Obsessive-compulsive disorder (OCD) is a common neuropsychiatric disorder, affecting 2% of adults and between 1% and 2% of children (1
). Although the DSM-IV-TR definition is straightforward, OCD is phenomenologically heterogeneous and etiologically complex (4
). OCD-affected individuals exhibit a wide variety of symptoms (e.g., contamination, sexual, religious, or aggressive fears, hoarding, checking behaviours, repeating rituals) and a range of comorbid neuropsychiatric conditions (including tic disorders, depression, generalized anxiety, grooming disorders, eating disorders and others) (10
In attempting to better understand OCD, investigators have used data reduction methods, most commonly factor analysis, to identify subgroups of symptoms (as defined by the Yale Brown Obsessive Compulsive Scale (YBOCS) symptom checklist) that may be amenable to etiological and treatment studies (6
). However, while factor analysis seeks the underlying structure in variables
, other approaches, such as latent class analysis (LCA), can be used to find latent homogeneous groups of individuals
and provide an additional dimension of analysis. This has the advantage of potentially refining the OCD phenotype, ultimately increasing our ability to identify the underlying causes of OCD, provide specific targeted treatments, and predict outcomes . While LCA has been used on symptom data from other neuropsychiatric disorders, most notably attention deficit-hyperactivity disorder (33
), studies examining the latent class structure of OCD are scarce, and there has only been one study published that examines the latent class structure of obsessive-compulsive symptoms (21
In a pioneering application of LCA to OCD published in 1991, Thomsen and Jensen (27
) used four variables in an LCA (i.e. neurological signs, EEG abnormalities, attention deficit and developmental disorder) to examine birth complications and neurological abnormalities in individuals with OCD and controls. With only four measures, they were limited to at most a two-class model, which they identified as an “organic class” and a “non-organic class”; individuals with OCD primarily fell into the “non-organic class”, suggesting that OCD is not likely to be the result of organic brain disease.
Two further studies of OCD using LCA were published in 2003 and 2008 by Nestadt et al., who conducted two analyses examining latent classes in OCD-affected individuals and their family members based on patterns of comorbidity (21
). The first study examined 80 OCD-affected individuals, 73 control subjects, and family members from both the cases and controls (total N=450), and identified four latent classes, a minimal disorders class, a recurrent major depression and generalized anxiety class, a highly comorbid class, consisting of individuals with multiple comorbid psychiatric disorders, and a tic disorder, panic, and agoraphobia class (21
). The authors suggested that the first three classes represented a single subgroup distributed ordinally along a severity spectrum, while the fourth class, the tic disorders class, represented a separate subgroup. In the second study, Nestadt et al. assessed a larger group of OCD-affected individuals (N=706), again using comorbid disorders, and examined the relationship of the resulting latent classes to specific clinical characteristics such as gender, age at onset, and OCD symptom type (41
). This study identified two possible solutions, a two-class solution that was characterized by lesser and greater comorbidity classes (similar to the classes identified in the 2003 study), and a three-class solution that consisted of an OCD-only class (±major depression), an OCD+tics class, and an OCD+affective disorders (highly comorbid) class (41
). When plotted according to prevalence of comorbid disorders, the three-class solution was also consistent with a severity spectrum profile, with individuals in class 1 endorsing low rates of comorbid disorders and individuals in class 3 endorsing high rates of comorbid disorders. The exception to this was the pattern of tic disorders, which were more prevalent in class 2 than in the other classes (41
). In general, individuals in the more severe classes had a younger age at onset, more OCPD features, and more ordering/symmetry and taboo symptoms; individuals in class 2 were more likely to be male compared with the other classes.
The most recent study, published in 2009 by Althoff et al., used LCA to examine the latent structure of the eight-item obsessive-compulsive scale (OCS) of the Child Behavior Checklist (CBCL) in several unselected community-based samples of children, including twin pairs (42
). The authors hypothesized that the LCA would parallel the known factor structure of the CBCL-OCS, and would identify two latent classes. However, the results instead suggested a four-class solution, including a “no symptoms” class, with low endorsement across all eight items, a “worries and has to be perfect” class, which the authors hypothesized consists of individuals with anxiety unrelated to OCD, a “thought problems” class, consisting of individuals who endorsed repeating behaviors, strange ideas, strange behaviors, and obsessions, and an “OCS items” class, which consisted of individuals who highly endorsed all eight items. All of these classes were heritable in the twin samples, and the authors concluded that the OCS items class represented a latent class for obsessive-compulsive behavior that may be a useful alternative to DSM-based diagnoses of OCD for genetic studies. The paucity of OCD studies using a person-centered analysis is surprising given the wide range of symptom manifestations that fall under OCD as such an analysis would indicate which patients tend to have similar symptoms. One reason for this may be that few studies have had sufficient sample sizes for such work.
Given this lack of research on the grouping of OCD-affected individuals by symptomatology, the primary aim of this study was to examine latent groups based on obsessive-compulsive (OC) symptoms using LCA in a large heterogeneous sample, and to examine the relevance of the derived latent classes by assessing their relationships to clinically relevant variables such as gender, symptom severity, presence or absence of comorbid tic disorders, family history of OCD symptoms, and treatment response. We hypothesized that the LCA would parallel the observed clinical heterogeneity as well as the results of the variable-centered data reduction analyses (i.e., factor analyses), that is, that we would identify latent classes characterized by groups of items which were highly endorsed for one class and not for the others (i.e, a cleaning/contamination class, a hoarding class, a taboo[aggressive, sexual, and religious obsessions] class, a doubts and checking class, and a repeating rituals and superstitions class) (10
). Further, we hypothesized that the latent classes would have specific clinical and/or demographic profiles. That is, the hoarding class would have a poorer treatment response, the contamination class would be more prevalent in women, and the taboo class would have an earlier age of onset (16