Evidence Accumulation and Neural Correlates of Uncertainty in Obsessive-Compulsive Disorder

Abstract

Background

Decision making is frequently associated with risk taking under uncertainty. Elevated intolerance of uncertainty is suggested to be a critical feature of obsessive-compulsive disorder (OCD). However, impairments of latent constructs of uncertainty processing and its neural correlates remain unclear in OCD.

Methods

In 83 participants (24 OCD patients treated with capsulotomy, 28 OCD control participants, and 31 healthy control participants), we performed magnetic resonance imaging using a card gambling task in which participants made decisions whether to bet or not that the next card would be larger than the current one. A hierarchical drift diffusion model was used to dissociate speed and amount of evidence accumulated before a decisional threshold (i.e., betting or no betting) was reached.

Results

High uncertainty was characterized by a smaller amount of evidence accumulation (lower thresholds), thus dissociating uncertainty from conflict tasks and highlighting the specificity of this task to test value-based uncertainty. OCD patients exhibited greater caution with poor performance and greater evidence accumulation overall along with slower speed of accumulation, particularly under low uncertainty. Bilateral dorsal anterior cingulate and anterior insula distinguished high- and low-uncertainty decision processes in healthy control participants but not in the OCD groups, indicating impairments in anticipation of differences in outcome variance and salience network activity. There were no behavioral or imaging differences relating to capsulotomy despite improvements in OCD symptoms.

Conclusions

Our findings highlight greater impairments particularly in more certain trials in the OCD groups along with impaired neural differentiation of high and low uncertainty and suggest uncertainty processing as a trait cognitive endophenotype rather than a state-specific factor.

Decision-making processes largely depend on the evaluation of risk, a process involving the weighted assessment of the likely cost-benefit ratio of the outcome. This likelihood or probability of the outcome can lead to uncertainty or unsureness and potentially raise anxiety and doubt and worsen preexisting psychiatric symptoms. Abnormalities in processing uncertainty have been suggested to underlie some psychiatric disorders (1,2). Among these, obsessive-compulsive disorder (OCD), which possesses a core feature of elevated intolerance of uncertainty (3), may be a paradigmatic example.

The process of reaching a decision requires the accumulation of noisy information and evaluation of evidence until a decisional threshold is crossed, at which point the response or decision is executed. Differing probabilities of uncertainty can result in different speed and amount of evidence accumulation. A theoretical mechanism underlying obsessive and compulsive behaviors in OCD is that of pathological evidence-gathering behavior intended to reduce uncertainty (4,5), but paradoxically resulting in greater doubt and uncertainty (6) with greater repetitive checking behavior or obsessions (7) leading to the inability to commit to a solid decision.

Studies of evidence accumulation in OCD have had mixed results. Using the Beads task, early results suggested that participants with OCD accumulated more evidence than healthy control (HC) participants before making a final decision (1,8,9) although a more recent study failed to replicate this finding (10). Similarly, a study using the Information Sampling Task did not show differences in evidence accumulation (11). The outcome measures of evidence accumulation in these tasks based on the numbers of beads or boxes revealed before a final decision are relatively discrete variables and may have limited sensitivity. Furthermore, in both laboratory-based and daily decisions, many circumstances require binary choices such as “left or right” in the random-dot motion (RDM) task and “bet or don’t bet” in gambling games, and decision-making processes in these tasks involve factors of conflict and/or uncertainty, which can also be confounded. Approaches such as computational models that take both responses and the distribution of reaction times into account are essential to decompose the latent evidence accumulation process in decision making.

The drift diffusion model is a well-defined model with strong explanatory power for human behavior. The basic assumption of the drift diffusion model is that information is accumulated over time until a criterial amount is reached, at which point a corresponding decision is made. The separation between the two choices or boundaries indexes the response caution, which can be simplified as the amount of evidence needed for reaching one of the two choices and is termed threshold. Another parameter of primary interest is the drift rate, which represents the strength of evidence gathered from existing information. A larger value of drift rate indicates stronger decision evidence, which leads to faster evidence accumulation toward one of the decision boundaries (12,13). An advantage of using the drift diffusion model is that it has the capacity to decompose the evidence accumulation process into parameters such as speed (drift rate) and amount (threshold), which might allow for greater sensitivity and modeling of hidden variables.

