A social affective neuroscience lens on placebo analgesia

Abstract

Pain is a fundamental experience that promotes survival. In humans, pain stands at the intersection of multiple health crises: chronic pain, the opioid epidemic, and health disparities. The study of placebo analgesia highlights how social, cognitive, and affective processes can directly shape pain and identifies potential paths for mitigating these crises. In this review, I examine recent progress in the study of placebo analgesia through affective science. I focus on how placebo effects are shaped by expectations, affect, and the social context surrounding treatment, and discuss neurobiological mechanisms of placebo, highlighting unanswered questions and implications for health. Collaborations between clinicians and social and affective scientists can address outstanding questions and leverage placebo to reduce pain and improve human health.

Pain: A biopsychosocial phenomenon that requires integrative research

Pain lies at the core of multiple health crises. It is central to the opioid epidemic [1,2], there are substantial health disparities in pain [3–5], over 11% of American adults report experiencing daily pain [6], and chronic pain is highly comorbid with other central nervous system conditions, including mental health illnesses [7] and substance use disorders [8]. Health care practitioners have reduced opioid prescriptions in response to the opioid epidemic [9], yet we lack effective alternatives, including non-addictive and nonpharmacological therapies.

Hope can be found in the progress the field has made in isolating mechanisms that contribute to pain through animal models, as the systems that process pain ([10]; see Glossary) and nociception are highly conserved across species. Yet in humans, pain is ultimately a subjective, conscious experience that involves the integration of sensory processing, emotion, context, and decision making. Understanding human pain requires an integrative approach that draws on psychology, neuroscience, and medicine. In this review, I focus on placebo analgesia, a phenomenon that lies at the intersection of these diverse fields. In the following sections, I review the critical influences of expectations, affect, and the social context surrounding treatment (see Figure 1, Key Figure). I then discuss the brain mechanisms underlying placebo effects and the extent to which placebo analgesia provides insights on domains other than pain. Throughout, I emphasize that placebo research has important implications for the urgent crises surrounding opioid use disorder and health disparities in pain.

Figure 1, Key Figure: Expectations, affect, and social context shape pain and lead to placebo analgesia or nocebo hyperalgesia.

Laboratory models of placebo analgesia combine acute noxious stimuli (e.g. heat administered through a thermode) with experimental manipulations to measure how expectations, affect, and social factors influence perceived pain and pain-related neurobiological responses. Expectations may be driven by associative learning from cues associated with treatment or from verbal instructions about the anticipated effects of treatment. Placebos may influence affect by reducing anxiety or increasing positive emotion, whereas nocebo effects are associated with increases in anxiety. Social factors also shape placebo effects through the patient-provider interaction. Placebo analgesia refers to reduced pain as a result of expectations, affect, and the social context surrounding treatment, whereas nocebo hyperalgesia refers to the exacerbation of pain due to these psychosocial processes.

Placebo analgesia: A window on social, cognitive, and affective influences on pain

The placebo effect refers to a positive treatment outcome that is attributable to the psychosocial context surrounding treatment, rather than pharmacological or biological properties of the treatment itself. In clinical trials, placebo effects can be differentiated from placebo responses by comparing placebo administration to a natural history or waitlist control, which allows researchers to differentiate between true placebo effects and other non-specific factors. Meta-analyses of clinical trials including natural history control groups indicate that placebo effects are strongest in pain and subjective outcomes [11,12]. This finding has motivated researchers across diverse fields to move beyond self-report and study the mechanisms of placebo. The placebo effect’s negative counterpart, the nocebo effect, can also be measured by studying increases in pain or side effects or pain in response to inert treatments.

Laboratory experiments can isolate mechanisms of placebo analgesia and nocebo hyperalgesia by pairing acute pain stimulation (e.g. heat administered through a device called a thermode; see Figure 1) with an inert substance (e.g. skin cream) that a participant believes is a potent analgesic treatment and measuring associations between pain and neurobiological responses (see Figure 2). Studies implicate the critical role of expectations that are driven by instructed knowledge or learned through experience [13,14]. But expectations about treatment are not cold cognitions; they are laden with meaning and emotional significance, leading to the proposal that the placebo effect be called a “meaning” response [15]. The placebo effect also depends on the unique psychosocial context surrounding treatment, including patient-provider interactions. In the following sections, I review recent progress on affective, social, and cognitive influences on placebo analgesia.

