Assessing the effect of subthalamic stimulation on metaperception in obsessive-compulsive disorder

Abstract

Most human studies on perceptual decision-making are correlational, and few addressed the causal role of the neural markers they identify. Here, we took the opportunity offered by deep brain stimulation to study the causal role of subthalamo-cortical networks on perceptual decision-making and perceptual monitoring. We asked participants with obsessive-compulsive disorder (OCD) treated with subthalamic stimulation to detect visual stimuli and provide confidence reports while we recorded scalp electroencephalography. Following a preregistered plan, we compared behavioural and neural results between patients off-stimulation and matched healthy controls and among patients on and off-stimulation to identify the effect of subthalamic stimulation on perceptual choices and metacognitive judgments. Our results indicate that patients off-stimulation detected visual stimuli similarly to matched healthy controls but provided poorer confidence judgments, indicative of a potential metacognitive alteration. Moreover, chronic subthalamic stimulation significantly reduced clinical symptoms, but acute stimulation changes did not impact visual detection, confidence, nor corresponding event-related potentials at the scalp level. We conclude that acute changes in subthalamic stimulation are insufficient to modify the potential metacognitive alterations observed in OCD.

Introduction

Obsessive-compulsive disorder (OCD) is characterised by intrusive thoughts (obsessions) and irrepressible repetitive behaviours or mental acts (compulsions), responsible for severe effects on functioning and quality of life [1]. Individuals with OCD have a heightened sense of doubt [1], with reduced confidence in their decisions based on perception or memory [2, 3], suggesting the existence of a metacognitive alteration. However, the physiopathological origins of this alteration remain poorly understood. Subcortical contributions to confidence are particularly relevant in OCD, as numerous studies point to a disruption of cortical-subcortical circuits in this disorder [4], and because chronic deep brain stimulation (DBS) of the subthalamic nucleus (STN) is now recognised as a promising treatment for drug-resistant OCD [5]. During speeded discrimination tasks, the STN is assumed to affect choices by modulating the threshold at which accumulated evidence leads to commitment to a decision [6–8]. This modulation can impact confidence, given that confidence is defined relative to the decision threshold, as assumed by numerous models based on signal detection theory and evidence accumulation (for reviews, see [9, 10]). Little is known regarding a possible subthalamo-cortical origin of metacognitive alterations in OCD, as most of the literature about the effects of STN stimulation concerns individuals with Parkinson’s disease and motor symptoms [11].

Here, we sought to assess whether STN stimulation could modulate decision thresholds and alleviate metacognitive alterations in OCD. To this end, we asked individuals with OCD and control participants to detect visual stimuli and provide confidence judgments in their responses while we recorded scalp EEG. Following a randomised, double-blind procedure, patients with OCD were tested on and off subthalamic DBS. The following predictions were preregistered before data collection (https://osf.io/qcvsd). First, considering the perceptual and metacognitive alterations in OCD [2, 12], we expected higher detection thresholds and lower metacognitive performance among patients off-stimulation than control participants. Second, assuming that subthalamic DBS lowers perceptual thresholds [13, 14], and that confidence is defined as the distance between sensory evidence and the threshold [15], we expected more liberal detection behaviour and higher confidence for seen stimuli and lower confidence for unseen stimuli in patients on-stimulation than off-stimulation.

Materials and methods

Participants

We recruited individuals with OCD implanted with bilateral 8-lead directional electrodes (Boston Scientific, Marlborough, MA, USA) in the anteromedial part of the STN (Fig. 1A) at least 12 months before the study. Healthy participants with no current or past psychiatric or neurological conditions served as controls. They received 20€ as compensation for their time. Behavioural analyses were conducted on 14 participants with OCD (11 female; age = 44.1 ± 10.0 years; mean ± 95% confidence interval) and 18 healthy controls (13 female subjects; age = 45.5 ± 11.7 years). Supplementary Table 1 and Table 2 provide groups’ demographic information and clinical scores. The protocol was approved by an ethics committee (2012-A00490-43), and all participants provided written informed consent before the study.

