Predictors for Persistence of Functional Somatic Symptoms in Adolescents Karin A. M. Janssens, PhD, Sandor Klis, MSc, Eva M. Kingma, MD, PhD, Albertine J. Oldehinkel, PhD, and Judith G. M. Rosmalen, PhD Objective To identify risk factors for persistence of functional somatic symptoms (FSS; ie, somatic symptoms that cannot be sufficiently explained by underlying organic pathology).
Study design The first (N = 2230, mean age = 11.1 years [SD 0.6], 50.8% girls), second (N = 2149, mean age = 13.7 years [SD 0.5], 51.0% girls), and third (N = 1816, mean age = 16.3 years [SD 0.7], 52.3% girls) assessment waves of the general population study TRacking Adolescents’ Individual Lives Survey were used. FSS were assessed with the Youth Self-Report and the Child Behavior Checklist. Growth mixture models were used to identify different subgroups of adolescents on the basis of the developmental trajectory of their symptoms. Adolescents with persistent symptoms were compared with adolescents with decreasing symptoms with a multivariable logistic regression analysis. Results In our general population cohort, 4.1% of adolescents suffered from persistent FSS. Risk factors for persistent FSS were being a girl (OR 4.69, 95% CI 2.17-10.12), suffering from depressive symptoms (OR 5.35, 95% CI 1.46-16.62), poor self-rated health (OR 1.56, 95% CI 1.02-2.39), and high parent-reported FSS (OR 4.03, 95% CI 1.20-13.54). Anxiety, parental overprotection, school absenteeism, and diversity of symptoms did not predict persistence of FSS. Conclusions This study identified risk factors for persistence of FSS in adolescents. Future studies might study effects of coping strategies and iatrogenic factors on symptom persistence. (J Pediatr 2014;164:900-5).
F
unctional somatic symptoms (FSS) are somatic symptoms that cannot sufficiently be explained by underlying organic pathology.1 FSS, such as pain and fatigue, are common among adolescents. FSS are known to cause substantial impairment in a subgroup of children and adolescents by resulting in school absenteeism and social problems.2,3 This impairment is especially true for adolescents suffering from persistent symptoms.4 To prevent symptoms from becoming persistent, early intervention is important. However, although FSS are persistent in a subgroup of adolescents,5 FSS often are self-limiting. Thus, it is probably not necessary to (extensively) intervene in all adolescents with FSS. Therefore, insight in whether an adolescent is at risk for persistence of FSS is important. In previous studies, investigators examined risk factors for persistence of either pain or fatigue in adolescents.6-9 It is probably more clinically relevant to examine symptom persistence for a full range of FSS because adolescents who have multiple symptoms are especially at risk for long-lasting psychological and social problems.1 Moreover, symptoms experienced by adolescents might change over time, and their recovery from specific symptoms does not automatically mean that they are symptom-free. Another problem with previous studies is that all, except for one,6 used cut-off scores to determine whether an adolescent had a good or a poor symptom prognosis. Classifying adolescents by use of developmental trajectories might be a more reliable method, with more realistic subdivisions, because complex trajectories like exponential growth trajectories can be taken into account. This study differs from previous studies in combining 3 important aspects: (1) it included several types of FSS; (2) it identified risk factors based on developmental trajectories; and (3) it was performed in the general population. We chose to study risk factors for persistence of FSS that can be easily assessed by clinicians or have previously been found to predict poor prognosis of pain or fatigue, or to perpetuate FSS. The risk factors we hypothesized to be risk factors for persistent FSS were being a girl8,10 and having a high number of different symptoms, poor self-rated health, anxiety,7,11 depression,6,7,9,11 school absenteeism,7,12 parental overprotection,13 and high parent-reported FSS. Our hypotheses were tested in 2210 adolescents of a prospective population-based cohort study.
