Introduction

Obsessive–compulsive disorder (OCD) is an often chronic and severely disabling disorder consisting of recurrent and persistent thoughts, urges, or images (obsessions) which for most individuals cause anxiety and distress. Repetitive behavior and/or mental acts are performed to relieve associated anxiety or prevent the occurrence of feared experience [1]. Between 0.25 and 4 % of all children and adolescents are affected [2, 3]. More than half of early diagnosed patients experience a chronic course with moderate to severe symptoms that last into adulthood [4, 5]. Common treatments for OCD include exposure and response prevention therapy [6, 7] and serotonin reuptake inhibitors [8].

OCD is a heterogeneous syndrome that consists of multiple overlapping elements [9, 10]. Children’s Yale-Brown Obsessive Compulsive Scale (CY-BOCS) is the most commonly used method both to assess OCD symptoms in children and adolescents, and to monitor the treatment outcome [11]. The CY-BOCS checklist is currently divided into 19 symptom categories that measure each their variety of obsession or compulsion symptoms. This structure has, however, been challenged by factor analytic studies [12]. Interestingly, there is a difference between those who examine structure at the subscale level and those who examine at the item level. In a meta-analysis, studies based on subscales find a four-factor solution similar to that of Leckman et al. [10]: (1) symmetry; (2) forbidden thoughts including aggression, sexual and religious obsessions; (3) cleaning; and (4) hoarding. On the other hand, studies that examine structure at the item level showed a five-factor solution: (1) symmetry obsessions and repeating, ordering, and counting compulsions, as well as hoarding obsessions and compulsion; (2) aggression, sexual, and religious obsessions; (3) cleaning compulsions and contamination obsessions; (4) aggression obsessions and checking compulsions; and (5) somatic obsessions [12]. Few studies have been conducted with child and adolescent samples, and no studies have explored the structure based on individual symptoms in this age group. The importance of investigating the symptom structure in children and adolescents is clearly being highlighted by findings of developmentally heterogeneous OCD symptoms between age groups [13, 14]. The first two studies to explore the OCD symptom structure by means of CY-BOCS in a sample of children and adolescents were published in 2006. They concluded that a four-factor structure provided the best fit but with considerable difference in factor composition with one study confirming factors found in adult studies [15], and the other unable to corroborate adult findings [16]. Later studies of this age group included larger sample sizes and resulted in factor structures similar to the four-factor structure found in adults [17–19]. One of these studies conducted by Stewart et al. compared several factor structures between children, adolescents, and adults and found the four-factor structure described by Leckman [10] to be adequate for use in all age groups in spite of its imperfections [19].

The current study is the first one to look at the OCD factor structure at the item level in this age group. Hoarding is often seen in combination with OCD, and in some studies of OCD symptom structure, hoarding symptoms emerge as a factor of its own, providing some support for the concept that these symptoms belong to an etiologically distinct subgroup. Consistent with evidence from many studies—that hoarding forms a factor on its own, DSM-5 includes hoarding as a distinct disorder even though it is acknowledged that hoarding symptoms are frequently present in OCD [20].

In childhood and adolescence, OCD frequently co-occurs with other mental disorders, the most common of which are neuropsychiatric disorders, including tics, anxiety disorders, and depression [21]. Tics are a part of the familial OCD phenotype [22], and tic disorder with OCD appears to be a more severe type than tic disorder without OCD [23]. Tic disorder has been shown to occur with symmetry/ordering and aggressive/checking symptoms [24–26], and with impulsive symptoms [27]; anxiety and depression with aggressive, religious, and sexual symptoms [28]. Adult studies demonstrate that males have a tendency to religious, sexual, checking, and repeating symptoms and females to contamination symptoms [29, 30]. Identification of symptom subtypes in OCD is an important step in understanding the phenomenology of the disorder as well as being useful regarding diagnosis and treatment of the disorder, as symptom subtypes have been shown to impact treatment outcome [31].

The aims of this study were (1) to determine the factor structure of CY-BOCS at the item level in a large multinational sample of children and adolescents; (2) to test the invariance of the factor structure across age and gender; and (3) to examine the association of factors with age, gender, and comorbid psychopathology.