Fig. 1. Experimental setting and procedure.

A. Implantation of subthalamic electrodes in participants with OCD. The three territories of the STN are represented in orange (motor territory), blue (associative territory), and yellow (limbic territory). The electrodes target the blue-yellow area, where most contacts are located. B Experimental procedure. Participants in the OCD group were tested either with or without STN stimulation in a randomised, double-blind way. Session order was counterbalanced across participants. On day 0, a clinical evaluation was performed together with a presentation of the task, a detection threshold calibration procedure, and a baseline behavioural assessment. Then, the DBS status was modified according to the double-blind randomisation. On day 1, the EEG and behaviour were recorded after a 16-h washout, followed by a blind clinical evaluation. Depending on the randomisation, the DBS status was switched on or off in the late afternoon. The last recording happened on day 2, after a similar 16-h washout period. After the recording, a final clinical evaluation was performed, and the DBS status was switched back to the initial on status. C Clinical scores before and after long-term subthalamic stimulation in participants with OCD. In blue, the Y-BOCS score was measured at the time of the surgery, before the stimulation electrode implantation, i.e., without any subthalamic stimulation. In red, the Y-BOCS score was measured on day 0 of the protocol, with chronic subthalamic stimulation for at least one year. D Task. A four-second stream of visual noise was presented. A 200 ms face stimulus could be embedded in this visual stream. The intensity level of the stimulus was adjusted to one of five intensity levels around the participant’s detection threshold (50%, 75%, 100%, 125%, 150% and catch trials). After four seconds, the participant reported the number of detected stimuli (0, 1, more than 1) and their confidence level (not sure, moderately sure, very sure) using a keyboard press.

Procedure

Participants were exposed to a 4 s stream of visual noise frames, each presented for 200 ms. A face stimulus could be embedded within a frame at a pseudo-random time ranging from 0.6 to 3.6 s from trial onset. Stimulus intensity was manipulated by changing the contrast between the stimulus and the noise frame (0, 0.5, 0.75, 1, 1.25, or 1.5 times each participant’s detection threshold, measured before the experiment). After each trial, participants were asked to report whether they detected no face, one face, or more than one face. A 0.5 to 0.75 s uniformly random delay between the last noise frame and the report prompt ensured that EEG signals after the stimulus onset were not contaminated by motor activity due to the button answer associated with the report. Next, participants reported their confidence in their detection report on a 3-level scale corresponding to: “not sure”, “moderately sure”, and “very sure” (Fig. 1D). The experiment lasted 1 h maximum, corresponding to 5 to 7 blocks of 30 trials each. Participants in the OCD group were tested for two sessions, either with or without STN stimulation, in a randomised, double-blind way. Session order was counterbalanced across participants, and a washout period of 16 h was respected between sessions (Fig. 1B).

Behavioural analysis

We analysed detection and confidence as a function of stimulus intensity and group (controls, patients off-stimulation, patients on-stimulation) using Bayesian mixed-effects regressions (see SI). Detection reports (yes/no) were analysed with a Bayesian mixed-effects binomial regression with group (control vs off, on vs off), stimulus intensity (0, 0.5, 0.75, 1, 1.25, 1.50 detection threshold), and stimulus onset (0.6 to 3.6 s) as fixed effects:

$$\left(model1\right)detection~group*intensity+onset+\left(group+intensity\mid participant/block\right)$$

Confidence was analysed with a Bayesian mixed-effects ordinal regression with group, stimulus intensity, detection answer (yes/no), and stimulus onset as fixed effects:

$$\left(model2\right)confidence~group*intensity*detection+onset+\left(group+intensity+detection\mid participant/block\right)$$

Electroencephalography

EEG was acquired with 64 electrodes at a sampling rate of 5000 Hz. All the preprocessing steps were performed using the Matlab Fieldtrip toolbox and in-house scripts. The main steps were low-pass filtering (45 Hz), downsampling to 500 Hz, high-pass filtering (0.5 Hz), additional notch filters at 50 Hz and half the stimulation frequency, epoching around stimulus onset, re-referencing to a common average, and independent component analysis (see SI). We first identified clusters of responsive electrodes. An electrode was considered responsive when it presented any significant effect using the following models, during at least 50 ms in a row, adjusting for false discovery rate. The models were linear mixed-effects regressions for each electrode and each time sample, excluding catch trials in which no stimulus was presented:

$$\left(model3\right)amplitude~intensity*detection*group+\left(1\mid participant\right)$$

