Subtypes of Severe OCD and Treatment Outcome: A Latent Profile Analysis of the Yale-Brown Obsessive-Compulsive Severity Scale

Abstract

Background:

The Yale-Brown Obsessive Compulsive Scale (Y-BOCS) is the gold standard measure of OCD and the most common tool to assess OCD severity and treatment outcome. Relying on Y-BOCS total scores likely to capture overall severity well, but may obscure important qualitatively different OCD profiles. The current study aimed to identify profiles of OCD and their association to obsession/compulsion content domains (e.g., contamination), and treatment outcomes.

Methods:

Participants were adults 759 (49% women) seeking partial/residential treatment for severe OCD. The sample was on average 29.81(SD=11.95) years old and predominantly White (85%). Latent profile analysis was used to identify patterns of OCD symptoms using the self-reported Y-BOCS severity scale (Y-BOCS-SS). Profiles were validated using generalized linear models to capture the association between profiles and obsession/compulsion content and changes in OCD severity, depression, and quality-of-life.

Results:

Three profiles were identified: a “Severe with Lower Resistance” Profile (69% of sample) characterized by high severity with the greatest relative effort to resist symptoms, a “Moderate” profile (25%) characterized by uniform endorsement of items in the moderate range, a “Low Compulsion” (6%) profile characterized by high mean endorsement of obsession items and low endorsement of compulsion items. The profiles varied significantly in terms of endorsement of different obsession/compulsion domains but did not vary significantly in terms of treatment outcomes as measured by changes in OCD, depression, and quality-of-life.

Conclusions:

Relying on Y-BOCS total score may fail to capture qualitatively different, albeit rare, presentations of OCD. However, these profiles were not predictive of treatment response.

Introduction

Obsessive-compulsive disorder is defined by obsessions and/or compulsions that elicit significant distress and impairment (American Psychiatric Association, 2013). Obsessions are recurrent unwanted thoughts, images, or urges that elicit significant distress, and compulsions are repetitive behaviors completed to decrease obsessive thoughts and/or alleviate distress (American Psychiatric Association, 2013). Epidemiological studies have ranked OCD the fourth most common mental disorder in the United States (Karno et al., 1988; Rasmussen and Eisen 1992) with a lifetime prevalence of 2.3% (Ruscio et al. 2010). It is estimated that OCD is the tenth leading cause of disability in the world (Murray et al. 1996). Half of individuals with OCD, about 50.6%, report having serious impairment compared to 34.8% with moderate impairment and 14.6% with mild impairment (Ruscio et al., 2010).

Although there have been great advances in the treatment for OCD, with studies demonstrating the efficacy of medication (Skapinkais et al., 2016) and behavior therapy (Skapinkais et al., 2016), and their combination (Mao et al., 2022), about 40–50% of individuals show minimal to no response from treatment (Fisher and Wells, 2005). Therefore, more research is needed to better understand severe and treatment-resistant OCD. One step toward this end is to maximize the utility of current, well-established measures that assess for OCD severity. In the current study, we re-examine conventional approaches to utilizing the Yale-Brown Obsessive-Compulsive Scale (Y-BOCS; Goodman et al., 1989) and propose a new approach for analyzing such data, which may hone clinicians’ and researchers’ abilities to categorize severe OCD, and in turn, improve individualized treatment options for treatment-resistant and treatment refractory individuals.

The Y-BOCS is one of the most widely used tools to assess the presence and severity of OCD and to measure treatment response (Goodman et al., 1989). It is an evidence-based, clinician administered interview made up of two parts: the checklist and the severity scale. The Y-BOCS checklist (Y-BOCS-CL) assesses the presence and theme of obsessions and compulsions over the past week, and it is composed of several symptom dimensions, including but not limited to, contamination, aggressive, and sexual obsessions and washing, checking, and ordering compulsions. The Y-BOCS-Severity Scale (Y-BOCS-SS) measures obsession and compulsion symptom severity over the past week, based on the symptoms endorsed on the checklist. On the Y-BOCS-SS, obsessions and compulsions are rated on 0–4 point-scales, ordered in severity, in five dimensions (time, interference, distress, resistance, control). In this way, the Y-BOCS-SS is multidimensional (as evidenced by the five different components that make up obsession and compulsion severity) and offers answers about how an individual experiences their symptoms. However, in terms of scoring and interpreting the Y-BOCS-SS, the traditional conceptualization of OCD severity results in an overall summation of items for obsession severity (0–20), compulsion severity (0–20), and total severity (0–40) with greater scores indicating greater level of severity of OCD symptoms. This results in a total of 50 different response options across 10 questions. Yet, most of this information is lost when reducing 10 data points to one total score. This problem is highlighted when two different individuals have the same severity score, but markedly different answers at the item level. Patients with the same total score may have distinct profiles or “patterns” of severity.

