Eur Child Adolesc Psychiatry DOI 10.1007/s00787-014-0656-2

ORIGINAL CONTRIBUTION

Modelling trajectories of psychosomatic health complaints in children and adolescents: results of the BELLA study Claus Barkmann · Christiane Otto · Gerhard Schön · Michael Schulte‑Markwort · Robert Schlack · Ulrike Ravens‑Sieberer · Fionna Klasen · The BELLA study group

Received: 28 April 2014 / Accepted: 19 November 2014 © Springer-Verlag Berlin Heidelberg 2014

Abstract  Psychosomatic health complaints (PHC) can significantly impair psychosocial development of children and adolescents and are therefore of considerable interest in health sciences and public health surveillance. Questions addressed the type of function that describes individual trajectories best, potential differences between these, and corresponding predictors from the perspective of both children and their parents. Based on the German population-based and representative BELLA cohort sample, 2,857 children and adolescents between 7 and 17 years of age at baseline were analysed over a period of 3 years with yearly followups using mixed growth curve analyses. PHC were measured in accordance with the health behaviour in school-aged children-symptom checklist. The mean level of PHC was rather low, slightly lower for the parent report than for the Members of the BELLA study group: Ulrike Ravens-Sieberer and Fionna Klasen, Hamburg (Principal Investigators); Claus Barkmann, Hamburg; Monika Bullinger, Hamburg; Manfred Döpfner, Köln; Beate Herpertz-Dahlmann, Aachen; Heike Hölling, Berlin; Franz Resch, Heidelberg; Aribert Rothenberger, Göttingen; Sylvia Schneider, Bochum; Michael SchulteMarkwort, Hamburg; Robert Schlack, Berlin; Frank Verhulst, Rotterdam; Hans-Ulrich Wittchen, Dresden. C. Barkmann (*) · C. Otto · M. Schulte‑Markwort · U. Ravens‑Sieberer · F. Klasen  Department of Child and Adolescent Psychiatry, Psychotherapy, and Psychosomatics, Clinic for Child and Adolescent Psychiatry, Psychotherapy, and Psychosomatics, University Medical Center Hamburg-Eppendorf, Martinistraße 52, 20246 Hamburg, Germany e-mail: [email protected] G. Schön  Department of Medical Biometry and Epidemiology, University Medical Center Hamburg-Eppendorf, Hamburg, Germany R. Schlack  Robert Koch Institute, Berlin, Germany

self-report and significantly different between subjects. Concerning the parent report, the 2-year course is best described by a slowly increasing linear trend that decelerates somewhat over time. The increasing linear trend was more pronounced in the self-report from 11 to 17 years of age, but was significantly different for each subject and correlated with baseline scores. Trajectories could be explained by known predictors, most importantly by mental health problems of the child or adolescent. The results confirm the findings of previous studies and provide representative data about the individual shortterm development of PHC in children and adolescents in Germany. Keywords  Psychosomatic symptoms · Children and adolescents · Development · Public health · Growth curve analysis

Introduction Most children and adolescents experience psychosomatic health complaints (PHC) such as abdominal pain, headache or nausea every now and then. Usually, these are mild and transient in nature and have no lasting impact on psychosocial function. However, long-lasting and more severe symptoms can have serious effects on individuals and their families and may—in the absence of successful and timely treatment— result in long-term impairment of a child’s psychosocial development [1]. Psychosomatic health complaints may manifest as physical perceptions including pain and other unpleasant physical sensations, such as burning, prickling, itching, and irritating changes in bodily functions. When no clear organic aetiology can be found, PHC are often described with terms such as “dissociative”, “functional”, “somatoform”, or “medically

13



unexplained”, depending on the etiological presuppositions [2].1 PHC are primarily perceived by the patient, whereas external assessment necessarily relies on verbal communication or behavioural observations. PHC do not necessarily manifest with objective somatic symptoms, but tend to be emotionally coloured, vague, and temporary in nature [2]. Reports of PHC by individuals depend on their body image and cognitive processing as well as their ability and willingness to report. In addition, characteristics such as developmental age, temperament, and personality as well as sociocultural norms and values are involved [1, 2]. Parent-reported PHC are also influenced by the child–parent relationship and special intentions of the parent which leads to a low to medium correlation between both reports [2]. The neurophysiological mechanisms underlying perception of the body’s internal environment have long been known (interoception [3]), but many different theories have been put forward for the cause and subjective evaluation of PHC (psychoanalytic theory, learning theory, systemic theory, psycho-physiological theories, specificity concepts, etc. [4]). PHC function as alarm signals that help to protect one’s health and fulfill particular psychological functions for the individual (primary and secondary gain). They are the starting point for medical and psychological diagnosis and may initiate the use of healthcare services. Some PHC play a special role because of their occurrence in clusters and/or established etiological concepts like somatoform and dissociative symptoms, chronic pain, or chronic fatigue [4]. To date, the natural course of PHC has been investigated in about 10 longitudinal studies with population-based samples [5–14]. In addition, two trend studies (repeated cross-sectional designs) have been carried out [15, 16]. In Germany, two studies have been conducted so far, both focused on somatoform disorders [9, 13]. In most studies, simple repeated-measurement designs were implemented using regional, school-based samples from Northern European and German-speaking countries. Sample sizes ranged from 90 to 2,246 cases (median  = 1,000), with a median duration of 4.5 years (min.  = 1.2 years, max. = 15 years). The subjects were predominantly adolescents, and the surveys were carried out with self-administered checklists that were analysed for changes in the mean using analyses of variances (ANOVA). Means and baseline scores are not directly comparable due to differences in the instruments. However, a rather low mean level is being described as practically significant in all studies. The time course of the PHC total score is reported

1   The term “Psychosomatic Health Complaints” in this article is used in its general sense and does not implicate any etiological preassumptions.

