Exploring response inhibition and error monitoring in obsessive-compulsive disorder

Abstract

Behavioral evidence of impaired response inhibition (RI) and hyperactive error monitoring (EM) in obsessive-compulsive disorder (OCD) is inconsistent. Recent neuroimaging work suggests that EM plays a role in RI impairments in OCD, but this has rarely been investigated using behavioral measures. The aims of this study were to (1) compare RI and EM performance between adults with OCD and non-psychiatric controls (NPC) while investigating possible moderators, and (2) assess whether excessive EM influences RI in OCD. We compared RI and EM performance on the Stop-Signal Task (SST) between 92 adults with OCD and 65 NPC from two Brazilian sites. We used linear regression to investigate which variables (group, age, medication use, clinical symptomatology) influenced performance, as well as to examine possible associations between RI and EM. OCD and NPC did not differ in RI and EM. However, age moderated RI performance in OCD with a medium effect size, reflecting differential effects of age on RI between groups: age was positively associated with RI in OCD but not NPC. Further, OCD severity predicted EM with a medium to large effect size, suggesting that more symptomatic patients showed greater monitoring of their mistakes. Finally, group moderated the relationship between RI and EM with a small effect size. Our findings suggest that demographic factors may influence RI, whereas clinical factors may influence EM. Further, we found preliminary behavioral evidence to indicate that impaired RI and excessive EM are related in OCD.

Introduction

Obsessive-compulsive disorder (OCD) is a psychiatric disorder characterized by recurrent and intrusive thoughts, mental images or urges that cause great anxiety (obsessions) and compel the individual to engage in repetitive rituals to ease discomfort (compulsions) (American Psychiatric Association, 2013). The lifetime prevalence of OCD in the US population is approximately 2% (Ruscio et al., 2010), and in Brazil the prevalence ranges from 0.3% in São Paulo (Andrade et al., 2002) to 2.1% in Porto Alegre (Almeida-Filho et al., 1992). Individuals with OCD have high levels of psychiatric comorbidity, particularly symptoms of depression and anxiety, are more likely to be unemployed (Torres et al., 2006) and have lower educational attainment (Pérez-Vigil et al., 2018) than individuals without OCD.

Response inhibition (RI) is the ability to inhibit inappropriate or no longer needed responses following changes in the internal or external environment (Verbruggen and Logan 2008). One of the paradigms for the evaluation of RI is the Stop-Signal Task (SST). In the SST, participants perform a go task, which occasionally is followed by a stop-signal that instructs the participant to withhold the response. The time required to inhibit the response is the Stop-Signal Reaction Time (SSRT), which is a reliable measure of RI (Verbruggen and Logan, 2008).

Impaired RI has been suggested to be a core feature of OCD, contributing to symptomatology (e.g. Bannon et al., 2002). Yet, while many case-control studies have found that OCD patients perform worse in inhibition tasks when compared to non-psychiatric controls (NPC) (Menzies et al., 2007; McLaughlin et al., 2016; de Wit et al., 2012), others have found no differences in RI between these groups (Krishna et al., 2011, Kalanthroff et al., 2017). Meta-analyses have pointed to significantly worse performance on RI tasks in participants with OCD, with effect sizes of either small or medium magnitude, but with significant heterogeneity across studies (Lipszyc and Schachar, 2010; Abramovitch et al., 2013; Shin et al., 2014; Snyder et al., 2015). Together, these findings suggest that some, but not all, individuals with OCD experience difficulties with RI.

An important avenue for further research in this area is to elucidate which factors contribute to the heterogeneity of RI findings in OCD. Better knowledge of which individuals with OCD exhibit RI deficits will be important for understanding the neurocognitive basis of this disorder, since several models propose that impaired inhibitory control is crucially involved in the generation and/or maintenance of OCD symptoms (see Robbins et al., 2019). Further, the ability to withhold inappropriate/unwanted behaviors is a key component employed in cognitive behavioral therapies for OCD; understanding which individuals with OCD may have difficulty with such treatments due to RI deficits (and may require adaptations to be made to cognitive behavioral therapies) could have clinical utility (Kalanthroff et al., 2018). Thus, it is important for studies examining RI in OCD to investigate factors that may influence findings.

While there is evidence that factors such as age (Borella et al., 2008; Smittenaar et al., 2015), medication (Skandali et al., 2018), depression (Snyder et al., 2013) and anxiety symptoms (Robinson et al., 2013; Grillon et al., 2017) influence either behavioral RI or its neural correlates outside the context of OCD, it is still unclear which factors play a differential role in OCD.