Most studies investigating the underlying measurement structure of the Y-BOCS-SS have utilized factor analysis, which is a variable-centered approach that identifies a continuous latent factor(s) that can be used to inform cut off scores along the continuum(s) to classify individuals into risk groups. Cut off scores are used clinically to classify patients into mild, moderate, severe, and extreme severity categories, and in research studies to determine eligibility into an experimental condition. This approach suggests that OCD severity is a continuous construct varying by level. Factor analytic studies of the severity scale have found support for two- and three- factor solutions made up of an obsessions severity factor, a compulsions severity factor, and, in some cases, a resistance factor (Anholt, 2010; Castro-Rodrigues et al., 2018; Garnaat & Norton, 2010; McKay et al., 1995; Moritz et al., 2002; Seol et al., 2013). Other factor analyses of the Y-BOCS-SS found support for a different two factor structure where the interference and distress items loaded onto a factor labeled “disturbance” and the time spent on symptoms, resistance, and control items loaded onto a factor labeled “severity” (Amir et al., 1997; Deacon & Abramowitz, 2005). McKay and colleagues (1998) found evidence for both factor structures. However, Fatori and colleagues (2020) recently showed that factor analysis models of the Y-BOCS-SS provide mediocre to fair fits to the data and they recommended further improvements to the Y-BOCS. In addition to changes to the measure, different data analytic approaches to the Y-BOCS-SS may help researchers gain new perspectives into their data.

Person-centered approaches to data analysis, such as latent profile analysis (LPA), are appropriate for multidimensional constructs and identify distinct subgroups of individuals based on patterns or profiles of item endorsement rather than cut-off scores. In general, there are two types of latent profile solutions. An ordered solution consists of similar patterns between items within each profile. In this case, profiles mainly differ by the level or frequency of item endorsement, but the pattern between items is relatively equal across profiles. An ordered solution in LPA may suggest that a variable is best represented as a continuous construct, as in factor analysis. An unordered solution is often signified by lines crossing and represents distinct patterns between items within each class. In summary, in ordered solutions, profiles differ by level or frequency, whereas, in unordered solutions, profiles differ by type.

Whereas latent class analyses have been conducted on the Y-BOCS checklist (Atli et al., 2014; Delucchi et al., 2011) and on other measures of OC symptoms (Althoff et al., 2009), no study has conducted a latent class or profile analysis of the Y-BOCS-SS. Given that not all individuals with severe OCD experience their symptoms in the same way, latent profile analysis can be used to identify discrete subtypes of severe OCD that vary not only by level, but also type. Discrete subtypes of severe OCD may inform specific treatment targets (e.g., distress tolerance, functional impairment) and provide a basis for comparing different treatments for distinct groups of patients.

The first aim of the current study is to identify distinct profiles of OCD symptoms using the gold-standard measure of OCD severity, the Y-BOCS-SS, in adults seeking partial hospital/residential treatment for severe OCD. Our second aim was to characterize the identified profiles based on obsession/compulsion content, treatment course (i.e., length of treatment), and treatment outcomes. Overall, our hypotheses were largely exploratory as no research has investigated profiles of OCD severity using the Y-BOCS-SS.

1. Material and methods

1.1. Transparency and Openness

This study was not preregistered and data was collected within the context of routine care. Data and analysis code will be made available upon request to investigators.

1.2. Participants

Participants included 759 individuals with OCD diagnoses consecutively admitted to the OCD Institute at McLean Hospital between 2008 and 2017. The McLean Hospital OCD Institute is a partial/residential treatment program for OCD and related disorders with an average length of stay of 30 to 90 days. As part of this treatment program, all participants receive daily ERP, group therapy, and individual meetings with behavioral therapists, family therapists, and psychiatrists. Of note, 89% and 93% of individuals were receiving medication upon admission and at discharge respectively. Mean Y-BOCS-SS score was 25.04 (SD = 5.94), which falls in the severe range per traditional severity classification (Goodman et al., 1989). The sample was on average 29.81 (SD = 11.95) years old. 85% of the sample was White, 4% Asian, 2% Latino, 2% Black, and 2% endorsing multiple or other racial/ethnic backgrounds. Lastly, 49% of participants endorsed being a woman, 50% being a man, and 1% reported being gender non-conforming, non-binary, or transgender.