Methods

Participants

The sample included 854 patents with OCD recruited from USA (n = 400), Sweden (n = 184), Norway (n = 111), Denmark (n = 101), and The Netherlands (n = 58). All participants had an OCD DSM-IV diagnosis [32] based on either the Kiddie Schedule for Affective Disorders and Schizophrenia Present and Lifetime version (KSADS-PL) or the Anxiety Disorders Interview Schedule (ADIS). The mean age was 12.29 years (range 7–18), and 53 % were males. Exclusion criteria varied between sites and are presented in Online Resource Table 1; details about sample are found elsewhere [6, 33–39]. The study was approved by the national ethical committees and data authorities in all contributing countries.

Measures

OCD symptoms were measured by the use of CY-BOCS [11], a semi-structured interview designed specifically for the purpose. The interview consists of two parts. The first part poses 10 questions about obsessions and compulsions (5 of each) with the aim of measuring their severity on a five-point scale; the second part is a 74-item checklist that aims at assessing a broad range of current and past obsessions and compulsions. The CY-BOCS has been proved reliable and valid when used with samples of children [40]. Other disorders have been measured by means of the K-SADS-PL [41], among these the Swedish, Norwegian and Danish samples, and one part of the USA sample. The anxiety disorders interview schedule (ADIS) [42] was used in the Netherlands and in some of the USA sample as well.

Analysis

A calculation of Cronbach’s alpha (α) was used to estimate the reliability of the standard CY-BOCS subscales [43]. In line with previous item-level exploratory factor analyses (EFA) in adults, we included all items in the analysis except for the last one in each category, the one labeled “other” (61 items included; Katerberg et al. [44]). First the data were subjected to exploratory principal axis factoring with direct oblimin rotation so as to allow the factors to be correlated, consistent with theoretical models of OCD. This approach was chosen because it is less affected by data skewing [45]. EFA was also conducted with a varimax rotation to evaluate possible differences and to compare to previous research. A factor analysis was carried out on a tetrachoric correlation matrix suited for binary variables [46], in STATA version 13 [47]. Missing item values were imputed from the mean of significantly correlated items (p > 0.05) rounded to 0 or 1. The final number of factors was chosen according to eigenvalues, scree-plot, and fit values from the CFA solutions. Cronbach’s alpha was calculated for each new factor. In a second stage CFA was conducted by the use of Mplus Version 7.3 [48] on factor models 1-5 in order to determine which exploratory model would give the most parsimonious solution, and to estimate the invariance of the solution across gender, and child (7–12 years) and adolescent (13-18 years) groups. This approach is similar to the one used in a previous study of CY-BOCS [44]. Items were excluded from the CFA model if they loaded below 0.3 on all factors from the EFA. Cross loading items in the EFA were assigned to the factor with the highest loading. Estimates of fit were based on a range of indices—such as likelihood-ratio χ 2 test, comparative fit index (CFI), Tucker-Lewis index (TLI) [49], the root mean square error of approximation (RMSEA) [50], and weighted root mean square residual (WRMR) [51].

Factor scores, calculated by use of regression, were used in a multiple regression model with age and gender as predictors and correlated to co-morbid disorders using Pearson partial correlation, with pairwise deletion of missing data, controlled for the effect of age and gender. Comorbid disorders were selected on basis of relevance and availability and pooled from K-SADS and ADIS (Table 6). Co-morbid diagnoses were only available for a part of the sample (n = 607–665) and scored as a binary variable because we did not have measures on their severity.

Results

Demographic and clinical characteristics for each site are presented in Table 1. The Mean CY-BOCS severity score was similar to one seen in other pediatric OCD studies [32, 52, 53]. Norway had the highest proportion of males (62 %); Denmark had the lowest (37 %). There were also site-specific differences in CY-BOCS scores (Table 1).

Table 1 Demographics and clinical characteristics by country
Full size table

Many of the CY-BOCS subscales had an unacceptably low level of internal consistency ranging from 0.34 to 0.74 (Online Resource Table 2). The three-factor solution was the most parsimonious when judged on the basis of a scree-plot, Eigenvalues, fit values for the CFA, and clinical evaluation of the factor structures. Scree-plot of eigenvalues can be seen in Online Resource diagram 1. Based on their item loadings they were labeled (1) harm/sexual, (2) symmetry/hoarding, and (3) contamination/cleaning (Table 2). Two items had a loading of less than 0.3 on all factors: “fear of saying certain things” and “other self-damaging or self-mutilating behaviors”. Cross loading items were “checking tied to somatic obsession” and “fear of not saying just the right thing”. Cronbach’s alpha values for the three factors were between 0.79 and 0.81 (Table 2). When we used a varimax rotation instead of an oblimin we found no major changes aside from “fear of not saying just the right thing” now loaded on both factor one (0.33) and factor two (0.33).

