Psychometric evaluation of a parent-rating and self-rating inventory for pediatric obsessive-compulsive disorder: German OCD Inventory for Children and Adolescents (OCD-CA)

Obsessive-compulsive disorder (OCD) is a severe mental disorder, characterized by obsessions, compulsive rituals, or both. Its prevalence rate in childhood and adolescence lies at approximately 1 to 4% [1, 2], and up to half of adult patients diagnosed with OCD report an onset of the disorder during childhood or adolescence [3]. To identify symptoms and treat the disorder as early as possible, appropriate assessment instruments for pediatric OCD are needed. OCD symptoms lead to a high psychological strain, distress and psychosocial impairment in children and adolescents [4], and considerably interfere with quality of life [5]. These serious consequences of the disorder have encouraged clinicians and researchers to develop new assessment instruments [6].

Several pediatric OCD-specific measures have been developed, which assess the self-report of children and adolescents only [7‐10]. Most of these measures showed satisfactory internal consistencies and there is at least some support for their convergent and/or divergent validity. However, there is a need to assess OCD symptoms as rated by parents and children separately, because younger children may be unable to report their OCD symptoms accurately. Moreover, some children and adolescents may not report their symptoms accurately due to shame and embarrassment about their OCD [11]. On the other hand, parent reports may give underestimations because some symptoms (e.g. recurrent thoughts) are more difficult for parents to notice [12].

Overall, correlations between parent ratings and self-ratings have usually been found to be low, both in the assessment of mental health problems in children and adolescents generally (e.g. [13]) and in the assessment of OCD symptoms in particular [11]. Thus, to achieve a comprehensive clinical picture of the disorder, a multiple-informant assessment is required.

Therefore, researchers have recently developed questionnaires which encompass both self- and parent reports (child-report version and parent-report version of the CY-BOCS, CY-BOCS-CR, CY-BOCS-PR [14]; Children’s Obsessional Compulsive Inventory, CHOCI/CHOCI-R [15, 16]. Satisfactory internal consistencies have predominantly been reported for these questionnaires. However, analyses in a community sample revealed poor internal consistency for the Obsession and the Compulsion subscales and the Total scale of the CY-BOCS-CR [17]. Support for convergent and/or divergent validity was found for both instruments. However, only global scores for OCD symptoms or obsessive symptoms and compulsive symptoms were derived from these rating scales, while scales assessing different domains (e.g. controlling, washing) are not provided. This is also true for the only self- and parent-rated instrument developed for the German-speaking countries—the SBB-ZWA (Selbstbeurteilungsbogen für Zwangsspektrum-Störungen and the FBB-ZWA (Fremdbeurteilungsbogen für Zwangsspektrum-Störungen) [18].

Overall, none of these self-rated or parent-rated scales fulfill the criteria for a well-established assessment tool according to the criteria for evidence-based assessment (EBA; i.e.: reliability and validity must have been presented in at least two peer-reviewed articles by different investigators [19, 20]. Currently, the clinician-rated Children’s Yale-Brown Obsessive-Compulsive Scale (CY-BOCS [21]) is the only pediatric OCD-specific measure that can be classified as a well-established assessment according to these criteria [22].

In sum, despite the variety of self-report and parent-report forms for the assessment of pediatric OCD symptoms and severity/impairment, there is, to the best of our knowledge, only one measure, the Obsessive Compulsive Inventory-Child Version (OCI-CV) [7], that focuses on symptom frequency across symptom domains. However, The OCI-CV only exists in a self-report form. Clearly, there is a lack of instruments assessing symptoms across common OCD domains, and there are no measures that record both self- and parent report regarding OCD symptom domains. To gain a comprehensive clinical picture of the child or adolescent, however, the assessment should encompass multiple informants and perspectives.

Therefore, the current study examined an inventory to assess OCD symptoms in children and adolescents across common OCD domains, the OCD-CA (OCD Inventory for Children and Adolescents) [23], which is rated by children and parents separately and is based on the Padua Inventory-Washington State University Revision [24].

The main goals of the study are to: (1) identify the factor structure of the self-report and the parent-report form of the OCD-CA, (2) assess internal consistency of the subscales and the Total scale derived from factor analyses, (3) assess the correlations between the subscales for each informant, (4) assess the correlations between parent ratings and self-ratings, and (5) evaluate convergent and divergent and discriminant validity of the scales.