Candidate moderators of the relationship between OCD and RI have been investigated through meta-regressions in meta-analytic work. Demographic variables such as age and sex were generally non-significant for RI (Shin et al., 2014; Snyder et al., 2015). OCD symptom severity was associated with worse RI performance in a recent meta-analysis by Abramovitch et al. (2019) with a medium effect size, but SSRT specifically explained only a small portion of the variance in OCD severity, while the relationship between RI performance and OCD severity was non-significant in previous studies (Abramovitch et al., 2013; Shin et al., 2014; Snyder et al., 2015). Depression symptoms have not been found to influence the relationship between RI and OCD (Abramovitch et al., 2011; Abramovitch et al., 2013), but Koorenhof and Dommett (2019) recently reported findings suggesting that comorbid depression might improve RI in OCD. Although anxiety was not a significant moderator in the study by Abramovitch et al. (2011), anxiety symptoms have rarely been considered in RI studies. The influence of medication has also been examined. Snyder et al. (2015) found a borderline effect for the use of medication on inhibitory performance in OCD and Abramovitch et al. (2013) found an effect for the use of neuroleptics, which was non-significant after correcting for multiple analyses. However, most studies show no effect of medication on neuropsychological performance in OCD (Abramovitch et al., 2011; Shin et al., 2014; Simpson et al., 2006), with Kalanthroff et al. (2017) reporting no difference in response inhibition when comparing medicated and unmedicated patients. Although most factors have not shown a moderating effect, there are important inconsistencies, especially regarding OCD severity, which warrant further research.

In addition to RI, the SST allows for the analysis of error monitoring (EM), which refers to the process of monitoring one’s own behaviour and adjusting performance following errors (Gehring et al., 1993; Holroyd and Coles, 2002). In OCD, neuroimaging and electrophysiological studies indicate that individuals have difficulty effectively disengaging from observing their own mistakes due to a hyperactive monitoring mechanism, reflected in enhanced error-related neural activity such as the error-related negativity (ERN) event-related potential (ERP) component (Endrass et al., 2008; Riesel et al., 2019) and increased activity in cortical regions involved in EM, including the anterior cingulate cortex (Norman et al., 2019). However, evidence for excessive EM as assessed with behavioural indices such as post-error slowing (PES), the slowing of performance following an error (Verbruggen and Logan, 2008), as opposed to neural measures, is mixed in OCD: there have been studies reporting enhanced PES indicative of hyperactive EM (Fitzgerald et al., 2005), no differences in PES between OCD and NPC (Endrass et al. 2010; Carrasco et al. 2013) and decreased PES in OCD indicative of impaired EM (Agam et al. 2014; Modirrousta et al. 2015). Moreover, while meta-analyses in OCD have found no influence of factors such as age, sex, OCD severity, depression symptoms and the use of medication on neural correlates of EM (Norman et al., 2019; Riesel 2019), there have been few investigations of which factors may contribute to the heterogeneity of behavioural EM in OCD. Anxiety and depression severity were not found to influence EM (Modirrousta et al., 2015) and, while Berlin et al. (2018) found no effect for OCD severity, PES was a predictor of compulsion severity with a medium effect size. Nevertheless, factors such as age (Ruitenberg et al., 2014), sex (Fischer et al., 2016), depression (Schroder et al., 2013; Ladouceur et al., 2012), anxiety (Hajcak et al., 2003) and medication (Fischer et al., 2015) have also been shown to influence EM in individuals without OCD, and further systematic investigations of these potential moderators may help explain the inconsistency of findings in OCD.

Interestingly, a recent meta-analysis examining neural activity underlying RI and EM in OCD suggested that impaired RI and excessive EM may be related and together contribute to symptoms in OCD, with the EM system continually detecting compulsive errors and an impaired RI system unable to correct those erroneous compulsions, resulting in a loop of compulsive behaviour (Norman et al., 2019). The effect found was of small magnitude and the authors suggested that future studies examining trial-to-trial modulations in neural activation following error processing were needed to confirm this effect. When it comes to behavioral measures, RI and EM were not found to be related in a recent study by Berlin and Lee (2018), which examined data from the SST. In this study, the correlation between SSRT and PES presented a small effect size which did not reach significance. Nonetheless, investigations of this association are scarce and further work may improve understanding not only of the formation of OCD symptoms but also the heterogeneity of RI and EM findings in OCD if RI and EM influence one another’s relationship with this disorder.