Method The TRacking Adolescents’ Individual Lives Survey (TRAILS) is a prospective cohort study of Dutch adolescents, approved by the Dutch Central Committee
BIC CBCL FSS LMR-LRT TRAILS YSR
Bayesian Information Criterion Child Behavior Checklist Functional somatic symptoms Lo-Mendell-Rubin likelihood ratio test TRacking Adolescents’ Individual Lives Survey Youth Self-Report
From the University of Groningen, University Medical Center Groningen, Interdisciplinary Center Psychopathology and Emotion Regulation, Groningen, The Netherlands This research is part of TRAILS, which has been financially supported by various grants from the Netherlands Organization for Scientific Research, Zorgonderzoek Nederland Medische Wetenschappen, Gebied Maatschappij-en Gedragswetenschappen, the Dutch Ministry of Justice, the European Science Foundation, Biobanking and Biomolecular Research Infrastructure, the participating universities, and Accare Center for Child and Adolescent Psychiatry. The authors declare no conflicts of interest. 0022-3476/$ - see front matter. Copyright ª 2014 Mosby Inc. All rights reserved. http://dx.doi.org/10.1016/j.jpeds.2013.12.003
900
Vol. 164, No. 4 April 2014 on Research Involving Human Subjects. Participating centers of TRAILS include various departments of the University Medical Center and University of Groningen, the Erasmus University Medical Center Rotterdam, the University of Utrecht, the Radboud Medical Center Nijmegen, and the Parnassia Bavo group, all in The Netherlands. The current study used data from the first 3 assessment waves, wave 1 in 2001-2002 (N = 2230, mean age = 11.09 years [SD 0.56], 50.8% girls); wave 2 in 2003-2004 (N = 2149, mean age = 13.65 years [SD 0.53], 51.0% girls), and wave 3 in 20052007 (N = 1816, mean age = 16.27 years, [SD = 0.73], 52.3% girls). Informed consent was obtained from both parents and subjects. Detailed information about sample selection and analysis of non-response bias has been reported elsewhere.14,15 To assess FSS, adolescents filled out the Somatic Complaints scale of the Youth Self-Report (YSR16) at each assessment wave. This scale contains 9 items, which refer to somatic complaints without a known medical cause (aches/ pains, headaches, nausea, eye problems, skin problems, stomach pain, and vomiting) or without obvious reason (overtiredness and dizziness). Factor analyses (Table I; available at www.jpeds.com) indicated that 2 items (eye problems and skin problems) had low factor loadings, suggesting that these items did not represent the underlying construct well in the TRAILS sample; therefore, these items were excluded. The remaining 7 items showed good internal consistency (Cronbach a at T1: 0.76; at T2: 0.77; at T3: 0.76). Each somatic complaint was rated on a 3-point scale with 0 = never or not at all true, 1 = sometimes or a bit true, and 2 = often or very true. A mean item score was computed by adding the scores of the 7 FSS items and dividing the sum score by 7, which resulted in a scale score that could range from 0 (all items rated as “never or not at all true”) to 2 (all items rated as “often or very true”).
scale of the YSR.16 One item (overtiredness) was excluded from this scale to prevent overlap with the Somatic Complaints scale. The internal consistency was adequate (Cronbach a at T1: 0.69; at T2: 0.74; at T3: 0.75). Again mean scale scores were used, which could range from 0 to 2.
Parental Overprotection. Parental overprotection at baseline was measured by use of the overprotection subscale of the Egna Minnen Betr€affande Uppfostran, Child Version (Swedish for “my memories of upbringing”17), which contains 12 items referring to children’s perception of parental overprotection, which can be rated on a 5-point scale ranging from 0 = never to 4 = always (Cronbach a = 0.84). Adolescents filled out this questionnaire for both their mothers and fathers. In line with our previous research, in which girls were found to be especially vulnerable for maternal and boys for paternal overprotection, an overall overprotection score was computed that used maternal overprotection scores for girls and paternal scores for boys.13 Our previous study showed that the effect of overprotection on FSS was equally strong for boys and girls. The low number of participants suffering from persistent FSS did not allow subgroup analyses for boys and girls in the current study. The mean item score was calculated, which could range from 0 to 4.
School Absenteeism. Parents answered the following question about school absenteeism: “How often has your child been absent from school during the past six months because of illness?” Answer categories were: “Never,” “Seldom,” “Sometimes,” “Often,” and “Mostly.” Because of low cell count, the last 2 categories were combined. Self-Rated Health. To assess self-rated health, adolescents answered the question “How did you perceive your health during the past year?” with: “Very good (1),” “Good (2),” “Fair (3),” “Moderate (4),” and “Bad (5).”
Potential Risk Factors
Symptom Diversity. The 7 somatic symptoms of the YSR also were used to construct a baseline symptom diversity score, which takes the diversity of symptoms into account without paying attention to the symptom severity. This symptom diversity score could range from 1 to 7 symptoms. A symptom was considered present if it was rated as “sometimes or a bit true” or “often or very true”.
Parent-Reported FSS. Parents completed the Child Behavior Checklist (CBCL), which contains the same items and response categories as the YSR. A mean item score was computed for the aforementioned 7 somatic symptoms (Cronbach a at T1: 0.71; at T2: 0.72; at T3: 0.75). Of all parents, 88% (N = 2017) completed the CBCL at T1, 82% (N = 1883) at T2, and 66% (N = 1509) at T3.