Confirmatory factor analyses in the CLIN (patients with OCD and patients with other psychological disorders) and the OCDS based on the factor structure found in the analyses of Waclawiak (2006, unpublished) did not reveal any satisfactory model fit. In none of the samples were all cut-off criteria for an acceptable model fit achieved (see Additional file 1).

Thus, exploratory principal component analyses with varimax rotation were conducted on the OCD-CA in the CLIN, separately for the parent form and the self-report form (Additional file 2). Data of the OCD-CA parent form consistently met criteria for conducting a factor analysis (Kaiser–Meyer–Olkin (KMO) = .90, measure of sampling adequacy coefficient: .76 ≤ MSA ≤ .96, Bartlett’s test of sphericity: x² = 7077.69, df = 630, p < .001). The MAP test and parallel analysis determined four factors to be extracted. Therefore, an exploratory principal component analysis extracting four factors was applied. The four extracted factors (Additional file 2) had eigenvalues greater than 1.95 and explained 54.04% of the variance. The first factor explained 17.40% of the variance (.57 ≤ factor loadings ≤ .88) and included nine items, which describe contamination obsessions and washing compulsions (Contamination & Washing). The second factor explained 14.30% of the variance (.43 ≤ factor loadings ≤ .75) and consisted of 11 items describing obsessions and compulsions concerning catastrophes and injuries (Catastrophes & Injuries). The third factor explained 11.39% of the variance (.36 ≤ factor loadings ≤ .73) and contained seven items describing checking compulsions (Checking); item 22, describing hoarding and saving, also loads highly on this factor. The fourth factor explained 10.96% of the variance (.43 ≤ factor loadings ≤ .69) and contained five items describing ordering/arranging and repeating compulsions (Ordering & Repeating). Three further items regarding counting (items 20–21) and not getting ready (item 23) also load highly on the fourth factor. An additional exploratory principal component analysis with four extracted factors excluding items 20–23, which did not fit to any of the described factors in terms of content, showed the same results.

Data of the OCD-CA self-report form also met criteria for conducting a factor analysis (Kaiser–Meyer–Olkin (KMO) = .88, measure of sampling adequacy: .76 ≤ MSA ≤ .93, Bartlett’s test of sphericity: x² = 3956.82, df = 630, p < .001). The MAP test suggested that five factors should be extracted. The five-factor solution did not show any meaningfully interpretable result. Parallel analysis determined four factors to be extracted. Thus, in line with the parent form, an exploratory principal component analysis extracting four factors was applied. The four-factor solution of the OCD-CA self-report form (Additional file 2) showed the following results: The four extracted factors had eigenvalues greater than 1.83 and explained 50.05% of the variance. The first factor explained 14.80% of the variance (.26 ≤ factor loadings ≤ .75) and contained six items regarding checking compulsions. A further eight items also had substantial loadings on the first factor. The second factor explained 13.67% of the variance (.54 ≤ factor loadings ≤ .78) and included nine items which describe contamination obsessions and washing compulsions. The third factor explained 10.91% of the variance (.40 ≤ factor loadings ≤ .72) and included five items describing ordering/arranging and repeating compulsions. Items 18, 20, 21, and 25, which describe compulsions regarding checking, counting and compulsions concerning catastrophes and injuries, also load (highly) on this factor. The fourth factor explained 10.67% of the variance (.45 ≤ factor loadings ≤ .74) and contained four items which describe obsessions and compulsions regarding catastrophes and injuries. Item 17 (“count and recount money”) and item 22 (“hoarding and saving”) also load highly on this factor. Although six further items describing obsessions and compulsions concerning catastrophes and injuries load on the fourth factor, all six actually load higher on other factors.

To sum up, the self-report form showed a less clear factor structure than the parent form. The factor structure of the parent form was broadly found in the self-report (see Additional file 2). For this reason, the factor structure of the parent form was used for scale formation. As items 20–23 (regarding “counting”/“certain number”, “hoarding and saving” and “not getting ready”) did not match to any of the described factors in terms of content, they were not included in any of the subscales but were included in the Total scale.

Exploratory principal component analyses with varimax rotation were also conducted in the OCDS, showing the same factorial solution as described for the CLIN above. Furthermore, exploratory principal axis factoring with varimax rotation also revealed no differences in the results.

