Does subthalamic stimulation change perceptual confidence in obsessive-compulsive disorder?

A causal test of subthalamic control over confidence

Most human work on perceptual decision-making is correlational: it links brain signals to choices and confidence without showing that the signals cause them. This study used deep brain stimulation (DBS) of the subthalamic nucleus (STN) as a reversible switch. Because the electrodes were already implanted in patients with treatment-resistant OCD, the researchers could compare the same person with stimulation on and off, in randomized, double-blind order, and ask whether subthalamic activity causally shapes perception and metacognition.

Metacognition here means monitoring one's own perceptual decisions, measured through confidence reports. The STN was a plausible target because it is thought to set the threshold at which accumulated sensory evidence triggers a decision, and confidence is often modeled relative to that threshold. If stimulation lowered the threshold, the preregistered prediction was more liberal detection and higher confidence for seen stimuli, with lower confidence for unseen ones.

Detecting a face in noise, then rating certainty

Participants watched a four-second stream of visual noise frames, each shown for 200 ms. A face stimulus could appear in one frame at a pseudo-random time between 0.6 and 3.6 s, with contrast set to 0, 0.5, 0.75, 1, 1.25, or 1.5 times that participant's own detection threshold. After the stream, they reported whether they saw no face, one face, or more than one face, then rated confidence as not sure, moderately sure, or very sure. A short random delay before the report kept motor activity from contaminating the stimulus-locked EEG.

Behavioral analyses covered 14 patients with OCD (11 female; mean age 44.1 years) and 18 healthy controls (13 female; mean age 45.5 years). Patients had been implanted at least 12 months earlier and had received chronic stimulation for a mean of 5.7 years. Each patient completed two sessions, on and off stimulation, separated by a 16-hour washout, while scalp EEG was recorded.

Intact detection but weaker confidence tracking in OCD

Against the first preregistered prediction, patients off stimulation did not show clearly worse detection than controls. The group effect on detection probability (estimate −0.37, 95% interval −1.89 to 1.14) and the group-by-intensity interaction (estimate 0.56, −0.94 to 2.02) both had Bayes factors below 1 (0.38 and 0.41), which the authors treat as inconclusive rather than as proof of equality. Absolute detection thresholds also did not differ detectably between patients at day 0 and controls.

Confidence behaved differently. A three-way interaction of group, response, and stimulus intensity on confidence had an estimate of 0.90 (0.14 to 1.65) with a Bayes factor of 2.56, just under the preregistered threshold of 3. The authors call this weak evidence, but a sensitivity analysis using the SExiT framework put the probability that the effect is positive at 99.06% and that it is large (above 0.30) at 93.97%. Substantively, confidence tracked stimulus intensity less well in patients off stimulation than in controls. Post hoc contrasts at individual intensity-detection combinations showed no significant difference, and separate low- and high-intensity models found no group effect, pointing to an alteration spread across intensity levels rather than tied to one condition.

The confidence difference did not appear in scalp EEG

The study also asked whether the behavioral difference had a visible cortical signature. EEG amplitude around electrode Pz showed an interaction between stimulus intensity and confidence, but follow-up Bayesian analyses found no main effect of group (estimate −0.1, −1.38 to 1.15; Bayes factor 0.30) and inconclusive evidence for interactions involving group. In other words, the confidence-related evoked response that was present did not differ detectably between patients off stimulation and controls.

The authors read this as a clue about where the alteration might sit. They suggest the neural locus could be downstream in frontal regions associated with OCD rather than in the centroparietal signal linked to evidence accumulation and confidence. That is the authors' interpretation, not a measured localization: the study recorded scalp EEG only, so it cannot identify a frontal source directly.

Symptom relief without an acute cognitive shift

Chronic stimulation clearly helped clinically. Yale-Brown Obsessive Compulsive Scale scores fell from a mean of 33.36 before surgery, an extreme symptom level, to 13.43 after at least a year of chronic stimulation, a mild level, with a Bayes factor above 8000. That clinical benefit is the backdrop for the experimental comparison, not a result of the on-off manipulation.

Acute switching produced no measurable change in the task. Detection reports showed evidence for an absence of a stimulation effect (estimate 0.34, −0.54 to 1.21; Bayes factor 0.25) and no stimulation-by-intensity interaction (estimate −0.04, −0.80 to 0.72; Bayes factor 0.18). False alarm rates and reports of more than one face also did not differ between on and off sessions. Confidence showed no interaction between stimulation and intensity (Bayes factor 0.12), detection (0.25), or their combination (0.18). The one neural exception was at occipital electrode PO8, where an interaction between stimulation and confidence appeared (estimate −2.36, −4.85 to 0.1; Bayes factor 3.57), but this did not translate into a behavioral confidence change.

Why an acute null does not rule out the STN

The authors argue the null result is informative rather than empty. They propose that STN modulation may mainly affect rapid, motor-linked decision thresholds, not slower metacognitive monitoring, which would imply a functional separation between decision and metacognition circuits. They also note that prior causal effects of STN stimulation on decision thresholds came from immediate-response tasks, often with dorsal stimulation in Parkinson's disease, whereas this task used delayed reports and stimulation aimed at the associative-limbic territory.

Several alternative explanations are raised and remain unresolved. The 16-hour washout may be too short for cognitive effects, since OCD symptoms can take much longer than anxiety to respond to stimulation, though longer washouts would burden patients. Because all patients were chronically stimulated and clinically improved, any preexisting metacognitive alteration may already have been partly normalized, so an acute on-off comparison could miss a stable treatment-induced recalibration. Stimulation parameters varied across individuals, and the sample of 14 patients limits power to link metacognition to symptom dimensions.

What the confidence measure can and cannot show

The preregistered plan called for second-order signal detection theory and meta-d′, a measure intended to separate metacognitive sensitivity from response bias. The authors report that this approach was incompatible with the multifactorial design and limited trial numbers, producing unstable estimates. They instead used Bayesian ordinal regression models that predict confidence from evidence strength with participant-level random effects. The limitation is explicit: this ordinal approach may not fully isolate metacognitive ability from criterion effects, so the confidence finding should be read as a potential alteration rather than a clean estimate of metacognitive sensitivity.

A second planned analysis, fitting bounded evidence accumulation models to ERP components modulated by stimulation, could not be run because no ERP component showed significant modulation by stimulation after correction for multiple comparisons. Reaction times in the delayed-response task were also not interpretable for behavioral diffusion modeling. These gaps mean the mechanistic account of how the STN might set a decision threshold was not directly tested here.

The priority is chronic versus acute stimulation effects

The paper's most decision-relevant split is between chronic and acute stimulation. Chronic STN stimulation was associated with a large symptom reduction, from a mean Yale-Brown score of 33.36 before surgery to 13.43 on chronic stimulation, yet acute on-off changes over a 16-hour washout left detection, confidence, and scalp ERPs essentially unchanged. For readers weighing whether subthalamic stimulation can address metacognitive doubt in OCD, this study says the acute switch is the wrong lever to test; the open question is whether long-term stimulation, different electrode contacts, or longer washout periods would reveal an effect that a single on-off comparison cannot.

The confidence result also sets a priority for measurement. The group difference in how confidence tracked stimulus intensity was statistically weak by the preregistered Bayes factor criterion (2.56 against a threshold of 3) and was not accompanied by a detectable scalp EEG difference, so the next step is a design with more trials and a cleaner separation of metacognitive sensitivity from response bias. Until then, the strongest supported claim is a dissociation: preserved visual detection alongside altered confidence tracking in OCD off stimulation, with acute subthalamic stimulation insufficient to modify either.

Sources used in this answer