Use of Factor-Analyzed Symptom Dimensions to Predict Outcome With Serotonin Reuptake Inhibitors and Placebo in the Treatment of Obsessive-Compulsive Disorder

The efficacy of both exposure therapy and serotonin reuptake inhibitor antidepressant treatment in obsessive-compulsive disorder (OCD) has been well established in placebo-controlled studies. Although many have benefited from those treatments, 10% to 40% of patients do not respond to an adequate trial (1, 2). However, the existing literature on reliable predictors of treatment outcome in OCD is sparse and inconsistent. The predominance of compulsive symptoms, cleaning rituals, earlier age at onset, longer illness duration, and chronic course was found to be associated with poor response to serotonin reuptake inhibitors in some studies (3–5). In two retrospective studies, the presence of schizotypal personality was a negative outcome predictor for pharmacological (6) and behavioral (7) treatments. The presence of schizotypal, avoidant, and borderline personality disorders predicted poorer treatment outcome with clomipramine (8). However, in one study the presence of a personality disorder was not related to improvement with fluoxetine (9). Similarly, others found that the presence of various personality disorders was not related to outcome with behavior therapy (10).

The symptoms of OCD are heterogeneous, so that any given patient with this diagnosis may present with only one or, more commonly, many of these symptoms (11). Studies are, however, conflicting about whether any particular subtype of OCD is easier to treat or more likely to benefit from a particular treatment. For instance, there is some evidence that OCD patients with comorbid chronic tic disorders, and possibly those with concurrent psychotic spectrum disorders, are more likely to require the addition of a neuroleptic medication to treatment with a serotonin reuptake inhibitor (12). Recently, Jenike et al. (13) reported that monoamine oxidase inhibitors (MAOIs) may be preferentially useful for the treatment of symmetry and unusual somatic obsessions. Similarly, the presence of symmetry obsessions, ordering compulsions, and hoarding rituals predicted better response in refractory patients treated with cingulotomy (14). Patients with cleaning and checking symptoms may respond best to exposure methods, but other subgroups, such as patients with ordering compulsions, hoarding rituals, or slow patients, have rarely been included in trials of behavior therapy (15). Some studies have suggested that patients with washing symptoms may do better with exposure therapy (16, 17) and worse with serotonin reuptake inhibitors (3, 5). Checking rituals predicted poorer outcome in some behavior therapy studies (16, 18), but others have found no differences in treatment response between patients with washing and checking symptoms (19). In addition, anecdotal evidence suggests that patients with predominantly obsessive symptoms (ruminations) might respond better to medication and worse to conventional behavioral techniques, but no empirical evidence is available.

Baer (20) has proposed searching for OCD symptom dimensions, which may be present in varying degrees and combinations in any given patient, rather than categorizing patients into mutually exclusive subgroups. His factor-analytic results, later replicated in a larger independent sample (21), confirmed the multidimensional and heterogeneous nature of OCD and suggested that the identified symptom dimensions may prove useful in future research into the etiopathogenesis, genetics, and treatment outcome, overcoming the limitations of the categorical approach of earlier studies (22). The elucidation of a putative relationship between symptom dimensions and treatment response addresses an important aspect of treatment specificity by characterizing the type of patients for whom a treatment is more appropriate and might, eventually, help to reduce the percentage of treatment failures.

In an effort to replicate and extend the findings of previous studies (20, 21), we factor analyzed data on the Yale-Brown Obsessive Compulsive Scale Symptom Checklist from a large patient group (N=354). The identified dimensions were then used as predictors of outcome in a subgroup of 150 patients who had completed six placebo-controlled trials with a serotonin reuptake inhibitor. Our research questions were as follows: 1) What symptom dimensions best summarize the heterogeneous phenomenology of OCD? 2) How are those dimensions related to variables such as sex, age at onset, and comorbidity with chronic tic disorders? 3) How are those dimensions related to treatment outcome with serotonin reuptake inhibitors? and 4) Are there any significant predictors of placebo response?

