Action-Outcome Knowledge Dissociates From Behavior in Obsessive-Compulsive Disorder Following Contingency Degradation

Abstract

Background: In obsessive-compulsive disorder (OCD), actions persist despite being inappropriate to the situation and without relationship to the overall goal. Dysfunctional beliefs have traditionally been postulated to underlie this condition. More recently, OCD has been characterized in terms of an imbalance between the goal-directed and the habit systems. To test these competing hypotheses, we used a novel experimental task designed to test subjective action-outcome knowledge of the effectiveness of actions (i.e., instrumental contingency), together with the balance between goal-directed and habitual responding.

Methods: Twenty-seven patients with OCD and 27 healthy control subjects were tested on a novel task involving the degradation of an action-outcome contingency. Sensitivity to instrumental contingency and the extent to which explicitly reported action-outcome knowledge guided behavior were probed by measuring response rate and subjectively reported judgments.

Results: Patients with OCD responded more than healthy control subjects in situations in which an action was less causally related to obtaining an outcome. However, patients showed intact explicit action-outcome knowledge, as assessed by self-report. In patients, the relationship between causality judgment and responding was altered; therefore, their actions were dissociated from explicit action-outcome knowledge.

Conclusions: These findings indicate reduced sensitivity to instrumental contingency in OCD, reinforcing the notion of a deficient goal-directed system in this disorder. By showing a dissociation between subjectively reported action-outcome knowledge and behavior, the data provide experimental evidence for the ego-dystonic nature of OCD.

Keywords: Action-outcome; Frontostriatal; Goal-directed; Habit; Learning; Obsessive-compulsive disorder.

Figure 1

Contingency manipulation. To degrade the contingency, once agents have learned to perform an action to receive a reward with a certain probability, a schedule of noncontingent outcome delivery is superimposed. By increasing the frequency of noncontingent outcomes, the overall contingency (i.e., the causal association between an action and its consequences) is degraded or becomes negative. If guided by the goal-directed system, an agent should stop responding in the face of contingency degradation. (A) Diagram illustrating a schedule with a positive contingency, in which outcome is delivered on performance of an action with a given probability [P(O|A)]. (B) Contingency is degraded by also delivering outcomes in the absence of an action with a given probability [P(O|∼A)]. If the contingency is degraded to the extent that the two probabilities are equal, the causal status of the action is nil, and the probability of the reinforcer is the same regardless of any response. (C) When P(O|∼A) is higher than P(O|A), the contingency becomes negative, and the action reduces the probability of reinforcer delivery. P(O|A), probability of outcome given an action; P(O|∼A) probability of outcome given the absence of an action; violet filled circles, contingent outcomes; green empty circles, noncontingent outcomes.

Figure 2

Experimental paradigm. (A) Subjects had to complete an experimental session of 12 blocks of 2 minutes each. At the end of each block, subjects had to judge to what extent pressing the space bar caused the occurrence of the reward, on a scale from −100 (pressing the space bar always prevented reward) to 100 (pressing the space bar always caused reward). During the experimental session, the participant was presented with a white triangle and could decide whether to press the space bar or not. Rewards were delivered contingently on pressing of the space bar or noncontingently in the absence of a response. In addition, a running total of the amount of money earned within a block was continuously displayed in the upper corner of the screen (not shown in figure). In cases where the participant was not pressing the space bar for multiple (hidden) 1-second bins in a row, the white triangle was continuously displayed on the screen, unless a nonresponse contingent reward occurred. In those cases, a reward was displayed on the screen noncontingently. (B) Each block was divided into 120 unsignaled time periods (bins) of 1 second. When a response occurred within each bin, the triangle turned yellow until the bin ended. If a response was recorded during the bin, a contingent reward was delivered at the end of that bin according to the applicable probability of outcome delivery given a response [P(O|A)]. If no response occurred during the bin, a noncontingent reward was delivered according to the applicable probability of outcome delivery given the absence of a response [P(O|∼A)]. (C) By varying P(O|A) and P(O|∼A), different levels of contingencies were achieved so that each experimental session included positive, degraded, and negative contingency blocks. O, outcome; P(O|A), probability of outcome given an action; P(O|∼A), probability of outcome given the absence of an action.