Anxiety and Depression. Symptoms of anxiety were measured by the 6 items of the Anxiety scale of the YSR (Cronbach a at T1: 0.63; at T2: 0.63; at T3: 0.65).16 Depression was measured by the 13 items of the Affective Problems
Statistical Analyses Growth mixture modeling was used to identify distinct developmental trajectories of FSS. Growth mixture modeling was conducted in Mplus, version 5.2 (Muthen & Muthen, Los Angeles, California).18 Trajectories were determined by latent growth factors, which model the intercepts and slopes (linear and quadratic) of the individual growth trajectories. First, single-class latent growth models were estimated to determine whether linear or quadratic growth curves fitted the data best. This was based on the smallest Bayesian Information Criterion (BIC),19 as well as the significance of the quadratic slope growth factor. Second, models (linear or quadratic) with increasing numbers of classes were fitted. The data were rearranged as a function of chronological age instead of clustered by wave of data collection, resulting in 8 (age 10 years until age 17 years) instead of 3 assessment points, which enabled modeling more complex pathways. Trajectories were estimated on the basis of full information maximum likelihood with robust standard errors, which is robust regarding non-normality of the scores, and adjusts 901
THE JOURNAL OF PEDIATRICS
www.jpeds.com
for uncertainty associated with missing data.20 Rules to determine the number of trajectory classes were formulated a priori, and were in increasing order of importance: (1) a significant Lo-Mendell-Rubin likelihood ratio test (LMRLRT),21 which indicates that a model with k classes is preferred over k-1 classes; (2) a BIC lower than the BIC of the model with k-1 classes; and (3) an entropy greater than the entropy of k-1 classes. The latter is an indicator of how good the model can subdivide participants into different classes because it summarizes the posterior probabilities (ie, the probability to be a member of a specific class) for all classes. The entropy can range from no distinction between classes (0) to perfect distinction between classes (1).22 Please note that conventional fit indices like the root mean square error of approximation are not available for growth mixture models.23 Estimating class specific variances for the intercept, linear slope, and quadratic slope resulted in convergence problems, which is often an indication of overfitting (ie, the fitted model is overly complex). Therefore, random slope variances were fixed to zero. All adolescents were allocated to one of the FSS trajectories on the basis of their greatest posterior probability score. Adolescents who were not adequately fitting one of the classes (probability score #0.50) were excluded from subsequent analyses. To compare CBCL-based developmental patterns with YSR-based developmental patterns, the parental CBCL data were plotted for the different classes. Again, data were rearranged as a function of chronological age instead of clustered by waves to ease comparing both plots. Four groups were identified (see results in Figure 1). To indicate risk factors for persistence of FSS, adolescents who had high FSS at baseline but for whom symptoms decreased (dummy coded 0) were compared with adolescents with high FSS at baseline who had persistent FSS (dummy coded 1). Adolescents of both other groups had low baseline FSS scores and were therefore excluded from this analysis. A multivariable logistic regression analysis was performed to examine whether sex, symptom
Vol. 164, No. 4 diversity, self-rated health, anxiety, depression, school absenteeism, parental overprotection, and parent-reported FSS, all at baseline, were risk factors for poor prognosis of FSS.
Results The mean item score of FSS decreased during subsequent waves and was 0.47 (SD 0.35) at T1 (n = 2115), 0.39 (SD 0.35) at T2 (n = 2015), and 0.34 (SD 0.34) at T3 (n = 1636). A total of 2210 adolescents had valid data on FSS on at least one assessment wave (Figure 2; available at www. jpeds.com). Developmental pathways were examined for these adolescents. Figure 2 shows a flowchart with number of adolescents included in subsequent analyses. Group Identification A one-class model that included a quadratic term was significantly worse than a one-class model that consisted of a linear relationship only. Moreover, the quadratic term was not statistically significant. Therefore, the different class solutions were fitted for models that included only linear and not quadratic relationships. The BICs, LMR-LRTs, and entropies for these different class-solutions are given in Table II. The model with the greatest number of classes that still had a significant LMR-LRT was the 4-class solution. Moreover, the BIC of this 4-class solution was lower than that of the 3-class solution. On the basis of these fit indices, 4 classes were considered to fit the data best. The course of FSS during adolescence in these 4 classes is given in Figure 1. The first class, 4.1% (n = 90) of the adolescents, had persistent FSS. The second class, 10.5% (n = 233) of the adolescents, started high on FSS, and these symptoms decreased when adolescence proceeded. The third class, 22.0% (n = 487) of all adolescents, had moderate symptoms at 10 years of age, and these symptoms slightly increased over the years. The fourth class, that constituted the majority, ie, 63.3% (n = 1400), had a low symptom