METHOD

Patients

Three hundred fifty-four ambulatory outpatients with a primary diagnosis of OCD, seen at the OCD clinic of the Massachusetts General Hospital, were included in the principal components analysis. Of those, 181 patients had participated in several placebo-controlled drug trials, 72 were part of the DSM-IV field trial, and 101 were obtained through chart review. Participants consisted of 190 men (53.7%) and 164 women (46.3%); the mean age was 36.2 years (SD=12.3, range=18–76). All subjects had had symptoms of OCD for at least 1 year; mean age at onset was 15.5 years (SD=10.6). Mean scores on the Yale-Brown Obsessive Compulsive Scale were as follows: total, 21.8 (SD=6.3); obsessions subscale, 10.9 (SD=3.5); and compulsions subscale, 11.0 (SD=4.0). Data comparability analyses showed no differences in sex distribution, age, age at onset, Yale-Brown total score, and Yale-Brown obsessions subscale score. However, the patients who had participated in drug trials scored significantly higher than the chart review patients on the Yale-Brown compulsions subscale (Mann-Whitney U=7429.5, p=0.008). The diagnosis of OCD was confirmed in all cases by an experienced psychiatrist or psychologist through use of a structured diagnostic interview. Coexisting axis I disorders such as depression were not excluded, provided that obsessive-compulsive symptoms predated the onset of the coexisting conditions.

Of the patients who had participated in drug trials, 150 had completed them and were eligible for the predictors analysis. They had participated in five published double-blind, placebo-controlled trials with a serotonin reuptake inhibitor—clomipramine (23), fluvoxamine (24), fluoxetine (13, 25), and sertraline (26)—and in a sixth, unpublished study with paroxetine (M.A. Jenike et al., 1991). Since some patients had participated in more than one of those trials, only the first study they had participated in was retained. All patients underwent a physical examination and gave a complete medical and psychiatric history, as well as written informed consent, before entering those trials. No patient met criteria for major depression according to clinical interview, and each patient had a baseline score of less than 20 on the 17-item Hamilton Depression Rating Scale (27). Patients with a history of other axis I psychiatric disorders within 1 year and pregnant or lactating women were excluded from these trials. More details can be obtained in the original publications. Pretreatment demographic and clinical data for this subgroup are shown in table 1.

Clinical Measures

For the direct assessment of OCD symptoms we used the Yale-Brown Obsessive Compulsive Scale symptom checklist, which provides more than 50 examples of obsessions and compulsions and also includes a target symptom list of the four most prominent symptoms for each patient. The major outcome measures were the 10-item clinician-rated Yale-Brown Obsessive Compulsive scale (28, 29) and the National Institute of Mental Health (NIMH) OCD scale (30). In addition, the 17-item Hamilton depression scale (27) was administered. History of chronic motor and vocal tic disorders was available from all patients in the DSM-IV field trial and chart review (N=173) and was assessed with the modules designed for that purpose in the DSM-IV field trial.

Analyses

Principal components analysis was performed by using pooled data from the three different data sources (N=354), following the original methodology of Baer (20). Briefly, for each of the 13 major symptom categories of the Yale-Brown symptom checklist, if a patient did not endorse any of the specific symptoms under that heading, then that category was assigned a score of 0. If a patient endorsed at least one of the specific symptoms but did not consider it a principal problem, that category was assigned a score of 1. If a patient identified at least one of the specific symptoms as a principal or major problem, that category was assigned a score of 2. Thus, a score of 0, 1, or 2 was assigned to each of the seven major obsessive symptom categories and to each of the six major compulsive categories (miscellaneous obsessions and compulsions were excluded because each contained many heterogeneous symptoms). This method ensured that all 13 categories would have equal representation, regardless of how many specific symptom examples were provided under each. In the principal components analysis, criteria for retention of factors were eigenvalue greater than 1 (Kaiser’s criterion), interpretability of the factors, and Cattell’s Scree test. The initial factor solutions were followed with a Varimax rotation in order to facilitate their interpretation. The relationship between the rotated OCD factors and the miscellaneous obsessions and compulsions, severity of OCD symptoms, demographic variables, and comorbid diagnoses was studied by means of multiple correlation analyses, chi-square tests, one-way analyses of variance, and Mann-Whitney U tests. For multiple correlations, the conservative Bonferroni significance criterion was adopted to control for type I error.

Efficacy and predictors analyses were conducted by using pooled data from the six trials with serotonin reuptake inhibitors (N=150; N=84 for drug group; N=66 for placebo group). Since clomipramine is not classified as a selective serotonin reuptake inhibitor (SSRI), analyses were repeated excluding this study (N=27) in order to ascertain whether results were generalizable to all serotonin reuptake inhibitor medications (N=123; N=71 for drug group; N=52 for placebo group). Analyses included all patients who had at least one measurement after the baseline visit (end-point-carried-forward method). The primary analyses were multiple linear regressions with the posttreatment scores on each outcome scale (Yale-Brown scale, NIMH OCD scale) as dependent variables. The baseline Yale-Brown scale scores were forced first into the equations as a covariate in order to control for symptom severity (31). For predictors analyses, scores on each OCD factor were then entered into the equations. Separate regression models were tested for the drug and placebo groups. The stepwise method was used. A two-tailed significance level of 0.05 was set. All statistical analyses were conducted by using the SPSS statistical package (version 6.0).