Figure 2. Brain mechanisms of placebo analgesia and relationship with salience, interoception, and nociceptive processing.

Placebo analgesia is associated with variations in brain responses to noxious stimuli that can be visualized in humans using FMRI or PET imaging. Left: The regions that are most commonly activated by painful stimuli include the anterior cingulate cortex (ACC), insula, primary and secondary somatosensory cortex (S1 and S2), and thalamus. Each of these are targeted by afferent nociceptive pathways and contain nociceptive neurons. FMRI studies that compare responses to placebo administration with a control treatment (i.e. no expected pain relief) indicate that pain-related responses in the ACC, thalamus, and anterior insula are reduced with placebo (blue), whereas placebos elicit increases in activation in modulatory regions (gold) including the dorsolateral prefrontal cortex, rostral ACC, and the opioid-rich periaqueductal gray. Upper middle: Meta-analysis of studies of placebo analgesia indicate reliable reductions in the dorsal ACC, anterior insula, and thalamus [75]. Lower middle: Despite placebo-induced reductions in pain-evoked responses in a subset of pain-related regions, placebos do not elicit reliable modulation of the Neurologic Pain Signature [30,92], which is a brain-based pattern that can reliably distinguish responses to painful and non-painful stimuli and is sensitive and specific to pain. This suggests that placebos might modulate non-specific affective and cognitive processes rather than affecting nociception. Right: Studies of placebo analgesia and other forms of pain modulation must carefully distinguish pain from salience processing (upper right) and interoception (lower right) since pain is highly salient and requires interoception, and the brain networks that process pain overlap substantially with the salience network and interoceptive processing. Maps depict term-based meta-analyses from Neurosynth [98] (red and yellow = association tests, green and purple = uniformity tests). Images of placebo-induced reductions (upper middle) were adapted with permissions from [75] and images of the Neurologic Pain Signature (lower middle) were adapted with permissions from [30].

Expectations and the predictive coding model of placebo analgesia

Much work has focused on the role of expectation and prediction in placebo analgesia. Expectancy-based models of placebo assume that individuals develop expectations about clinical outcomes through experience (i.e. conditioning or associative learning) and explicit knowledge (e.g. through verbal suggestion or instruction), and that these expectations lead to placebo effects. Studies suggest that conditioning and verbal suggestion may have differential impacts on different outcomes. For example, one team of researchers treated different groups of participants with active treatments over several days [16]. Participants who underwent a tourniquet pain challenge received the analgesic ketorolac tromethamine, patients with Parkinson’s Disease received subthalamic nucleus stimulation, and healthy volunteers received sumatriptan, which decreases cortisol and increases growth hormone. After this conditioning phase, subsets of participants were instructed that they would receive a new treatment with opposite effects from the conditioned treatment. All participants received placebo, and researchers measured whether placebo effects mimicked instructions (i.e. reversed with the placebo) or conditioning (no reversal). Placebo effects on pain and motor performance reversed with instructions, while growth hormone and cortisol mirrored conditioning. This provides evidence that a) placebo effects depend on both prior experience and explicit beliefs; and b) outcomes may differ in their sensitivity to explicit expectations mediated by verbal instruction. We recently combined this type of instructed reversal with computational models of aversive learning and identified dissociable effects of instructions on brain regions involved in learning, such that the striatum and orbitofrontal cortex updated with instructions, whereas the amygdala learned purely from experience [17,18]. Future work should test whether similar mechanisms underlie dissociable effects of instructions and learning on placebo effects across domains.

Computational frameworks can indeed provide insights on the mechanisms of placebo. Some researchers have argued that placebo effects and pain modulation can be explained through predictive coding [19–23], which draws on Bayesian integration [24]. This framework predicts that the extent to which perception is biased toward prior expectations depends on the expectation’s precision: The more certain the expectation, the stronger the influence on perception. Conversely, if an expectation is vague or variable, perception will be driven by the experience itself (in the case of experimental pain, the noxious stimulus). This framework links placebo analgesia with other areas of perception and new interest in computational psychiatry and provides theoretical models that can be formally tested. How well do these models account for placebo effects in practice?