1.2. Measures

1.2.1. Demographic and Diagnostic Information.

Participants completed a brief questionnaire to obtain information on demographics.¹ Diagnostic information was obtained using Structured Clinical Interview for DSM-IV Axis I Disorders, Research Version, Patient Edition With Psychotic Screen (SCID-I/P with Psychotic Screen; First, Gibbon, Spitzer, & Williams, 2002) or the Structured Clinical Interview for the DSM-5 (SCID-5; First et al., 2015). The SCID is a semi-structured diagnostic interview; the version administered depended on the year of administration. All assessments were completed by trained research staff or graduate students under the supervision of a licensed clinical psychologist.

1.2.1. Treatment Outcomes Variables

Yale-Brown Obsessive-Compulsive Scale- Severity Scale, (Y-BOCS-SS; Goodman et al., 1989) measures OCD severity and related impairment. The participants in this study completed the self-report version of the Y-BOCS-SS, which has been highly correlated with the clinician-administered version (Federici et al., 2010). Hamilton Depression Rating Scale (HAMD; Hamilton, 1960) and Quality of Life and Satisfaction Questionnaire (Q-LES; Endicott et al., 1993) were given to capture changes in depression and overall quality of life during the course of treatment. These measures were given at admission and upon discharge. Given the procedures of the clinic, discharge could happen at different lengths of time for different patients. Identified profiles were also compared based on their length of stay in the program, measured in days, to examine whether some profiles required more or less time in treatment.

1.2.2. Obsession and Compulsion Content

The Dimensional-Obsessive Scale (DOCS; Abramowitz et al., 2010) is a self-report measure that assesses the degree to which common categories of obsessional content and compulsion are present. The measure assesses four categories: (1) Germs/Contamination, such as worries about becoming sick, (2) Being responsible for harm, such as being worried that a mistake will or has harmed someone, (3) Unacceptable Thoughts such as unpleasant thoughts related to sex, immorality or violence, (4) “Just Right” Symmetry/Completeness such as feeling something isn’t “just right” or wanting things to be in “order”. Subscales scores for these categories were used to characterize identified profiles.

1.3. Procedures

All participants completed written informed consent, as approved by Mass General Brigham’s Institutional Review Board, to provide clinical data for ongoing research studies in the McLean Hospital OCD Institute. Participants were administered the SCID-I/P or SCID-5 and completed the self-report version of the Y-BOCS, HAM-D, QLES, and DOCS, at admission and the Y-BOCS, HAM-D, and QLES at discharge.

1.4. Data Analytic Plan

1.4.1. Latent Profile Analyses

Latent profile analysis was conducted on the 10 items of the self-report YBOCS-SS, which capture the severity of different OCD symptom domains. LPA were estimated using Mplus, version 7.4 (Muthén & Muthén, 2012). Multiple imputation with robust standard errors (MLR) was used to account for missing data and skew. A series of nested equal volume, equal shape (EEI in Mplus) and equal volume, equal shape, and equal orientation (EEE in Mplus) LPA models (e.g., 2-class, 3-class, 4-class, etc.) were fit and compared to one another. Traditional EEI LPA models assume that there is no remaining covariation between items after accounting for profile membership. Although this can lead to parsimonious solutions, it also can lead to the identification of redundant classes that only vary by overall severity. In contrast, EEE LPA models allow for covariation of items within profiles, which can in some cases lead to more useful and varied profiles (Wardenaar, 2021). The indices used to evaluate LPA model fit were: (1) Bayesian Information Criterion (BIC) with lower values indicating better fit. BIC is a measure of model fit that applies a penalty for increasing numbers of estimated parameters to identify a parsimonious and generalizable solution., (2) Vuong-Lo-Mendell-Rubin (VLMR) test to compare nested models (McLachlan & Peel, 2000; Schwarz, 1978), and (3) entropy, which ranges from 0 to 1, with values closer to 1 indicating clear delineation between classes (Celeux & Soromenho, 1996). We also considered the interpretability of the solution when selecting the best model, in addition to the size of the identified profiles (Muthén, 2003).