The first aim of the current study was to compare behavioral indices of RI and EM between adults with OCD and NPC in one of the largest samples collected to date. Next, we explored how factors which have been shown to influence RI and EM outside of the context of OCD impact these functions in OCD. These factors include age, medication and co-occurring anxiety and depression, as well as biological sex for EM. We also included OCD symptom severity as there have been mixed results regarding the role of this factor. Since previous findings have been inconsistent in terms of whether RI and EM differ between OCD and NPC and how the aforementioned factors influence these functions, we did not make specific directional hypotheses. Finally, the last aim of this study was to investigate how RI and EM are related to one another in OCD and which clinical variables affect this relationship, namely, OCD severity, depression, anxiety and medication. In line with meta-analytic findings (Norman et al., 2019), we predicted that greater impairments in RI would be associated with more hyperactive EM in OCD.

Methods

Participants

Ninety-two patients with a primary diagnosis of OCD and 65 NPC were included in this study. Sixty-five patients were recruited from the OCD clinic of the Institute of Psychiatry of the Clinics Hospital of the Medical School of the University of São Paulo (IPq-HC-FMUSP) in São Paulo, Brazil, and 33 NPC were recruited from the local community of São Paulo. The remaining participants were 27 patients and 32 NPC recruited from the OCD clinic of the Institute of Psychiatry of the Federal University of Rio de Janeiro (IPUB-UFRJ), Rio de Janeiro, Brazil, and from the D’Or Institute for Research and Education (IDOR). At both sites, NPC were screened and patients were diagnosed by board certified psychiatrists with 15 years or more of experience with OCD using the Brazilian version of the Structured Clinical Interview for DSM-IV (SCID; First et al., 1998; Del-Ben et al., 2001). Comorbid psychiatric conditions were assessed by the same clinicians using the SCID and are reported in Table 1. Exclusion criteria for patients were: age younger than 18 or older than 60 years, current psychoactive substance abuse, current or past psychoactive substance dependence, intellectual disability (defined as IQ scores < 70), neurological disorders and current or past psychotic disorders. Exclusion criteria for NPC were the same as for patients as well as the presence of any current or past psychiatric diagnosis. Patients were medication free or stably medicated for at least 6 weeks before participating in the study. Pharmacotherapy at time of assessment is described in Table 2. All participants provided written informed consent to take part in the study and the research was approved by both local ethical review boards (Clinics Hospital, SP: 42447814.3.000.0068 and Federal University of RJ: 05089412.2.1001.526).

Table 1.

Psychiatric comorbidities according to DSM-IV-R among the 92 patients with OCD

Comorbidities		N (%)
Depressive Disorders	Major Depression	48 (52.2)
	Dysthymic Disorder	16 (17.4)
Anxiety Disorders	Generalized Anxiety Disorder	32 (34.8)
	Social Anxiety Disorder	14 (15.2)
	Specific Phobia	6 (6.5)
	Panic Disorder	6 (6.5)
	PTSD	5 (5.4)
ADHD	-	9 (9.8)
Substance Related Disorders	Alcohol	2 (2.2)
	Other	2 (2.2)
Tic Disorders	Tourette Syndrome	2 (2.2)
Impulsive Disorders	Trichotillomania	1 (1.1)

OCD = Obsessive Compulsive Disorder; PTSD = Post-Traumatic Stress Disorder; ADHD = Attention Deficit Hyperactivity Disorder.

Table 2.

OCD and NPC demographics and clinical characteristics

	OCD (N = 92)	NPC (N = 65)	t-value	p-value
Demographics
Sex – M/F	40 (43.5%)/52 (56.5%)	28 (43.1%)/37 (56.9%)	-	0.909
Age – mean (SD)	38.4 (11.1)	35.8 (10.7)	1.49	0.138
Years of education – mean (SD)	13.9 (3.5)	14.8 (4.2)	−1.32	0.188
IQ – mean (SD)	98.9 (13.3)	102.1 (13.8)	−1.42	0.158
Clinical characteristics – mean (SD)
Y-BOCS	27.9 (7.4)	-	-	-
BAI	18.4 (9.9)	11.1 (10.4)	4.43	< 0.001
BDI	18.2 (11.3)	3.4 (3.9)	11.66	< 0.001
Medications at time of study
SSRI	47
Antipsychotics	8
Benzodiazepines	8
Tricyclic antidepressants	5
Drug free	48

OCD = obsessive-compulsive disorder; NPC = non-psychiatric controls; M = male; F = female; IQ = total intelligence quotient; Y-BOCS = Yale-Brown Obsessive Compulsive scale total score; BAI = Beck Anxiety Inventory score; BDI = Beck Depression Inventory; SSRI = selective serotonin reuptake inhibitor.