Table 2 Rotated solution of exploratory factor analysis with three factors: (1) harm/sexual, (2) symmetry/hoarding, and (3) contamination/cleaning
Full size table

Fit indices for CFA models can be seen in Table 3. The three-factor solution provided the best fit measured on CFI, TLI, and χ 2 values. The four-factor solution was superior on RMSEA and WRMR measures. Items loading under 0.3 on any factor in the EFA were omitted, while all cross loading variables were kept in the CFA models. The three-factor solution was significant on the χ 2 test (p < 0.001). However, it is quite common that the χ 2 index reports unacceptable goodness-of-fit (p < 0.05), even if the data are highly compatible with the structure; therefore other indices are reported as well [54]. The four and five-factor solutions showed slightly poorer fit values than the three-factor solution, while the single and two-factor (obsessions/compulsions) solutions fitted poorly to the sample. Regarding item loading and cross loading differences between models, the four-factor solution had two items that loaded below 0.3 on all factors and 10 items loaded above 0.3 on more than one factor. In the five-factor solution three items loaded below 0.3 on any factor, and 20 items loaded above 0.3 on more than one factor. The four and five-factor solutions are presented in the Online Resource Tables 3 and 4. For comparison, a CFA was conducted on the five-factor solution provided by Katerberg and colleagues [44], resulting in a similar fit as our three- and four-factor solutions. Based on the slightly better fit indices, less cross loadings and clinical evaluation of the factor composition, we chose to proceed with the three-factor model.

Table 3 Confirmatory factor analysis of EFA models, variables loading under 0.3 on any factor excluded
Full size table

The factor solution was invariant across age and gender groups (Table 4).

Table 4 Confirmatory factor analysis of the 3-factor model for gender and age groups, variables loading under 0.3 on any factor excluded
Full size table

Females had higher score on factor two (symmetry/hoarding). Older patients had higher scores on all factors (Table 5).

Table 5 Multiple regression analysis of factor scores using gender and age at assessment as predictors
Full size table

There was no difference in factor scores between age groups on factor one (harm/sexual) (Myounger = 0.08, SD = 0.33; Molder = 0.13, SD = 0.39); t(851) = −1.899, p = 0.058. The younger group had significantly lower scores than the older group on factor two (symmetry/hoarding) ((Myounger = 0.32, SD = 0.32) Molder = 0.44, SD = 0.38); t(851) = −5.175, p = 0.000 and factor three (contamination/cleaning) (Myounger = 0.44, SD = 0.42; Molder = 0.51, SD = 0.44); t(851) = −2.386, p = 0.017.

Tics and ADHD were correlated negatively with contamination/cleaning; depression was correlated with symmetry/hoarding factor scores. Anxiety was correlated almost equally to these two factors. Correlations found were small or insubstantial (Table 6).

Table 6 Correlation of co-morbid disorders and factor scores controlled for age and gender
Full size table

