Participants with a primary OCD/ARD diagnosis within the previous six months, based on DSM-5 (American Psychiatric Association, 2013) or ICD-10 (World Health Organization, 1992) criteria, or those undergoing treatment due to OCD/ARD during this period, were recruited from collaborating clinics in Germany and another research project at the LMU Munich (https://osf.io/8gkjc). The diagnosis of OCD/ARD was given by healthcare providers (for participants recruited through cooperating clinics) or with a structured interview (Mini-DIPS, for participants recruited through another project; Margraf & Cwik, 2017).

For OCD participants, inclusion required a Yale-Brown Obsessive-Compulsive Scale (Y-BOCS; Hand & Büttner-Westphal, 1991) total score > 12 or a subscale score ≥ 8 for obsessions or compulsions as an indicator for clinically relevant OCD symptoms (see also Külz et al., 2014, 2019). The Y-BOCS was administered as self-rating (Y-BOCS-SR; Baer, 1993) for participants recruited through cooperating clinics (n = 64) and as interview version (Hand & Büttner-Westphal, 1991) for participants recruited through another project (n = 38). The final sample comprised 102 participants with OCD with Y-BOCS scores indicating moderate symptoms (M = 22.14, SD = 6.08; M_{ObsessionSubscale} = 11.25, SD_{ObsessionSubscale} = 3.39, M_{CompulsionSubscale} = 10.88, SD_{CompulsionSubscale} = 3.89).

ARD participants were excluded if they had a lifetime diagnosis of OCD. In total 69 participants with ARD fulfilled the inclusion criteria and completed the assessment. The diagnoses were as follows: 28.99% social anxiety disorder, 13.04% generalised anxiety disorder, 31.88% panic disorder, 5.8% agoraphobia, 10.14% post-traumatic stress disorder, 42.03% specific phobia¹1

As multiple anxiety disorders could be present at the same time, the percentages exceed 100%.

Non-clinical participants were recruited via the German online panel PsyWeb (Universität Münster, 2025). Participants were screened for major psychological disorders with the simple version of the Web Screening Questionnaire (WSQ; Donker et al., 2009) and excluded if they exceeded any cut-off. Of 906 participants that gave informed consent to participate in the study and publication of their data, 383 filled out the screening questions and fulfilled the inclusion criteria. Of those, a total of 248 participants completed the first assessment, with 163 eligible for test-retest analyses after completing both assessments.

In summary, the final sample consisted of N = 419 participants, including n = 102 in the OCD group, n = 69 in the ARD group, and n = 248 in the NCC group. Sample characteristics are presented in Supplement B.

The dataset had also been used in a prior publication (Müller et al., 2025), which investigated the psychometric properties of the four-item ultra-brief Obsessive-Compulsive Inventory (OCI-4) by extracting the items from the OCI-12.

The OCI-12 is a 12-item self-report questionnaire measuring OCD symptoms and associated distress on a five-point Likert scale [ranging from 0 (not at all) to 4 (extremely)]. Each scale is assessed by three items, such as “I get upset if objects are not arranged properly.” for the ordering subscale, “I repeatedly check doors, windows, drawers, etc.” for the checking subscale, “I sometimes have to wash or clean myself simply because I feel contaminated.” for the washing subscale, and “I frequently get nasty thoughts and have difficulty in getting rid of them.” for the obsessing subscale. The German wording of each item and the associated subscales are displayed in Supplement D.

The Y-BOCS was assessed as a 10-item interview (Hand & Büttner-Westphal, 1991) for participants recruited through another project and as a 10-item self-report measure (Y-BOCS-SR; Baer, 1993) for participants recruited through cooperating clinics. The Y-BOCS assesses the severity of obsessions and compulsions over the past week. Each item is rated on a five-point scale (0 to 4), with higher scores indicating higher symptom severity. While previous studies proposed that the two versions can be used interchangeably (Steketee et al., 1996), more recent investigations showed slightly higher scores in the clinician administered version (Federici et al., 2010; Hauschildt et al., 2019). In the current study, the Y-BOCS total scores did not differ significantly between the two administration modalities: Y-BOCS: M = 22.29, SD = 6.44 (completed by 38 participants recruited through another project); Y-BOCS-SR: M = 22.05, SD = 5.91 (completed by 64 participants recruited through cooperating clinics); t(71.61) = -0.19, p = .85.

The DOCS (Fink-Lamotte et al., 2021) assesses OCD symptom severity over the past month across four dimensions (i.e., contamination, responsibility for harm and mistakes, symmetry, and unacceptable/taboo thoughts). Each dimension incorporates five items rated on a five-point scale (0 to 4), with higher scores representing higher symptom severity.

The ASI-3 (Kemper et al., 2011) assesses anxiety sensitivity with 18 items rated on a five-point Likert scale [0 (very little) to 4 (very much)], with higher scores representing higher anxiety sensitivity.