13

Eur Child Adolesc Psychiatry

as more or less stable, with a moderate age-related linear increase, and a peak at the onset of puberty in some studies. This also accounts for the most common single symptoms, namely abdominal pain and headache [5, 6, 11, 14]. Both symptoms are a research field of its own and current reviews [e.g. 17, 18] confirm what is known from studies on general PHC. Inconsistent time courses have not been attributed to possible inter-individual differences as yet. The predictors used varied substantially with the different paradigms of the studies. The most commonly identified factors apart from age included gender (higher scores for girls during puberty), mental health problems of the child and their parents (especially emotional problems such as depression or anxiety measured by DSM-orientated standardised diagnostic interviews or syndrome-orientated questionnaires), somatic problems of the child and their parents (symptom-sharing, [4]), family and school problems, and problems with peer relations. Factors such as migration, socio-economic status (SES), self-efficacy, and the use of healthcare facilities also had small effects in some studies [1]. PHC have been investigated less intensively than mental health problems in children and adolescents [1, 4]. In particular, there has not been a longitudinal observation of individual trajectories in a representative, population-based sample, including consideration of a broad age range and diverse relevant predictors from the perspective of the parents and the child or adolescent. In summary, the aim of this study was to provide accurate and representative information about growth parameters. The following questions were addressed: 1. What are the trajectories of PHC in German children and adolescents over a 2-year period? 2. How are these trajectories influenced by age, gender, and other relevant predictors? 3. How do the parent report and the self-report differ?

Methods Design The BELLA study (in German: BEfragung zum seeLischen WohLbefinden und VerhAlten, study of wellbeing and behaviour of children and adolescents) is the mental health module of the German Health Interview and Examination Survey for Children and Adolescents (KiGGS), prepared and conducted by the Robert-Koch Institute (Berlin) and funded by the German Federal Ministry of Health. It is a population-based, nationwide cohort study with four measurement points, of which the first three provided data for the present analyses: baseline (t0) between 2003 and 2005,

Eur Child Adolesc Psychiatry

and t1 and t2 at intervals of 1 year (for more details see [19], in this issue). Parent reports for children and adolescents from 7 to 17 years of age and self-reports from adolescents from 11 to 17 years of age were obtained by means of standardised psychometric questionnaires. Data were analysed using a mixed growth curve model, in which the PHC trajectories of the individual subjects and the differences between these trajectories were modelled over time. Variables and instruments Outcome Evaluation of PHC was based on the German version of the health behaviour in school-aged children (HBSC) symptom checklist [20]. This instrument was chosen due to its internationally widely spread use and economy. The eight items of this checklist (headache, abdominal pain, backache, feeling low, irritability or bad mood, feeling nervous, sleeping difficulties, and dizziness) were extended by six symptoms that are particularly relevant in childhood and were available in the data (breathlessness, exhaustion, fidgetiness, lack of appetite, nausea, tremble, [1, 4]). Parents and children had three options with which to describe symptoms for the last 3 months (0, 1, 2; never, sometimes, always). The unweighted sum of the 14 item scores resulted in a PHC score ranging from 0 to 28. Internal consistency at t0, t1 and t2 was Cronbach’s α = 0.72, 0.70, and 0.74 for the parent report and 0.73, 0.68 and 0.70 for the self-report. The distribution of the resulting PHC scale was slightly skewed to the left and leptokurtic for both perspectives and all measurement points. Predictors Predictors were selected in accordance with current information (see “Introduction” section) as available from the BELLA data for the parent and child perspectives. For the analyses, the distinction between time-varying Level 1 (L1) and constant Level 2 (L2) predictors was especially relevant. L1 variables such as mental health problems or somatic problems were measured repeatedly to explain discontinuities in the trajectories within subjects. L2 variables such as age or gender were only measured at t0 to explain differences in the trajectories between subjects. All variables refer to the child if not otherwise stated. Level 1 (within-child level, repeated measurement from t0 to t2) 1. Time in exact years, set 0 for baseline. 2. Mental health problems of the child: Total Difficulties Score of the Strengths-and-Difficulties-Questionnaire