Discussion

This paper reports the first item-level factor analysis of CY-BOCS in a large multinational sample of children and adolescents with OCD. The reliability of the standard CY-BOCS subscales was low for most categories, which supports the need for a revision of these subscales by means of item-level analysis. Similar to the three-factor structure found in the first studies on adults [55, 56], a three-factor solution proved to be the best model here too. Fit indices from CFA did not effectively differentiate between the three- and four-factor solution, but Screeplot and investigation of the factors suggested the three-factor solution. The four-factor solution had ten items that cross-loaded between factors making the solution unclear. Additionally, the four factor solution extracted a forth factor containing only five items which most were closely related to the items in factor one or two, as indicated by their factor loadings. This further supported the strength of the tree-factor solution over the four-factor solution. Finally, the three-factor solution was rated to be the most clinically sound solution by clinicians involved in this study. Direct comparison to previous childhood studies is difficult as the previous studies report on analyses at the subscale level. Stewart et al. [19] found the factor structure of child and adolescent OCD to fit Leckman’s four-factor model better than Baer’s original three-factor model, but the opposite is true for this study. However, our three-factor solution is highly comparable to most four-factor solutions presented in earlier studies on children and adolescents, with the exception that our symmetry/hoarding factor represents a combination of two factors—seemingly identical with the separate symmetry and hoarding factors [17–19]. Only McKay et al. found hoarding and ordering/arranging symptoms to combine as one factor in a sample of children and adolescents [16]. The reason for this difference is unclear, but it could in part be explained by the fact that hoarding is only represented by two items and will, therefore, influence the solution less than in a category analysis. The result is, however, not surprising, since a shared symmetry/hoarding factor has been evidenced in several previous studies [16, 44, 55, 57]. An alternative possibility is that hoarding symptoms are not fully developed at younger ages, which is why hoarding does not separate as a factor on its own before adulthood. Surprisingly, we found the item “Fear will be responsible for something terrible to happen” to have a higher loading on the symmetry/hoarding factor than on the harm/sexual factor, indicating that the item is too vague and may refer to two distinct types of OCD symptoms. Alternatively, a sub-category of “terrible” things are specifically related to symptoms of symmetry/hoarding. Our contamination/cleaning factor corresponds well to that found in most previous studies on children and adolescents [15, 17, 19]. However, one study found contamination obsessions and washing/cleaning compulsions to belong to separate factors [16]. In our study, somatic symptoms are divided between harm/sexual and contamination/cleaning factors and seem to be tied to both these factors, possibly indicating two distinct sub-categories of somatic symptoms. Symptoms of checking are distributed between all three factors, indicating that checking behaviors are linked with various distinct obsessional themes. Also, this could explain the varying correlation of checking symptoms to factors in previous studies based on the CY-BOCS original symptom categories.

Due to category heterogeneity, items belonging to the miscellaneous categories in both obsessions and compulsions were excluded from most previous studies. We included these items in our study and found twelve out of fourteen of them belonging to two of the extracted factors, indicating a current misplacement in the miscellaneous categories and providing further support for our three-factor model. Only one of the miscellaneous items (“Fear of not saying just the right thing”) cross loaded between factors, loading above 0.3 on harm/sexual and symmetry/hoarding factors. This may possibly represent generalized anxiety disorder (GAD) thoughts in patients with both OCD and GAD, which is a common combination. The following items correlated with factor one (harm/sexual): “Trichotillomania”, “The need to know or remember”, “Intrusive (non-violent) images” and “Excessive telling, asking or confessing”. The following items were found to correlate with factor two (symmetry/hoarding): “Measures to prevent harm to self”, “Intrusive sounds, words, music or numbers”, “Ritualized eating behaviors”, “Excessive list making”, “Excessive touching, tapping and rubbing”, “Need to do things until it feels just right”, and “Rituals involving blinking or staring” (Table 2). In this study only two miscellaneous items (“fear of saying certain things” and “other self-damaging or self-mutilating behaviors”) failed to correlate significantly to any factor. These items may need to be split and/or re-worded, as it is possible that they refer to other obsessions. “Fear of saying certain things” can, for example, refer not only to sexual content but also to magical events happening in case the word is expressed (thought-action fusion). Contrary to the notion of the miscellaneous items as less “true” OCD symptoms in the Katerberg study, where many of these items formed a factor on their own [44], locating miscellaneous items on specific symptom factors in this study supports the view that these items are in fact true OCD symptoms. However, the Katerberg study was based on an adult sample and a slightly different methodology which may explain the different outcome. Since one of the samples included in this study contains children diagnosed with Pediatric Autoimmune Neuropsychiatric Disorders Associated with Streptococcal Infections (PANDAS), we chose to exclude this sample when conducting the EFA so as to investigate whether a possible difference in symptom structure in PANDAS might have influenced our results. This was, however, not the case, and the factor structure remained stable with the sample excluded.