Figure 1. Overview of the developmental patterns of FSS in the 4 classes that have been identified with growth mixture modeling (n = 2210). 902
Janssens et al
ORIGINAL ARTICLES
April 2014
Table II. Fit criteria of different fitted models Number of clusters (k)
BIC
LMR-LRT
Entropy
1 2 3 4 5
3471.2 3256.1 3159.8 3051.7 3027.5
238.3 (P = .01) 119.3 (P < .001) 131.3 (P < .001) 47.3 (P = .53)
0.71 0.76 0.69 0.67
count at 10 years of age, and this amount slightly decreased over time. Differences between Classes Adolescents were allocated to one of the 4 classes on the basis of their probability scores (ie, >0.50). Adolescents with low probability scores for all classes were excluded (n = 186). After allocation, the CBCL-based developmental trajectories of adolescents’ FSS were plotted for the four separate classes. Taking into account that parent-reports of FSS are generally lower than child-reports, the developmental patterns based on parental data were comparable with the self-reported data (Figure 3; available at www.jpeds.com), which increases the robustness of the findings. Baseline differences between the classes in sex ratio, symptom diversity, self-rated health, anxiety, depression, school absenteeism, parental overprotection, and parent-reported FSS can be found in Table III. Overall, adolescents who had persistent FSS had the greatest baseline levels of all risk factors, followed by adolescents who started high, but had decreasing symptoms. Adolescents with persistent low symptoms had the lowest baseline levels of all risk factors (Table III). The only exception is that boys were more likely to belong to the class with decreasing symptoms, and girls to the class with slightly increasing symptoms (Table III). Risk Factors for Persistence of FSS Approximately one- third (class 1) of the adolescents with high baseline symptoms (class 1 and 2) continues to have high symptoms, and about two-thirds (class 2) of the adolescents with high baseline symptom scores had rapidly decreasing symptoms. Being a girl, poor self-rated health,
number of parent-reported FSS, and suffering from depressive symptoms significantly predicted persistence of FSS (Table IV). Symptom diversity, anxiety, parental overprotection, and school absenteeism did not significantly predict a persistence of FSS (Table IV).
Discussion Our results suggest that 4.1% of adolescents suffer from persistent FSS from age 10 up to age 17. Of all adolescents that suffered from many FSSs at baseline (14.6%), 32.4% had persistent FSS (ie, having persistent instead of decreasing symptoms). Risk factors for persistence of FSS were being a girl, poor self-rated health, depressive symptoms, and high parent-reported FSS. Symptom diversity, anxiety, school absenteeism, and parental overprotection were not predictive of symptom persistence. In keeping with our hypotheses and previous studies,6-8,11 depressive symptoms and sex predicted persistence of FSS in adolescents. One reason for depression being a risk factor for persisting FSS might be that depressed adolescents are more bothered by their symptoms and more focused on them, which might make it more difficult for symptoms to resolve.24,25 Although this could apply to anxious adolescents as well, this effect was not found. This might be explained by the lack of positive affect that is common in depressed but not in anxious adolescents.26 This lack of positive affect might lead to withdrawal of activities in depressed adolescents and thereby result in less distraction from their symptoms. Another reason for depression being identified a risk factor for persistence of FSS and not anxiety, might be that the YSR anxiety scale covers only 6 anxiety items and not the full spectrum of anxiety disorders. When we used the more extensive anxiety scale of the Revised Child and Adolescent Depression and Anxiety scale, anxiety was predictive of the persistence of FSS; however, the effect disappeared after inclusion of depression into the model. Why girls are more vulnerable for developing persistent FSS than boys is not clear. In a previous study, we did not
Table III. Average scores for adolescents with persistent high, decreasing, increasing, and low symptom levels
Girl Anxiety* Depression* Parental overprotection† School absenteeismz Low self-rated healthx Parent-reported FSS* Symptom diversity{
Persistent high symptom level (class 1, n = 84)
Decreasing symptom level (class 2, n = 175)
Slightly increasing symptom level (class 3, n = 416)
Persistent low symptom level (class 4, n = 1349)
82.1% 0.40 (0.38) 0.58 (0.31) 1.99 (0.42) 2.12 (0.68) 2.65 (0.83) 0.49 (0.37) 5.66 (1.28)
49.7% 0.38 (0.30) 0.47 (0.26) 1.95 (0.41) 2.09 (0.66) 2.34 (0.80) 0.38 (0.27) 5.67 (1.08)
72.4% 0.33 (0.31) 0.33 (0.24) 1.85 (0.41) 1.97 (0.69) 2.12 (0.88) 0.27 (0.25) 3.45 (1.82)
42.6% 0.29 (0.29) 0.24 (0.21) 1.83 (0.41) 1.85 (0.65) 1.90 (0.72) 0.18 (0.22) 2.07 (1.71)
*Mean item score (range 0-2). †Mean item score for paternal overprotection for boys, and maternal overprotection for girls (range 0-3). zSchool absenteeism (range 1-4). xLow self-rated health (range 1-5). {Symptom diversity (range 1-7).