The predictive coding model has been directly tested in a series of studies by Büchel and colleagues [25–27]. In one task [25], individuals saw cues that predicted different intensities of noxious heat with different probabilities. Pain-predictive cues often serve as a model of placebo analgesia since they easily isolate effects of learning and instructions on pain (see Box 1). The researchers measured whether responses to noxious stimuli tracked a) stimulus intensity alone; b) additive effects of expectation & stimulus intensity; or c) a weighted sum of prediction and prediction errors. Autonomic responses and responses in the anterior insula integrated predictions and prediction errors consistent with predictive coding, while the posterior insula, a region thought to be fundamental for pain perception [28,29], and the neurologic pain signature, a pain-predictive brain classification pattern (NPS; [30]) tracked stimulus intensity. Büchel’s team subsequently built on this approach in a placebo analgesia experiment by manipulating the precision, or variability, of the expectation [26]. During conditioning, healthy volunteers either experienced consistent relief (i.e. temperature reduction) on the placebo-treated skin site, or experienced variable relief, which reduces the precision of the expectation. Placebo effects were larger in the group with high precision, and pain ratings conformed to Bayesian integration. In addition, activation in the opioid-rich periaqueductal gray (PAG) correlated with the attraction weight, a quantity that is sensitive to the precision of both the prior and the likelihood. This builds on work demonstrating that the PAG tracks uncertainty [31] and prediction errors [32] during pain prediction and plays a key role in descending modulation of pain [33,34], as discussed later in this review. Together, this work indicates that instructions and experiential learning combine to shape expectations, which in turn directly impact pain and pain-related processing.

Box 1. Stimulus expectancies vs treatment expectancies.

Both pain-predictive cues and placebo analgesia paradigms can lead to pain modulation through predictive coding. However, it is unknown how well cue-based pain modulation paradigms capture the construct of placebo analgesia, or whether these approaches should be evaluated independently. Pain-predictive cues provide information about the intensity of an upcoming noxious stimulus, inducing stimulus expectancies. Both the cue and the stimulus are usually brief phasic signals. In contrast, a typical analgesic treatment is long lasting, has a variable onset, and makes people less sensitive to the noxious stimulus itself. For example, a patient on morphine cares little about whether they will receive an innocuous finger prick, a shot, or a more invasive procedure. We refer to cue-based expectations as stimulus expectancies and placebo-related expectations as treatment expectancies, similar to Kirsch’s distinction between stimulus and response expectancies [118], although both stimulus and treatment expectancies are likely to induce response expectancies (i.e. expectations for reduced pain).

We have previously shown that cue-based pain modulation is additive with placebo and opioid analgesia [119], supporting the idea that stimulus and treatment expectancies engage independent mechanisms. We have hypothesized that the two are mediated by separable neuromodulatory mechanisms [13], whereby placebo analgesia engages tonic μ-opioid release to engage long-lasting descending inhibition [120], while cue-based pain modulation is likely to involve phasic dopamine prediction errors to signal the expected value of the upcoming stimulus [121]. Consistent with this hypothesis, placebo analgesia is intact during haloperidol dopamine blockade [122], while cue-based pain modulation is not influenced by the μ-opioid receptor antagonist naltrexone [123]. One study that directly compared treatment and stimulus expectancies indicated that the framing did indeed influence both prediction error processing and expectancy-related brain responses, such that prediction errors were only observed in response to stimulus expectancies, while the striatum and prefrontal cortex showed greater coupling in response to treatment expectancies [124]. We also conducted a meta-analysis that indicated preliminary distinctions between the brain mechanisms of stimulus and treatment expectancy [75], such that treatment expectancies were more likely to elicit decreases in the left insula and increases in the ventral striatum, rACC, PAG, and other regions, while stimulus expectancies were more associated with increases in the cerebellum, inferior parietal lobule, and lateral prefrontal cortex. To determine the validity of our paradigms, researchers should continue to assess whether stimulus and treatment expectancies do indeed engage the same mechanisms, or whether they reflect distinct pain-modulatory processes.