Studies have shown increased dorsal anterior cingulate cortex (dACC) activity as a function of uncertainty, along with other brain regions such as orbitofrontal cortex, anterior insula (AI), and ventral striatum (14, 15, 16, 17). One critical white matter structure in OCD is the anterior limb of the internal capsule, which links multiple prefrontal and subcortical regions relevant to OCD (18,19). Capsulotomy, or ablative lesions of the anterior limb of the internal capsule, reduces resting-state functional connectivity between dACC and ventral striatum, and the preoperative connectivity between dACC and dorsal caudate can predict changes in Yale-Brown Obsessive Compulsive Scale (Y-BOCS) scores, thus highlighting the potential influence of capsulotomy on dACC function (20). In this study, we secondarily sought to assess the effects of capsulotomy in OCD on uncertainty processing.

Here, we performed functional magnetic resonance imaging (fMRI) using a card gambling task predicting different probabilities toward the outcome, thus modeling uncertainty as defined by the greatest outcome variance in which the probability is unclear (21). In this task, participants decided whether to bet that the next playing card would be larger than the current one shown on the screen. We used the hierarchical drift diffusion model (HDDM) to decompose different components of evidence accumulation in high versus low uncertainty predicting higher thresholds and slower drift rates as well as a role for aberrant dACC activity in participants with OCD relative to HC participants and a potential influence of capsulotomy on the processing of uncertainty and cingulate activity.

Methods and Materials

Participants

A total of 83 participants (28 OCD control participants [OCD contr group], 24 OCD patients treated with capsulotomy [OCD cap group] and 31 HC participants [HC group]) were recruited. Y-BOCS scores of the OCD contr and OCD cap groups were assessed by a psychiatrist. Beck Depression Inventory and Obsessive Compulsive Inventory–Revised were self-rated by participants. The study was approved by the ethics committee of Ruijin Hospital, and written informed consent was obtained. Participants were compensated for their time. Inclusion and exclusion criteria can be found in the Supplement.

Stimuli and Procedure

This task was coded with PsychToolbox-3 (http://psychtoolbox.org/) on MATLAB, version R2018b (The MathWorks, Inc.). We used a card gambling task adapted from Critchley et al. (17). In this task (Figure 1A), a playing card numbered from 1 to 10 was presented on the left screen for 2.5 seconds. Participants decided whether to bet or not that the next card would be higher than the present one. Participants responded when the blue outline was shown, followed by an outcome phase showing the next card on the right screen for 1 second. Then, a 1.5-second monetary feedback was presented on the screen indicating whether the participant won ¥10 (chose to bet with lower second card), lost ¥10 (chose to bet with higher second card), or received nothing (chose not to bet). The intertrial intervals ranged from 1 to 4 seconds. Increasing risk is defined as the decreasing likelihood of winning money and the increasing likelihood of losing money. In any trial, the outcome card was either higher or lower than the first card, and the probability of the outcome card being higher or lower met the true likelihood of a random set, meaning that card 1 had the lowest risk to bet (largest chance to win). The participants were asked to maximize monetary gain. Only 3 trials were run for cards 1 and 10 as their outcomes were certain. For the other 8 cards, 9 trials were run resulting in a total of 78 trials.

Figure 1.

Experiment paradigm and behavioral results. (A) Card gambling task. The choice phase was presented on the screen for 2.5 seconds with a card ranging from 1 to 10 on the left side of the screen. Participants were asked to make a decision whether to bet or not to bet that the next card would be higher after they saw the blue outline border as a response cue. Outcome with the second card and monetary feedback were then presented subsequently for 1 second and 1.5 seconds separately. (B) Behavioral results. Main effects of cards were significant for all 3 dependent variables. Group differences were found for winning rate and reaction time, both driven by better performance in the healthy control (HC) group than the other 2 groups. Interaction was found only for reaction times driven by significant differences of high vs. low uncertainty in obsessive-compulsive disorder control (OCD contr) group and OCD capsulotomy (OCD cap) group, but not the HC group.