Predictors for Persistence of Functional Somatic Symptoms in Adolescents
903
THE JOURNAL OF PEDIATRICS
www.jpeds.com
Table IV. Multivariable logistic regression analysis for poor prognosis of FSS Girl Anxiety* Depression* Parental overprotection† School absenteeismz Low self-rated healthx Parent-reported FSS* Symptom diversity{
OR
95% CI
P value
4.69 0.91 5.35 1.04 0.86 1.56 4.03 0.87
2.17-10.12 0.31-2.68 1.46-16.62 0.46-2.35 0.46-1.35 1.02-2.39 1.20-13.54 0.63-1.18
Study design The first (N = 2230, mean age = 11.1 years [SD 0.6], 50.8% girls), second (N = 2149, mean age = 13.7 years [SD 0.5], 51.0% girls), and third (N = 1816, mean age = 16.3 years [SD 0.7], 52.3% girls) assessment waves of the general population study TRacking Adolescents’ Individual Lives Survey were used. FSS were assessed with the Youth Self-Report and the Child Behavior Checklist. Growth mixture models were used to identify different subgroups of adolescents on the basis of the developmental trajectory of their symptoms. Adolescents with persistent symptoms were compared with adolescents with decreasing symptoms with a multivariable logistic regression analysis. Results In our general population cohort, 4.1% of adolescents suffered from persistent FSS. Risk factors for persistent FSS were being a girl (OR 4.69, 95% CI 2.17-10.12), suffering from depressive symptoms (OR 5.35, 95% CI 1.46-16.62), poor self-rated health (OR 1.56, 95% CI 1.02-2.39), and high parent-reported FSS (OR 4.03, 95% CI 1.20-13.54). Anxiety, parental overprotection, school absenteeism, and diversity of symptoms did not predict persistence of FSS. Conclusions This study identified risk factors for persistence of FSS in adolescents. Future studies might study effects of coping strategies and iatrogenic factors on symptom persistence. (J Pediatr 2014;164:900-5).
F
unctional somatic symptoms (FSS) are somatic symptoms that cannot sufficiently be explained by underlying organic pathology.1 FSS, such as pain and fatigue, are common among adolescents. FSS are known to cause substantial impairment in a subgroup of children and adolescents by resulting in school absenteeism and social problems.2,3 This impairment is especially true for adolescents suffering from persistent symptoms.4 To prevent symptoms from becoming persistent, early intervention is important. However, although FSS are persistent in a subgroup of adolescents,5 FSS often are self-limiting. Thus, it is probably not necessary to (extensively) intervene in all adolescents with FSS. Therefore, insight in whether an adolescent is at risk for persistence of FSS is important. In previous studies, investigators examined risk factors for persistence of either pain or fatigue in adolescents.6-9 It is probably more clinically relevant to examine symptom persistence for a full range of FSS because adolescents who have multiple symptoms are especially at risk for long-lasting psychological and social problems.1 Moreover, symptoms experienced by adolescents might change over time, and their recovery from specific symptoms does not automatically mean that they are symptom-free. Another problem with previous studies is that all, except for one,6 used cut-off scores to determine whether an adolescent had a good or a poor symptom prognosis. Classifying adolescents by use of developmental trajectories might be a more reliable method, with more realistic subdivisions, because complex trajectories like exponential growth trajectories can be taken into account. This study differs from previous studies in combining 3 important aspects: (1) it included several types of FSS; (2) it identified risk factors based on developmental trajectories; and (3) it was performed in the general population. We chose to study risk factors for persistence of FSS that can be easily assessed by clinicians or have previously been found to predict poor prognosis of pain or fatigue, or to perpetuate FSS. The risk factors we hypothesized to be risk factors for persistent FSS were being a girl8,10 and having a high number of different symptoms, poor self-rated health, anxiety,7,11 depression,6,7,9,11 school absenteeism,7,12 parental overprotection,13 and high parent-reported FSS. Our hypotheses were tested in 2210 adolescents of a prospective population-based cohort study.