The role of affect in placebo analgesia and nocebo hyperalgesia

While most studies of placebo mechanisms have focused on manipulating expectations and learning, placebo effects do not depend strictly on expectations, and expectations are not purely cognitive. Patients seek treatment when they are suffering, and regardless of the mechanisms of treatment or even the specific disease or condition, the treatment ritual might relieve suffering by a) reducing negative affect and emotions such as anxiety, fear, and stress and b) increasing positive affect through feelings of trust, hope, and being supported. Likewise fear or anxiety about side effects or negative outcomes might directly lead to nocebo effects, since negative emotions can exacerbate pain [35,36]. Thus, emotion and affect may be as important as beliefs and expectations, yet relatively few studies have measured the relationships between placebo effects and mood or emotion. Do placebos act by reducing negative affect and increasing positive affect?

Several studies indicate that placebos do reduce negative emotion [37]. For example, placebo analgesia is associated with reduced fear and anxiety [38,39] relative to no-treatment control conditions in healthy volunteers, and placebo analgesia is inversely related to negative affect in patients with neuropathic pain [40]. Brain imaging studies also support links between placebo analgesia and affective processing. Individual differences in the magnitude of placebo analgesia can be predicted based on anticipatory activity in emotion-related brain networks, but not regions involved in pain or cognitive control [41], and placebo effects for pain and anxiety are associated with overlapping activity in the rostral anterior cingulate cortex (rACC), lateral orbitofrontal cortex (OFC), and ventrolateral prefrontal cortex (VLPFC) [42], suggesting common mechanisms of action in value-related circuits.

Studies of nocebo hyperalgesia also implicate anxiety. Nocebo administration enhances anticipatory anxiety [39] and nocebo analgesia can be blocked by the anxiolytic diazepam [43] and the cholecystokinin (CCK) antagonist proglumide [43]. CCK mediates anxiety-induced hyperalgesia and stress-induced hyperalgesia primarily through the PAG, where it also can inhibit opioid-related analgesia [44]. Interestingly, pre-existing anxiety does not seem to be a risk factor for nocebo effects [45–47], suggesting anxiety influences outcomes insofar as it emerges in response to the treatment ritual itself. Recent behavioral work also indicates that positive mood inductions can reduce the nocebo effect for headache [48,49], indicating a direct link between affect and nocebo. Fewer studies have examined the neural and psychological mechanisms of nocebo, relative to the large literature on placebo. We need more research on the specific factors that give rise to nocebo effects so that clinicians can mitigate nocebo effects in clinical practice. For a comprehensive review of the role of affect in placebo and nocebo, see [50].

Social influences on placebo analgesia

The role of affect in placebo and nocebo likely stems not only from internal shifts in mood and expectations, but also through the psychosocial treatment context, including patient-provider interactions. Having a positive therapeutic alliance is associated with beneficial outcomes, including in chronic pain [51,52]. Studies of the placebo effect indicate that the physician’s attitude and behavior directly influence clinical outcomes. Patients with irritable bowel syndrome show improvements with placebo acupuncture when administered by a warm, empathic provider, relative to a neutral doctor or waitlist control [53], and experimental models indicate that placebo effects on allergic reactions are largest when a provider is both warm and competent [54]. Qualities of the health practitioner are likely to induce shifts in patients’ mood and emotions that in turn maximize placebo, although studies rarely measure how providers influence affect per se. Social factors may also influence expectations and health outcomes even before interactions. Recent work indicates that first impressions of health care providers based on facial features alone can influence pain and expectations about pain [55,56].

Of course, actual clinical encounters involve bidirectional interactions between patient and provider, and these may be critical for determining whether a patient will show placebo or nocebo effects [57]. Simulated clinical interactions indicate that perceived similarity [58] and perceived empathy [59] influence patients’ responses. Importantly, patient-provider interactions and the social context surrounding treatment are not likely to be the same for all patients. For instance, experienced discrimination is associated with increased stress and reduced physical and mental health [60], and individuals who have experienced negative outcomes or discrimination in a healthcare environment have different expectations than individuals who have had positive experiences [61] (see also Box 2).