Model 4 was similar to model 3 but considered trials with detected stimuli only (HIT), and confidence instead of detection:

$$\left(model4\right)amplitude\left(HIT\right)~intensity*confidence*group+\left(1\mid participant\right)$$

Effects of interest were further examined with Bayesian models, including a full random effect structure (see [16] for a similar approach). Namely, we selected the electrode with the most prolonged effect for each model. We formed a cluster with its surrounding electrodes showing significant effects for at least 50 milliseconds at any time during this time window. Then, we averaged the amplitudes over electrodes and over the corresponding time window to fit the following models:

$$\left(model5\right)amplitude~detection*group*intensity+\left(detection*group*intensity\mid participant\right)$$

$$\left(model6\right)amplitude\left(HIT\right)~confidence*group*intensity+\left(confidence*group*intensity\mid participant\right)$$

Categorical variables were dummy-coded, with the following references: control group or DBS-off group, low confidence, no stimulus detection.

Results

Metacognitive alteration in individuals with OCD

We first compared demographic and clinical data between control participants and patients with OCD at the beginning of the protocol, on-stimulation. The groups did not differ significantly in age, education, anxiety, or depression (SI Table 1). Patients had received chronic, continuous STN stimulation for at least one year (mean stimulation duration = 5.7 years ± 1.2 years), resulting in a significant reduction of symptoms score as measured on the Yale-Brown Obsessive Compulsive Scale (mean Y-BOCS before surgery = 33.36 ± 2.61, corresponding to an extreme level of symptoms; after chronic stimulation = 13.43 ± 5.82 – mild symptoms, BF > 8000, Fig. 1C).

We next sought to assess if OCD was associated with altered perceptual abilities by comparing the probability of a stimulus being detected (stimulus detectability) between groups (control participants vs patients off-stimulation) as a function of sensory evidence (stimulus intensity). Contrary to what we had pre-registered, we found inconclusive evidence concerning the main effect of group (estimate −0.37 [−1.89 1.14]; BF = 0.38) and interaction between group and stimulus intensity (estimate 0.56 [−0.94 2.02], BF = 0.41, Fig. 2A, B, SI Table 4). These results suggest that participants with OCD off-stimulation and control participants had qualitatively similar perceptual performances. We also checked that there was no difference between the absolute value of the detection threshold measured in participants with OCD at day 0, compared to control participants (t[29.54] = −1.29, BF = 0.59). At the neural level, we found equivalent EEG responses between groups. A Bayesian mixed-effects regression (model 5) on a cluster of electrodes around CPz showing an interaction between detection and intensity revealed no conclusive difference between groups (Fig. 2C, D and SI Tables 5-6 for full model results and electrode selection). Together, these results indicate that participants with OCD showed no impairment in perceptual decision compared to control participants, making a direct comparison of confidence between the two groups possible.

Fig. 2. Participants with OCD off-stimulation (in blue) compared to control participants (in green): detection.

A Detection reports. The dots show the mean probability of detection for each stimulus intensity level. The line and ribbon show the Bayesian model prediction and 95% credible interval. B Posterior distributions from the Bayesian model predicting detection reports. Each row shows the posterior distribution for one of the model parameters, whose name is indicated at the beginning of the row. C Detection-related event-related potential (ERP). The topography represents the mean t-statistic of the most significant effect during the time window, i.e., mean effect of intensity x detection, between 0.38 and 0.66 s after stimulus presentation. The dots represent the electrodes of interest, i.e., electrodes responding to intensity x detection for more than 50 ms in a row, within the time window of interest, and surrounding the most responsive electrode, CPz. Below is the amplitude of the evoked ERP on the cluster of interest. The time window of interest is shown in grey. D Posterior distributions from the Bayesian model predicting EEG amplitudes depending on detection reports. Each row shows the posterior distribution for one of the model parameters, whose name is indicated at the beginning of the row.

