance were also assessed at baseline and one week after treatment completion (weeks 0 and 11). After treatment, the participants in the treatment as usual-group were crossed over to Internet-CBT. Therefore it was not possible to compare the mediated effects after treatment completion.

Statistical analysis
The primary analytic models used to test direct and indirect effects (mediation, moderated mediation) were univariate and multivariate growth models with random effects [28,29]. The mediators of interest were child-assessed GI-anxiety and GI-avoidance and the outcome was parent-assessed abdominal symptoms. We also performed the analyses with child-assessed abdominal symptoms as the outcome, reported in Appendix C. Models were fitted using Mplus Version 7.4 and incorporated all available data [30]. Effect sizes in the form of standardized mean differences (Cohen's d) were computed based on estimated means at the endpoint [31]. Univariate growth models were combined into a multivariate parallel process growth model to test for mediation following the recommendations provided by Cheong, MacKinnon and Khoo [32]. The parallel process growth model was subsequently extended to moderated mediation by including the baseline scores of the mediator variable as a moderator of the effect of treatment on the mediator [33]. Mediation was formally evaluated by a bootstrapped 95% confidence interval for the mediated effect (i.e., ab-product) and the moderated mediated effect [34]. The formula ab/(ab+c') was used to estimate the proportion mediated effect [35]. The statistical analyses are described in detail in Appendix B.

Results
There were 90 children (69% girls) with an average age of 10.2 (SD = 1.4) years included in the study. For each child one parent was included, of which 86% were mothers. The proportion missing data in the entire sample was 6.5%. This did not include the planned missing data of the proposed mediators, which according to the design was 50% on weeks 1-10 leaving a maximum of 45 participants each week instead of 90. Table 1 presents descriptive statistics for the outcome measure and the proposed mediators for both conditions. Univariate growth models revealed statistically significant differences in average trajectories on the primary outcome parent ratings of abdominal symptoms (PedQL) favoring Internet-CBT relative to treatment as usual: estimate = 1.71, SE = 0.83, P = 0.04, Cohen's d = 0.65. There was also a statistically significant difference between conditions on the proposed mediators GI-avoidance (BRQ-C) and GI-anxiety (VSI-C) favoring Internet-CBT over treatment as usual: estimate = − 2.28, SE = 0.67, P = 0.001, Cohen's d = 0.48; estimate = − 1.54, SE = 0.50, P = 0.002, Cohen's d = 0.35, respectively. In all univariate growth models, there was a significant unexplained heterogeneity in individual growth trajectories after adjusting for treatment effects (all P's < 0.01). This made it possible to evaluate mediation using parallel process growth models.

Mediation analysis
The main results from parallel the process growth models are shown in Fig. 2. Condition significantly predicted the slope of the mediator in favor of Internet-CBT (a-path) and the slope of the mediator was significantly correlated with the slope of the outcome (b-path) for GIavoidance (BRQ-C) and GI-anxiety (VSI-C). The ab-product was statistically significant (i.e., did not include zero) as evaluated with the bootstrapped confidence interval method both when testing GIavoidance (BRQ-C) ab = 1.43, 95% CI [0.42, 3.23] and GI-anxiety (VSI-C) ab = 1.58, 95% CI [0.43, 3.62] as mediators. The proportion mediated effect for the mediators was 79.8% for BRQ-C and 78.6% for VSI-C. Mediation analyses based on the child-rated outcome showed similar results, see Appendix C.

Moderated mediation
The estimate of moderated mediation was statistically significant for GI-avoidance (BRQ-C) mod-ab = 0. 16 [33]. As shown in Figs. 3 and 4, the mediated effects of GI-avoidance (BRQ-C) and GI-anxiety (VSI-C) on abdominal symptoms (PedQL) increased with higher baseline scores on the mediators. For children scoring low on the mediators at baseline (1 SD below the grand mean), the mediated effect was small and not statistically significant different from zero, whereas as for those with a high baseline scores on the mediators (1 SD above the grand mean) the mediated effect was statistically significant and large (Figs. 3 and 4). Moderated mediation analyses based on the child-rated outcome showed similar results, see Appendix C. PedQL, PedsQL Gastrointestinal Symptom Scale measuring abdominal symptoms; BRQ-C, Behavioral Responses Questionnaire Child-adapted version, measuring gastrointestinal-specific avoidance behaviors; VSI-C, Visceral Sensitivity Index Child-adapted version, measuring gastrointestinal-specific anxiety; n, Number of observations; M, Mean; SD, Standard deviation; Total obs, total number of observations; Missing obs, number of unplanned missing observations.