1.4.2. Profile Validation

After an LPA solution was selected, participants were assigned profile membership and compared based on obsession/compulsion content, treatment outcome, and length of treatment. For obsession/compulsion content, generalized linear models were used to examine whether profile membership is predictive of DOCS subscale scores. In these models, DOCS total score was included as a covariate to control for overall obsession/compulsion severity to examine whether profiles are associated with specific domains of obsession/compulsion content. To examine treatment outcomes, generalized linear models were used to test whether profile membership is predictive of OCD severity, depression severity, and self-reported quality of life upon discharge. Baseline pretreatment measures of these variables were included as covariates to capture change in these domains. Lastly, a generalized linear model was used to examine whether profiles differ in terms of length of treatment.

2. Results

2.1. Extractions of Latent Profiles

Table 1 presents fit indices for the tested LPA models. We selected the EEE LPA 3-class solution based on fit statistics, class size, and interpretability. Figure 1 depicts the mean item responses for the 3 identified profiles. The first profile was labeled “Moderate” (25% of sample) as this class was characterized by moderate severity across all items, the second was labeled “Severe with Lower Resistance” (69% of the sample) as this profile was characterized by high severity in the majority of items, but with lower severity on the two items related to resistance of obsessions and compulsions. Lastly, the third group was labeled “Low Compulsion” (6% of the sample) as the profile was characterized by high severity on items related to obsessions with only mild severity for items related to compulsions. The four class EEE solution fit the data significantly better, but resulted in a fourth class that only contained 2% (n = 18) of the sample, which was deemed too small to examine in subsequent analysis. Interestingly, this class appeared to be a “Low Obsession” profile characterized by low endorsement of obsession items, with moderate endorsement of compulsion items (see supplementary Figure 1 for a plot of the four class EEE solution). Lastly, the traditional EEI models, in addition to worse model fit, resulted in redundant profiles that only varied by severity, specifically breaking the “Severe with Lower Resistance” profile into 3 smaller profiles with minor variation in overall severity, but nearly identical endorsement patterns.

Table 1.

Latent Profile Fit Indices

Model	# par	BIC	Log likelihood	VLMRT	Entropy	Size of Profiles (%)
EEI
2	31	18508.965	−9151.687	p<.001	.85	65, 35
3	42	18154.691	−8938.074	p=.05	.80	51, 32, 17
4	53	17916.425	−8782.464	p=.04	.83	50, 25, 19, 6
5	64	17762.298	−8668.925	p=.23	.84	49, 24, 17, 6, 4
EEE
2	76	17105.576	−8300.772	p<.001	.95	93, 7
3	87	17072.047	−8247.531	p=.01	.87	69, 25, 6
4	98	17025.306	−8187.685	p=.01	.89	68, 24, 6, 2

Model column lists the type of model EEI = equal volume and equal shape, EEE = equal volume, equal shape, and equal orientation and the number of fitted profiles. # par = number of parameters in the model. BIC = Bayesian Information Criteria, V-LMRT = the p-value from the Vuong-Lo-Mendell-Rubin test to compare nested models.

Figure 1.

Mean values of item responses in 3-Profile EEE solution

O = an obsession item, C = compulsion item.

2.2. Profiles Association with Obsession/Compulsion Content

Table 2 displays the mean DOCS subscale scores broken down by profile. In models to compare DOCS subscale scores, the “Moderate” profile group served as the reference category. Compared to the “Moderate” group, the “Low Compulsion” group had lower scores on the Germs/Contamination subscale (B = −2.90, p < .001, Cohen’s d = −1.03), higher scores on the Unacceptable Thoughts subscale (B = 4.06, p < .001, d = .44), and had no significant difference on the Harm subscale (p = .95) or “Just Right”/Symmetry/Completeness subscale (p = .15). Compared to the “Moderate” group, the “Severe with Lower Resistance” group had lower scores on the Harm subscale (B = −.87, p = .02, d= −.35), higher scores on the Unacceptable Thoughts subscale (B = 1.80, p < .001, d = .73), and no significant difference with the Germs/Contamination (p = .75) or “Just Right”/Symmetry/Completeness subscales (p = .06).