Contrary to stimulus detection, we found an interaction between group, response, and stimulus intensity on confidence (estimate = 0.90 [0.14 1.65], BF = 2.56). We interpret this as weak evidence as the credible interval did not overlap with zero, but BF was smaller than 3 (SI Table 10). A SExiT framework sensitivity analysis, which quantifies evidence in probabilistic terms and is less dependent on arbitrary priors [17], showed that the effect has a 99.06% probability of being positive, a 98.66% probability of being statistically significant, and a 93.97% probability of being large (>0.30), which suggests that the effect is not only present but also meaningful in magnitude (SI Fig. 1). This interaction implied that confidence tracked the intensity of detected stimuli less well in the OCD than in the control group, in line with a metacognitive alteration in the OCD group (Fig. 3A, B). A visual inspection of the data (Fig. 3A, SI Fig. 2) suggested that participants with OCD off-DBS may have completely avoided high-confidence ratings when the stimulus was detected at low intensity, whereas control participants reported some trials with high-confidence ratings. To unpack the main regression results in detail and assess whether this apparent difference was significant and specific to any particular stimulus intensity, we ran additional post hoc analyses. Predicted confidence values across intensity levels and detection responses are reported in SI Table 11. However, no post hoc contrast for a given stimulus intensity-detection combination showed a difference between OFF and CTR (SI Table 12). Additionally, we ran models on low- and high-intensity trials separately, which failed to provide any evidence in favor of a group effect at a given stimulus intensity, pointing toward a metacognitive alteration that is distributed across stimulus intensity levels and across detection responses rather than specific to a stimulus intensity (SI Tables 13-14). Time-wise analyses of EEG amplitude revealed an interaction between intensity and confidence over a cluster of electrodes around Pz (SI Table 10). Follow-up Bayesian analyses at the cluster level revealed an absence of a main effect of group (estimate = −0.1 [−1.38 1.15], BF = 0.30) and inconclusive evidence concerning the interaction between group and other independent variables (Fig. 3C, D, SI Table 15-16). Together, these results indicate that the potential metacognitive alteration observed behaviorally was not observable in terms of EEG amplitude at the scalp level.

Fig. 3. Participants with OCD off-stimulation (in blue) compared to control participants (in green): confidence.

A Confidence reports. Probability of confidence level for each stimulus intensity level when the stimulus is detected. B Posterior distributions from the Bayesian model predicting confidence reports. Each row shows the posterior distribution for one of the model parameters, whose name is indicated at the beginning of the row. Only effects related to the group are shown here. C Confidence-related evoked potential. The topography represents the mean t-statistic of the most significant effect during the time window, i.e., mean effect of intensity, between 0.38 and 0.51 s after stimulus presentation. The dots represent the electrodes of interest, i.e., electrodes responding to intensity for more than 50 ms in a row, within the time window of interest, and surrounding the most responsive electrode, Pz. Below is the amplitude of the evoked ERP on the cluster of interest. The time window of interest is shown in grey. Low and medium-confidence reports have been pooled into a single Low level to increase plot readability, and all stimulus intensity levels are pooled together. D Posterior distributions from the Bayesian model predicting EEG amplitudes depending on confidence reports. Each row shows the posterior distribution for one of the model parameters, whose name is indicated at the beginning of the row. All parameters are shown here.