RESULTS

Frequencies of the major symptom categories of the Yale-Brown symptom checklist in our study group are listed in table 2. As would be expected on the basis of previous studies (32), the most frequent specific categories included obsessions about harm, contamination, and symmetry, as well as checking, cleaning, and repeating compulsions.

The principal components analysis yielded a five-factor solution. The five factors accounted for 65.5% of the total variance and were named as follows: symmetry/ordering, hoarding, contamination/cleaning, aggressive/checking, and sexual/religious obsessions. The first factor accounted for 19% of the variance and included symmetry obsessions and ordering, counting, and repeating compulsions; loadings ranged from 0.41 to 0.82. The hoarding factor accounted for 13.8% of the variance and included hoarding obsessions and compulsions, both with a loading of 0.90. The third factor (12.7% of the variance) included contamination obsessions and cleaning/washing compulsions. Both symptom categories had a loading of 0.91 on this factor. The aggressive/checking factor explained a 10.4% of the variance and included aggressive obsessions and checking compulsions, with loadings over 0.82. The fifth factor accounted for 9.7% of the total variance. Sexual and religious obsessions had strong loadings (>0.71) on this factor. The five-factor solution is highly congruent with other factor-analytic studies following a comparable methodology (table 3).

After Bonferroni correction for 55 tests (0.05/55=0.0009), the symmetry/ordering dimension was robustly associated with the compulsive need to know (r=0.20; for this and the following correlations, N=354, p<0.0009), fear of not saying the right thing (r=0.24), lucky numbers (r=0.25), and touching compulsions (r=0.20). The need to know was also associated with higher scores on hoarding (r=0.22). Fear of saying certain things (r=0.22), intrusive images (r=0.24) and sounds (r=0.20), lucky numbers (r=0.23), colors with special significance (r=0.20), and the need to tell, ask, or confess (r=0.28) were all positively associated with higher scores on sexual/religious obsessions.

After correction for multiple tests (0.05/15=0.003), the symmetry/ordering dimension was significantly correlated with Yale-Brown scale total score (r=0.16, N=354, p<0.003) and compulsions subscale score (r=0.22, N=354, p<0.003) but not with obsessions subscale score (r=0.04, N=354). The contamination/cleaning dimension was also significantly correlated with Yale-Brown scale total score (r=0.20, N=354, p<0.003) and compulsions subscale score (r=0.23, N=354, p<0.003) but not with obsessions subscale score (r=0.10, N=354).

Men (N=190) scored significantly higher than women (N=164) on symmetry/ordering (men: mean=0.12, SD=1.09; women: mean=–0.14, SD=0.85) (Mann-Whitney U=13598.5, p=0.03). Women scored higher than men on contamination/cleaning (men: mean=–0.09, SD=1.00; women: mean=0.10, SD=0.99) (Mann-Whitney U=13642.5, p=0.04) and aggressive/checking (men: mean=–0.11, SD=0.96; women: mean=0.12, SD=1.03) (Mann-Whitney U=13429.5, p=0.02).

For the 172 patients for whom age at onset was available, it was significantly negatively correlated with symmetry/ordering (r=–0.16, N=172, p=0.02), aggressive/checking (r=–0.18, N=172, p=0.01), and sexual/­religious obsessions (r=–0.21, N=172, p=0.005). In analyses by gender, age at onset was correlated with symmetry/ordering only in men (r=–0.26, N=90, p=0.01) and with aggressive/checking (r=–0.25, N=82, p=0.01) and sexual/religious obsessions (r=–0.31, N=82, p=0.004) only in women.