To confirm the four-factor solution found in exploratory factor analyses, confirmatory factor analyses were conducted once again. Correlation paths between the factors were allowed. The x² test was significant for the parent form in the CLIN (\({\text{x}}^{2}_{{\left( {df = 458} \right)}}\) = 1503.170, p = .001) and OCDS (\({\text{x}}^{2}_{{\left( {df = 458} \right)}}\) = 1024.023, p = .001). Further fit-indices (except for the TLI in the OCDS) indicated an acceptable factorial validity of the model (CLIN: RMSEA = .08, SRMR = .08, CFI = .83, TLI = .82; OCDS: RMSEA = .08, SRMR = .09, CFI = .80, TLI = .78).

Except for the SRMR (CLIN: .08, OCDS: .09), no fit indices met cut-off criteria for the self-report (CLIN: \({\text{x}}^{2}_{{\left( {df = 458} \right)}}\) = 1285.319, p = .001, RMSEA = .09, CFI = .74, TLI = .72; OCDS: \({\text{x}}^{2}_{{\left( {df = 458} \right)}}\) = 1013.752, p = .008, RMSEA = .09, CFI = .71, TLI = .69).

Table 2 shows the internal consistency (Cronbach’s alphas) and the ranges of the item-total correlations for the OCD-CA subscales and the Total scale (parent form and self-report form) across the CLIN, OCDS and COS. The Cronbach’s alpha values of the subscales and the Total scale (regarding both age groups) in the parent form were acceptable to excellent across the samples (CLIN: .78 ≤ α ≤ .94; OCDS: .74 ≤ α ≤ .93; COS: .77 ≤ α ≤ .93). The self-report form also had acceptable to excellent internal consistency, with the exception of the subscale Ordering & Repeating in the COS (CLIN: .74 ≤ α ≤ .93; OCDS: .70 ≤ α ≤ .92; COS: .55 ≤ α ≤ .91). Item-total correlations were generally satisfactory. Although several items had low item-total correlations (rit < .30), excluding any of these items did not noticeably change the Cronbach’s alpha.

The intercorrelations of the subscales in the parent form (Additional file 3) yielded different results across the samples. In the CLIN, the subscales showed low to high intercorrelations (.23 ≤ r ≤ .61). All intercorrelations were significant at a level of .01 (except for the intercorrelation between the subscale Contamination & Washing and the subscale Checking, p < .05). In the OCDS, low and moderate intercorrelations emerged (.05 ≤ r ≤ .51, partially significant at a level of p < .01 or p < .05). High intercorrelations were found in the COS (.55 ≤ r ≤ .71, p < .01). The intercorrelations of the subscales in the self-report form (Additional file 4) yielded similar, comparable results across the samples. Subscales showed moderate to high significant intercorrelations (.32 <= r <=.71, p < .01), with the exception of the subscales Contamination & Washing and Ordering & Repeating in the OCDS (r = .28, p < .01, low and significant correlation).

The correlations between the corresponding OCD-CA subscales and Total scores of the parent form and self-report form (Table 3) were generally moderate to high and significant (.32 ≤ r ≤ .68, p < .01), with the exception of the correlations of the corresponding subscales Contamination & Washing (r = .27, p < .01) and Catastrophes & Injuries (r = .29, p < .01) in the COS, which were significant but low.

The aim of this study was to examine the psychometric properties of a new parent-rated and self-rated inventory for pediatric obsessive-compulsive disorder, the OCD-CA, across a clinical sample comprising an OCD subsample and a non-OCD clinical subsample, as well as a community sample. For the total clinical sample and the OCD subsample, confirmatory factor analyses were unable to replicate the factor structure found in a community sample in a previous study (Waclawiak 2006, unpublished). Thus, exploratory principal component analysis with varimax rotation was conducted, resulting in a four factor-solution: (1) Contamination & Washing, (2) Catastrophes & Injuries, (3) Checking, and (4) Ordering & Repeating. Internal consistency was acceptable to excellent for all subscales (except for the self-report subscale Ordering & Repeating in the COS) and for the Total scale across the samples (CLIN, OCDS, COS). Therefore, internal consistency is comparable to that of other OCD-specific assessment instruments examined in OCD patients (e.g. Scahill et al. [21]; Storch et al. [14]). In contrast to the CY-BOCS-CR [17], but in line with the OCI-CV [7, 42‐44], good internal consistency was also confirmed in a community sample.

Intercorrelations between the subscales mainly lay at r ≤ .70, with the exception of those between the subscales Catastrophes & Injuries and Checking (r = .71) and Checking and Ordering and Repeating (r = .71) in the COS (parent form: 11–18 years old), and between Catastrophes & Injuries and Checking (r = .71) in the CLIN (self-report). The intercorrelations of the self-report subscales in the OCD subsample were similar to or higher than those found in analyses of the OCI-CV [42].

