The aim of the present study was to empirically derive a multidimensional model of OCD symptom structure using individual YBOCS-SC items, extending previous work by this collaborative group using YBOCS-SC a priori
categories (Hasler et al., 2007
). The new item-based model was then compared to a factor solution based on a traditional principal components analysis of YBOCS-SC categories, adjusted according to a method developed by Pinto et al. (in press)
. Data were collected from a large (n = 485), well-defined sample of adults with lifetime OCD, from 218 multiply affected OCD families, recruited as part of the OCGS. To our knowledge, this is the first exploratory DFA of individual OCD symptoms, sensitive to the dichotomous, non-normal nature of these data.
The results provide compelling evidence for a multidimensional model of OCD. The item-and category-level analyses yielded nearly identical 5-factor solutions, consisting of Symmetry/Ordering, Taboo Thoughts, Hoarding, Doubt/Checking, and Contamination/Cleaning. These homogeneous symptom clusters correspond to widely accepted and long held themes of OCD. There were few differences between the two solutions. While counting and repeating compulsions loaded on Symmetry/Ordering in the category-level analysis, items pertaining to counting and repeating routine activities did not load on any of the factors in the item-level analysis. The item that deals with symmetry obsessions with magical thinking also did not load on any factor in the item-level analysis, unlike its counterpart without magical thinking (which loaded on Symmetry/Ordering). In the item-level analysis, the household cleaning item loaded on both Contamination/Cleaning and Symmetry/Ordering, likely due to the clinical impression that this item is endorsed both by patients with contamination/harm avoidant concerns and those with “incompleteness.” Meanwhile, in the category-level analysis, the cleaning compulsions category loaded strongly on Contamination/Cleaning.
Our item-level analysis yielded a five-factor solution, as compared to the four factors derived in our previous category-level analysis in the OCGS sample (Hasler et al., 2007
). The key difference in the solutions is the separation at the item level of “taboo” (aggressive, sexual, religious) obsessions from “overresponsibility for harm” obsessions, which have traditionally been lumped in the a priori
aggressive obsessions category in category-level analyses. Over the years, the placement of aggressive, sexual, and religious obsessions in category-level factorial studies has been quite variable, likely due to the heterogeneity of the a priori
aggressive obsessions category, which contains both items targeting fear of aggressive impulses and overresponsibility for harm. Two category-level analyses (Baer, 1994
; Hantouche and Lancrenon, 1996
) reported a “pure obsessions” factor that is consistent with Taboo Thoughts. However, in both Leckman et al. (1997)
and Hasler et al. (2007)
, aggressive, sexual, and religious obsessions load with somatic obsessions and checking compulsions. Cullen et al. (2007)
reported an obsessions factor that consists of aggressive, sexual, religious, and somatic obsessions. Mataix-Cols et al. (1999)
reported separate aggressive/checking and sexual/religious dimensions in their 5-factor solution. When symptoms are factor analyzed at the item-level, as in the present study and two others (Summerfeldt et al., 1997
; Denys et al., 2004
), the fear of aggressive impulse items load with the other “taboo” obsessions (Taboo Thoughts) while the overresponsiblity for harm items load with checking compulsions (Doubt/Checking). Taboo Thoughts and Doubt/Checking are phenomenologically more homogeneous than the corresponding combined aggressive/sexual/religious/somatic/checking factor derived in Hasler et al. (2007)
and previously in Leckman et al. (1997)
The present category-level analysis (adjusted to include a category for overresponsibility for harm obsessions) yielded a solution that matches almost exactly the category-level factors retained by Pinto et al. (in press)
, using the same statistical methods, in a separate nonfamilial sample of 293 individuals with OCD from the Brown Longitudinal Obsessive Compulsive Study (Pinto et al., 2006
). The only differences between the solutions are the order of the factors in each solution and the lack of a robust loading for the somatic obsessions category in the current study. Replicating the results of previous factor analyses is important in demonstrating the generalizability of dimensions across samples (Gorsuch, 1983