For the behavioral performance, we calculated bet rate, reaction time, and winning rate as a function of uncertainty. Bet rate was defined as the ratio of choosing to bet following the first card, which indicated a risk-taking strategy. Winning was defined as choosing to bet when the second card was higher or choosing not to bet when the second card was lower.

By defining uncertainty as a function of outcome variance, high-uncertainty cards were 4 to 7, and low-uncertainty cards were 1 to 3 and 8 to 10. We further divided low-uncertainty cards into low-uncertainty–low risk (cards 1–3) and low-uncertainty–high risk (cards 8–10) when calculating bet rates as betting behavior was highly related to risk, which caused a strong tendency to choose to bet with small-number cards and choose not to bet with higher-number cards. Repeated-measures analyses of variance were used for the 3 dependent variables with least significant difference in post hoc analyses.

To rule out the influence of medication status, we conducted control analyses with 1) medication or antidepressant on/off as covariates and 2) overall patients medicated/unmedicated or antidepressant/non-antidepressant comparison on task effects.

Hierarchical Drift Diffusion Model

The HDDM uses Bayesian methods to estimate model parameters including threshold, drift rate, starting bias, and nondecision time. This Bayesian-based model estimates parameters as joint posterior distributions based on reaction times and choices. A Python package by Wiecki et al. (22) was used for parameter estimation, in which the Markov chain Monte Carlo sampling method was utilized to approximate the posterior distributions (generating 11,000 samples, discarding 1000 samples as burn-in period). In the model, we focused on the parameters of threshold (the amount of evidence accumulation) and drift rate (the speed of evidence accumulation) (2,5). Note that for drift rate, similar to the bet rate calculation, low-uncertainty–low-risk (cards 1–3) and low-uncertainty–high risk (card 8–10) were separately fitted. The 2 boundaries of the HDDM were “bet” at the top and “no bet” at the bottom. Trials with reaction times less than 300 ms were excluded as outliers because these fast trials force the fitted reaction time distributions to shift toward zero, which may impact the model fitting (23). For statistical analyses, both Bayesian repeated-measures analysis of variance and posterior distribution comparison were used (detailed methods are reported in the Supplement).

General Linear Model Analyses

MRI data acquisition and preprocessing are reported in the Supplement. To assess brain regions contrasting high and low uncertainty, we built a general linear model in which high- and low-uncertainty trials were regarded as 2 regressors and time locked to the onset of the choice phase. The outcome phase was divided into bet–win, bet–lose, no bet–win, and no bet–lose, and the feedback phase was divided into win, lose, and nothing. Thus, the general linear model included 9 task-related regressors and 6 head movement–related regressors.

We first conducted a whole-brain analysis comparing high versus low uncertainty in the HC group to obtain regions of interest (ROIs) sensitive to uncertainty. Then, after false discovery rate correction (p < .05), we extracted the ROI β values of all participants in the 3 groups and conducted 2 (uncertainty) × 3 (group) repeated-measures analyses of variance for each ROI to examine the differences between groups. Control analyses of medication were also done for ROIs.

Results

Demographics and Clinical Characteristics

The demographic and clinical characteristics are summarized in Table 1. There were no differences in gender, age, and years of education. The Beck Depression Inventory scores were highest in the OCD contr group and lowest in the HC group (post hoc: OCD contr vs. OCD cap: p =.007; OCD contr vs. HC: p < .001; OCD cap vs. HC: p = .001). The duration of disorder and age of onset of the 2 OCD groups were not different, but Y-BOCS and Obsessive Compulsive Inventory–Revised scores were higher in the OCD contr group than in the OCD cap group, indicating the treatment improvement effect of the capsulotomy. No differences of medication status were found using the χ² test.

Table 1.