Box 2. Leveraging the study of placebo to mitigate health disparities.

While we have learned a great deal about mechanisms of placebo, not all placebo effects are equal, and not all individuals are placebo responders. Individual differences have been studied from a psychological and neurobiological perspective, yet social determinants of health are likely to play a critical role. For instance, in the US, non-White individuals are less likely to receive adequate treatment for their pain [4,125], despite the fact that they experience more pain in both clinical and laboratory settings [3,126], and even health care practitioners endorse false beliefs about Black people’s pain [127]. Disparities are present even in simple decision-making tasks: White perceivers are slower to identify pain in Black faces than White faces [128] and are less likely to prescribe pain treatment for the same level of pain in Black relative to White faces. Inequities may also directly influence pain perception in racial and ethnic minority patients: A recent fMRI study demonstrated that African Americans experienced increased pain and increased heat-related activation in the VMPFC and VS relative to White and Hispanic Americans [129], and these associations were larger in participants who had experienced higher levels of discrimination.

These inequities are likely to directly impact placebo responses, but placebo research also indicates that leveraging the patient-provider interaction might help reduce disparities. Individuals whose racialized groups have suffered trauma and unfair treatment are less likely to report trust in their physician [130,131], but reported satisfaction is higher when patients see providers from their own background [132,133]. Consistent with this, in a simulated clinical experiment, concordance between experimenter and participant reduced pain within Black participants, but not Hispanic or non-Hispanic White Participants [134], and the protective effect of concordance was largest for participants who reported higher levels of experienced discrimination and worry about discrimination. Across multiple studies, perceived similarity is associated with stronger placebo analgesia [58,134] and lower expected pain [55], whereas greater experienced discrimination has been linked to higher pain and stronger brain responses to pain in Black / African American participants in a study of acute pain [129]. Non-Hispanic Black participants have also been shown to exhibit reduced μ-opioid receptor binding during painful stimulation [135], which might contribute to these effects. Placebo researchers and health disparities researchers should work together to understand how placebo and nocebo effects might differ across sociocultural groups, while taking a cultural neuroscience perspective into account [136]. For further perspectives on paths to racial equality in pain science, see [137].

A recent study of acupuncture analgesia makes an important first step toward illuminating behavioral and neurobiological processes that underlie patient-provider interactions [62]. Researchers used fMRI hyper-scanning to simultaneously measure neural responses in fibromyalgia patients undergoing experimental acute pain and clinicians. Half the dyads had a preliminary interaction to establish rapport, and researchers measured the patient-provider coherence of both facial and neural responses while clinicians applied real or sham electroacupuncture in response to patients’ pain. Analgesic effects of real and sham electroacupuncture were similar, indicating the study provides more insight into placebo effects than acupuncture per se. Patient analgesia was associated with numerous indicators of the therapeutic alliance, including: 1) the strength of the self-reported relationship with the provider; 2) providers’ estimated treatment efficacy; 3) how much patients and providers mirrored one another’s facial expressions during pain anticipation; and 4) patient-provider concordance during pain anticipation in the right TPJ and other regions of the mentalizing network [63]. These exciting findings provide initial insight on links between therapeutic rapport and acute pain relief in chronic pain. Future work should extend this approach to other patient populations and measure interpersonal processes in the context of active treatments that differ from placebo.

Brain mechanisms of placebo analgesia and specificity to pain

The potency of placebo analgesia provides evidence that pain is modulated by the joint influences of expectation, affect, and the social context surrounding treatment. How are these effects mediated, and are these mechanisms unique to pain? Placebo analgesia is likely to engage similar psychological processes as other placebo effects, since most treatments should influence expectations, shape mood, and occur in the clinic’s unique psychosocial context. Neural mechanisms that support these psychological processes may be similar across clinical outcomes, although downstream mediators of placebo effects are likely to differ across conditions. While only a few studies have compared placebo effects across modalities [42,64–67], a larger literature focuses on whether brain mechanisms that process pain (and show reductions with placebo) are unique to pain. Here, I review brain mechanisms of placebo analgesia and evaluate evidence regarding pain specificity.