Method The TRacking Adolescents’ Individual Lives Survey (TRAILS) is a prospective cohort study of Dutch adolescents, approved by the Dutch Central Committee
BIC CBCL FSS LMR-LRT TRAILS YSR
Bayesian Information Criterion Child Behavior Checklist Functional somatic symptoms Lo-Mendell-Rubin likelihood ratio test TRacking Adolescents’ Individual Lives Survey Youth Self-Report
From the University of Groningen, University Medical Center Groningen, Interdisciplinary Center Psychopathology and Emotion Regulation, Groningen, The Netherlands This research is part of TRAILS, which has been financially supported by various grants from the Netherlands Organization for Scientific Research, Zorgonderzoek Nederland Medische Wetenschappen, Gebied Maatschappij-en Gedragswetenschappen, the Dutch Ministry of Justice, the European Science Foundation, Biobanking and Biomolecular Research Infrastructure, the participating universities, and Accare Center for Child and Adolescent Psychiatry. The authors declare no conflicts of interest. 0022-3476/$ - see front matter. Copyright ª 2014 Mosby Inc. All rights reserved. http://dx.doi.org/10.1016/j.jpeds.2013.12.003
900
Vol. 164, No. 4 April 2014 on Research Involving Human Subjects. Participating centers of TRAILS include various departments of the University Medical Center and University of Groningen, the Erasmus University Medical Center Rotterdam, the University of Utrecht, the Radboud Medical Center Nijmegen, and the Parnassia Bavo group, all in The Netherlands. The current study used data from the first 3 assessment waves, wave 1 in 2001-2002 (N = 2230, mean age = 11.09 years [SD 0.56], 50.8% girls); wave 2 in 2003-2004 (N = 2149, mean age = 13.65 years [SD 0.53], 51.0% girls), and wave 3 in 20052007 (N = 1816, mean age = 16.27 years, [SD = 0.73], 52.3% girls). Informed consent was obtained from both parents and subjects. Detailed information about sample selection and analysis of non-response bias has been reported elsewhere.14,15 To assess FSS, adolescents filled out the Somatic Complaints scale of the Youth Self-Report (YSR16) at each assessment wave. This scale contains 9 items, which refer to somatic complaints without a known medical cause (aches/ pains, headaches, nausea, eye problems, skin problems, stomach pain, and vomiting) or without obvious reason (overtiredness and dizziness). Factor analyses (Table I; available at www.jpeds.com) indicated that 2 items (eye problems and skin problems) had low factor loadings, suggesting that these items did not represent the underlying construct well in the TRAILS sample; therefore, these items were excluded. The remaining 7 items showed good internal consistency (Cronbach a at T1: 0.76; at T2: 0.77; at T3: 0.76). Each somatic complaint was rated on a 3-point scale with 0 = never or not at all true, 1 = sometimes or a bit true, and 2 = often or very true. A mean item score was computed by adding the scores of the 7 FSS items and dividing the sum score by 7, which resulted in a scale score that could range from 0 (all items rated as “never or not at all true”) to 2 (all items rated as “often or very true”).
scale of the YSR.16 One item (overtiredness) was excluded from this scale to prevent overlap with the Somatic Complaints scale. The internal consistency was adequate (Cronbach a at T1: 0.69; at T2: 0.74; at T3: 0.75). Again mean scale scores were used, which could range from 0 to 2.
Parental Overprotection. Parental overprotection at baseline was measured by use of the overprotection subscale of the Egna Minnen Betr€affande Uppfostran, Child Version (Swedish for “my memories of upbringing”17), which contains 12 items referring to children’s perception of parental overprotection, which can be rated on a 5-point scale ranging from 0 = never to 4 = always (Cronbach a = 0.84). Adolescents filled out this questionnaire for both their mothers and fathers. In line with our previous research, in which girls were found to be especially vulnerable for maternal and boys for paternal overprotection, an overall overprotection score was computed that used maternal overprotection scores for girls and paternal scores for boys.13 Our previous study showed that the effect of overprotection on FSS was equally strong for boys and girls. The low number of participants suffering from persistent FSS did not allow subgroup analyses for boys and girls in the current study. The mean item score was calculated, which could range from 0 to 4.
School Absenteeism. Parents answered the following question about school absenteeism: “How often has your child been absent from school during the past six months because of illness?” Answer categories were: “Never,” “Seldom,” “Sometimes,” “Often,” and “Mostly.” Because of low cell count, the last 2 categories were combined. Self-Rated Health. To assess self-rated health, adolescents answered the question “How did you perceive your health during the past year?” with: “Very good (1),” “Good (2),” “Fair (3),” “Moderate (4),” and “Bad (5).”
Potential Risk Factors
Symptom Diversity. The 7 somatic symptoms of the YSR also were used to construct a baseline symptom diversity score, which takes the diversity of symptoms into account without paying attention to the symptom severity. This symptom diversity score could range from 1 to 7 symptoms. A symptom was considered present if it was rated as “sometimes or a bit true” or “often or very true”.
Parent-Reported FSS. Parents completed the Child Behavior Checklist (CBCL), which contains the same items and response categories as the YSR. A mean item score was computed for the aforementioned 7 somatic symptoms (Cronbach a at T1: 0.71; at T2: 0.72; at T3: 0.75). Of all parents, 88% (N = 2017) completed the CBCL at T1, 82% (N = 1883) at T2, and 66% (N = 1509) at T3.