). This replication suggests that the same underlying structure of OCD symptoms holds in both familial and nonfamilial affected samples, despite differences in study recruitment. In Pinto et al. (in press)
, the nonfamilial sample was collected at a single site and inclusion criteria included a “primary” diagnosis of DSM-IV OCD (defined as the disorder participants considered their biggest lifetime psychiatric problem) with participants having sought treatment for OCD. In contrast, the current familial sample was collected at six sites. Affected participants all had lifetime DSM-IV OCD but the disorder was not necessarily “primary,” and they may or may not have sought treatment. While early symptom onset was a requirement for probands in the current sample, this was not the case in Pinto et al. (in press)
. Furthermore, YBOCS-SC data consisted of lifetime symptoms in the present study and current symptoms in Pinto et al. (in press)
Like replication, familiality also affirms the validity (Robins and Guze, 1970
) of the distinctive components that make up the complex clinical presentation of OCD and supports their use in genetic studies. Access to a family sample afforded the opportunity to test the familiality of our symptom dimensions in 145 independent affected sib pairs. Significant sib-sib associations for 4 of the 5 factors provided evidence for a considerable familial effect. Hoarding (hoarding obsessions and compulsions) and Taboo Thoughts (aggressive, sexual, and religious obsessions) were the most robustly familial. These findings suggest that our factor solution describes familial components of the disorder but does not allow us to draw any conclusions about whether the source of this familiality is genetic or environmental.
Hoarding has emerged as an independent factor in most previous factorial studies (Mataix- Cols et al., 2005
). The familiality of hoarding has been previously demonstrated using data from the OCGS and the Johns Hopkins OCD Family Study, which predated the OCGS. Samuels et al. (2002)
reported that the first-degree relatives of hoarding probands had a greater prevalence of hoarding behavior than the relatives of non-hoarding probands, and both Hasler et al. (2007)
and Cullen et al. (2007)
noted a significant intrafamilial sib-sib correlation for the hoarding factor. When used as a predictor in treatment studies, the hoarding factor also stands out as being associated with poorer response to selective serotonin reuptake inhibitors (SSRIs) (Black et al., 1998
; Mataix-Cols et al., 1999
; Saxena et al., 2002
) and greater likelihood of drop out from cognitive behavior therapy (Mataix-Cols et al., 2002
), although a recent study found that hoarding and non-hoarding OCD patients responded equally well to paroxetine (Saxena et al., 2005
). Neuroimaging results indicate that the severity of hoarding is negatively correlated with glucose metabolism in the dorsal anterior cingulate cortex (Saxena et al., 2004
). In addition, hoarding is the only subphenotype for which genetic linkage results have been reported. Zhang et al. (2002)
noted significant allele sharing for the hoarding factor for loci at 4q34, 5q35.2, and 17q25.
The significant sib-sib association for the Taboo Thoughts factor is consistent with the findings of two prior segregation analyses (Alsobrook et al., 1999
; Leckman et al., 2003
), though those studies used the Leckman et al. (1997)
factor which also includes somatic obsessions and checking compulsions. As compared to the rICC
for Taboo Thoughts in the present study, Hasler et al. (2007)
reported a lower, yet still statistically significant, sib-sib correlation for their more heterogeneous obsessions/checking factor (same as the Leckman et al. (1997)
factor). Cullen et al. (2007)
reported a sib-sib association of 0.215 for their obsessions factor (which also included somatic obsessions), comparable to the rICC
for Taboo Thoughts in the present study, but the correlation was not statistically significant due to limited power (analyses based on only 35 independently affected sib pairs). Further support for this subphenotype comes from studies of treatment response. According to Alonso et al. (2001)
, high scores on a sexual/religious dimension were associated with poorer long-term outcome with SSRIs and behavior therapy. Similarly, Mataix-Cols et al. (2002)
reported that high scores on this dimension were associated with poorer behavior therapy response.