Brain mechanisms of placebo analgesia

The first evidence that placebo analgesia is accompanied by real biological processes came when the μ-opioid receptor (MOR) antagonist naloxone was administered along with placebo in patients who underwent wisdom tooth extraction [68]. Patients who received naloxone reported higher post-surgical pain than the those who received only placebo, suggesting that naloxone blocked the placebo effect. This implied that the placebo effect reduced pain by engaging endogenous MORs, mimicking the effects of pharmacological opioid analgesics. Findings were later confirmed with meta-analysis [69] and PET studies that measured MOR binding under placebo [38,70,71] or showed that placebo analgesia and opioid analgesia engaged similar rACC glucose release during painful stimulation [72]. In addition, fMRI studies demonstrated that brain responses to noxious stimuli in pain-related regions were diminished with placebo [73] (see Figure 2), and that these reductions were abolished with naloxone administration [74]. Furthermore, the regions that show consistent placebo-induced increases in fMRI activation, including the dorsolateral prefrontal cortex (DLPFC), rACC, and PAG [75], have high MOR density [70,71]. Together, this work provides strong evidence for the involvement of endogenous MOR binding in placebo analgesia.

MOR engagement indicates that placebo analgesia engages some of the same mechanisms as pharmacological analgesics (see Box 3). MORs are critical for descending pain modulation, i.e. inhibiting nociceptive signals from reaching the brain [33,34]. Thus placebos might block otherwise painful signals from reaching the brain, consistent with the gate-control theory of pain [76,77]. Importantly, the effects of MOR binding are not specific to pain. MOR agonists can produce euphoria and affective shifts [78] and directly influence social processing [79,80]. Preclinical models indicate that chronic pain reduces opioid receptor availability, and this is associated with reduced sucrose preference [81]. Therefore, simply showing MOR involvement in placebo analgesia is not sufficient for proving that placebo blocks nociceptive signals through descending modulation. Instead, MORs might be more closely associated with affective and social influences discussed above. A meta-analysis of brain imaging studies of placebo analgesia indicates reliable reductions in responses to noxious stimuli within the anterior cingulate, insula, and thalamus [75] (see Figure 2). These regions have been associated with pain’s affective components (i.e. pain unpleasantness [82,83]). However, they are also collectively known as the “salience network”, due to their reliable recruitment by any salient stimulus, including pain [84–86], and are implicated generally in interoception [87–89] (see Figure 2). Thus, placebos and other forms of pain modulation might reduce the salience or aversiveness of the stimulus without changing nociception per se [85]. To determine whether placebos modulate the earliest pain signals, we need tests that are specific to pain.

Box 3. Combining placebos with pharmacological treatment: Implications for reducing opioid dependence.

Studies of placebo analgesia isolate the psychological factors that shape pain in response to inert treatment. However, these same factors also directly influence responses to active treatment, which is why clinical trials compare a treatment arm with a placebo arm to isolate the pure effects of the treatment. This subtraction logic assumes that placebo effects and treatment effects are additive and independent. We and others have directly tested these assumptions using a balanced placebo design [138], in which instructions about treatment – which manipulate expectations – are crossed with actual drug delivery in a classic 2 × 2 factorial design. Thus, participants receive an active treatment under both open (expect drug) and hidden (expect placebo) administration and receive an inert treatment under both placebo (expect drug) and control (expect placebo) conditions. This allows researchers to directly test for pure drug effects, placebo/instruction effects, and whether they interact to affect clinical outcomes. We previously observed additive effects of placebo and the μ-opioid analgesic remifentanil on subjective pain [93,119], but pure drug effects on responses to heat in pain-related regions [30,93], regardless of instruction. Importantly, different patterns of placebo-drug combinations are likely with different doses, pain measures, and analgesics. For instance, expectancy-drug interactions have been observed with the peripherally acting analgesic lidocaine [139], while cannabidiol shows additive effects on conditioned pain modulation and interactive effects on offset analgesia, pain threshold, and pain tolerance [140]. Continued research using the balanced placebo design can provide a path forward to understanding how to leverage psychological factors that underlie placebo and effectively combine them with active treatments to enhance patient outcomes and perhaps even reduce the need for pharmacological treatments.