The PSWQ (Glöckner-Rist & Rist, 2014) assesses excessive and unrealistic worry using 16-items that are rated on a five-point Likert scale [1 (not at all typical of me) to 5 (very typical of me)], with higher scores indicating higher worry.

The PHQ-9 (Löwe et al., 2002) assesses the severity of depressive symptoms throughout the past two weeks with nine items rated on a four-point scale [0 (not at all) to 3 (nearly every day)]. Higher scores indicate more severe depressive symptoms.

The adapted simple version of the WSQ (Donker et al., 2009) contains 13 questions that screen for the most common psychological disorders and acute suicidality. The original version of the WSQ has been validated and deemed as an appropriate screening tool (Donker et al., 2009; Meuldijk et al., 2017).

The factor structure of the OCI-12 was investigated by confirmatory factor analyses (CFA) using the lavaan package (version 0.6-19; Houben et al., 2015). Both, the four-factor structure (washing, checking, ordering, obsessing) and the four-factor structure including a higher-order factor of general OCD symptoms were investigated. We evaluated goodness of fit using the standardised root-mean-square residual (SRMR), root-mean-square error of approximation (RMSEA), the comparative fit index (CFI), and the Tucker-Lewis index (TLI). The following criteria as indicator for good model fit (Hu & Bentler, 1999; Schmitt, 2011): RMSEA ≤ 0.06; SRMR ≤ 0.08; CFI ≥ 0.95; TLI ≥ 0.95. As the multivariate normality assumption was violated (for mean and variance), we decided to use the “Maximum Likelihood with Robust Standard Errors and Mean-Variance Adjusted Test” in our CFAs. Therefore, all fit indices reported are robust fit indices. We further investigated in separate linear regression models whether each factor of the OCI-12 could predict the corresponding subscale of the DOCS.

To examine construct validity, correlation analyses were conducted between OCI-12, Y-BOCS, and DOCS (convergent validity) and between OCI-12, ASI-3, PSWQ, and PHQ-9 (discriminant validity). Pearson’s correlation coefficients were interpreted according to Cohen (1988).

For internal consistency, both Cronbach’s α and McDonald’s ω were calculated (Dunn, 2014; McDonald, 1999) and interpreted according to Hair (2009). The test-retest reliability of the OCI-12 was investigated with correlation analyses (interpreted according to Cohen, 1988), paired t-tests, and the two-way mixed effect intraclass correlation coefficient (ICC; interpreted according to Koo & Li, 2016) between T₁ and T₂ in the NCC sample.

We investigated group differences in the OCI-12 total and subscale scores by means of univariate (ANOVA) and multivariate analysis of variance (MANOVA). Furthermore, we conduced post-hoc Tukey Honest Significant Difference (Tukey HSD).

We investigated the diagnostic accuracy of the total score and each subscale with receiver operating characteristic (ROC) analyses. The area under the curve (AUC) was interpreted according to the criteria by Carter et al. (2016). Cut-off scores were established with the Youden Index (J; Youden, 1950).

Figure 1A displays the CFA examining the four-factor solution. The Chi-square test, χ²(48, N = 102) = 70.985, p = .017, rejected the hypothesis of a perfect fit. Furthermore, the TLI (0.945) did not support a good model fit while the RMSEA indicated a marginal model fit (0.081; MacCallum et al., 1996). The remaining goodness-of-fit indices supported a good fit of the four-factor model: SRMR = 0.059; CFI = 0.960.

The four-factor model including a higher-order factor of general OCD symptoms is shown in Figure 1B. Aside from the Chi-square test, χ²(50, N = 102) = 71.181, p = .026, and the RMSEA (0.076; reasonable fit; MacCallum et al., 1996), all fit indices support a good model fit: SRMR = 0.058; CFI = 0.963; TLI = 0.951. The first-order factors loaded weakly to strongly on the higher-order factor of general OCD symptoms. The higher-order factor accounted for a significant proportion of variance in the first-order factors checking, ordering, and washing (RChecking2 = .791, ROrdering2 = .286, RWashing2 = .148), but not obsessing (RObsessing2 = .023).

The path model of the four-factor solution in the total sample is displayed in Figure 2A. As in the OCD sample, the Chi-square test was significant, χ²(48, N = 419) = 80.558, p = .002) and the RMSEA showed only a reasonable fit (0.065; MacCallum et al., 1996). The remaining goodness-of-fit indices supported a good fit of the data: SRMR = 0.043; CFI = 0.979; TLI = 0.972.