Materials and procedures

Demographics and clinical assessment –

Each hospital collected sociodemographic and clinical data from their respective participants using a structured interview administered by a board certified psychiatrist or a psychologist with 15 years or more of experience in OCD. Demographic variables included age, sex, years of education and total intelligence quotient (IQ), evaluated with the Brazilian version of the Wechsler Abbreviated Scale of Intelligence (WASI) (Trentini, 2014; Wechsler, 1999). The severity of OCD symptoms was assessed using the Yale-Brown Obsessive Compulsive Scale (Y-BOCS; Goodman et al., 1989). Since there is no formal validation of Y-BOCS in Portuguese, we used a translated version of the original scale that has been widely used in research in Brazil (Asbahr et al., 1992; Miguel et al., 2008). Severity of co-occurring anxiety and depression symptomatology was assessed using the Beck Anxiety Inventory (BAI) and the Beck Depression Inventory (BDI), translated and validated in Brazilian Portuguese with adequate psychometric properties (Beck et al., 1988a; 1988b; Cunha, 2001; Gorenstein and Andrade, 1996).

Stop-Signal Task –

The neuropsychological evaluations were conducted in each hospital by the same psychiatrists or psychologists within one week of the demographic and clinical interview. To assess RI in patients and NPC, we used the SST from the Cambridge Neuropsychological Test Automated Battery (CANTAB ®; Logan, 1994). An arrow pointing either to the left or to the right served as the stimulus for the go task. On a randomly selected 25% of the trials, the go stimulus was followed by an auditory tone (100 ms, 300 Hz), the stop stimulus, which indicated that the response should be suppressed. The time between the presentation of the arrow and the auditory tone creates the stop signal delay (SSD), which varied across trials according to the participant’s performance in order to ensure that the overall proportion of successful inhibition was around 50%. Successfully inhibited responses increased the SSD by 50 ms on the next trial, whereas failed inhibitions decreased the SSD by 50 ms on the next trial. To ensure unpredictability of the SSD, each block began with an SSD of either 100, 200, 400 or 500 ms, after which the tracking procedure began. Participants first completed a training block consisting of 16 go trials. Subsequently, 5 experimental task blocks were presented, each of which was divided into 4 sub-blocks. Each of these subblocks was composed of 12 go trials and 4 stop trials, for a total of 240 go trials and 80 stop trials across the entire task. Participants were instructed at the beginning of the task to respond as quickly and as accurately as possible and received visual and verbal performance feedback at the end of each block. Responses were made using a controller (press pad), which they held in both hands. They responded with the right thumb to right-pointing arrows and the left thumb to left-pointing arrows.

RI was indexed by the SSRT, which represents the time between the presentation of the stop signal and the point at which the stop process occurs. The SSRT was calculated automatically within the CANTAB programme by subtracting the mean SSD from the mean reaction time (RT) on go trials and given as a direct output in milliseconds (ms). Other outputs from the SST were mean RT on go trials, total direction errors, SSD and proportion of successful inhibitions. Following recommendations from an SSRT reliability study (Congdon et al., 2012), we excluded participants with proportions of successful stops smaller than 25% and greater than 75%. We also excluded participants with outlying numbers of total direction errors (defined as 3 standard deviations above the group mean).

To assess EM, for each participant we first calculated mean RTs of go trials after each of three possible outcomes of a trial: correctly responded-to go trials after a go trial (post-go trials – pG), correctly responded-to go trials after a successful stop trial (post-stop-success – pSS) and correctly responded-to go trials after a failed stop trial (post-stop-error – pSE). EM was indexed by PES = pSE – pSS, which reflects the magnitude of slow-down in RT after a failed inhibition as compared to after a successful inhibition. A large PES indicates a larger amount of EM and behavioural adjustment following failed stops when compared to successful ones. The EM analyses included a smaller number of participants (85 OCD and 63 NPC) because of missing data.

Statistical methods

We assessed group differences in demographic and clinical variables using chi-square tests for sex and independent-samples t-tests for continuous variables, which included age, years of education and IQ.