Forty-six (26.6%) of 173 patients met criteria for a lifetime history of chronic motor or vocal tic disorders or both. They were compared with the 127 patients without that diagnosis on all demographic and clinical variables. Male patients were significantly more frequent in the tic group (N=30 [65.2%] of 46) than in the nontic group (N=59 [46.5%] of 127) (χ²=4.76, df=1, p=0.03). Groups did not differ on age. Patients with comorbid tic disorders had a significantly earlier age at onset (tic group: mean=12.51, SD=7.95; nontic group: mean=16.90, SD=10.10) (F=4.05, df=1, 90, p=0.04) and scored higher on the Yale-Brown scale total (tic group: mean=23.65, SD=6.89; nontic group: mean=20.89, SD=6.57) (F=5.77, df=1, 171, p=0.01) and compulsions subscale (tic group: mean=11.73, SD=4.25; nontic group: mean=10.08, SD=4.32) (F=4.98, df=1, 171, p=0.02). Patients with tics scored significantly higher on symmetry/ordering (tic group: mean=0.02, SD=0.83; nontic group: mean=–0.36, SD=0.81) (F=7.52, df=1, 171, p=0.006). Separately by gender, men with tics scored significantly higher on symmetry/ordering (tic group: mean=0.12, SD=0.89; nontic group: mean=–0.34, SD=0.85) (F=5.89, df=1, 87, p=0.02). Women with tics scored significantly higher than women without tics on the Yale-Brown total (tic group: mean=24.81, SD=6.93; nontic group: mean=20.67, SD=6.37) (F=5.28, df=1, 82, p=0.02) and obsessions subscale (tic group: mean=12.68, SD=3.36; nontic group: mean=10.44, SD=3.61) (F=5.12, df=1, 82, p=0.03).

Patients treated with serotonin reuptake inhibitors were significantly improved after treatment on all measures as compared to placebo-treated patients, after adjustment for initial severity scores (Yale-Brown scale: beta=0.19, R²=0.35, p=0.004; NIMH scale: beta=0.23, R²=0.21, p=0.001). This pattern was also observed after the exclusion of the clomipramine study (Yale-Brown scale: beta=0.23, R²=0.39, p=0.001; NIMH scale: beta=0.15, R²=0.29, p=0.05).

Regression analyses showed that severity at baseline predicted posttreatment severity for drug and placebo conditions on all clinical measures. After control for symptom severity, higher scores on the hoarding dimension predicted poorer outcome at posttreatment on all outcome measures. No predictors of placebo response were obtained. Exclusion of clomipramine recipients did not modify the overall results, suggesting a cross-serotonin reuptake inhibitor effect. Table 4shows the beta coefficients and the proportion of variance in the dependent variables accounted for by the independent variables (R² change).

DISCUSSION

To our knowledge, this is the first study to report a relationship between factor-analyzed symptom dimensions and response to serotonin reuptake inhibitors in OCD. Previous research had failed in the attempt to relate OCD symptom subtypes to clinical and demographic variables and to treatment outcome with serotonin reuptake inhibitors. The five dimensions obtained in the present study are consistent with prior factor-analytic research that used a similar methodology (20, 20) and were found to be differentially related to sex, age at onset, comorbid tic-related disorders, and response to serotonin reuptake inhibitors.

The first dimension was named symmetry/ordering and had high factor loadings from symmetry obsessions and ordering, counting, and repeating compulsions. As in previous studies (20, 21), patients with comorbid tic disorders scored significantly higher on this factor. As many as 27% of the patients met criteria for a lifetime history of chronic motor or vocal tics or both; this finding is consistent with previous research (33). These patients had an earlier age at onset, and most were men. In fact, analyses by gender revealed that only male patients with tics scored significantly higher on symmetry/ordering. Scores on this dimension were correlated with total score on the Yale-Brown scale and the compulsions subscale score but not with the obsessions subscale score. Similarly, Holzer and associates compared the phenomenological features of 35 OCD patients with comorbid tics to 35 age- and sex-matched OCD patients without tics and found that patients with tics had more touching, tapping, rubbing, blinking, and staring rituals and fewer cleaning rituals but did not differ on obsessions (34). This tic-like factor corresponds to Janet’s classic description of patients who were tormented by an inner sense of imperfection and felt that their actions were never completely achieved to their satisfaction (35). This feeling of incompleteness is also experienced by patients with Tourette’s syndrome and trichotillomania (32). It also corresponds to the form of OCD that is thought to be genetically related to Tourette’s syndrome, as family and genetic data have suggested (36). McDougle and associates found that serotonin reuptake inhibitor monotherapy was less effective in OCD patients with comorbid chronic tics than in those without tics (37) and that those patients may benefit from the combination of a serotonin reuptake inhibitor and a dopamine antagonist (12). In light of these findings, it would be expected that symmetry/ordering symptoms would be related to poorer outcome with serotonin reuptake inhibitors. Nonetheless, in the current study, this dimension was unrelated to treatment outcome. A putative explanation is that comorbid tic disorders were entry exclusion criteria for the studies included in our predictors analysis.