No effect of STN stimulation on detection and confidence

Next, we assessed the clinical effect of DBS by comparing clinical outcomes on and off-stimulation. As detailed in the previous section, OCD symptoms measured with the Y-BOCS were significantly lower on-stimulation before the experiment than before surgery (BF > 8000). Because some Y-BOCS items account for symptoms over the last course of days, it was uninformative to collect a complete Y-BOCS questionnaire on days 0, 1 and 2 of the protocol. Instead, we collected a complete Y-BOCS on day 0; then, for day 1 and 2 (which could be either ON or OFF DBS according to randomization), we focused on Y-BOCS items that were deemed more susceptible to rapid changes due to the stimulation status over the last 24 h, namely : Resistance against Obsessions (item 4) and Compulsions (item 9; both hereafter summed as YBOCS-R, ranging from 0 – absence of symptom to 8 – high level of symptoms) and Control over Obsessive Thoughts (item 5) and over Compulsive Behavior (item 10; hereafter summed as YBOCS-C ranging from 0 – absence of symptoms to 8 – high level of symptoms). Y-BOCS sub-scores on- and off-stimulation collected after each experimental session did not differ significantly (mean YBOCS-R off-stimulation = 3.38 ± 2.07, mean on-stimulation = 2.53 ± 1.61, t[12] = 1.8, BF = 0.99 ; mean YBOCS-C off-stimulation = 3.92 ± 2.51, mean on-stimulation = 3.23 ± 2.51, t[12] = 1.6, BF = 0.74 ; see SI Table 2 for additional comparisons and SI Table 3 for individual values).

To evaluate the impact of STN stimulation on stimulus detectability, we compared detection reports on- and off-stimulation. We found evidence for an absence of a stimulation effect (estimate 0.34 [−0.54 1.21], BF = 0.25), and absence of interaction between stimulation and stimulus intensity (estimate = −0.04 [−0.80 0.72], BF = 0.18, Fig. 4A, B, SI Table 7). False alarm rates did not differ significantly between groups (t[13] = 0.26, BF = 0.28), nor did the proportion of trials in which more than one face was reported (t[13] = 0.77, BF = 0.38). At the neural level, we found that EEG amplitude co-varied with stimulus intensity and detection irrespective of STN stimulation on a cluster of electrodes around CP1, from 0.38 to 0.56 s after stimulus presentation. Follow-up Bayesian analyses at the cluster-lever (model 5) confirmed the interaction between intensity and detection (estimate 2.65 [1.66 3.61], BF > 16,000). However, they revealed weak evidence in favor of a main effect of the stimulation and inconclusive evidence of an interaction effect of stimulation (Fig. 4C, D, SI Table 8-9). In sum, DBS did not significantly alter stimulus detectability.

Fig. 4. Participants with OCD off-stimulation (in blue) and on-stimulation (in red): detection.

A Detection reports. The dots show the mean probability of detection for each stimulus intensity level. The line and ribbon show the Bayesian model prediction and 95% credible interval. B Posterior distributions from the Bayesian model predicting detection reports. Each row shows the posterior distribution for one of the model parameters, whose name is indicated at the beginning of the row. C Detection-related evoked potential. The topography represents the mean t-statistic of the most significant effect during the time window, i.e., the interaction between intensity and detection on CP1 (190 ms) from 0.38 to 0.56 s after stimulus presentation. The dots represent the electrodes of interest, i.e., electrodes responding to intensity for more than 50 ms in a row, within the time window of interest, and surrounding the most responsive electrode, CP1. Below is the amplitude of the evoked ERP on the cluster of interest. The time window of interest is shown in grey. D Posterior distributions from the Bayesian model predicting EEG amplitudes depending on detection reports. Each row shows the posterior distribution for one of the model parameters, whose name is indicated at the beginning of the row.

Next, we checked if DBS modulated confidence and possibly modulated the metacognitive alterations we observed among participants with OCD off-DBS, relatively to controls. To do so, we examined confidence ratings depending on the intensity of detected and undetected stimuli in patients on- and off-stimulation. Contrary to the control-patients analysis, and against what we had pre-registered, we found evidence in favour of an absence of interaction between STN stimulation and intensity (estimate = 0.08 [−0.38 0.54], BF = 0.12), detection (estimate = 0.26 [−0.49 1.02], BF = 0.25) or intensity and detection (estimate = 0.06 [−0.68 0.80], BF = 0.18; Fig. 5A, B, SI table 17). At the neural level, we found significant EEG amplitude modulations by stimulus intensity and confidence over only one occipital scalp location (PO8). Follow-up Bayesian analyses at this location revealed evidence in favour of an interaction between stimulation and confidence (estimate = −2.36 [−4.85 0.1], BF = 3.57), irrespective of stimulus intensity (Fig. 5C, D and SI tables 18-19). Together, these results indicate that subthalamic stimulation affected cortical evoked potential related to confidence. However, within participants with OCD and under the present experimental conditions, the stimulation failed to modify the metacognitive alteration identified among participants with OCD off-DBS compared to control participants.