To test for group differences in RI and EM between OCD and NPC and to analyze which factors influenced RI and EM performance, we used two linear regression models, one for SSRT and one for PES. The following variables and interactions were included as predictors: group, age, medication, and Y-BOCS, BAI and BDI scores. Interactions between group and age, BAI score and BDI score were also included to investigate moderating effects of these variables on group differences in RI. Sex and an interaction between group and sex were included only for PES, as this has been shown to influence EM but not RI. To test for influences of each type of medication in SSRT and EM, we used ANOVAs comparing these variables between users and non-users of each medication. We did not include individual medications in the regression models because of small sample sizes.

To investigate associations between RI performance and EM and potential differences in this association between groups, a third regression model was conducted with PES and group, and the interaction between group and PES, predicting SSRT in a first block. In a second block of this model, we added interactions between PES and Y-BOCS, BDI and BAI to investigate potential moderators of the relationship between RI and EM.

Significant interactions between group and moderating factors were further investigated by computing Pearson correlation coefficients between RI or EM and the moderating variable within each group. Subsequently, we applied Fisher’s Z transformations using the ‘cocor’ package for R (Diedenhofen and Musch, 2015) to test whether the strength of those correlations differed significantly between groups. For all analyses, the alpha level for statistical significance was set to p < 0.05. R 3.5.0 (R Development Core Team 2016) and IBM Statistical Package for the Social Sciences 23.0 (IBM Corp. Released 2015) were used for the analyses.

Results

Demographics and clinical assessments

Descriptive statistics of demographic and clinical characteristics can be seen in Table 2. Groups did not differ significantly regarding sex (p = 0.960), years of education (p = 0.188), IQ (p = 0.158) or age (p = 0.138). Age ranged from 18 to 60 in the OCD group and from 21 to 60 in the NPC group, and both distributions differed significantly from normality as assessed by Shapiro-Wilk tests (W = 0.964, p-value = 0.013 for OCD; W = 0.928, p-value = 0.001 for NPC). Medicated and unmedicated OCD participants also did not differ significantly regarding age, sex, years of education or IQ (all p > 0.101). The OCD group showed significantly greater clinical symptoms as measured by BAI and BDI scales (both p < 0.001).

Group differences and factors influencing RI and EM in OCD

Descriptive statistics of SST results are shown in Table 3, while the results for regression models predicting SSRT and PES are shown in Table 4. The regression model used to predict SSRT included group, age, medication, and Y-BOCS, BAI and BDI scores, as well as interactions with group, and explained 9.68% of the variance in SSRT (F(9,144) = 2.822, p = 0.004). Examination of the beta coefficients showed that, while OCD presented a slightly larger SSRT when compared to NPC with a coefficient of medium to large effect size, group was not a significant predictor of SSRT (p = 0.075, ß = −0.645). While there was no association between age and SSRT (p = 0.854, ß = 0.022), the interaction between group and age predicted SSRT with a medium effect size (p = 0.046, ß = 0.611). Investigating this further, we found that age was correlated with SSRT in OCD (r = 0.332, p = 0.001) but not in NPC (r = 0.024, p = 0.850). An one-tailed Fisher’s Z transformation confirmed the significance of the difference between these correlations (p = 0.026, z = 1.941). The moderating effect of age can be seen in Figure 1. Since the age distributions of the two groups were not exactly the same (with quartile splitting points at 30, 37 and 47 years in the OCD group and 27, 33 and 43 years in the NPC group), we also checked whether the presence of the association between age and SSRT in the OCD group but not in the NPC group could have reflected differences in the age distributions of the two groups. We did this by selecting a subset of the OCD group that was matched on age distribution to that of the NPC group and repeating the regression model in this subset. The interaction between age and group remained a predictor of SSRT in this age-distribution matched subgroup with a large effect size (p = 0.015, ß = 0.791), there was a correlation between age and SSRT in OCD (r = 0.366, p = 0.001) but not NPC (r = 0.024, p = 0.850) and these coefficients were significantly different after applying a Fisher’s Z transformation (p = 0.002, z = 2.049), suggesting that the presence of the association was not due to a difference in the age distribution in the OCD group compared to the NPC group. In addition, we ran independent-samples T-tests comparing SSRT between the groups at different age-quartiles to assess whether a group difference restricted to older or younger participants may have driven the interaction. For all age sections, we found no difference between groups: first quartile t(32.9) = −0.829, p = 0.412; second quartile t(19.1) = −1.131, p = 0.272; third quartile t(30.9) = 1.100, p = 0.2798; fourth quartile t(39.4) = 1.004, p = 0.321. Y-BOCS (p = 0.565, ß = 0.138), BDI (p = 0.810, ß = 0.089) and the interaction between group and BDI (p = 0.437, ß = −0.309) were not significant in predicting SSRT. BAI was the only clinical scale to significantly predict SSRT (p = 0.002, ß = −0.397). While the interaction between group and BAI was not significant (p = 0.087, ß = 0.369), the magnitude of the beta coefficient was moderate, which warranted further exploration. Thus, we computed within-group correlation coefficients and found a negative correlation between SSRT and BAI in NPC (r = −0.403, p < 0.001) but not in OCD (r = −0.155, p = 0.140). These coefficients were close to but not significantly different after applying a Fisher’s Z transformation (z = 1.64, p = 0.051). Medication use (p = 0.852, ß = 0.018) was not associated with SSRT; type of medication was also not related to SSRT (see Table S1, supplementary materials).

