Underpredicting Pain: An Experimental Investigation Into the Benefits and Risks

Kaya J. Peerdeman; Andrew L. Geers; Delia Della Porta; Dieuwke S. Veldhuijzen; Irving Kirsch


Pain. 2021;162(7):2024-2035. 

In This Article


We investigated the effects of strong underpredictions of pain. Contrary to our hypothesis, both studies showed that strong underpredictions of pain reduced experienced pain intensity. In other words, despite large expectation–experience discrepancies, assimilation of pain experiences to expectations occurred instead of contrast effects. Notably, study 2 indicated that pain intensity reductions upon strong underpredictions were not considerably larger than upon medium underpredictions. Also, expected pain, fear of pain, and pain unpleasantness were generally found to assimilate to the strong underpredictions of pain. Study 2, using a slightly larger and more noticeable expectation–experience discrepancy than study 1, further provided some indications, although inconsistent, for reduced heart rate and skin conductance levels (ie, assimilation), but increased skin conductance responses (ie, contrast effects). Finally, aside from these generally positive findings, we observed important counteractive effects of strong underpredictions of pain, namely, increased uncertainty and lowered trust in the person providing the instructions, although again not considerably more so than medium underpredictions.

Participants of both studies who were given a verbal suggestion that subsequent heat stimuli would be nonpainful, which in fact were moderately (study 1) or highly (study 2) painful, reported substantially less intense pain than participants who were correctly informed about the stimulus intensity. This assimilation of pain experiences to the suggestion was observed for a first heat pain trial directly following the underprediction and was maintained for the 7 subsequent trials, also upon repetition of the suggestion. These findings are in line with the majority of the literature on expectancy and placebo effects demonstrating expectations to generally be self-confirming.[2,3,8,10,34,35,44,48–50,59,62] However, they contradict our hypothesis of contrast effects (ie, increased pain), which we formulated based on the literature on contrast effects for diverse outcomes in cases of large expectation–experience discrepancies.[1,7,11,18,19,36,44,45,62] A lack of awareness of the discrepancy is unlikely to explain our current findings, as we intentionally drew participants' attention to the expectation–experience (mis)match. Participants' resultant awareness was indicated by how they compared their experiences with what they expected and were instructed. Nonetheless, it might be that their focus on their experiences made participants aware of the ambiguous nature of pain and thereby actually facilitated assimilation.[10,17] As such, it appears that assimilation of pain experiences to expectations is the rule.

Notably, study 2 showed no reliably significant differences in pain intensity upon a strong vs medium underprediction of pain intensity. This finding complements a study demonstrating smaller placebo effects of decaffeinated coffee on alertness upon suggestions of high vs moderate caffeine doses[36] and recent experiments demonstrating a tipping point at which large expectation–experience discrepancies led to less assimilation of pain experiences than moderate discrepancies.[25] Thus, there might be a limit to the effects of underpredicting pain intensity in terms of the benefits for pain relief. Possibly the relatively smaller assimilation upon experiencing a larger expectation–experience mismatch is due to increased expectation uncertainty because of the larger prediction error,[10,27] although certainty differed significantly only upon repetition of the suggestion in our study.

We conducted secondary and exploratory analyses to gain a broader understanding of the effects of strongly underpredicting pain. Both studies demonstrated that expected pain intensity, fear of pain, and pain unpleasantness also generally reduced, that is, assimilated, upon a strong underprediction as compared to a correct prediction. Again, study 2 demonstrated that the strong underprediction generally exerted no larger effects than the medium underprediction. Assimilation of these pain responses is in line with the larger placebo literature[48] but contradicts a study in which underprediction of pain led to increased expectations and fear of pain.[3] Aside from these arguably positive effects of underpredicting pain, it should be noted that participants were also mostly less certain of their expectations upon strong underprediction, particularly in study 2, but not more so than upon a medium underprediction. Predictive coding models posit that next to the magnitude, the certainty of expectations is core in determining to which degree expectations influence pain experiences; lower certainty reduces their influence.[10,27,33] Note, however, that participants in our studies continued to assimilate their experienced pain intensity when the underprediction was repeated for a second series of trials, despite reduced certainty and apparent reduced effects of the suggestion on expectations.