(SDQ, [21], 20 items scored 0–2, corrected for psychosomatic items2), parent report and self-report. 3. Somatic problems of the child: acute or chronic somatic illness, disability or accidents (single item, not present vs. present), parent report. 4. Mental health problems of the parents: nine-item short form of the Symptom Checklist by Derogatis (SCL-K9, scored 1–5, [22]; SCL-90-R [23]), parent report. 5. Somatic problems of the parents: acute or chronic somatic illness, disability or accidents (single item, not present vs. present), parent report. 6. Use of healthcare services: single item, not present versus present, parent report. 7. Medication: single item, not present versus present, parent report. 8. Stress: problems in family, school and peer relations, study-specific screener with 11 items, cumulative index (not present versus present by median split), parent report. 9. Self-efficacy: General Self-Efficacy Scale (GSES [24], 10 items scored 1–4), parent report and self-report. Level 2 (between-child level, t0). 1. Age at t0: exact years, interaction with time and gender is expected. 2. Gender (0 = female, 1 = male), interaction with time and age is expected. 3. Immigrant: at least one parent is an immigrant, single item, not present versus present. 4. SES: Winkler-Index, based on internationally established classifications of education, profession and income of adults [25], cumulative index, score between 3 (low) and 21 (high). Sample The BELLA participants were randomly recruited from the nationally representative KiGGS sample of German children and adolescents aged 4–17 years. The stratified, two-stage probability sampling procedure is described in detail in this issue [19]. Of the 2,863 families in BELLA, 2,857 (99.8 %) gave valid responses in the outcome of this study for at least one measurement point (5,274 measurements altogether, see Table 1). These included 2,806 parent reports for children and adolescents 7–17 years of age (67.4 % with three, 17.6 % with two, and 15.1 % with one measurement) and 2,183 self-reports for adolescents 2

 The SDQ items such as 3 headaches and 10 fidgetiness were excluded without changing the score range of 0–20 to avoid an artificial correlation with the PHC scale.

13



Eur Child Adolesc Psychiatry

Table 1  Descriptive statistics Parent report (n = 2,806 7–18 years) t0

a

Self-report (n = 2,183 11–18 years)

t1

t2

t0

t1

t2

n = 1,669 (76.5 %)

n = 1,464 (67.1 %)

n = 1,644 (75.3 %)

Sample size

n = 2,778 (99.0 %)

n = 2,175 (77.5 %)

n = 2,126 (75.8 %)

Physical health complaints (PHC scaleb)

M = 4.68 (SD = 2.57)

M = 5.31 (SD = 2.98)

M = 5.51 M = 6.62 M = 6.30 M = 6.29 (SD = 3.09) (SD = 3.23) (SD = 3.44) (SD = 3.63)

Mental health problems of the child (SDQ)

M = 10.13 (SD = 4.92)

M = 9.81 (SD = 5.02)

M = 9.60 M = 11.94 M = 10.40 M = 10.18 (SD = 4.88) (SD = 4.23) (SD = 4.44) (SD = 4.34)

Somatic problems of the child

n = 1064 (38.3 %) n = 702 (32.3 %) n = 643 (30.2 %)

n = 656 (39.3 %)

n = 496 (33.9 %)

n = 500 (30.4 %)

Mental health problems of parents (SCL-K9) M = 14.328 (SD = 4.66) Somatic problems of parents n = 1,526 (54.9 %) Healthcare utilisation n = 232 (8.4 %) Medication n = 405 (14.6 %) Psychosocial stress n = 821 (33.7 %)

M = 14.21 (SD = 4.66)

M = 13.44 M = 14.44 M = 14.27 M = 13.52 (SD = 4.37) (SD = 4.73) (SD = 4.62) (SD = 4.41)

n = 748 (34.4 %)

n = 1,446 (68.0 %)

n = 772 (46.3 %)

n = 532 (36.3 %)

n = 1,072 (65.2 %)

n = 133 (6.1 %)

n = 128 (6.0 %)

n = 102 (6.1 %)

n = 74 (5.1 %)

n = 91 (5.5 %)

n = 279 (12.8 %)

n = 239 (11.2 %)

n = 251 (15.0 %)

n = 196 (13.4 %)

n = 180 (10.9 %)

n = 772 (35.5 %)

n = 1,024 (48.2 %)

n = 485 (34.3 %)

n = 511 (36.2 %)

n = 643 (48.0 %)

Self-efficacy (GSES)

M = 32.02 (SD = 4.42)

M = 31.92 M = 31.33 M = 31.32 M = 31.40 (SD = 4.33) (SD = 3.89) (SD = 4.58) (SD = 4.43)

M = 32.00 (SD = 4.29)

Notes a t0 between 2001 and 2003; b14 items, scored 0–2, total score between 0 and 28, see “Methods” section for further instruments; (n = 2,857 German children and adolescents)

11–17 years of age (47.2 % with three, 24.5 % with two, and 28.4 % with one measurement). 23.6 % were rated exclusively by parents and 1.8 % was only represented by the self-report. As described earlier [19], families with a migration background and low SES were slightly underrepresented at baseline. Statistics Data were analysed by means of an individual growth curve model [26]. This model allows simultaneous consideration of all units of observation and their organisation within children, even with varying time points and missing values. First, a multiple linear regression equation with L1 predictors is calculated for each subject. (To remember: The intercept is the outcome score at baseline, the slope represents the increase or decrease of the outcome score per year.) A fixed effect is the mean of all subject-specific intercepts or slopes in these equations, measured in units of the outcome, namely PHC scores. For example, in Model 1, the equation for determining the PHC score for case i is Yi = 4.69 + tim ei  × 0.80 + time2i   × (−0.19). Random effects are simply the variances of the parameters in all L1 regression equations, i.e. the variance of intercepts and slopes for all cases. These variances can be explained by means of regression equations with L2 predictors in turn (for a deeper understanding see [27]). The predictors were analysed blockwise