Table 3.

Descriptive statistics of SST results for OCD and NPC groups.

	OCD		NPC
	Mean	SD	Mean	SD
	N = 92		N = 65
SSRT (ms)	204.44	58.75	201.26	53.12
Mean RT on go trials (ms)	552.23	139.04	551.49	147.10
Direction errors	2.42	2.81	2.30	2.53
Proportion of successful inhibitions	0.53	0.10	0.53	0.08
SSD (50%) (ms)	323.14	100.83	332.88	118.51
	N = 85		N = 63
PES (ms)	52.74	113.52	18.60	98.07

SST = stop-signal task; OCD = obsessive-compulsive disorder; NPC = non-psychiatric controls; SSRT = stop-signal reaction time; RT = reaction time; SSD = stop-signal delay; PES = post-error slowing.

Table 4.

Summary of regression models predicting SSRT and PES from selected variables and analyses of moderators.

	SSRT		PES
	R2 = 9.68%, F(9,144) = 2822, p = 0.004		R2 = 12.13%, F(11,133) = 2807, p = 0.002
	Beta coefficient	p-value	Beta coefficient	p-value
Group	−0.645	0.075	−0.537	0.237
Age	0.022	0.854	0.116	0.352
Age * Group	0.611	0.046	0.398	0.211
Sex	-	-	0.123	0.314
Sex * Group	-	-	−0.124	0.666
BAI	−0.397	0.002	−0.057	0.652
BAI * Group	0.369	0.087	0.097	0.658
BDI	0.089	0.810	0.640	0.149
BDI * Group	−0.309	0.473	−0.881	0.082
Y-BOCS	0.137	0.565	0.720	0.016
Medication	0.018	0.852	−0.147	0.150

SSRT = stop-signal reaction time; PES = post-error slowing; BAI = Beck Anxiety Inventory; BDI = Beck Depression Inventory; Y-BOCS = Yale-Brown Obsessive Compulsive Scale.

Figure 1.

Scatter plot shows the moderating effect of age in the relationship between group and response inhibition. There was an age-related increase of stop-signal reaction time (SSRT) in obsessive-compulsive disorder (OCD) that was not present in the non-psychiatric controls (NPC).

The model used to predict PES included group, age, sex, medication, and Y-BOCS, BAI and BDI scores, as well as interactions with group, and explained 12.13% of the variance in PES (F(11,133) = 2.807, p = 0.002). Group was not a significant predictor of PES (p = 0.237, ß = −0.534). Age did not predict PES (p = 0.351, ß = 0.116), and neither did the interaction between group and age (p = 0.211, ß = 0.399). Y-BOCS was the only clinical scale associated with PES with an effect approaching a large magnitude (p = 0.016, ß = 0.720). BDI (p = 0.149, ß = 0.640) and BAI (p = 0.652, ß = −0.057) as well as the interactions between group and BAI (p = 0.659, ß = 0.097) and between group and BDI (p = 0.081, Beta = −0.882) were non significant. However, due to the large magnitude of the beta coefficient for the interaction between group and BDI, we further investigated this by computing within-group correlation coefficients; there was no correlation between BDI and PES in OCD (r = 0.001, p = 0.931) or in NPC (r = 0.162, p = 0.205). We found no influence of the use of medication on PES (p = 0.150, ß = −0.147); type of medication was also not related to PES (see Table S1, supplementary materials).