Fig. 5. Participants with OCD off-stimulation (in blue) and on-stimulation (in red): confidence.

A Confidence reports. Probability of confidence level for each stimulus intensity level, when the stimulus is detected. B Posterior distributions from the Bayesian model predicting confidence reports. Each row shows the posterior distribution for one of the model parameters, whose name is indicated at the beginning of the row. Only effects related to the subthalamic stimulation are shown here. C Confidence-related evoked potential. The topography represents the mean t-statistic of the most significant effect during the time window, i.e., the main effect of confidence, between 0.28 and 0.38 s after stimulus presentation. The dots represent the electrode of interest, i.e., PO8 electrode responding to intensity for more than 50 ms in a row, within the time window of interest. Below is the amplitude of the evoked ERP on the cluster of interest. The time window of interest is shown in grey. Low and medium-confidence reports have been pooled into a single Low level to increase plot readability, and all stimulus intensity levels are pooled together. D Posterior distributions from the Bayesian model predicting EEG amplitudes depending on confidence reports. Each row shows the posterior distribution for one of the model parameters, whose name is indicated at the beginning of the row. All parameters are shown here.

Discussion

We used a visual detection and confidence judgement task to assess perceptual and metacognitive performance in OCD. We found that individuals with OCD have preserved visual detection abilities but altered metacognitive performance (when off stimulation) compared to control participants.

In the absence of pre-DBS baseline measures and given chronic treatment, no causal inference about restoration or normalization of metacognitive performance can be made. However, altogether, our findings provide evidence that metacognitive monitoring could be altered in OCD under chronic stimulation, even when first-order perceptual processing is intact and symptoms reduced. This dissociation indicates that metacognitive alteration is not merely a byproduct of perceptual deficits, as is commonly the case in psychiatric disorders [3, 18] but may represent a core cognitive feature of OCD. Such a selective alteration supports the view that compulsive behaviors may stem from altered monitoring of one’s own cognitive states, rather than from low-level sensory or attentional dysfunctions.

We hypothesised two behavioural differences between the OCD and control groups. First, contrary to our hypothesis, we found no evidence for better stimulus detectability among control participants. Second, we confirmed the existence of a metacognitive alteration that was previously found in individuals with sub-clinical OCD traits such as high compulsivity [19], compulsive behaviour, and intrusive thoughts [20]. This result contrasts with a recent study which showed — with a different methodology — that the metacognitive alteration found in highly compulsive controls was not present in individuals with clinically relevant OCD [21]. This apparent discrepancy may stem from differences in task structure and measurement of confidence, but taken together, these studies suggest that metacognitive dysfunction may depend on task demands or symptom dimensions (e.g., checking vs. intrusive thoughts). Our findings therefore help reconcile inconsistencies in the literature by showing that metacognitive alterations in OCD are observable under conditions emphasizing perceptual uncertainty rather than memory-based decisions.

Exploratory analyses in our study revealed no over- or under-confidence in the OCD group, which is found in some but not all memory studies [3]. Interestingly, we found evidence against a corresponding scaling of the ERP correlate of detection over the centroparietal cortex, previously linked to evidence accumulation and confidence [22, 23]. This absence of effect suggests that the neural locus of this putative metacognitive alteration could be downstream, in frontal regions that are commonly associated with OCD. Although this result is clinically relevant and can contribute to better management of OCD, the comparison of a population with a psychiatric disorder with a so-called control population is undermined by numerous pitfalls [24]. Among others, the comparison with a healthy population is too unspecific to study precise, circumscribed clinical dimensions. Here, we took advantage of the fact that patients were treated with subthalamic DBS to evaluate - in a within-subject design - the role of the STN in perception and metacognition. Contrary to what we had preregistered, we found evidence in favour of the absence of an effect of subthalamic DBS on both visual detection and confidence.