13

forward to obtain information about interdependencies between the different groups of variables (null model without predictors, model 1 with time only, model 2 with all L1 predictors, model 3 with L2 predictors and model 4 with interactions added). Model improvement was tested using the likelihood-ratio test and the Bayesian information criterion (BIC). Analyses were performed separately for both perspectives with SPSS 20 mixed models (ML algorithm). Missing values were rare: for L2, 12 parent reports and 18 self-reports of the SES score were lacking. For L1, 12.4 and 15.2 % of the stress reports at t1 were missing. In addition, 3–4 % of the SCL scores for the self-report were absent at all measurement points. Whilst missing values for L2 were imputed by an Expectation–Maximisation algorithm, missing values for L1 could be left unchanged because the individual regression estimation allows for varying missing patterns [27]. Apart from age, all metric predictors were centred by the group mean, so that corresponding estimated fixed effects referred to the average subject.

Results Descriptive statistics Table  1 shows the distribution of time-varying variables (L1) over the three measurement points (M  = 1.0,

Eur Child Adolesc Psychiatry

SD  = 0.10 years between t0 and t1 and M  = 2.1, SD  = 0.17 years between t0 and t2). The case numbers ranged between 67.1 % for t1 (self-report) and 99.0 % for t0 (parent report). The mean PHC score was constantly higher for the self-report, but only the parent report indicated an increase of complaints over time. Most predictors were rather stable over time: only somatic problems of the parents and psychosocial stress varied (compare Table 1). The PHC parent reports correlated with r  = 0.52 for t0– t1, 0.50 for t0–t2 and 0.59 for t1–t2; correlations of selfreports were comparable (r = 0.54, 0.46 and 0.61). Predictors correlated in the range of small effects: the strongest association was found between mental health problems of parents and self-efficacy (r = −0.33) for the parent report, and between mental health problems of the child and selfefficacy (r = −0.38) for the self-report. The distribution of L2 predictors was as follows: age at t0 was evenly distributed with 10 % per year, 51.5 % of the subjects were boys, 14.2 % had a migration background, and the mean SES was M = 11.76, SD = 4.22. The self-report showed similar values (age was also evenly distributed with 12.5 % per year, 51.3 % were boys, 14.1 % had a migration background, and the mean SES was M = 11.6, SD = 4.23).

the mental health problems of the child had a large effect, the other significant effects were small or very small. The subject-related variance was reduced by 44 % ((3.15– 1.75)/3.15 × 100). Overall, one-third of the total variance was explained by the subjects and improvement of the model was highly significant. Consideration of L2 predictors in Model 3 resulted in very little change. The fixed effects included up to this step remained stable. Although age and gender contributed significantly, the migration background and SES stayed insignificant. Accordingly, the random effects hardly changed and the additional explained variance continued to be low. Cross-level interactions in Model 4 also had only marginal effects: the main effects of age and gender lost their significance. However, interactions of time and age, age and gender, as well as the three-way interaction of these predictors became significant, albeit with very small effects. Although the log-likelihood test indicated a significant improvement, the random effects were not substantially reduced. Both the within-subject and between-subject variances remained significant. Further exploratory analyses revealed that there were several possible interaction terms that slightly improved the model, but did not contribute to a significant increase in the explained variance.

Parent report Self‑report The null model without predictors (random intercept only) showed an overall mean PHC score of M  = 5.11 (95 % CI 5.02, 5.20), a subject-bound variance of τ = 4.21 (95 % CI 3.90, 4.54) and a residual variance of σ = 4.05 or more (95 % CI 3.88, 4.22). This results in an intraclass correlation of ICC = .51, i.e. 51 % of the total variance in the outcome could be explained by differences between the subjects. Exploration of the individual growth trajectories in Model 1 revealed fixed effects for the linear and squared trend (slope parameter) and a random effect for the linear trend (variance of the individual slopes, Table 2). The mean PHC score increased linearly by 0.8 points per year, accompanied by a slight deceleration over time. The baseline scores (intercept parameter) and linear trends differed significantly between subjects (significant random effects). In addition, intercepts and linear trends were correlated, i.e. higher starting values were associated with a greater slope and vice versa. The subject-related variance was reduced by five percentage points (ICC = .46). The log-likelihood test showed a highly significant improvement compared to the null model. Entering more time-varying predictors in Model 2, the random effect of the linear trend had to be abandoned and a uniform linear growth for all children had to be assumed to ensure conversion of the model. All added predictors showed significant fixed effects except somatic problems in children and parents, as well as psychosocial stress. Whilst

The null model indicated a slightly higher mean PHC score than the parents’ judgment (M = 6.46, 95 % CI 6.33, 6.59). The subject-bound variance was τ  = 6.43 (95 % CI 5.89, 7.03), the residual variance σ = 5.47 (95 % CI 5.18, 5.77). Corresponding with the parent report, this led to an intracluster correlation of ICC = .54, i.e. the differences in mean PHC scores between subjects explained 54 % of the total variance. The exploration of growth curves in Model 1 only revealed a significant fixed and random effect for the linear slope (Table 3). However, instead of an increase, the corresponding linear slope showed a decline in PHC levels over time. Again, both intercepts and linear slopes differed significantly between subjects. A correlation between both parameters was not observed. The log-likelihood test showed a highly significant improvement compared to the null model. With ICC = 0.58, more than half of the total variance was again due to differences between subjects. In Model 2, in contrast to the parent report, the linear trend lost its significance, i.e. the observed trend in the previous model could be explained by the new L1 predictors. In addition, only self-efficacy and the mental health problems of the child and parents showed significant effects. The residual variance did not change, but the variance of the intercepts was reduced by 45.6 %. Accordingly, the intracluster correlation decreased to ICC = .43. Although