Demographic and Clinical Characteristics of Participants

	OCD Contr Group, n = 28	OCD Cap Group, n = 24	HC Group, n = 31	Statistics
				F/t/χ2	p Value
Gender, Female/Male	11/17	6/18	14/17	2.418	.299
Age, Years	31.30 (7.25)	33.06 (6.46)	34.75 (8.45)	1.554	.218
Education, Years	14.14 (3.65)	12.92 (3.15)	14.48 (3.30)	1.551	.218
BDI	23.75 (14.97)	15.21 (11.91)	4.48 (4.40)	22.257	<.001
Y-BOCS Total	23.39 (9.46)	12.92 (9.83)	–	3.908	<.001
Obsessions	13.25 (5.52)	7.00 (5.68)	–	4.016	<.001
Compulsions	10.14 (6.23)	5.82 (5.37)	–	2.597	.012
OCI-R	21.89 (13.11)	14.46 (12.27)	–	2.099	.041
Duration, Years	12.05 (6.77)	13.63 (6.52)	–	−0.849	.400
Onset Age, Years	19.11 (5.33)	18.38 (4.86)	–	0.514	.609
Medication, On/Off	22/6	14/10	–	2.485	.115
Antidepressant, On/Off	20/8	12/12	–	2.507	.113

Values are n/n or mean (SD).

BDI, Beck Depression Inventory; HC, healthy control; OCD cap, obsessive-compulsive disorder capsulotomy; OCD contr, OCD control; Y-BOCS, Yale-Brown Obsessive Compulsive Scale; OCI-R, Obsessive Compulsive Inventory–Revised.

Behavioral Results

All 3 analyses revealed robust main effects of uncertainty (ps < .001, η_p²s > 0.248) (Figure 1B). For bet rate, no group difference (F_2,80 = 0.215, p = .807, η_p² = 0.005) or interaction (F_4,160 = 2.173, p = .074, η_p² = 0.052) was found. For winning rate, a group difference was shown (F_2,80 = 4.255, p = .018, η_p² = 0.096) with the HC group higher than the OCD contr (p = .046) and OCD cap (p = .007) groups, while interaction was not significant (F_2,80 = 0.360, p = .699, η_p² = 0.009). Reaction times also showed a group difference (F_2,80 = 7.866, p = .001, η_p² = 0.164), with faster responses in the HC group than in the OCD contr (p = .010) and OCD cap (p < .001) groups. An interaction was revealed for reaction times (F_2,80 = 3.311, p = .042, η_p² = 0.076) driven by significant differences of high versus low uncertainty in the OCD contr (p < .001) and OCD cap (p < .001) groups, but not in the HC group (p = .322). No difference was found after controlling for medication status (ps > .067). Thus, OCD participants irrespective of surgical procedure responded more slowly with less optimal (indicated by lower winning rate) betting behavior than HC participants.

HDDM Results

Drift rate (Figure 2) showed extreme evidence that across all groups participants accumulated evidence faster in the low-uncertainty condition than in the high-uncertainty condition (Bayes factor [BF₁₀] > 100). Moderate evidence of group difference was found (BF₁₀ = 3.92), and interaction evidence was extremely strong (BF₁₀ > 100). A post hoc Bayesian t test of the interaction showed that in the low-uncertainty condition, but not in the high-uncertainty condition, the HC group had higher drift rate or faster speed toward the decision boundary than the OCD contr (BF₁₀ = 39.36) and OCD cap (BF₁₀ > 100) groups. A posterior distribution comparison similarly showed that drift rates of the HC group were higher than the OCD contr (p = .002) and OCD cap (p = .001) groups in the low-risk–high-uncertainty condition and lower (note that here lower meant more negative and thus also indicated faster speed) than the OCD contr (p < .001) and OCD cap (p = .001) groups in the low-uncertainty–low-risk condition. No differences were found between the OCD contr and OCD cap groups in low-uncertainty conditions (ps > .174) or between all 3 groups in the high-uncertainty condition (ps > .528).

Figure 2.

Hierarchical drift diffusion model results. (A) Posterior probability distribution of model-fitting results of drift rate and decisional threshold. (B) Schematic representation of decision process in high and low uncertainty, respectively. ∗∗∗Bayes factor (BF₁₀) 30–100; ∗∗∗∗Bayes factor (BF₁₀) > 100. H, high uncertainty; HC, healthy control; L, low uncertainty; LH, low risk–high uncertainty; LL, low risk–low uncertainty; OCD cap, obsessive-compulsive disorder capsulotomy; OCD contr, OCD control.