Discussion
The aim of this study was to assess if a reduction in symptom-specific fear and avoidance would mediate symptom change for children with FAPDs receiving exposure-based Internet-CBT compared with treatment as usual. We used data from a randomized controlled trial including 90 children aged 8-12 with FAPDs. In line with our hypotheses, we observed that a decrease in children's GI-anxiety and GI-avoidance mediated the improvement in parent-reported abdominal symptoms for children receiving Internet-CBT compared with children receiving treatment as usual. Further analyses showed that for children in Internet-CBT, higher baseline scores on the proposed mediators were associated with stronger mediated effects. The results of this study can be interpreted in the light of the fear and avoidance model [35]. According to this model fear of symptoms and avoidance of activities that may elicit symptoms will increase symptom severity and disability, while approaching symptoms and feared situations leads to decreased symptoms. Approaching feared situations and provoking symptoms is the key element of exposure-based CBT and the results of this mediation analysis corroborate the hypothesized mechanism of change: by breaking the vicious circle of fear and avoidance the symptoms decreases and function and quality of life increases.
Previous studies have shown that reductions in GI-anxiety and GIavoidance mediate symptom improvement in adults and that GIavoidance mediate symptom improvement in adolescents during exposure-based cognitive behavioral therapy for irritable bowel syndrome [18,19]. The adult study also demonstrated that the mediated effect of GI-avoidance was more pronounced for patients with high values of GI-avoidance at baseline [18]. This study confirms these results and expands them to young children with a broader range of FAPDs. The moderated mediation shows that the indirect effect of the mediators on outcome seen in Internet-CBT is more pronounced for children with higher baseline values of the proposed mediators, which gives further support to GI-anxiety and GI-avoidance as key mediators in Internet-CBT for children with FAPDs. The moderated mediation also indicates that Internet-CBT may be particularly beneficial for children high in symptom-specific fear and avoidance.
Future mediation studies should assess if parental responses to children's symptoms affect the children's symptoms, avoidance behaviors, and fear of symptoms. Further, it would be interesting if mediators  that have been assessed in other studies of pediatric FAPDs, such as pain reactivity and coping strategies [20,21], were compared to the mediators found in this study. Another recommendation for future studies is to include experimental manipulation of potential mediators, to be able to meticulously investigate mechanisms of change.

Strengths and limitations
Strengths of the study include the randomized study design as well as the use of multiple assessment points during the course of treatment for both proposed mediators and outcome. The mediation analyses were performed using parallel process growth models that assess changes in both mediators and outcome at the individual level, which is generally a preferred approach for testing longitudinal mediation [32]. Another strength of the study is the use of the two theoretically relevant and distinct mediators: GI-anxiety and GI-avoidance. A limitation of the study is that temporal order was not established between mediators and outcome. Other limitations are the fact that potential confounding in the b-path (affecting both mediator and outcome) is not handled by the randomization and that the sample size was in the lower acceptable range for the models. The use of the parent-reported measure of abdominal symptoms as outcome in the mediation analysis instead of the child-reported measure is a limitation. The decision to use the parent-reported outcome was made to maximize the statistical power of the mediation analysis. The effect size of the parent report of their child's abdominal symptoms was larger than for the child-reported abdominal symptoms, as described in the main outcome study [12]. Thus, using the parent-rated measure as the outcome reduced the risk for making a Type II-error. However, we also performed the mediation and moderated mediation analyses based on the child-reported measure and report these in Appendix C. The results from these analyses were in line with the primary analysis. Another limitation is that the treatment as usual comparison group did not control for attention or expectation during Internet-CBT. Also, it is important to note that the high percentages of the effect explained by the mediators (79.8% for BRQ-C and 78.6% for VSI-C) should be interpreted with caution because of the small sample size as the effect is regarded as unstable in sample sizes below 500 [35].

Conclusion
This is the first mediation study of exposure-based CBT in FAPDs to clearly demonstrate both mediation and moderated mediation of symptom-specific fear and avoidance. The results show that GI-anxiety and GI-avoidance are mediators of change for children with FAPDs in exposure-based Internet-CBT compared with treatment as usual. Children with high baseline values of GI-anxiety and GI-avoidance likely benefit more from exposure-based Internet-CBT than children with low symptom-specific fear and avoidance.

Financial support
This study was supported by grants from the Jane and Dan Olsson Foundation None of the funding bodies had any influence on study design, execution or publication of the study.

Ethical standards
The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008.

Declaration of Competing Interest
All authors have completed the ICMJE uniform disclosure form at www.icmje.org/coi_disclosure.pdf and declare that they have no competing interests.