Examinations of autonomic responses to the heat stimuli indicated no significant effects of the suggestions in study 1, but study 2 provided some indications for lower heart rate and skin conductance level during few early trials upon the strong underprediction as compared to the correct prediction (ie, assimilation effects). By contrast, there were some indications, although inconsistent, for increased skin conductance responses upon the strong underprediction as compared to the correct prediction and medium underprediction (ie, contrast effects). Interpretation of these findings is hampered by their inconsistency, high variability, and relatively weak responses to the heat stimuli as compared to rest (see Tables Supplemental digital contents 1 and 2, available at http://links.lww.com/PAIN/B269), but they suggest that effects of the verbal suggestions may go beyond self-report effects. Confidence in the latter interpretation is strengthened by the general absence of evidence for between-group differences in response bias measures assessed in the exit questionnaire and the lack of moderation of the effects of pain underpredictions on pain intensity by participants' ability to modify self-presentation and sensitivity to expressive behaviors of others as assessed with the Self-Monitoring Scale. Moreover, the observed psychophysiological responses might indicate that different measures of autonomic responses can highlight different underlying processes. One might, for example, speculate that upon an initial startled response when pain is higher than expected, responses are regulated so that on average, even over the course of seconds, autonomic responses align with self-reported pain reductions, if any response is visible at all. Findings of previous research are also largely equivocal, with some providing evidence for placebo responses on heart rate[6,51] and skin conductance responses,[15,21,43] whereas other studies showed no significant effects (heart rate: Refs. 4,20,41 and skin conductance response: Ref. 31).

Further explorations of participants' responses to the underprediction of pain led to the novel and potentially highly clinically relevant finding that despite assimilation of pain, even upon repetition of the suggestion, participants reported substantially lower trustworthiness of the experimenter who provided the suggestion upon strong underprediction as compared to correct prediction of pain and not considerably more than upon medium underprediction. Thus, there appears to be a trade-off between reducing pain and fostering trust. Trust is critical in good clinical communication and is also thought to be key in placebo effects,[22] as such trust is an important determinant of treatment outcomes. Hence, one might advocate for providing honest, correct information about upcoming pain (eg, during a medical procedure), even when short-term pain relief opportunities might then be missed. However, replications and extensions of these findings, also in clinical contexts, are required before conclusions can be drawn and clear-cut advice for clinical implementation can be formulated.

Finally, we examined whether individual differences in initial responses to strong underpredictions of pain could be explained by several personal characteristics. We found some indications for moderation by state anxiety, optimism, and pain catastrophizing of the effects of the strong underprediction vs correct prediction, but the effects were small, inconsistent, and—in the case of state anxiety—contradictory within our studies. Also, the effects generally do not align with previous studies in which, for example, higher optimism predicted larger placebo effects on pain.[20,42] Hence, adding to the previous literature,[26,49,60] it might be inferred that no consistent, readily identifiable profile of responders to suggestions exists.

When considering the implications of the current findings, it should be kept in mind that the current studies were conducted with verbal suggestions about few brief heat stimuli in healthy samples. This allows for stringent experimental control of stimulus intensity and participants' expectations about the stimuli, for manipulations that might maximize the chance of contrast effects, and for assessing many relevant variables, but is also associated with limitations. For example, rating expectations and instructing participants to attend to and compare their sensations could bias participants' pain ratings. As such, the current design does not directly translate to clinical practice. Previous findings do suggest that experimental placebo studies with acute experimental pain translate well to acute clinical pain, but this appears less so for chronic pain outcomes,[48] and this might not apply for large expectation–experience mismatches. Moreover, trust in a healthcare provider might less readily be affected by a simple suggestion than trust in an experimenter, especially when vulnerability and, hence, the need for trust is high and/or when a long-established trusting relationship is established.[22] On the other hand, when personal relevance is high, violations of trust might lead to particularly strong responses, such as a sense of betrayal. Future research should examine the studied effects in more clinically relevant contexts, including long-term outcomes. Of further interest for future investigations is that the effects of expectation–experience mismatches might also depend on (1) the certainty or precision of both expectations and sensations,[7,10,33] (2) the direction of the mismatch (underpredicting pain may have a larger impact than overprediction[3]), (3) attention and affect,[19,28] (4) the number and nature of prior experiences,[18] (5) whether expectations pertain directly to the pain stimulus (ie, stimulus expectations) or to the effects of (placebo) treatments on pain (ie, treatment outcome expectations),[55] (6) how expectations are induced (eg, using verbal suggestions or visual cues), and (7) interpersonal relationship characteristics (eg, communicated warmth and competence) in case expectations are communicated verbally.[24,46] These factors may be manipulated independently in future research.

In conclusion, even strong underpredictions of pain can reduce pain, although not considerably more so than medium underpredictions. However, such overly positive instructions may cause uncertainty and undermine trust. Thus, underpredicting pain has both benefits and risks. Although further research is required, these findings suggest that healthcare providers may wish to be cautious with providing overly positive information about painful medical procedures, despite pain-relieving effects.