13



Eur Child Adolesc Psychiatry

Table 2  Course of PHC for 7- to 17-year-old German children and adolescents between 2001 and 2005 (parent report) Model 1

Fixed effects Intercept

Model 2

Model 3

Model 4

Effect

95 % CI

Effect

95 % CI

Effect

95 % CI

Effect

95 % CI

4.69***

4.60;4.79

4.47***

4.35;4.59

4.07***

3.76;4.38

4.04***

3.54;4.55

0.96*** −0.21*** 0.26*** 0.06 0.11*** 0.04 0.53*** 0.35*** −0.07 −0.02*

0.77;1.16 −0.30;−0.12 0.25;0.27 −0.07;0.19 0.10;0.13 −0.08;0.15 0.30;0.77 0.18;0.52 −0.18;0.04 −0.03;−0.00

1.00*** −0.23*** 0.27*** 0.06 0.11*** 0.06 0.57*** 0.35*** −0.07 −0.02**

0.80;1.19 −0.32;−0.14 0.25;0.28 −0.07;0.20 0.10;0.12 −0.06;0.17 0.33;0.81 0.18;0.52 −0.18;0.04 −0.03;−0.00

0.47** −0.23*** 0.27*** 0.08 0.11*** 0.03 0.53*** 0.33*** −0.06 −0.02**

0.11;0.83 −0.32;−0.14 0.25;0.28 −0.06;0.21 0.10;0.12 −0.08;0.15 0.29;0.76 0.16;0.50 −0.16;0.05 −0.03;−0.01

0.04*** −0.27*** 0.12 −0.00

0.02;0.06 −0.41;−0.13 −0.09;0.32 −0.02;0.01

0.03 0.64 0.13 −0.00

−0.00;0.07 −0.04;1.32 −0.08;0.33 −0.02;0.02

0.05*** 0.18 −0.06* −0.03*

0.03;0.08 −0.25;0.61 −0.11;−0.02 −0.07;0.00

3.54*** 1.71***

3.39;3.69 1.54;1.90

L1 slopes Time (years) 0.80*** 0.61;1.00 Time × time −0.19*** −0.28;−0.10 Mental health problems child Somatic problems child Mental health problems parents Somatic problems parents Health service use Medication Stress Self-efficacy L2 slopes Age at t0 (years) Gender (male) Migration background SES Interactions Time × age at t0 Time × gender Age at t0 × gender Time × age at t0 × gender Random effects Residuum Intercept Intercept × time Time ICC −2*log-likelihood Χ 2/df

3.72*** 3.15*** 0.58*** 0.14*** .46 33,181.3 366.6/4

p BIC

3.50;3.96 2.78;3.56 0.39;0.76 0.05;0.43

3.58*** 1.75***

3.43;3.74 1.58;1.95

3.57*** 1.72***

3.42;3.73 1.54;1.91

.33 29,796.4 3,384.8/6

.32 29,658.9 137.5/4

.33 29,602.7 56.2/4

.000

.000

.000

.000

33,243.3

29,911.0

29,808.7

29,787.7

Note MRC model (ML); n  = 2,806 children and adolescents with up to 3 measurements, PHC raw score (14 items, scored 0–2, total score between 0 and 28) BIC Bayesian information criterion

there was no mean time trend for all subjects, individual subjects showed significantly different linear trends. If L2 predictors were entered in Model 3, the linear trend became significant again, but now with a positive sign. The other L1 predictors remained stable compared to the previous model. Of the L2 predictors, only age at t0 and gender showed a significant impact, whilst migration background and SES were irrelevant. Although the random effects remained unchanged, the improvement of the model became significant.

13

Consideration of the interaction terms in Model 4 increased the time trend significantly. The main effect of gender lost its significance. Of the four added interaction effects, only the interaction of age and gender was significant. The intercept and slope parameters showed a slight negative correlation (r = −.25). All other effects remained the same. The random effects did not change substantially and the intracluster correlation remained stable. Although the model improved significantly, the BIC value indicated no improvement in the economical sense. Both within

Eur Child Adolesc Psychiatry Table 3  Trajectories of PHC for 11- to 18-year-old German children and adolescents between 2001 and 2005 (self-report) Model 1

Fixed effects Intercept L1 slopes Time (years) Time × time Mental health problems child Somatic problems child Mental health problems parents Somatic problems parents Health service use Medication Stress Self-efficacy

Model 2

Model 3

Model 4

Effect

95 % CI

Effect

95 % CI

Effect

95 % CI

Effect

95 % CI

6.65***

6.50;6.80

6.15***

5.97;6.33

5.01***

4.37;5.65

3.24***

1.95;4.54

−0.37** 0.11

−0.65;−0.10 −0.2;0.23

0.26 −0.04 0.34*** 0.10 0.06*** −0.03 0.09 0.27* −0.02 −0.07***

−0.02;0.55 −0.17;0.09 0.31;0.36 −0.10;0.30 0.03;0.08 −0.21;0.14 −0.30;0.47 0.01;0.52 −0.18;0.14 −0.09;−0.04

