The present study was conducted with the aim of identifying sex-specific trajectories of paediatric headache in a longitudinal study with a population-based sample. The statistical method of LCGA was used in order to classify children according to their initial headache level and their pathways across time. The identification of genetic, somatic and (socio-)psychological predictors can be the first step for planning effective preventive actions and considering possible treatment programmes. This is the first study to conduct sex-specific longitudinal analyses in a paediatric headache sample of children and adolescents (9 to 14 years old at first wave) using this kind of method and a combination of predictors.
Results from LCGA
Concordant to our hypotheses we did not find more than four classes for either subsample. For girls a four-class solution was identified. This is in line with Dunn et al. [
33]. Here also four classes were extracted.
In accordance with our hypotheses, one class was found with girls reporting no headache across all four assessment points (‘no pain class’; 30.5 %). The ‘moderate decreasing pain class’ (32.5 %) was comprised of girls showing moderate to rather low headache frequencies across time, with significant changes. Post-hoc tests revealed that the majority of girls belonging to the ‘no pain class’ and the ‘moderate decreasing pain class’ were not disabled by headache in their daily or school activities and were able to pursue their school or occupational tasks. This result corresponds to comparable prior studies that found no or a low incidence of paediatric headache in the majority of children of population-based samples [
33,
56,
61].
Consistent with our hypotheses, we also identified one group of girls with a high headache frequency at first wave and across all other time points (‘high pain class’; 20.8 %). In post-hoc analyses, we carried out several t-tests to compare the headache-related functional disability between the extracted classes at wave four. As compared to the ‘no pain class’ and the ‘moderate decreasing pain class’ girls of the ‘high pain class’ showed significantly higher impairments in daily and school activities due to headache (all p < 0.001).
The ‘increasing pain class’ (16.2 %) was defined by a very low starting level and a significant increase across time. In this respect this class differed from the other classes with rather stable trajectories and only slight changes. As compared to the ‘no pain class’ and the ‘moderate decreasing pain class’ this group of girls reported significantly higher impairments in daily (all p < 0.05) and school activities (all p < 0.001) and suffered from significantly more disability days due to headache (all p < 0.01).
Whereas for most of the cases belonging to the ‘no pain class’ and ‘moderate decreasing pain class’, preventive or therapeutic interventions would probably not be necessary, girls of the ‘increasing pain class’ and ‘high pain class’ seem to be in special need of attention.
In the boys’ subsample, a two-class solution was identified as best-fitting model. In accordance with our hypotheses the ‘no pain class’ (48.6 %) included boys that reported to have had no headache across all time points. In this way, this class is comparable to the ‘no pain class’ of the girls’ subsample. Additionally, we found one class with a moderate headache frequency and significant but small changes across time (‘moderate stable pain class’; 51.4 %). Post-hoc analyses showed that the majority of boys of both classes negated headache-related disability concerning school or social activities at wave four. Referring to the observed time period of four years, there seems to be no need of involving boys from the current study in preventive programs.
The number as well as the patterns of change of the extracted classes in boys and girls reflect sex-specific differences in prevalence of headache and associated pain-related disability between boys and girls. These sex-discrepancies are in accordance with many prior findings [
5,
62].
Results from multinomial logistic regression analyses
In the girls’ subsample, migraine, parental headache and depression/anxiety were significant predictors for the categorisation in the ‘moderate decreasing pain class’ and ‘high pain class’ as compared to the ‘no pain class’. The relevance of these three predictors in the development and maintenance of headache was repeatedly confirmed in several prior studies [
21,
63,
64]. Hence, it seems a stable and reliable finding. Furthermore, several studies showed that being diagnosed with migraine is in general associated with diverse somatic, psychological, behavioural and social impairments [
11,
63,
65,
66].
Moreover, prior studies reported the importance of parental headache in the development and maintenance of paediatric headache [
23,
67]. Besides a familial transfer of a genetic disposition for headache, pain-associated learning mechanisms have repeatedly been discussed [
68]. It is assumed that pain-related modelling, i.e., the parental perception of and coping with pain, influence the child’s cognitive, emotional, and behavioural coping with pain and thus can exert an influence on the child’s pain intensity, frequency, and pain-related disability [
68].
Furthermore, the probability of belonging to the ‘moderate decreasing pain class’, the ‘high pain class’ and the ‘increasing pain class’ as compared to the ‘no pain class’ was significantly elevated for girls that showed depressive/anxiety symptoms at the first wave. It can be assumed that the reported depressiveness/anxiety is associated with a higher sensitivity for pain symptoms, a biased attention towards pain and a rather dysfunctional evaluation of pain (pain catastrophizing) [
69,
70]. The association between headache and depression/anxiety was repeatedly confirmed in several studies [
11,
71].
Life events and dysfunctional stress coping turned out to be significant predictors for the ‘high pain class’ but not for the ‘moderate decreasing pain class’ in girls. Post-hoc analyses revealed significantly higher mean values for life events as well as dysfunctional stress coping in the ‘high pain class’ as compared to the ‘moderate decreasing pain class’ (all
p < 0.001). It may be assumed that girls of the ‘high pain class’ were more often confronted with stressful life events without being able to use functional coping strategies. This interpretation is supported by former studies [
72,
73].