Relationship between RI and EM

The regression model with PES, group and an interaction between group and PES predicting SSRT explained only 4.73% of the variance in SSRT (F(3,144) = 3.436, p = 0.019). We found no effect for group (p = 0.220, ß = −0.104) or PES (p = 0.557, ß = −0.080), but the interaction between group and PES was significant (p = 0.020, ß = 0.331) and presented a small effect size. Pearson correlation coefficients computed between PES and SSRT within groups revealed a positive correlation in OCD (r = 0.335, p = 0.002) and no correlation in NPC (r = −0.072, p = 0.572). These coefficients were significantly different (z = 2.48, p = 0.013) after applying a Fisher’s Z transformation. This association is shown in Figure 2. When interactions between PES and clinical scales were included, the model explained 9.34% of the variance in SSRT (F(9,135) = 2.649, p = 0.007). The interactions between PES and Y-BOCS (p = 0.212, ß = 0.617), BAI (p = 0.924, ß = 0.019) and BDI (p = 0.741, ß = 0.072) did not predict SSRT, and in this model the interaction between group and PES became non-significant (p = 0.417, ß = −0.352).

Figure 2.

Post-error slowing (PES) moderated the relationship between stop-signal reaction time (SSRT) and group. While patients with obsessive-compulsive disorder (OCD) presented increased PES in accordance to higher SSRT, this was not found in non-psychiatric controls (NPC).

Discussion

The models used to predict SSRT and PES from selected demographic and clinical variables explained only a small portion of their variance, but our analyses of moderators yielded interesting results. We did not find a significant difference in SSRT between OCD and NPC. Considering prior findings of meta-analyses and the fact that, in our study, group presented a coefficient of medium to large effect size when predicting SSRT, however non-significant, we did not rule out a difference between groups. In addition, although OCD participants showed a PES almost three times larger than NPC, this difference was not significant. While there have been recent reports of reduced PES in OCD (Agam et al., 2014; Modirrousta et al., 2015), increased PES in OCD would be consistent with evidence of enhanced error-related neural activity indicating excessive EM in OCD (Norman et al., 2019; Riesel et al., 2019).

Nevertheless, our findings revealed associations between RI and EM and demographic and clinical characteristics. First, contrary to previous findings in meta-analyses (Shin et al., 2014; Snyder et al., 2015), we found age to be a significant moderator of the relationship between RI and OCD with a medium effect size. While studies in non-psychiatric populations have shown an age-related decline in RI performance in the SST (Borella et al., 2008; Smittenaar et al., 2015), we found that age was positively correlated with SSRT in OCD but not in NPC. Since our study excluded participants older than 60 years old, which was the age from which the decline was more pronounced in the referred studies, we do not believe our findings of no effect of age in NPC to be contradictory with the previous literature. The association in the OCD group remained when we selected a subset of the sample that was matched in age-distribution to the NPC group, suggesting that the presence of the association in the OCD group was not a result of differences in the age-ranges between the two groups. Instead, we believe our findings indicate that the age-related decline in SSRT may be present at an earlier age in OCD compared to NPC. A possible explanation for this may be a lack of cognitive resources in OCD because of the need to control symptoms, which might become exacerbated with decline in cognitive ability with age. Further investigation of this effect by comparing the OCD and NPC groups at different age quartiles indicated that this interaction did not reflect a difference in SSRT between groups that was dependent on age. Rather, our results reflected a difference in associations between age and SSRT in the two groups, with the OCD group showing a significant decline in SSRT with age and the NPC group showing no such association.

However, it is important to point out limitations regarding this specific finding. First, ours is a cross sectional study which does not permit confirmation of an enhanced age-related decline of RI in OCD, and further longitudinal studies to establish this effect are necessary. In addition, it would be important to investigate whether a longer duration of illness and an earlier age of onset are associated with underperformance in RI tasks in OCD, as this would be expected if OCD was associated with an exacerbation of the age-related decline of RI. Thus, these are preliminary results which should be considered with caution and require further investigation.

On the other hand, while there have been studies showing an age-related increase of PES (Ruitenberg et al., 2014), we found neither an overall increase of PES with age nor a differential age-related increase of PES between groups. While we did not find a difference between groups in EM, mean PES was almost three times larger in the OCD group, and group presented a medium effect size when predicting PES. We believe that a hyperactive EM in the OCD group may result in PES being more robust to age-related changes, which may help explain the presence of a moderating effect of age in SSRT but not in PES.