Figure 2B presents the path model including the general OCD factor. Except the Chi-square test, χ²(50, N = 419) = 84.168, p = .002, and the RMSEA (0.065; reasonable fit; MacCallum et al., 1996), all fit indices support a good model fit: SRMR = 0.046; CFI = 0.979; TLI = 0.972. In the total sample, the first-order factors loaded strongly on the higher-order factor of general OCD symptoms. The higher-order factor accounted for a significant proportion of variance in all first-order factors (RChecking2 = .796, ROrdering2 = .433, RWashing2 = .588, RObsessing2 = .557).

The linear regression models of the four OCI-12 factors predicting each DOCS subscale are presented in Supplement E. Each DOCS subscale significantly and strongly predicted by the corresponding OCI-12 subscale (βs ranging from β = 0.44 for checking and ordering to β = 0.84 for washing).

The internal consistency for the OCI-12 subscales in the OCD group ranged from good (obsessing) to excellent (washing; see Table 1). The OCI-12 total showed good internal consistency. In the ARD subgroup, the internal consistency ranged from acceptable (washing) to good (ordering), while the OCI-12 total demonstrated good internal consistency. The NCC group showed the lowest internal consistency, ranging from not satisfactory (checking) to acceptable (obsessing), with the OCI-12 total showing acceptable internal consistency.

The test-retest reliability of the OCI-12 was assessed over an interval of M = 13.42 (SD = 3.64) days in the NCC group. Results are displayed in Table 2. Results of t-tests indicated no significant changes over the test-retest interval for OCI-12 total and all subscales except for washing, which significantly increased from T₁ to T₂. A strong positive correlation between T₁ and T₂ was shown for the OCI-12 total and a moderate (checking) to strong (ordering, washing, obsessing) positive correlation for the OCI-12 subscales. The two-way mixed effect ICC demonstrated moderate (checking) to good (ordering, washing, obsessing) reliability for the subscales and good reliability for the OCI-12 total score.

Descriptives and group differences on the OCI-12 are presented in Table 4. The total scores of the OCI-12 significantly differed between groups, F(2, 416) = 373.1, p < .001, as indicated by a main effect of group in the ANOVA. Participants with OCD had significantly higher OCI-12 scores than participants with ARD and NCC, which showed significantly higher OCI-12 scores than the NCC group (all p’s < .001 in Tukey’s HSD tests).

When considering the OCI-12 subscales, the MANOVA showed a significant main effect of group across the subscales, Pillai’s Trace = 0.733, F(8, 828) = 59.837, p < .001. As for the OCI-12 total score, separate ANOVAs with post-hoc Tukey HSD tests revealed that participants with OCD had significantly higher scores on each subscale than participants with ARD, which showed significantly higher scores than the NCC group (all p’s < .001).

The diagnostic accuracy of the OCI-12 to discriminate participants with OCD from participants with ARD was good (AUC = .85, 95% CI [.795, .909]; see Figure 3A). The diagnostic accuracy for each subscale ranged from AUC = .59 (ordering) to AUC = .81 (obsessing; see Figure 4A).

Considering the diagnostic accuracy of the OCI-12 for distinguishing individuals with OCD from NCC’s, the OCI-12 total score evidenced excellent accuracy (AUC = .99, 95% CI [.976, .997]; see Figure 3B). The diagnostic accuracy of the subscales ranged from AUC = .75 (ordering) to AUC = .96 (obsessing). Overall, the OCI-12 total score evidenced the best diagnostic accuracy for discriminating OCD participants from both, ARD and NCC (see Figure 4B).

Table 5 summarises the Youden Indices, sensitivities, and specificities of the OCI-12 total and each subscale for discriminating participants with OCD from ARD and NCC participants. An OCI-12 total score ≥ 17 was considered optimal to discriminate participants with OCD from participants with ARD. When discriminating participants with OCD from NCC’s, a score of ≥ 11 was considered optimal. Out of the subscales, the washing subscale could best discriminate OCD from ARD participants, while the obsessing subscale best discriminated OCD from NCC participants. The ordering subscale was suited worst to discriminate participants with OCD from ARD and NCC participants.

OCI-12 severity benchmarks were investigated in severity-groups based on Y-BOCS cut-offs (Storch et al., 2015; see Supplement F for OCI-12 descriptives per severity group). The OCI-12 total score fairly discriminated mild from moderate cases (AUC = .73, 95% CI [.557, .900]), but only poorly distinguished moderate from moderate-severe cases (AUC = .69, 95% CI [.576, .807]) and moderate-to-severe from severe cases (AUC = .52, 95% CI [.059, .983]).

Given that the severe group included only three individuals, moderate-to-severe and severe cases were combined, but discrimination from moderate cases remained poor (AUC = .69; 95% CI [.577, .801]). An optimal cut-off ≥ 12 was suggested for mild vs. moderate cases (J = .38), and a cut-off ≥ 24 for moderate vs. moderate-severe cases (J = .38). Due to small sample sizes and limited discrimination, further research with a larger sample is needed to establish OCI-12 severity benchmarks.