0.35* −0.06 0.33*** 0.11 0.05*** 0.00 0.21 0.23 −0.01 −0.07***

0.06;0.64 −0.19;0.07 0.31;0.36 −0.09;0.31 0.03;0.07 −0.17;0.17 −0.17;0.59 −0.02;0.48 −0.17;0.15 −0.09;−0.04

1.04* −0.08 0.33*** 0.12 0.05*** 0.02 0.19 0.21 −0.01 −0.06***

0.21;1.86 −0.21;0.05 0.31;0.36 −0.08;0.32 0.03;0.07 −0.15;0.20 −0.19;0.57 −0.05;0.46 −0.17;0.15 −0.09;−0.04

0.10*** −0.84*** 0.26 −0.02

0.06;0.15 −1.04;−0.63 −0.05;0.57 −0.05;0.00

0.22*** 1.37 0.27 −0.02

0.13;0.30 −0.45;3.18 −0.04;0.59 −0.05;0.00

−0.03 −0.21 −0.13* −0.02

−0.08;0.02 −1.28;0.86 −0.26;−0.01 −0.05;0.00

4.42*** 3.11*** −0.34* 0.58*** .41 21,058.1 32.6/4

4.07;4.81 2.55;3.80 −0.68;−0.00 0.35;0.94

L2 slopes Age at t0 (years) Gender (male) Migration background SES Interactions Time × age at t0 Time × gender Age at t0 × gender Time x age at t0 × gender Random effects Residuum Intercept Intercept × time Time ICC −2*log-likelihood Χ2/df

4.49*** 6.10*** −0.14 0.88*** .58 24,266.3 90.0/4

p

.000

.000

.000

.000

BIC

24,325.6

21,388.6

21,250.1

21,250.1

4.14;4.86 5.39;6.90 −0.50;0.23 0.63;1.23

4.42*** 3.32*** −0.28 0.59*** .43 21,262.8 3,003.5/8

4.07;4.81 2.75;4.01 −0.62;0.06 0.36;0.95

4.43*** 3.16*** −0.37* 0.61*** .42 21,090.8 172.0/4

4.08;4.82 2.60;3.86 −0.71;−0.02 0.38;0.97

Note MRC model (ML); n  = 2,183 children and adolescents with up to 3 measurements, PHC raw score (14 items, scored 0–2, total score between 0 and 28) BIC Bayesian information criterion

and between subjects, significant variance remained to be explained with further predictors. Further exploratory analyses showed that there were several further interaction terms that slightly improved the model, but this did not result in a significant increase in the explained variance. Figure 1 shows the profiles of the different age cohorts over the observed age range for both genders and perspectives with age instead of time on the x-axis. Thus, the PHC score for the entire age range of 7- to 17 + 2-year olds in the parent report is linear and stable in the mean, albeit with a slight increase for girls. In the self-report for

the 11- to 17 + 2-year olds, a slightly s-shaped curve with higher scores for girls is observable. For both perspectives, the small but significant effect of the age cohort is evident, visible by the upward and downward cohort curves. Sensitivity Alternative transformations such as different age and SES categories or deriving the log PHC scores resulted in a significantly worse model fit. The slight drop-out bias regarding migration background and SES did not play a role, because

13



Eur Child Adolesc Psychiatry

Fig. 1  Mean course of psychosomatic health complaints with regard to age and gender (n = 2,806 parent reports for 7–19 years, n = 2,183 selfreports for 11–19 years, single lines = age cohorts)

these factors were considered in the model and showed no effect. The rates of missing data in the predictors were too small to have an impact. Technical variables such as the number of measurements per subject, the date of measurement, or the region showed no impact. Cross-validation in a subsample was not possible due to the limited sample size. A bootstrap analysis with k = 1,000 samples confirmed the results shown, with the exception that gender and migration reached significance for the perspectives of both parents and children. Other predictors such as the age of parents, the type of school, or family environment as well as the breakdown of cumulative predictors such as SES and psychosocial stress provided no further information. Exploration of the SDQ subscales revealed the best model fit and the largest effect for “Emotional problems” (d = 1.03 and 1.01 for the parent report and self-report, respectively).

parent report compared to the self-assessment and significantly different between subjects. In the parent report for the 7- to 17-year old, the two-year course of PHC could be best described by a slightly increasing linear trend which decelerates somewhat over time. In the self-report for the 11- to 17-year olds, the mean increasing linear trend was more pronounced, but significantly different for each subject. The trajectories could be explained to a significant proportion by predictors already known, the most important role played mental health problems of the subjects. With respect to the range of the PHC scale with 14 symptoms, the average complaint levels in children and adolescents were rather low, confirming the results of previous studies [e.g. 5, 11, 13]. The descriptive course of PHC in the parents’ report in Model 1 increased linearly (d  = 0.293) and decelerated slightly at the same time (d = 0.15). Baseline scores and slopes differed significantly and were correlated initially.

Discussion Key results could be summarised as follows: on average, the level of PHC was relatively low, slightly lower for the

13

3

 For the discussion, significant effects were standardised using Cohen’s d (d ≥ .20 small, .50 medium, .80 large, [27]).

