Learning to predict pain: differences in people with persistent neck pain and pain-free controls

Abstract

Background: Learning to predict threatening events enables an organism to engage in protective behavior and prevent harm. Failure to differentiate between cues that truly predict danger and those that do not, however, may lead to indiscriminate fear and avoidance behaviors, which in turn may contribute to disability in people with persistent pain. We aimed to test whether people with persistent neck pain exhibit contingency learning deficits in predicting pain relative to pain-free, gender-and age-matched controls.

Method: We developed a differential predictive learning task with a neck pain-relevant scenario. During the acquisition phase, images displaying two distinct neck positions were presented and participants were asked to predict whether these neck positions would lead to pain in a fictive patient with persistent neck pain (see fictive patient scenario details in Appendix A). After participants gave their pain-expectancy judgment in the hypothetical scenario, the verbal outcome (PAIN or NO PAIN) was shown on the screen. One image (CS+) was followed by the outcome "PAIN", while another image (CS-) was followed by the outcome "NO PAIN". During the generalization phase, novel but related images depicting neck positions along a continuum between the CS+ and CS- images (generalization stimuli; GSs) were introduced to assess the generalization of acquired predictive learning to the novel images; the GSs were always followed by the verbal outcome "NOTES UNREADABLE" to prevent extinction learning. Finally, an extinction phase was included in which all images were followed by "NO PAIN" assessing the persistence of pain-expectancy judgments following disconfirming information.

Results: Differential pain-expectancy learning was reduced in people with neck pain relative to controls, resulting from patients giving significantly lower pain-expectancy judgments for the CS+, and significantly higher pain-expectancy judgments for the CS-. People with neck pain also demonstrated flatter generalization gradients relative to controls. No differences in extinction were noted.

Discussion: The results support the hypothesis that people with persistent neck pain exhibit reduced differential pain-expectancy learning and flatter generalization gradients, reflecting deficits in predictive learning. Contrary to our hypothesis, no differences in extinction were found. These findings may be relevant to understanding behavioral aspects of chronic pain.

Keywords: Associative learning; Chronic pain; Contingency learning; Extinction; Fear avoidance; Fear avoidance model; Fear learning; Generalization; Neck pain; Predictive learning.

Figure 1. Conditioned stimuli (CS+ and CS−) and generalization stimuli (GS1-5) presented with CSs on the extremes and the intermediate GSs in between representing the generalization gradient.

Which stimulus served as the CS+ and the CS− (i.e., white vs. blue labels) was counterbalanced across participants.

Figure 2. Flow chart of the predictive learning task, with a CS−, CS+ and GS trial as examples.

Middle row displays the question: To what extent do you expect Alex to experience pain in this position? CS+, conditioned stimulus paired with “PAIN”; CS−, conditioned stimulus paired with “NO PAIN”; GS, generalization stimulus paired with “NOTES UNREADABLE”.

Figure 3. Mean pain-expectancy judgments for the CS+ and CS− both for the pain-free control group (n = 30) and persistent neck pain group (n = 30) during the familiarization phase (FAM), the four blocks of acquisition (ACQ1-4), and the five blocks of extinction (EXT1-5), separately per block.

Vertical bars represent 95% confidence intervals. CS+, conditioned stimulus paired with “PAIN” in 75% of the trials; CS−, conditioned stimulus always paired with “NO PAIN”.

Figure 4. Mean pain-expectancy judgments for CS+, GS1-5, CS− for the patient group (n = 30) as well as the pain-free controls (n = 30) during the generalization phase, averaged over blocks.

Vertical bars represent 95% confidence intervals. CS+, conditioned stimulus paired with “PAIN” in 75% of the trials; CS−, conditioned stimulus always paired with “NO PAIN”; GS1, Generalization Stimulus most closely resembling the CS+.