Anxiety and Depression. Symptoms of anxiety were measured by the 6 items of the Anxiety scale of the YSR (Cronbach a at T1: 0.63; at T2: 0.63; at T3: 0.65).16 Depression was measured by the 13 items of the Affective Problems
Statistical Analyses Growth mixture modeling was used to identify distinct developmental trajectories of FSS. Growth mixture modeling was conducted in Mplus, version 5.2 (Muthen & Muthen, Los Angeles, California).18 Trajectories were determined by latent growth factors, which model the intercepts and slopes (linear and quadratic) of the individual growth trajectories. First, single-class latent growth models were estimated to determine whether linear or quadratic growth curves fitted the data best. This was based on the smallest Bayesian Information Criterion (BIC),19 as well as the significance of the quadratic slope growth factor. Second, models (linear or quadratic) with increasing numbers of classes were fitted. The data were rearranged as a function of chronological age instead of clustered by wave of data collection, resulting in 8 (age 10 years until age 17 years) instead of 3 assessment points, which enabled modeling more complex pathways. Trajectories were estimated on the basis of full information maximum likelihood with robust standard errors, which is robust regarding non-normality of the scores, and adjusts 901
THE JOURNAL OF PEDIATRICS
www.jpeds.com
for uncertainty associated with missing data.20 Rules to determine the number of trajectory classes were formulated a priori, and were in increasing order of importance: (1) a significant Lo-Mendell-Rubin likelihood ratio test (LMRLRT),21 which indicates that a model with k classes is preferred over k-1 classes; (2) a BIC lower than the BIC of the model with k-1 classes; and (3) an entropy greater than the entropy of k-1 classes. The latter is an indicator of how good the model can subdivide participants into different classes because it summarizes the posterior probabilities (ie, the probability to be a member of a specific class) for all classes. The entropy can range from no distinction between classes (0) to perfect distinction between classes (1).22 Please note that conventional fit indices like the root mean square error of approximation are not available for growth mixture models.23 Estimating class specific variances for the intercept, linear slope, and quadratic slope resulted in convergence problems, which is often an indication of overfitting (ie, the fitted model is overly complex). Therefore, random slope variances were fixed to zero. All adolescents were allocated to one of the FSS trajectories on the basis of their greatest posterior probability score. Adolescents who were not adequately fitting one of the classes (probability score #0.50) were excluded from subsequent analyses. To compare CBCL-based developmental patterns with YSR-based developmental patterns, the parental CBCL data were plotted for the different classes. Again, data were rearranged as a function of chronological age instead of clustered by waves to ease comparing both plots. Four groups were identified (see results in Figure 1). To indicate risk factors for persistence of FSS, adolescents who had high FSS at baseline but for whom symptoms decreased (dummy coded 0) were compared with adolescents with high FSS at baseline who had persistent FSS (dummy coded 1). Adolescents of both other groups had low baseline FSS scores and were therefore excluded from this analysis. A multivariable logistic regression analysis was performed to examine whether sex, symptom
Vol. 164, No. 4 diversity, self-rated health, anxiety, depression, school absenteeism, parental overprotection, and parent-reported FSS, all at baseline, were risk factors for poor prognosis of FSS.
Results The mean item score of FSS decreased during subsequent waves and was 0.47 (SD 0.35) at T1 (n = 2115), 0.39 (SD 0.35) at T2 (n = 2015), and 0.34 (SD 0.34) at T3 (n = 1636). A total of 2210 adolescents had valid data on FSS on at least one assessment wave (Figure 2; available at www. jpeds.com). Developmental pathways were examined for these adolescents. Figure 2 shows a flowchart with number of adolescents included in subsequent analyses. Group Identification A one-class model that included a quadratic term was significantly worse than a one-class model that consisted of a linear relationship only. Moreover, the quadratic term was not statistically significant. Therefore, the different class solutions were fitted for models that included only linear and not quadratic relationships. The BICs, LMR-LRTs, and entropies for these different class-solutions are given in Table II. The model with the greatest number of classes that still had a significant LMR-LRT was the 4-class solution. Moreover, the BIC of this 4-class solution was lower than that of the 3-class solution. On the basis of these fit indices, 4 classes were considered to fit the data best. The course of FSS during adolescence in these 4 classes is given in Figure 1. The first class, 4.1% (n = 90) of the adolescents, had persistent FSS. The second class, 10.5% (n = 233) of the adolescents, started high on FSS, and these symptoms decreased when adolescence proceeded. The third class, 22.0% (n = 487) of all adolescents, had moderate symptoms at 10 years of age, and these symptoms slightly increased over the years. The fourth class, that constituted the majority, ie, 63.3% (n = 1400), had a low symptom
Figure 1. Overview of the developmental patterns of FSS in the 4 classes that have been identified with growth mixture modeling (n = 2210). 902
Janssens et al
ORIGINAL ARTICLES
April 2014
Table II. Fit criteria of different fitted models Number of clusters (k)
BIC
LMR-LRT
Entropy
1 2 3 4 5
3471.2 3256.1 3159.8 3051.7 3027.5
238.3 (P = .01) 119.3 (P < .001) 131.3 (P < .001) 47.3 (P = .53)
0.71 0.76 0.69 0.67
count at 10 years of age, and this amount slightly decreased over time. Differences between Classes Adolescents were allocated to one of the 4 classes on the basis of their probability scores (ie, >0.50). Adolescents with low probability scores for all classes were excluded (n = 186). After allocation, the CBCL-based developmental trajectories of adolescents’ FSS were plotted for the four separate classes. Taking into account that parent-reports of FSS are generally lower than child-reports, the developmental patterns based on parental data were comparable with the self-reported data (Figure 3; available at www.jpeds.com), which increases the robustness of the findings. Baseline differences between the classes in sex ratio, symptom diversity, self-rated health, anxiety, depression, school absenteeism, parental overprotection, and parent-reported FSS can be found in Table III. Overall, adolescents who had persistent FSS had the greatest baseline levels of all risk factors, followed by adolescents who started high, but had decreasing symptoms. Adolescents with persistent low symptoms had the lowest baseline levels of all risk factors (Table III). The only exception is that boys were more likely to belong to the class with decreasing symptoms, and girls to the class with slightly increasing symptoms (Table III). Risk Factors for Persistence of FSS Approximately one- third (class 1) of the adolescents with high baseline symptoms (class 1 and 2) continues to have high symptoms, and about two-thirds (class 2) of the adolescents with high baseline symptom scores had rapidly decreasing symptoms. Being a girl, poor self-rated health,
number of parent-reported FSS, and suffering from depressive symptoms significantly predicted persistence of FSS (Table IV). Symptom diversity, anxiety, parental overprotection, and school absenteeism did not significantly predict a persistence of FSS (Table IV).