Eur Child Adolesc Psychiatry

The most significant model improvement was achieved with time-varying factors in Model 2, which explained the differences in intercepts and the interaction of intercepts and slopes, especially for mental health problems of the child (d = 1.02, low or very low effect sizes for the other L1 predictors). This is basically in line with what has been identified in other investigations [e.g. 6, 12, 14] and can be explained by specific concepts, psycho-physiological theories, learning theory, etc. (see introduction). One can only speculate why somatic problems of children and parents as well as psychosocial stress showed no effect. Bivariate correlations with the outcome (r  = 0.08, 0.03, and 0.09) indicated that dichotomous implementation of these factors was not sensitive enough to achieve a higher effect, and that the remaining partitions were explained by the other predictors in the model. As expected, age at baseline and gender explained some differences of the trajectories between children [6, 13, 14], but without improving the model substantially (d  = 0.14 and 0.15). Inclusion of the interaction term showed that age at t0 and gender should not be considered without the linear time trend. Figure 1 illustrates the estimated development of PHC per age, gender and report form for an average child, although this interpretation is limited by some differences between the age cohorts. Even though many predictors were accounted for, a significant amount of information within and between subjects remained unexplained. This might be caused by measurement errors (only satisfactorily reliability of the PHC scale, rudimentary measurement of some predictors) and/or inability to include further factors such as genetic, biological, constitutional, and environmental characteristics [4]. The results for self-reports showed both, similarities and differences by comparison with the parent reports. The general PHC score was slightly higher in the self-assessment. Over time, the mean symptom level also varied considerably between subjects, but the linear trend was lower and had a negative sign (d = −0.12) indicating a slow decrease. In the self-report, the time-varying predictors explained the whole mean trend whereas they only explained individual discontinuities in the parent report. These results and the differences in baseline scores were mainly caused by self-reported mental health problems of the child again (d = 0.92). Such differences between self-reported and parent-reported PHC are well known [2] and can be explained by the factors mentioned in the introduction. The identification of differences in individual slopes without an existing mean trend is a special strength of the chosen statistical approach that leads to a more detailed understanding of the developmental pathways. Age and gender showed a slight impact over time with lower values for boys as expected (d  = 0.21 and −0.37, [1, 4]). The lack of effects with regard to migration and SES may be explained by a lack of sensitivity of the

measurement and mutual control of predictors. It is not obvious why the linear trend suddenly turns into a positive and significant effect under these conditions, but it may be that previously uncontrolled influences of age and gender masked the positive effect. This assumption is confirmed by the introduction of the interaction terms in the final model. Several methodological limitations need to be taken into account of which the most important are: –– The PCH scale only incorporated psychosomatic health complaints and did not take somatoform or any other disorder-related constructs into account. Due to the survey method, it was not possible to conduct the corresponding clinical interviews. –– The data comprised a broad range of relevant predictors, but some of these (see list of predictors in the “Methods” section) were only obtained by one dichotomous item and some were not available from both perspectives (ibd.). Both limitations may explain the substantial residual variances in the models. –– The number of measurement points and the length of time intervals were limited, although robust against period effects. Future studies should consider longer time intervals with more measurement points and an instrument for assessing PHC with (a) subdimensions to account for the clustering of certain symptoms and (b) a higher reliability to enable identification of latent cluster structures. Further factors for within and between child levels should be taken into account to explain more of the residual variance. In addition, the role of migration background, SES, and physical problems should be investigated with multiple analyses in more detail. Psychosomatic health complaints are an existential phenomenon and should not be seen as clinically relevant unless they have reached a duration and/or intensity that impairs the affected individual in the long term. Children and adolescents suffering from PHC should be diagnosed with care, considering age- and gender-specific differences, screening for influencing factors like mental health problems, and including the parent report as well as the self-report. The data reported here provide accurate and representative information about the individual shortterm development of psychosomatic health complaints in children and adolescents in Germany. Of course, results of epidemiological surveys like this do not necessarily have a direct impact on clinical practice. However, the description of the natural course of certain symptoms and the identification of influencing factors and their precise contribution lead to a gain of knowledge helpful in improving diagnostic accuracy, planning and further developing an effective primary prevention, regulating health services, and identifying future research priorities [28].

13



Key points 1. A longitudinal observation of individual trajectories of psychosomatic health complaints (PHC) in a representative, population-based sample of children and adolescents are missing. 2. The mean level of PHC is relatively low, slightly lower for the parent report compared to the self-assessment and significantly different between subjects. 3. In the parent report, the 2-year course of PHC could be best described by a slightly increasing linear trend which decelerates somewhat over time. 4. In the self-report, the mean increasing linear trend is more pronounced, but significantly different for each subject. 5. This study provides nationwide representative data on the individual short-term development of PHC in children and adolescents. Conflict of interest  The authors declare that they have no conflicts of interest.

