Background & Aims

Pain expectations are formed by learning (from past experience and environmental cues) the association between specific circumstances and the occurrence of pain.[1] These expectations are known to influence pain perception.[2]
Whether this learning is statistically optimal remains debated.[3] If so, the learning process of the participant should reflect their perception of the precision of each observation (stochasticity) and of the instability of the environment (volatility).
Further, it is unclear whether interindividual differences observed in pain learning experiments are due to temporary states or intrinsic traits. If they reflect traits, they could account for the wide variability in pain experiences across individuals and might be a feature of chronic pain conditions.
Using a novel pain learning task (with periods of high and low volatility) and computational modeling, we aimed to probe the learning strategy and temporal stability of the learning parameters of participants.

Methods

Fifty healthy volunteers participated in 2 sessions, 1 week apart. At each session, participants completed a reversal learning task in which they learnt probabilistic associations between 2 arbitrary visual cues and 2 stimulus intensities (selected to be reliably perceived as either non-burning or burning). The task included 160 trials and 10 reversals.
We constructed several alternative computational models, corresponding to specific hypotheses on how expectations are translated into predictions, the effect of expectations on stimulus recognition, and the algorithm used by humans to update their expectations. These models were fitted on the cue-stimulus contingencies, participant responses to the prediction, recognition, and rating questions, as well as reaction times. Once model fitting is complete, the best model will be selected through model comparison (loo).
We will use ICC to assess the test-retest reliability of individual learning parameters derived from the winning model.

Results

Data collection is complete but modeling is still ongoing.
Manipulation checks showed the following expected patterns (all p<10?): participants were able to predict the next stimulus above chance level based on the visual cue; they accurately identified the stimulus types above chance level; non-painful stimuli were consistently rated less intense than the painful ones; as per predictive coding, recognition accuracy improved when the stimulus matched the participant's prediction and intensity ratings were biased towards these predictions; and, finally, incorrect predictions or recognitions were associated with longer reaction times to the corresponding questions, indexing harder decision making. Preliminary results, derived from a reduced set of models, suggest that humans adapt their pain learning rate based on their perception of the sequence volatility (p=0.0005) and indicate poor-to-excellent test-retest reliability of pain learning parameters (ICC CIs ranging from <0.4 to 1).

Conclusions

Manipulation checks and preliminary results appear to suggest that human agents process expectations in a statistically optimal manner. This includes both the updating of expectations, which appeared to follow a Bayesian rule accounting for changes in the sequence volatility, and their integration with sensory evidence during stimulus recognition, which seemed consistent with predictive coding principles.
Another important finding is that only some of the learning parameters may be reliable over time. If confirmed, this would indicate that these parameters may reflect transient states rather than intrinsic traits.

References

1.Buchel, C., Geuter, S., Sprenger, C. & Eippert, F. Placebo analgesia: a predictive coding perspective. Neuron 81, 1223–1239 (2014).
2.Atlas, L. Y. How Instructions, Learning, and Expectations Shape Pain and Neurobiological Responses. Annual Review of Neuroscience 46, 167–189 (2023).
3.Mancini, F., Zhang, S. & Seymour, B. Computational and neural mechanisms of statistical pain learning. Nat Commun 13, 6613 (2022).

Presenting Author

Arthur S. Courtin

Poster Authors

Arthur Courtin

PhD

Aarhus University

Lead Author

E-mail

Melina Vejlø

Center of Functionally Integrative Neuroscience, Department of Clinical Medicine, Aarhus University

Lead Author

E-mail

Jesper Ehmsen

center of functionally integrative neuroscience

Lead Author

E-mail

Francesca Fardo

Aarhus University

Lead Author

E-mail

Micah G. Allen

Ph.d.

Center of Functionally Integrative Neuroscience, Department of Clinical Medicine, Aarhus University

Lead Author

E-mail

Topics

  • Placebo