Discussion Our results suggest that 4.1% of adolescents suffer from persistent FSS from age 10 up to age 17. Of all adolescents that suffered from many FSSs at baseline (14.6%), 32.4% had persistent FSS (ie, having persistent instead of decreasing symptoms). Risk factors for persistence of FSS were being a girl, poor self-rated health, depressive symptoms, and high parent-reported FSS. Symptom diversity, anxiety, school absenteeism, and parental overprotection were not predictive of symptom persistence. In keeping with our hypotheses and previous studies,6-8,11 depressive symptoms and sex predicted persistence of FSS in adolescents. One reason for depression being a risk factor for persisting FSS might be that depressed adolescents are more bothered by their symptoms and more focused on them, which might make it more difficult for symptoms to resolve.24,25 Although this could apply to anxious adolescents as well, this effect was not found. This might be explained by the lack of positive affect that is common in depressed but not in anxious adolescents.26 This lack of positive affect might lead to withdrawal of activities in depressed adolescents and thereby result in less distraction from their symptoms. Another reason for depression being identified a risk factor for persistence of FSS and not anxiety, might be that the YSR anxiety scale covers only 6 anxiety items and not the full spectrum of anxiety disorders. When we used the more extensive anxiety scale of the Revised Child and Adolescent Depression and Anxiety scale, anxiety was predictive of the persistence of FSS; however, the effect disappeared after inclusion of depression into the model. Why girls are more vulnerable for developing persistent FSS than boys is not clear. In a previous study, we did not
Table III. Average scores for adolescents with persistent high, decreasing, increasing, and low symptom levels
Girl Anxiety* Depression* Parental overprotection† School absenteeismz Low self-rated healthx Parent-reported FSS* Symptom diversity{
Persistent high symptom level (class 1, n = 84)
Decreasing symptom level (class 2, n = 175)
Slightly increasing symptom level (class 3, n = 416)
Persistent low symptom level (class 4, n = 1349)
82.1% 0.40 (0.38) 0.58 (0.31) 1.99 (0.42) 2.12 (0.68) 2.65 (0.83) 0.49 (0.37) 5.66 (1.28)
49.7% 0.38 (0.30) 0.47 (0.26) 1.95 (0.41) 2.09 (0.66) 2.34 (0.80) 0.38 (0.27) 5.67 (1.08)
72.4% 0.33 (0.31) 0.33 (0.24) 1.85 (0.41) 1.97 (0.69) 2.12 (0.88) 0.27 (0.25) 3.45 (1.82)
42.6% 0.29 (0.29) 0.24 (0.21) 1.83 (0.41) 1.85 (0.65) 1.90 (0.72) 0.18 (0.22) 2.07 (1.71)
*Mean item score (range 0-2). †Mean item score for paternal overprotection for boys, and maternal overprotection for girls (range 0-3). zSchool absenteeism (range 1-4). xLow self-rated health (range 1-5). {Symptom diversity (range 1-7).
Predictors for Persistence of Functional Somatic Symptoms in Adolescents
903
THE JOURNAL OF PEDIATRICS
www.jpeds.com
Table IV. Multivariable logistic regression analysis for poor prognosis of FSS Girl Anxiety* Depression* Parental overprotection† School absenteeismz Low self-rated healthx Parent-reported FSS* Symptom diversity{
OR
95% CI
P value
4.69 0.91 5.35 1.04 0.86 1.56 4.03 0.87
2.17-10.12 0.31-2.68 1.46-16.62 0.46-2.35 0.46-1.35 1.02-2.39 1.20-13.54 0.63-1.18