References 1. Eminson DM (2007) Medically unexplained symptoms in children and adolescents. Clin Psychol Rev 27:855–871 2. Barkmann C, Braehler E, Schulte-Markwort M, Richterich A (2011) Chronic somatic complaints in adolescents: prevalence, predictive validity of the parent report and association with social class, health status, and psychosocial distress. Soc Psychiatry Psychiatr Epidemiol 46(10):1003–1011 3. Pennebaker JW (1982) The psychology of physical symptoms. Springer, Berlin 4. Beck JE (2008) A developmental perspective on functional somatic symptoms. J Pediatr Psychol 33(5):547–562 5. Rauste-von Wright M, von Wright J (1981) A longitudinal study of psychosomatic symptoms in healthy 11–18 year old girls and boys. J Psychosom Res 25:525–534 6. Aro H, Haenninen V, Paronen O (1989) Social support, life events and psychosomatic symptoms among 14–16-year-old adolescents. Soc Sci Med 29(9):1051–1056 7. Egger HL, Costello EJ, Erkanli A, Angold A (1999) Somatic complaints and psychopathology in children and adolescents: stomach aches, musculoskeletal pains, and headaches. J Am Acad Child Adolesc Psychiatry 38:852–860 8. Zwaigenbaum L, Szatmari P, Boyle MH, Offord DR (1999) Highly somatizing young adolescents and the risk of depression. Pediatrics 103(6 Pt 1):1203–1209 9. Lieb R, Zimmermann P, Friis RH, Hoefler M, Tholen S, Wittchen HU (2002) The natural course of DSM-IV somatoform disorders and syndromes among adolescents and young adults: a prospective-longitudinal community study. Eur Psychiatry 17(6):321–331 10. Poikolainen K, Aalto-Setaelae T, Marttunen M, Tuulio-Henriksson A, Loennqvist J (2000) Predictors of somatic symptoms: a five year follow up of adolescents. Arch Dis Child 83(5):388–392

13

Eur Child Adolesc Psychiatry 11. Dhossche D, Ferdinand R, van der Ende J, Verhulst F (2001) Outcome of self-reported functional-somatic symptoms in a community sample of adolescents. Ann Clin Psychiatry 13:191–199 12. Isshiki Y, Morimoto K (2004) Lifestyles and psychosomatic symptoms among elementary school students and junior high school students. Environ Health Prev Med 9(3):95–102 13. Essau CA (2007) Course and outcome of somatoform disorders in non-referred adolescents. Psychosomatics 48(6):502–509 14. Steinhausen HC, Winkler Metzke C (2007) Continuity of functional somatic symptoms from late childhood to young adulthood in a community sample. J Child Psychol Psychiatry 48:508–513 15. Hagquist C (1990) Psychosomatic health problems among adolescents in Sweden—are the time trends gender related? Eur J Public Health 19(3):331–336 16. Santalahti P, Aromaa M, Sourander A, Helenius H, Piha J (2005) Have there been changes in children’s psychosomatic symptoms? A 10-year comparison from Finland. Pediatrics 115(4):e434–e442 17. Antonaci F, Voiticovschi-Iosob C, Di Stefano AL, Galli F, Ozge A, Balottin U (2014) The evolution of headache from childhood to adulthood: a review of the literature. J Headache Pain 15:15 18. Schulte IE, Petermann F, Noeker M (2010) Functional abdominal pain in childhood: from etiology to maladaptation. Psychother Psychosom 79:73–86 19. Ravens-Sieberer U, Otto C, Kriston L, Rothenberger A, Döpfner M, Herpertz-Dahlmann B, Barkmann C, Schön G, Hölling H, SchulteMarkwort M, Klasen F, the BELLA study group (2014) The longitudinal BELLA study: design, methods and first results on the course of mental health problems. Eur Child Adoles Psych. doi: 10.1007/ s00787-014-0638-4 20. Haugland S, Wold B (2001) Subjective health complaints in adolescence: reliability and validity of survey methods. J Adolesc 24:611–624 21. Goodman R (1997) The strengths and difficulties questionnaire: a research note. J Child Psychol Psychiatry 38(5):581–586 22. Klaghofer R, Brähler E (2001) Konstruktion und teststatistische Prüfung einer Kurzform der SCL-90-R. Z Klin Psychol Psych 49(2):115–124 23. Derogatis LR, Savitz KL (1999) The SCL-90-R, Brief symptom inventory, and matching clinical rating scales. In: Maruish ME (ed) The use of psychological testing for treatment planning and outcome assessment, 2nd edn. Erlbaum, Mahwah, pp 679–724 24. Schwarzer R, Jerusalem M (1995) Generalized Self-Efficacy scale. In: Weinman J, Wright S, Johnston M (eds) Measures in health psychology: a user’s portfolio. Causal and control beliefs. NFER-NELSON, Windsor, pp 35–37 25. Lampert T, Kroll LE, Mueters S, Stolzenberg H (2013) Messung des sozioökonomischen Status in der Studie zur Gesundheit Erwachsener in Deutschland (DEGS1). Bundesgesundheitsblatt (5/6):631–636 26. Raudenbush SW, Bryk AS (2002) Hierarchical linear models, 2nd edn. Sage Publications, Thousand Oaks 27. DeLucia C, Pitts SC (2006) Applications of individual growth curve modeling for pediatric psychology research. J Pediatr Psychol 31(10):1002–1023 28. Ahrens W, Krickekberg K, Pigeot I (2005) An introduction to epidemiology. In: Ahrens W, Pigeot I (eds) Handbook of epidemiology. Springer, Berlin, pp 3–40

Modelling trajectories of psychosomatic health complaints in children and adolescents: results of the BELLA study.

Psychosomatic health complaints (PHC) can significantly impair psychosocial development of children and adolescents and are therefore of considerable ...
483KB Sizes 1 Downloads 5 Views