The second factor, termed “hoarding,” had high factor loadings from hoarding/saving obsessions and hoarding compulsions. Baer’s original factor analysis (20) grouped this factor with symmetry/ordering. In fact, correlation analysis showed that these two factors were the most intimately linked. Hoarding symptoms were present in as many as 20% of the patients and were a major problem for about 5% of them, as previously reported (32). Different accounts of the hoarding phenomenon in OCD are evident. First, hoarding has been seen as an abnormal personality characteristic, and it constituted a diagnostic criterion for obsessive-compulsive personality disorder in DSM-III-R. Second, from an ethological perspective, hoarding symptoms can be regarded as inappropriately released fixed-action patterns (38). Third, hoarding can be conceptualized as the consequence of dysfunctional beliefs formed by prior learning experiences (39).

Hoarding obsessions and compulsions predicted poor response to serotonin reuptake inhibitors, regardless of the outcome measure used or the inclusion of the clomipramine trial. In our principal components analysis, somatic obsessions also loaded on this factor (loading=0.35) but were not listed because we set 0.40 as the cutoff point. This is especially interesting in light of the results of the study by Jenike et al., in which symmetry, hoarding, ordering, and unusual somatic obsessions were significantly more common in patients who responded to phenelzine than in those who responded to fluoxetine (13). Moreover, one study found that presence of symmetry obsessions, ordering compulsions, and hoarding rituals predicted better response in patients treated with cingulotomy (14). Therefore, hoarding, and perhaps somatic obsessions, which predicted poor outcome with serotonin reuptake inhibitors in this study, might better respond to alternative treatments such as MAOIs or cingulotomy. This statement is speculative and requires further examination.

Factors 3 and 4 correspond to the classic contamination and checking dimensions of the Maudsley Obsessional Compulsive Inventory (40) and, as in previous studies (32), were the most frequent symptoms in our group. Finally, the sexual/religious obsessions dimension corresponds to the pure obsessions factor in Baer’s study (20). These OCD symptoms are usually termed “pure obsessional disorder” when they appear in isolation (32).

The rationale for a dimensional model of OCD, according to which certain symptom dimensions—that, in most cases, coexist in the same patient—are differentially related to demographic and clinical characteristics and treatment response, reflects the existing contemporary neurobiological models for the disorder (22). It has been hypothesized that the heterogeneous phenomenology of OCD could be mediated by neuroanatomically different structures (41). The striatum has a well-described topographic organization with respect to its afferent cortical connections and connections to other subcortical structures, as well as cytoarchitecture and neurotransmitters (42). According to the model, the symptoms of OCD would parallel the topography of dysfunction within the corticostriatal network. The extent and location of the dysfunction would determine the heterogeneous clinical picture of OCD. This would explain, for example, why different symptoms can occur either alone or in combination with others in any given patient, the relationship between certain symptoms and comorbid diagnoses, and the differential treatment response. The model still has to explain why the clinical manifestations of OCD change over time, as suggested by Rettew et al. (43) in the only published study directly addressing this issue. These authors followed up 79 children and adolescents with OCD over 2 to 7 years and found that none maintained the same constellation of symptoms from initial assessment to follow-up. Analogous data have yet to be published for adults with OCD.

Possible limitations of the present study need to be considered. The retrospective acquisition of information from case records is not ideal, but histories were comprehensive and followed a semistructured format. Additional prospective research on outcome predictors is required both for behavior and drug therapies. We pooled data from different SSRIs. It may be premature to conclude that differences in outcome are attributable to symptom dimensions when there has been no control for which SSRI was given to which patient. Nonetheless, power was inadequate to analyze each drug separately. Another direction for future study will be to develop a self-rated version of the symptom checklist in order to use it as a standardized instrument in OCD research. Further research on genetics, neuroimaging, and neuropsychology is warranted to confirm the differential involvement of distinct neural elements with the identified symptom dimensions. It is possible that OCD may constitute a label that researchers and clinicians use to name multiple disorders with multiple etiologies, rather than a homogeneous disease. The factor-analytic approach used in the present study and in others has identified meaningful symptom dimensions to help guide future research.

Received June 19, 1998; revision received Feb. 26, 1999; accepted April 2, 1999. From the Department of Psychiatry and Clinical Psychobiology, Universitat de Barcelona; and the Departments of Psychiatry, Massachusetts General Hospital and Harvard Medical School, Boston. Address reprint requests to Dr. Mataix-Cols, Departament de Psiquiatria i Psicobiologia Clínica, Universitat de Barcelona, Passeig de la Vall d’Hebron, 171, E-08035, Barcelona, Catalunya, Spain; [email protected] (e-mail). Supported in part by the David Judah Research Fund and by the Spanish Ministerio de Educación y Cultura (Dr. Mataix-Cols).

TABLE 1

Enlarge table

TABLE 2

Enlarge table

TABLE 3

Enlarge table

TABLE 4

Enlarge table