Threshold (Figure 2) again showed extreme difference of the uncertainty level (BF₁₀ > 100), with participants accumulating more evidence under the low-uncertainty condition. Group difference was very strong (BF₁₀ = 42.51), driven by the lower threshold of the HC group compared with the OCD contr (BF₁₀ > 100) and OCD cap (BF₁₀ > 100) groups, but no difference was found between the OCD contr and OCD cap (BF₁₀ < 1) groups. Anecdotal evidence supported an interaction effect (BF₁₀ = 2.71). In the posterior distribution comparison, the HC group showed higher threshold in low uncertainty than in high uncertainty (p < .001), while this effect in the OCD control (p = .794) and OCD contr (p = .089) groups was not significant.

Put together, high uncertainty was characterized by a smaller amount of evidence accumulation (threshold) but slower drift rates. OCD participants irrespective of surgical intervention showed greater caution overall with greater evidence accumulation and slower drift rates.

fMRI Results

We first used whole-brain analysis to identify regions implicated in uncertainty processing in the HC group demonstrating greater dACC and bilateral AI activity in the high- versus low-uncertainty condition (p < .05 false discovery rate cluster corrected). Then, using these ROIs in subsequent analyses (Figure 3), we showed main effects of uncertainty level (ps < .001, η_p²s > 0.144) with no main group effects (ps > .333, η_p²s < 0.028). Significant interactions were found in all 3 ROIs (dACC: F_2,80 = 3.53, p = .034, η_p² = 0.081; left AI: F_2,80 = 4.09, p = .020, η_p² = 0.093; right AI: F_2,80 = 4.19, p = .019, η_p² = 0.095), which were driven by the differences between high and low uncertainty only in the HC group (ps < .001, least significant difference corrected) and not in the other groups (ps > .051, least significant difference corrected). No difference between the 2 OCD groups was found, and control analyses indicated that our findings were not influenced by medication status (ps > .051). Thus, compared with the HC group, both OCD groups had deficits in dACC and AI in the contrast of high- versus low-uncertainty levels.

Figure 3.

Brain activity of high vs. low uncertainty. Top row shows the brain activity of high vs. low uncertainty in the healthy control (HC) group. Numbers above brain maps represent the x-axis value in Montreal Neurological Institute space. Bottom row is the region-of-interest analysis results of all 3 groups. AI, anterior insula; dACC, dorsal anterior cingulate cortex; HC, healthy control; n.s. nonsignificant; OCD cap, obsessive-compulsive disorder capsulotomy; OCD contr, OCD control.

On an exploratory level, we then calculated the difference in activity (β value) of high versus low uncertainty in dACC and AI and correlated these values with disease-related scores including Beck Depression Inventory, Y-BOCS, and Obsessive Compulsive Inventory–Revised and duration and onset for all OCD participants. Results showed that brain activity in bilateral AI significantly positively correlated with onset age of the disease (left: r = 0.379, p = .042; right: r = 0.381, p = .042, false discovery rate corrected), meaning that the smaller the AI difference between high and low uncertainty, the younger the participants were at diagnosis of OCD (Figure 4). There were no significant correlations in the other measures.

Figure 4.

Correlation of brain activity and onset age. Results show that the difference in brain activity (β value) calculated from the difference between high and low uncertainty in bilateral anterior insula was correlated with the onset age of obsessive-compulsive disorder (OCD).

Discussion

In the current study, we assessed risk taking under uncertainty defined by outcome variance. We had hypothesized that uncertainty would be associated with higher thresholds and slower drift rates. However, using HDDM, we showed that participants generally make risky decisions under high uncertainty characterized by high outcome variance by less (lower threshold) and slower (slower drift rate) evidence accumulation. Based on HDDM modeling results, participants with OCD exhibited greater caution with greater (higher thresholds) but slower (slower drift rates) evidence accumulation, particularly under low uncertainty. Participants with OCD did not show any differences in risk-taking choices, but their performance was suboptimal as indicated by lower winning rates.

fMRI analyses revealed that in the HC group, dACC and bilateral AI showed stronger activity during the decision phase contrasting high versus low uncertainty. These differences were not found in the OCD groups. A smaller high- versus low-uncertainty difference in AI activity was correlated with earlier onset of OCD.