Thus, subscales of the OCD-CA are generally sufficiently independent of each other [45].

The correlations between the corresponding OCD-CA subscales and Total scale of the parent form and self-report form were generally moderate to high and statistically significant, which is in line with results reported by Shafran et al. [15], Uher et al. [16], and Storch et al. [8].

In the OCD subsample, self-rated and parent-rated corresponding scales only demonstrated significant mean differences on two scales with opposite tendencies, while Storch et al. [8] demonstrated significantly lower self-rated scores than parent-rated scores in an OCD sample. However, significant mean differences between informants were found across all scales in the COS, with children/adolescents providing higher scores than their parents. It might be assumed that children/adolescents from a mainly healthy population have not discussed the assessed OCD symptoms with their parents, while those affected by OCD (and who have already visited outpatient departments) are likely to have communicated with their parents about their obsessions and compulsions. This finding might also indicate that some of the symptoms of OCD (e.g. obsessions) might be more difficult for other people to detect [12].

With regard to convergent validity, the self-reported OCD-CA Total score correlated moderately with the clinician-rated CY-BOCS-D Total Score in the OCD sample. In other studies, moderate to large correlations between pediatric OCD assessments and the CY-BOCS were only found when the assessed instruments also focused on more global severity assessment, unrelated to the number and type of symptoms (e.g. CHOCI Impairment Scale [15]). Instruments assessing OCD symptoms in different domains usually found lower correlations with the CY-BOCS Rating Scale Total Score [7, 42, 46]. In contrast, parent ratings on the OCD-CA did not correlate with the CY-BOCS-D Total Score. This difference between parent ratings and self-reports on the OCD-CA may be due to the fact that the clinicians rated the CY-BOCS-D primarily based on an interview with the child or adolescent.

The correlations between the OCD-CA scales and the corresponding CY-BOCS-D Checklist scales (also focusing on OCD symptom dimensions) were higher than correlations with the Total scale of the CY-BOCS-D Rating Scale.

Correlations between the OCD-CA Total scores (parent- and self-reported) and measures of internalizing problems, depressive symptoms and anxiety symptoms were predominantly moderate to high across samples, which is in line with other studies [7, 8, 46].

To sum up, correlations between the OCD-CA and the CY-BOCS-D as well as measures of internalizing problems, depressive symptoms and anxiety symptoms provided support for convergent validity.

Discriminant validity of the OCD-CA was confirmed by (negative) low to moderate correlations between the self-report/parent form and the subscale Externalizing Problems of the CBCL and YSR. Other studies found exclusively low correlations between pediatric OCD measures and the subscale Externalizing Problems of the CBCL (e.g. Storch et al. [8]).

Regarding discriminant validity, in line with expectation, the OCD-CA scores in the OCD subsample were significantly higher than those in the non-OCD subsample and the COS sample.

The strengths of the current study include the evaluation of a new pediatric OCD-specific assessment, including a self-report and a parent-report form, across three samples (CLIN, OCDS, COS) with large sample sizes. However, some limitations should also be mentioned: First, with regard to the samples, the COS was not a representative sample, and the CLIN consisted mainly of patients with tic disorders and OCD as the data were collected at the corresponding outpatient departments of the described institutions. Second, the exploratory factor analysis did not show an adequate fit for any clearly interpretable model for the self-rated OCD-CA. Furthermore, except for the SRMR, the values resulting from the confirmatory factor analysis did not indicate goodness of fit of the model. Accordingly, the factorial validity of the self-report form could not be confirmed. Nevertheless, based on the parent report model, reliability and validity of the self-report form were confirmed. Overall, internal consistency, factorial validity (for the parent version only), and convergent und divergent validity of the new rating scale were confirmed. However, the OCD-CA should be examined further by other research teams based on the EBA criteria.

Due to the lack of instruments assessing self-rated and parent-rated symptoms across common OCD domains, this study aimed to evaluate a German version of the Padua Inventory-Washington State University Revision which enables to measure pediatric OCD and records both self- and parent report regarding OCD symptom domains. Accordingly, the OCD-CA supports multiple-informant assessment to achieve a comprehensive clinical picture of the disorder. Overall, the results of the present study show that the OCD-CA is a promising, valid and reliable instrument to assess self-rated and parent-rated pediatric OCD symptoms in clinical and non-clinical (community) populations.