Table 2.

Validating variables broken down by latent profile membership

Measure	Moderate	Severe with Lower Resistance	Low Compulsion
Germs/Contamination	5.61(5.60)	7.57(6.61)	1.02(2.18)
Harm	6.56(5.40)	8.58(6.20)	4.22(4.57)
Unacceptable Thoughts	6.78(4.93)	10.88(6.13)	9.09(5.35)
“Just Right”, Symmetry/Complete	5.65(4.75)	7.26(6.12)	2.73(3.60)
OCD (Y-BOCS-SS)
Admission	19.84(5.00)	27.63(4.39)	17.56(4.63)
Discharge	12.82(5.61)	16.37(6.31)	12.24(5.90)
Depression (HAMD)
Admission	6.56(4.61)	10.22(5.04)	9.58(4.27)
Discharge	4.15(4.59)	5.45(4.77)	4.74(4.69)
Quality-of-life
Admission	44.74(9.11)	40.30(8.54)	42.44(6.21)
Discharge	50.36(9.33)	48.56(9.23)	50.08(8.28)
Treatment Length (Days)	53.50(23.19)	58.27(23.09)	55.09(26.00)

Cell values are means with standard deviations in the parentheses

2.3. Profiles Association with Treatment Outcomes

Table 2 presents the mean OCD symptom severity, depression, and quality-of -life scores at both admission and discharge broken down by profile membership. Again, using the “Moderate” group as the reference category, there was not a significant association between membership in the “Severe with Lower Resistance” group and treatment outcome as measured by changes in the OCD symptoms (p = .35), depression (p = .90), or quality of life (p = .38). Similarly, compared to the “Moderate” group, membership in the “Low Compulsion” group was not significantly associated with treatment outcome as measured by OCD severity (p = .62), depression (p = .50) or quality-of-life (p = .57). There were also no significant differences between the “Low Compulsion” and “Severe with Lower Resistance” groups when adjusting the reference category to obtain this comparison. Compared to the “Moderate” group, membership in the “Severe with Lower Resistance” group was positively associated with length of stay in treatment (B = 4.76, p = .02, d = .21), in contrast, membership in the “Low Compulsion” group was not associated with length of treatment (p = .68). Lastly, there was no significant difference between length of stay between the “Low Compulsion” and “Severe with Lower Resistance” group.

3. Discussion

3.1. Subtypes of Severe OCD

This was the first study to classify individuals seeking intensive/residential/partial treatment for OCD based on the self-report Y-BOCS Severity Scale (Y-BOCS-SS) using latent profile analysis. Three distinct profiles of OCD severity emerged that varied both by level and type. These distinct profiles also varied in terms of obsession/compulsion content, but did not differ significantly in terms of treatment response. All profiles were associated with robust reduction in OCD symptoms demonstrating the ability of exposure-based treatments to adapt to varied OCD symptom presentations.

The “Severe with Lower Resistance” was the largest identified profile (69% of the sample) and appears to capture a group of individuals with OCD that report experiencing severe frequency, distress, and impairment from obsessions and compulsions, but make the greatest relative effort in resisting symptoms. These individuals may represent individuals with OCD who are trying hard to respond to obsessions adaptively and resist compulsions, but have little success. Lack of success may be due to the approach of “resisting” symptoms. For example, it is possible that this profile represents individuals with OCD whose attempts at suppressing obsessions, which can be an indication of insight, end up backfiring and resulting in a rebound effect. However, the resistance items have been controversial in the literature based on low correlations with overall OCD severity (Boyette et al., 2011; Storch et al., 2015), theoretical concerns (Storch et al., 2010), and confusion regarding scoring and interpretation. Therefore, it is possible that this pattern of response is driven by a measurement issue and is not represent of a theoretically meaningful OCD presentation. The second most common profile labeled “Moderate” (25% of the sample) can be characterized by uniformly endorsing items in the moderate range. This group can be said to be captured very well by a Y-BOCS-SS total score as there is minimal variability between item endorsement across the measure.