Risk Taking Under Uncertainty Dissociates From Conflict on Computational Measures

We hypothesized that high uncertainty would be associated with higher thresholds and slower drift rates consistent with greater evidence accumulation (2,24,25). However, we found that high uncertainty was instead associated with lower thresholds, which contrast with measures of conflict processing, which also similarly implicate the dACC.

Conflict is invoked between two incompatible responses when a task activates both options, but only one is correct. Conflict more broadly may also occur if the current features of the decision are more similar but might also involve greater similarity in the associated outcomes. Studies of conflict using conventional cognitive tasks such as the Flanker task show high conflict associated with higher thresholds and slower drift rates (12,26). We have previously used the RDM task with differing levels of coherence presuming it is a measure of perceptual uncertainty, but we believe it might also potentially be confounded by conflict (5). The degree of coherence of the RDM task can be modified systematically so that it appears to be moving more right or more left with differing degrees of perceptual randomness and uncertainty. This can elicit a response competition with one response being correct. In the RDM task, high perceptual uncertainty is associated with greater thresholds consistent with increased evidence accumulation similar to observations in high conflict. Thus, we were uncertain if our measure of perceptual uncertainty represented a pure measure of uncertainty without the confounder of conflict.

In our current task, the choice to bet or not bet is subjective and hence is neither right nor wrong. There is a more optimal and rational choice associated with higher long-term winnings. However, in our analyses, we have focused on subjective risk choice. Our HDDM is modeled using bet/no bet rather than optimal/suboptimal choices. In principle, an element of conflict could be elicited because the outcome variance increases, as response competition between “bet” and “no bet” can be elicited when the subjective risk preference becomes more ambiguous. There is, however, no correct or incorrect response, and thus this is inconsistent with the conventional definition of conflict. It also might suggest that given that the playing cards are associated with strong priors of win or lose outcomes, subjective risk decisions and the evaluation of cost-benefit ratio in the context of uncertain outcome might be associated with more rapid decision making. Our task highlights a cognitive process that is not confounded by conflict and more likely to be a pure value-based uncertainty process, which, to our knowledge, is the first time this has been tested in OCD.

Computational and Behavioral Differences in OCD

Our main findings relating to OCD showed greater caution with slower drift rate in participants with OCD only when the uncertainty level was low and greater threshold irrespective of uncertainty level. Our results indicate that individuals with OCD need to gather more evidence before a decision is made. When a decision is highly uncertain, OCD participants accumulate evidence at the same speed as HC participants, but with greater certainty, their ability or efficiency in reaching a decision is impaired. Low uncertainty may be particularly ecologically valid in OCD. For example, exposure to a dirty environment is associated with uncertainty of cleanliness, but repeated hand washing in OCD participants should conventionally be associated with lower outcome variance with greater certainty of cleanliness. Yet despite greater caution, OCD participants still showed impaired winning rates perhaps related to the neural impairment in the representation of uncertainty. These results echo a study in which OCD participants differed from HC participants only when the decision contained low objective uncertainty such that patients rated themselves as less subjectively certain and emphasized that this greater self-doubt might characterize the disorder (4).

Intriguingly, although participants with OCD needed more time and evidence before committing a final decision, they still performed suboptimally with lower winning rates than HC participants. Our task is a simple card gambling task with known probabilities and presumably very limited learning. OCD participants showed poor performance despite greater evidence accumulation, which suggests impaired performance in risk-taking behavior, and further highlights the dissociation of learning and performance. Our findings converge with our previous findings of impaired performance in participants with OCD despite intact learning in a sequential learning task (2).