Because of the steady increase of headache frequency in the ‘increasing pain class’ it may be hypothesised that these children will suffer from a further increase of headache frequency during their adolescence. Thus, especially for this group, it is necessary to identify risk factors which can explain this course of symptoms. Migraine did not turn out to be a significant predictor for belonging to the ‘increasing pain class’. This may easily be explained by the non-existence of headache at first wave. Moreover, not being affected by headache may explain the irrelevance of a potential parental pain model [
68]. Only if headache symptoms are relevant aspects of the child’s daily life a parental pain model can come into effect. However, internalising symptoms as well as school burden significantly increased the risk of belonging to the ‘increasing pain class’. It may be assumed that this group describes girls being especially sensitive for the beginning puberty with increased psychosocial stressors and somatic symptoms. This enhanced sensitivity may result in depressive/anxious symptoms, a biased attention towards pain and increased stress experienced in school. These variables are all associated with headache symptoms [
11,
24]. However, since most of the included variables are not significantly associated with this trajectory class, further research is necessary in order to define additional risk factors in an adequate methodological manner. Referring to the assumption of a higher sensitivity for psychosocial processes and changes in this group of girls, it may be hypothesised that variables like a lack of social support, stress in the peer group or psychosocial experiences of loss (e.g., break-up with friends or romantic partners) may be relevant factors for an increase of headache in that specific group [
58,
74].
For boys, again, migraine and parental headache were able to differentiate between the classes. In contrast to the girls’ subsample, depression/anxiety was not significantly associated with a pain class in the boys’ data set. Our results show higher depression/anxiety scores for girls as compared to boys. This supports former studies [
9,
39]. It may be possible that the general low level of depression/anxiety in boys explains the lacking significance of this variable in the boys’ headache trajectories. However, as for girls, life events and dysfunctional stress coping turned out to be significant predictors for a pain class in the boys’ subsample. Post-hoc analyses revealed significantly higher mean values for life events (
p = 0.005) and dysfunctional stress coping strategies (
p < 0.001) in the ‘moderate stable pain class’ as compared to the ‘no pain class’. Just as for the girls’ subsample, this finding may also be explained by a lack of effective coping strategies when being confronted with stressful situations, i.e., life events.
In sum, our results point to the necessity of early identification of children and adolescents at risk of becoming significantly affected by headache. Additionally, for children suffering from headache already at baseline, the identification of possible factors that contribute to the maintenance of headache is crucial. In the light of our results especially migraine and parental headache seem to be risk factors for syndrome patterns in both sexes. Against the background of a parental influence, not only a possible genetic predisposition should be considered, but also the influence of dysfunctional pain-related modelling mechanisms. Additionally, inadequate coping with headache as well as depressive cognitive styles, both associated with negative self-instructions, a rather anxious-negative and biased attention towards pain, an increased sensitivity for pain, and tendency to ruminate about pain seem to heighten the probability of increased headache frequency and a corresponding pain-associated disability [
70,
75].
Strengths and limitations of the study
One of the main strengths of the current work is the large sample size and the random choice that allowed insight into paediatric headache trajectories in a population-based sample. Moreover, by using a special statistical method, conducting sex-specific analyses and including various relevant predictors, the scope in this study field could be extended. Insight into genetic, somatic and (socio-)psychological variables that need to be considered in preventive plans and treatment programs could be facilitated.
Another advantage of our study is the self-report that was used in order to assess paediatric headache. Several studies showed that the inclusion of the child’s perception of pain symptoms is crucial and cannot be substituted by the parental report, since there is a systematic difference between children’s and parents’ estimation, i.e., an underestimation of the child’s headache prevalence by parents [
76,
77].
One limitation that needs to be considered is the assessment of headache frequency. Our question demanded the classification of headache days in relation to a specific period of time (either the last week, the last month or the last six months) depending on the precedent answer of the respondents. Especially for younger children, it may have been difficult to give a correct estimate of the number of headache days, particularly for the categories that included a greater time period. Moreover, in order to be able to extrapolate the answers of the questioned children and adolescents to the last six months, the answers were weighted correspondingly. This resulted in great intra-individual differences over the assessment points, especially for the ‘high pain class’ since this class was comprised of girls with significant intraindividual changes across time. Future studies should avoid the necessity of weighting by creating a scale for a differentiated and valid assessment of very low as well as very high headache levels, so that the valid assessment of severely affected patients can also be considered. One possible type of assessment would be the use of a pain diary. However, the effort that would be necessary (personal contact, filling out the diary over at least four weeks) would not be compatible with the type of study and the study sample (postal survey, high number of respondents, children).
The type of assessment and the chosen sample influence possible interpretations of our results. At each measurement point a cohort of a specific age range (9–14 years at wave one) was examined. Hence, we can make statements about changes in headache across the course of the study, but not particularly concerning changes across childhood or adolescence.
In our study we used migraine as a predictor and disregarded tension-type headache (TTH). This decision was mainly based on studies revealing a higher long-term psychosocial disability due to migraine in contrast to TTH [
11]. However, future studies should compare the influence of the two headache types on paediatric headache and related disabilities over time. Moreover, we used the ICHD-II criteria to diagnose migraine since the classification of headache was based on the ICHD-II at the time of the conduction of the study. Future studies should rely on ICHD-III beta when diagnosing migraine. It is unmistakably evident that a clinical relevant diagnosis of headache has to rely on a comprehensive set of instruments for clinical examination of patients. Our method of analysis rather served to describe and differentiate headache in the general population.