The only clinical scale that significantly predicted SSRT was the BAI measure of anxiety, which was negatively associated with RI with a medium effect size. This association between SSRT and anxiety is consistent with previous findings in the literature, which reported better response inhibition in association with induced (Robinson et al. 2013) and clinical (Grillon et al. 2017) levels of anxiety - an effect that was interpreted as reflecting an adaptive function related to harm avoidance. Moreover, although anxiety symptoms did not present a significant moderating effect on the relationship between group and RI, the magnitude of the interaction was moderate and further investigation showed that the negative correlation between RI and anxiety was only present in NPC and not in OCD. Although these findings require replication in future studies, we tentatively interpret the absence of a negative association between anxiety and RI in OCD as reflecting high anxiety levels (in the clinical range) in the OCD group preventing an adaptive effect of anxiety on performance. Symptoms of OCD and depression were not significant moderators of associations between group and RI, which suggests that these variables do not contribute to the heterogeneity in RI performance in OCD reported in previous metaanalyses.

In contrast, EM was associated with Y-BOCS score with an effect size approaching large magnitude, which indicates that more symptomatic patients presented a more active error processing mechanism. A possible explanation for this is that more symptomatic individuals dwell more on aversive outcomes, such as inhibition errors in the task. While there has been a recent report of reduced PES related to greater compulsion severity (Berlin and Lee, 2018), our findings add to a previous error-related neural activity study which found hyperactive EM to be associated with more severe symptoms (Agam et al., 2014).

As commonly used medications for OCD, i.e. selective serotonin reuptake inhibitors (SSRIs), have been shown to influence both RI and EM in individuals without OCD (Fischer et al., 2015; Skandali et al., 2018), we assessed whether medication use influenced RI and EM in OCD. Medication did not affect SSRT or PES. These findings are consistent with previous research reporting that increased error-related neural activity, such as enhanced ERN amplitude, is not altered by medication use (Stern et al. 2010), a finding that has contributed to the proposal that hyperactive EM is a stable, endophenotypic trait for OCD (Kathmann et al., 2016). We also performed analyses excluding patients with comorbid attention-deficit and hyperactivity disorder (ADHD, N = 9), and results were maintained for both models (SSRT and PES), indicating that ADHD symptoms did not influence the findings of this study.

Finally, we found that group was a significant moderator of the RI/EM relationship. There was a correlation between SSRT and PES in OCD but not NPC and these coefficients were significantly different. Although the size of this effect was too small to firmly conclude that EM contributes to worse RI performance of OCD but not NPC, our findings are in line with a recent meta-analysis of fMRI studies by Norman et al. (2019) which suggested that impaired RI and hyperactive EM together create a loop of compulsive behavior that characterizes the disorder and that OCD patients are unable to break. On the other hand, when accounting for possible influences of OCD severity, depression, anxiety and medication in this relationship, we did not find any significant moderators, and group was no longer a significant moderator of the RI/EM relationship. Thus, this should be considered a preliminary finding and further investigation is required to establish this effect.

The findings of this study should be considered in the context of some limitations. First, there were considerable sample size differences between the NPC and OCD groups, with the OCD group having 27 more participants, which is known to affect the statistical power of the analyses (Rusticus and Lovato, 2014). Second, we were unable to analyse several important variables, including age of OCD onset and duration of illness, which limits interpretation of our findings, particularly the finding that age moderated the relationship between RI and OCD. We do not possess these data and we recommend that future studies analysing RI performance in OCD take these variables into account in addition to age of participants. Third, although studies which have shown RI impairment in OCD have suggested that this is the case regardless of symptom dimension (Lei et al. 2015), our study could have benefited from considering different presentations of the disorder from a dimensional perspective, in order to better understand factors influencing performance in RI and EM. Since it has been suggested that OCD dimensions have distinct neural substrates (Van Den Heuvel et al. 2008), we encourage future studies to take this into consideration. Lastly, we used the original versions of the BDI (Beck et al., 1988a) and BAI (Beck et al., 1988b) because at the time the data were collected there was no Brazilian Portuguese translation of the newer versions. Nevertheless, these scales used were translated and validated in our environment (Cunha, 2001; Gorenstein and Andrade, 1996).

In conclusion, we identified factors that may influence SSRT and PES. Older age was related to lower RI in OCD with a medium effect size, and greater OCD severity was associated with hyperactive EM with a large effect size. We also found an association between RI and EM in OCD of small effect size, but no association in NPC, indicating that worse RI and excessive EM may be related in this disorder.