Imaging Results

Our fMRI results highlight the role of the dACC and AI in the processing of uncertainty in HC participants, which has been previously implicated (17,27), and impairments in the processing of high versus low uncertainty in these regions in participants with OCD. Uncertainty is related to elements of a range of cognitive processes such as anticipation and attention (17) and is intrinsically associated with error likelihood with enhanced error monitoring demonstrated as an endophenotype in OCD (28). Although the dACC has been suggested to respond to conflict according to the conflict monitoring hypothesis (29), dACC activity is associated with error likelihood even when there is no error or conflict (30), which was more likely to be the case in our study. The AI has been shown to represent the probability of risk and is implicated in the anticipation of risk before the risk outcome (31). The absence of dACC and AI activity for high versus low uncertainty in participants with OCD may reflect impairments in anticipation of differences in outcome variance associated with error likelihood (30).

Notably, the dACC and AI are central hubs of the salience network (32). Altered salience network functioning can lead to aberrant assignment of salience to innocuous external stimuli, which results in reality distortion (33), leading potentially to illogical thinking and pathological doubt (34). Uncertainty is related to salience (35). It is possible that our findings are related to the lack of salience sensitivity, which may also underlie OCD. This converges with a meta-analysis showing hypoactivity of salience network in inhibitory control in OCD (36), which is also a characteristic impairment of this disease (37).

Moreover, the dACC and insula regions have been highlighted in OCD treatment itself. A recent study including 4 cohorts of patients with OCD with different deep brain stimulation targets has shown that functional connectivity to the ACC and insula is clinically predictive regardless of stimulation target (38). Further, the dACC is also implicated as a prominent transcranial magnetic stimulation target for OCD treatment highlighted by linking a meta-analytic functional brain network of OCD and the deep brain stimulation connectivity maps (38).

Capsulotomy Effects

We did not show significant differences between participants in the OCD contr and OCD cap groups, indicating that the ablation of the anterior limb of the internal capsule does not affect decision making involving value-based uncertainty in OCD despite improvements in symptoms. Neuropsychological studies of OCD capsulotomy have shown mixed results. Some early studies have shown impaired cognitive functions that tend to improve in the long term (39,40). Other studies suggest no evidence of impairment (41,42) or even improved executive functions, which are related mostly to dorsolateral prefrontal and orbitofrontal cortices (43, 44, 45, 46). Two studies have assessed decision making in OCD with the Iowa Gambling Task implicating ventromedial and orbitofrontal cortices, and both have shown improved risk-taking strategy in the long term after capsulotomy (43,46). Our patient population fits into the long-term time window in the previous studies, but our task involves different brain regions. It is conceivable that capsulotomy may allow neural reprogramming of the lateral and ventromedial pathways but perhaps less so the dorsomedial prefrontal pathways implicated in the current study. This echoes a recent study from our group using the same population showing that capsulotomy-induced clinical improvement may be related to more rostral anterior cingulate activity in aversive processing (47). In addition, the lack of an effect of capsulotomy despite improvements in symptoms suggests that our findings may link to trait-related effects or the risk for the development of OCD rather than state-related symptom improvement, which is consistent with a previous study indicating decision making under ambiguity as a neurocognitive endophenotype for OCD (48). Impaired decision making has also been found in individuals with early-onset OCD and their first-degree siblings (49). Altered insula gyrification has been shown to be related to age of onset of OCD. These converge with our observation that AI difference between high versus low uncertainty correlates with age of onset rather than duration or current OCD severity.

Limitations

The current study is not without limitations. First, this is a cross-sectional study, not a longitudinal study, and thus individual effects of capsulotomy might be more apparent with longitudinal follow-up. Second, one of our inclusion criteria of the OCD cap group is a 6-month time window after surgery, while studies have suggested that the cognitive function changes might occur at least 1 year or even longer after capsulotomy (43,44). Indeed, in our study, only 2 of 24 patients were tested within the 6-month to 1-year time window, and 8 were tested between 1 and 2 years after capsulotomy. Thus, the negative findings between the two OCD groups are unlikely to have been driven by a short-term effect after surgery. Finally, our gambling task has only optimal/suboptimal rather than correct/incorrect answers, which may contribute to the failure of eliciting conflict in our study. Future studies will require direct design of tasks to disentangle the process of uncertainty and conflict.