Subthalamic stimulation did not alter either detection thresholds or metacognitive performance. This null result is informative: it suggests that STN modulation may primarily affect rapid, motor-linked decision thresholds rather than slower, metacognitive monitoring processes. In other words, while the STN contributes to the implementation of decision policies, our results indicate it does not directly modify metacognitive insight in OCD, highlighting a functional dissociation between decisional and metacognitive control circuits.

The causal effect of the STN on decision threshold was observed previously during immediate-response tasks with dorsal stimulation in Parkinson’s disease [6, 7], but also ventromedial stimulation in OCD [14, 25]. The reason why we found no effect may be that the STN modulates decision thresholds for immediate but not delayed responses such as in our paradigm. This possibility, however, seems at odds with the fast response of STN neurons to sensory and perceptual content even in the absence of an immediate response [26] or the fact that the effect of STN-DBS on response times is restricted to a short period post-stimulus [27]. Another possibility is that STN-DBS mediates the inhibition between two competing response accumulators [28]. In the case of a detection task with only one accumulator, the STN would have no causal effect on the decision. There may also be several other reasons why subthalamic DBS was ineffective in modulating decision thresholds or changing metacognitive performance in our study. We note that the effectiveness of STN DBS on OCD is not in doubt [5] since the symptoms were drastically lower when participants were enrolled in the research protocol (mean YBOCS: 13.4) compared to before DBS surgery (mean YBOCS: 33.4). However, symptoms before and after the 16 h wash-out period within our experimental protocol were similar, suggesting that not all cognitive processes relevant to OCD symptomatology are affected by acute changes in subthalamic stimulation. Indeed, while anxiety is rapidly influenced by stimulation, OCD symptoms can take much longer to be affected by stimulation [29]. Therefore, this wash-out period may be insufficient for observing behavioural changes in our task. Extending the washout period is disputable, given the discomfort it imposes on patients. Notably, this wash-out period was long enough to observe non-specific side effects that may have precluded the expected effects, including fatigue and headaches, consistent with the literature about STN-DBS for Parkinson’s Disease [30]. Additionally, subclinical depressive symptoms, even without group-level differences in BDI (Beck Depression Inventory) scores, may diminish metacognitive variability or reduce confidence calibration [20]. Moreover, since all patients were chronically treated with STN-DBS and showed significant clinical improvement, it is plausible that any pre-existing metacognitive alterations were partly normalized by long-term stimulation. This interpretation aligns with evidence that DBS can modify aspects of decision-making and self-monitoring over long periods [14]. In this context, our null effect of acute stimulation may reflect a stable, treatment-induced recalibration of metacognitive function rather than indicating that the STN has no causal role. Finally, the variability in stimulation parameters might also have contributed to the inconsistent effects observed across individuals (see SI Table 3). Considering this interindividual variability and linking metacognitive alterations with symptomatology requires working with larger cohorts of patients. This remains a significant challenge today, given that subthalamic DBS remains an experimental treatment with few beneficiaries for OCD [31]. Nevertheless, the rise of DBS in neurology and psychiatry portends better clinical management of metacognitive alterations and a window to causally explore the thalamocortical circuits involved in decision-making and metacognition.

Limitations

Several limitations should be acknowledged to properly contextualize our findings. First, although our study was preregistered, deviations from the original analysis plan occurred. The use of second-order signal detection theory and meta-d′ [32] proved incompatible with our multifactorial design and limited trial numbers, leading to unstable and sometimes uninterpretable estimates. We therefore adopted Bayesian ordinal regression models, which allowed us to model confidence as a function of evidence strength and participant-level random effects. While the distinction between metacognitive sensitivity and response bias remains debated in the literature [33–35], we recognize that the ordinal approach may not fully isolate metacognitive ability from criterion effects. Future studies with higher trial counts and simpler designs will be needed to compare these methods directly.