The “Low Compulsion” group (6% of the sample) is characterized by high mean endorsement of items related to obsessions with much lower mean endorsement of compulsion related items. Importantly, this presentation results in an average Y-BOCS-SS total score very similar to the “Moderate” group despite a very different endorsement pattern. Relatedly, although not analyzed due to the small sample size, in the four-profile solution (see supplementary Figure 1), the fourth identified profile (2% of the total sample) was characterized by moderate endorsement of compulsion items and only mild endorsement of obsessional items. Although these presentations were relatively rare, it is worth noting that these presentations are likely obscured by relying on Y-BOCS-SS total score. However, a two-factor Y-BOCS-SS model that provides subscale scores broken down by obsession versus compulsion items would succeed in characterizing these subtypes in a simple fashion.

The identified profiles also varied significantly by obsession/compulsion content. The “Low Compulsion” profile was characterized by significantly higher likelihood of endorsing obsession/compulsion content related to unacceptable thoughts (i.e., unwanted thoughts about immorality, violence, sexuality etc.) and significantly lower likelihood of endorsing obsession/compulsion content related to germs/contamination compared to the “Moderate” group. This is consistent with previous research that found that individuals whose OCD presentation is characterized by obsessional symptoms without compulsion symptoms (i.e., “Pure Obsessional Type”) are more likely to endorse obsessions related to unacceptable thoughts and harm compared to other content domains (Williams et al., 2011). Importantly, researchers have argued this “Pure Obsessional Type” is likely a misnomer because although many individuals may not endorse compulsions, they likely engage in more subtle (e.g., reassurance seeking) or mental compulsions (Williams et al., 2011).

Despite varied symptom profiles, there was not a significant difference in treatment response across the identified profiles as measured by changes in OCD symptoms, depression, or quality-of-life. On average, all groups saw a significant decrease in OCD symptoms across treatment with the “Moderate” and “Low Compulsion” group moving on average from the “moderate” to the “mild” symptom range and the “Severe with Lower Resistance” moving from the “severe” to “moderate” symptom range. The comparable success of the “Low Compulsion” profile may provide additional evidence that individuals with this presentation due indeed have compulsions that can be targeted with an ERP protocol, but that these compulsions may be more subtle or outside of the awareness of some patients. Psychoeducation in this area may be beneficial to patients, family members, and clinicians to learn to identify and target more subtle compulsions in these less common OCD presentations. Lastly, although individuals in all profiles, on average, saw significant improvement in OCD severity, there is still substantial variability within all profiles indicating many individuals do not respond or have a limited response. Future work may need to incorporate a larger variety of both internal (e.g., psychiatric comorbidity) and external (e.g., familial support, employment patterns) baseline patient characteristics to increase predictive power of such a complex outcome as treatment response (Perlis, 2013).

3.2. Limitations

We acknowledge several limitations of this study. First, this study included patients seeking partial/residential treatment for OCD and related disorders and may not generalize to patients with OCD in traditional outpatient settings. Second, OCD symptom severity was assessed by self-report and thus may not generalize to interviewer-rated Y-BOCS, although previous research suggests that both versions are highly correlated. Third, it is possible that the results of these analyses may vary across racial/ethnic groups, and our interpretation of this is limited given that 85% of the participants in this study identified as White. Future work should prioritize recruitment of more diverse samples to directly assess generalizability to other groups. The current sample was collected during routine care of the clinic, which included additional interventions besides ERP such as medication management. In fact, over 93% of patients were receiving medication upon discharge (rates of medication prescription did not differ between identified classes). It is possible that medication treatment may have influenced the lack of differences in treatment response across classes. Although this combination treatment is consistent with current standards of care, it makes assessing the effect of ERP alone on different symptom presentations more difficult.

3.3. Conclusions

To our knowledge, this was the first study to conduct a latent profile analysis on the Y-BOCS-SS, which assesses the severity of 10 different OCD symptom domains. Findings from the current study suggest the presence of distinct subtypes of severe OCD, in particular, the existence of rarer presentations characterized by obsessions without compulsions and compulsions without obsessions. These presentations are likely obscured when relying on Y-BOCS-SS total score, but are well characterized by a two-factor (i.e., obsession and compulsion subscales) Y-BOCS-SS structure. The identified subtypes did vary significantly in terms of obsession/compulsion content, most notably the “Low Compulsion” presentation being characterized by primarily obsession/compulsions related to unacceptable thoughts. Surprisingly, the different profiles were not associated with differential treatment response. Future work may need to consider more complex multivariable models to improve our prediction of ERP treatment response in severe OCD.