Figure 3

Mean response rate and causality judgments for control (CTL) group and obsessive-compulsive disorder (OCD) group. (A) Mean response rate by contingency (ΔP). Both groups responded more for higher contingencies. However, patients with OCD showed reduced sensitivity to instrumental contingency. (B) Subjective judgments of causality increased as a direct function of response-outcome contingency in both groups. Data are presented in ascending order of programmed contingency, but contingencies were experienced by each subject in a semirandomized order. Error bar indicates Fisher’s least significant difference to facilitate post hoc comparisons (error bars are ± 0.5 least significant difference). However, in the context of mixed designs, as in this case, this error bar can be used only for within-subject comparisons. The difference between OCD and CTL groups in mean causality judgments at ΔP = −0.6 was not significant. However, CTL subjects, but not patients with OCD, subjectively detected a difference between neighboring levels of negative programmed contingency between ΔP = −0.3 and ΔP = −0.6). (C) Response rate as a function of causality judgment by group. The two populations differentially employed action-outcome knowledge to guide their behavior. Points and error bars (SEMs) show values clustered by programmed contingency. As described in the main text, data were collapsed across blocks having equal contingencies (ΔP = −0.6, block 6; ΔP = −0.3, block 5, block 9; ΔP = 0.0, block 2, block 3, block 4, block 8, block 12; ΔP = 0.3, block 7, block 11; ΔP = 0.6, block 1, block 10. See Table 2 for naming of the blocks). Programmed contingency refers to the a priori experimentally programmed contingency resulting from the a priori programmed conditional probabilities. *p < .05, within-group comparison; ^##p < .01, interaction; ^#p < .05, group × quadratic causality judgment interaction. n.s., not significant.

Figure 4

Habit and goal-directed ratio scores for contingent and corresponding degraded-contingency conditions. The ratio score was computed by dividing the number of responses in the nondegraded block by the sum of the responses for the degraded and nondegraded sessions [i.e., (contingent/(contingent+degraded)]. The ratio score represents the proportion of responses in the contingent condition relative to the nondegraded condition, with a value ≤0.5 indicating a greater or equal number of responses in the degraded contingency (ΔP) condition and thus habitual behavior. (A) Ratio score for pairs of blocks for which the action-outcome relationship was contingent (ΔP = 0.6 [P(O|A) = 0.6, P(O|∼A) = 0.0], block 10) and then degraded to ΔP = 0.3 by superimposing a noncontingent schedule (ΔP = 0.3 [P(O|A) = 0.6, P(O|∼A) = 0.3], block 11). Patients with obsessive-compulsive disorder (OCD) displayed increased habitual behavior (t₅₂ = 3.350, p = .002). (B) Ratio score for pairs of blocks for which the action-outcome relationship was contingent (ΔP = 0.6 [P(O|A) = 0.6, P(O|∼A) = 0.0], block 10) and then completely degraded to ΔP = 0.0 by superimposing a noncontingent schedule (ΔP = 0.0 [P(O|A) = 0.6, P(O|∼A) = 0.6], block 12). Patients with OCD showed increased habitual behavior compared with control (CTL) subjects (t₅₂ = 2.23, p = .03). (C) Ratio score for pairs of blocks for which the action-outcome relationship was contingent (ΔP = 0.3 [P(O|A) = 0.3, P(O|∼A) = 0.0], block 7) and then completely degraded to ΔP = 0.0 by superimposing a noncontingent schedule (ΔP = 0.0 [P(O|A) = 0.3, P(O|∼A) = 0.3], block 8). Error bars: SEM. *p < .05; **p < .01. P(O|A), probability of outcome given an action; P(O|∼A), probability of outcome given the absence of an action.

Figure 5

Causality judgments when the contingency (ΔP) was zero. There were no group differences. For both the control (CTL) group and obsessive-compulsive disorder (OCD) group, causality judgments increased as a function of higher density of reinforcement even though there was no causal association between the action and the outcome (contingency ΔP = 0.0) in all three situations. Error bar indicates Fisher’s least significant difference to facilitate post hoc comparisons (error bars are ± 0.5 least significant difference). However, in the context of mixed designs, as in this case, this error bar can be used only for within-subject comparisons. ***p ≤ .001, main effect of density of outcome. P(O|A), probability of outcome given an action; P(O|∼A), probability of outcome given the absence of an action.