Second, our preregistration included an analysis of bounded evidence accumulation models fitted to ERP components modulated by DBS. Because no ERP component showed significant modulation by stimulation after correction for multiple comparisons, model fitting was not feasible. Additionally, reaction times were not interpretable in our delayed-response task, preventing behavioral diffusion modeling.

Third, the Bayes factor related to our main interaction (group × response × stimulus intensity) narrowly missed our preregistered criterion (BF = 2.56 vs. threshold = 3). Although supplementary analyses using the SExiT framework [17] support the presence and strength of the effect (SI Fig. 1 and Table 10, 11), we recognize that the statistical evidence for a metacognitive difference remains limited. Therefore, we describe this finding as a potential metacognitive alteration and encourage replication with independent samples and different analytical approaches.

Fourth, the limited sample size (n = 14) and variability in trial numbers for each participant limit the generalizability of our findings. Although Bayesian mixed-effects modeling partially addresses these issues by combining data across participants, smaller datasets can still produce unstable parameter estimates. Replication with larger cohorts will be crucial, especially to connect metacognitive performance with clinical symptom dimensions and DBS parameters.

Finally, our within-subject design focused on short-term DBS effects after a 16-h washout period. This protocol may not capture slower neuromodulatory changes that affect cognitive or metacognitive processes. Future studies using longer washout intervals or long-term assessments could better understand the timing of DBS effects on metacognition.

Together, these limitations underscore the preliminary nature of our findings and highlight the need for future studies with larger samples, refined task designs, and harmonized analytical approaches to further elucidate the neural mechanisms of metacognition in OCD.

Conclusion

This study suggests that a potential metacognitive alteration in patients with OCD persists during acute subthalamic stimulation, despite the therapeutic effect of chronic stimulation. Clinically, identifying a selective metacognitive alteration in OCD may have translational relevance for cognitive interventions that target metacognition. From a neuromodulatory perspective, improving metacognition in OCD may require targeting cortical–subcortical loops beyond the STN, including prefrontal or anterior cingulate regions. Clarifying the effect of DBS on OCD symptoms requires disentangling specific cognitive dimensions (here, decision and metacognition), which may depend on distinct cortical or subcortical loci that are differentially modulated by acute stimulation, within the broader circuits involved.

In summary, this study contributes to a growing body of evidence showing that metacognitive monitoring, rather than perceptual decision-making itself, is selectively altered in OCD. By combining behavioural and electrophysiological measures in a rare DBS cohort, we demonstrate that STN stimulation does not modify these alterations, thereby refining our understanding of the neural circuits supporting metacognition in psychiatric disorders. This work provides a foundation for future causal investigations of metacognitive mechanisms and their clinical modulation.

Author contributions

MPe and NF designed the study. AK, MPe, RM, DG, AP, MPo collected the data. MPe, AK, and NF analysed the data. SC performed DBS surgeries. MPo supervised all clinical aspects of the study. AK, MPe, and NF wrote the article, which all authors reviewed. All authors approved the final version of the manuscript. This work was presented at the Association for the Scientific Study of Consciousness in Amsterdam in July 2022.

Funding

Michael Pereira is supported by a Postdoc Mobility fellowship from the Swiss National Science Foundation (P400PM_199251). Nathan Faivre has received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (grant agreement No 803122).

Data availability

The experimental procedure and analysis plan have been preregistered (https://osf.io/qcvsd). All analysis scripts are publicly available (https://gitlab.com/nfaivre/ocd_dbs_public). Due to the small sample size and the risk of patient re-identification, the data is not publicly available but can be shared upon reasonable request.

Competing interests

The authors declare no competing interests.

Ethics declaration

The research protocol was approved by the Ethics Committee CPP Sud Est V (protocol N° ID/RCB: 2012-A00490-43/5) and the Research Ethics Committee of Grenoble Alpes University (CER Grenoble Alpes - Avis-2019-11-05-4) and was conducted in accordance with the declaration of Helsinki. All participants gave informed consent prior to participation.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41398-026-04423-6.