Overall our factor structure shows considerable similarity to the original factor solutions found in adult studies [55, 56], as well as to most studies on children and adolescents with the exception of hoarding which was associated with symptoms of symmetry in our study. For comparison, we conducted a CFA on the five-factor solution provided by Katerberg et al. [44], resulting in a CFI fit value of 0.856. This is, however, likely to be an overestimation since Katerberg et al.’s analysis included fewer CY-BOCS items because they excluded all items that cross loaded between factors in the initial EFA. The inclusion of cross loading items in our CFA could also explain the overall suboptimal fit indices. It is noticeable that the symmetry/hoarding factor is consistent with some previous item-level studies in which hoarding loads with symmetry, ordering, repeating and arranging items [44, 56, 58–61], and does not constitute a separate factor. Also, hoarding did not separate as a factor on its own in any of the solutions that we tested (supplementary Tables 3, 4). DSM-5 hoarding disorder is classified as a separate disorder and is as such no longer part of the OCD classification, although it should be noted that symptoms of hoarding are still recognized as a part of OCD if they are a consequence of typical obsessions [1]. It is possible that variation in samples regarding number of patients with co-morbid hoarding disorder and OCD related hoarding could affect the factor structure in both this and previous studies that have used category-based factor analysis which tends to isolate hoarding as a separate factor as this method is unable to detect any correlation to individual items from the CY-BOCS categories. There is, however, a limitation to CY-BOCS in that it does not distinguish between OCD-associated hoarding and hoarding disorder per se. To differentiate between hoarding disorder and OCD-related hoarding we suggest that OCD-related hoarding items should be rephrased.

In regard to the factor analysis, we consider the three-factor solution presented here to be generally compatible with factor solutions from previous studies although it suggests fewer factors than most of those. This could indicate a possible difference in symptom factor structure in child and adolescent samples, but if this is to be confirmed, more studies on individual symptoms are required. CFA model fit was acceptable for the three, four- and five-factor solutions, but since extracting more factors did not make the model stronger, the three-factor solution was considered the sturdiest (Table 3). CFA for groups of males, females, children, and adolescents did not reveal any difference in symptom structure, indicating no gender or age differences and confirming previous research [17]. Generally, the CFI and TLI values for model fit are below 0.95 which could be caused by the large number of items, their heterogeneity, or both.

We did not find support for gender predicting either contamination/cleaning or harm/sexual as would have been expected from previous research [29]. However, contrary to previous research, being female predicted higher scores on the symmetry/hoarding factor. Tic-related OCD has previously been found to correlate with aggressive, sexual, religious, and symmetrical obsessions and checking, ordering, counting, repeating, and symmetry compulsions [24–26], but this was not confirmed by our results. We did, however, find tics and ADHD to be slightly negatively correlated to the contamination/cleaning factor, thus providing some support for our factor structure. However, this is not compatible with the view of OCD as a disorder of impulsivity/compulsivity [62]. In adults, anxiety and depression have been found to correlate highly with OCD sexual and aggressive symptoms [63]. In this study, however, a low correlation was found with the symmetry/hoarding factor while generalized anxiety disorder and social anxiety disorder had a low or insubstantial correlation with the factors of harm/sexual and symmetry/hoarding. Neither anxiety disorder nor depression correlated with the contamination/cleaning factor, thereby underlining the different nature of these factors. The partial correlations between comorbid disorders and OCD symptom factors were generally low or insubstantial in our sample. Still, however, they provide some support for the observed factor structure, particularly in the need to separate the contamination/cleaning factor from the remaining two.

This study is the first to apply factor analysis to individual CY-BOCS items in a sample of children and adolescents. In addition, comorbidity data were used to support the factor structure, and a particular strength of the study strength was its ability to pool data from several clinics to obtain a large sample. On the other hand, multi-center studies pose a challenge in regard to sample homogeneity, as language and cultural differences can be associated with differential item functioning. Thus, this study’s methodology was limited by a possible selection bias, differences in exclusion criteria, and varying use of instruments between sites. In particular, heterogeneity in instruments used to evaluate co-occurring disorders may limit the conclusions that can be drawn from the comorbidity analyses. Invariance and inter-rater reliability across sites were not tested although rater interpretation for the CY-BOCS items may differ between sites. Identification of symptom dimensions in child and adolescent OCD can facilitate understanding of diverse etiological factors underlying each symptom dimension, as well as benefit future studies investigating treatment outcome, and prediction of immediate and longitudinal outcome of OCD.

Conclusion

This study found three symptom dimensions in an international sample of children and adolescents with OCD. The results are based on a very large sample and this is the first study to conduct an explorative factor analysis on CY-BOCS items in a sample of children and adolescents. The factor structure was similar across groups of gender and age. The solution is to some extent compatible with previous studies of this age group and of adults, although most previous studies found 4–5 factors. The results of this study may be of clinical importance as they strengthen the view of OCD being a heterogeneous disorder with possible differences in etiology and treatment demands that needs further research.