The low sib-sib association for Symmetry/Ordering, the only factor with a nonsignificant rICC
, contrasts the substantial familiality of this factor in two prior segregation analyses (Alsobrook et al., 1999
; Leckman et al., 2003
). However, unlike these other studies, the OCGS excluded probands with Tourette Syndrome. Leckman et al. (2003)
, on the other hand, focused on affected sibling pairs with Tourette Syndrome and included children/adolescents. Therefore the relatively weak association for Symmetry/Ordering in the present study, and in Hasler et al. (2007)
(also based on the OCGS sample), is likely due to our exclusion criteria since symmetry/ordering symptoms are a prominent feature of the familiality of Tourette Syndrome (Leckman et al., 2003
). The low sib-sib association for Symmetry/Ordering in the present study also contradicts the significant rICC
found for this factor in Cullen et al. (2007)
. Since the authors of that study included two symptoms from the miscellaneous compulsions category – sensory compulsions (touch, tap, rub) and motor compulsions (rituals involving blinking or staring) – in their symmetry factor, the higher sib-sib association for this dimension may have been due to the familiality associated with the added tic-like sensorimotor behaviors.
The present study overcomes many of the limitations of prior factorial studies with its large cohort of extensively characterized OCD patients, novel statistical methods, use of individual YBOCS-SC items, and the collaborative involvement of expert investigators. However, the study also has several limitations of its own. Factor analyses were conducted on a sample of 485 adults recruited as part of a family study of OCD that targeted affected sibling pairs and their first- and second-degree relatives. While the potential limitations of deriving factors from affected relatives were considered, the factor analyses were not corrected for nonindependence because when dealing with correlated data (e.g., family data, longitudinal data, etc.) that are large in number of clusters (or families) and small in cluster sizes, it is generally the case that ignoring the dependence will yield valid estimates of parameters (in this case, factor loadings) but incorrect standard error estimates. Since the standard error estimates were not of particular interest given the aims of this project, the consequences of not correcting for nonindependence in this case are expected to be minimal. The a priori
exclusion of open-ended and miscellaneous YBOCS-SC items in the factor analysis may have led to a biased representation of OCD symptoms. The proposed latent symptom groupings are limited to the manifest items available on the YBOCS-SC, though the checklist is the most comprehensive measure of OC symptoms to date and remains the standard in the field. New instruments, like the new edition of the YBOCS (YBOCS-II) and the Dimensional YBOCS (DY-BOCS) (Rosario-Campos et al., 2006
), will allow collection of symptom data in a dimensional manner, better facilitating the development of quantitative traits for genetic analyses. Though higher than previous item-level analyses, our subjects-to-items ratio limited power for the DFA. Despite the wide use of principal components analysis (applied here in the category-level analysis), this approach is sensitive to scaling and lacks a probability model. The variable decision rules for retaining factors and scoring the YBOCS-SC categories have led to discrepancies in the number of factors reported across prior studies. In addition, our sample was recruited from a variety of sources, including treatment settings and support groups, and was predominantly female (69.7%) and almost entirely Caucasian (97.1%). Therefore, the results may not generalize to community samples or more diverse groups. Finally, the retrospective assessment approach used, with an emphasis on lifetime OCD symptoms, is vulnerable to memory bias and may be confounded by age at OCD onset. However, the retrospective nature of data collection may not have been a problem considering that our findings replicated a factor structure (Pinto et al., in press), collected from a separate sample, that was based on current symptoms.
As this is the first item-level DFA for OCD symptoms, replication is required in a larger sample utilizing all YBOCS-SC items, including miscellaneous symptoms and open-ended items. Replication of affected sib pair associations for the 5-factor model is also strongly recommended. Further research on the temporal stability of these factors, currently under way, would be another important indicator of the validity of these symptom dimensions. Our sample was restricted to affected adults in the OCGS; upcoming studies should extend these findings to child/adolescent samples as well as individuals with subclinical symptoms.
In summary, based on the consistency between the item- and category-level analyses, as well as the significant familiality demonstrated for 4 of the symptom dimensions, our results represent a logical phenotypic starting point for future genetic studies of OCD. In fact, linkage analyses incorporating these dimensions are forthcoming from collaborators within the OCGS. The use of homogeneous, replicable, and familial subphenotypes will considerably increase the power of such genetic analyses.