As we learn how placebos combine with active analgesics, we may be able to safely integrate these approaches to lessen our reliance on pharmacological interventions. If we can continue to enhance the psychosocial context surrounding pain by providing a supportive patient-provider relationship, target patients’ expectations by avoiding nocebo effects and enhancing placebo effects, and reduce anxiety in patients, we may ultimately enhance pain relief non-pharmacologically and reduce the quantities of pharmacological treatments necessary to relieve suffering.

Fortunately, we can now use advanced neuroimaging analysis approaches and machine learning to develop reliable pain classifiers and evaluate pain specificity. In particular, the Neurologic Pain Signature [30] is a fMRI-based brain pattern (see Figure 2) that predicts whether a statistical brain map (i.e. a standard fMRI analysis result) represents pain or not, and which of two conditions is more painful. It is sensitive and specific to pain, highly reliable [90], and can predict acute pain accurately even in new studies [30] and in different types of noxious stimuli [91]. The NPS is strongly modulated by opioid analgesics [30]. Does it show similar reductions with placebo?

This was tested in a meta-analysis of twenty placebo analgesia fMRI studies [92]. NPS differences between painful stimulation and baseline had a large effect size, with activation in all studies and over 95% of individual participants. Across studies, placebo effects on subjective pain ratings were moderate. However, placebo analgesia was not accompanied by robust reductions in the NPS; effects were small even when restricted to placebo responders (i.e. individuals who reported large reductions in pain with placebo). Two studies in the meta-analysis permitted direct comparisons of placebo and the μ-opioid analgesic remifentanil [93,94]. While the magnitude of remifentanil analgesia and placebo analgesia was similar, the drug effect on the NPS was about ten times larger than the placebo effect [92].

Taken together, these findings suggest that placebo effects on acute pain are mediated at least in part by non-nociceptive networks. A follow up meta-analysis [95] indicated that the only reliable reductions across subjects were in the insula, habenula, and the cerebellum, as well as the ventral attention and somatomotor networks. Interestingly, most pain-related regions did not show consistent effects across studies, but were instead correlated with the magnitude of the placebo effect across participants. Most regions that had previously been implicated as showing increases with placebo analgesia, including the DLPFC and rACC, showed large heterogeneity across studies. This suggests that these modulatory regions may be sensitive to the various contextual factors reviewed earlier, such as expectation, affect, and the social context surrounding treatment, since these factors varied across studies.

Placebo effects across modalities

To what extent do studies of placebo analgesia generalize to other modalities? We can consider both the neural and psychological processes identified in placebo analgesia and studies that directly measured placebo and expectancy effects across modalities.

As mentioned in the previous section, placebos engage MOR binding, which is associated with descending pain modulation [96] as well as affective shifts [33] and social processing [97]. Rather than being specific to placebo analgesia, placebo effects on MOR binding might mediate the affective effects that accompany placebo administration. As mentioned above, placebos do not elicit reliable reductions in the NPS, and the strongest reductions are observed in regions that are not specific to pain but are instead part of the so-called “salience network.” In fact, the brain regions that make up the so-called “pain matrix” and show the strongest evidence of pain-related responses during neuroimaging include the insula, dACC, and thalamus [98,99], as depicted in Figure 2, leading to a wide debate on whether activation of these regions is indicative of pain [100,101]. Pain is highly salient and arousing, and therefore placebo effects on regions like the insula and cingulate may not be specific to pain but may instead reflect reductions in the aversiveness or salience of the noxious stimulus. If this is the case, alterations in salience responses should be seen in domains other than pain, and the brain mechanisms of placebo analgesia should generalize across domains.

Several recent studies compared noxious stimuli to salient stimuli from other sensory modalities [102–105] to measure the specificity of pain-related brain responses. When stimuli were matched on subjective intensity/salience [102], spatial patterns of brain activation distinguished painful stimuli from non-nociceptive stimuli, and nearly all nociceptive regions, with the exception of the ACC, showed stronger activity for painful stimuli relative to equally salient touch, sound, or vision. Similarly, when painful heat and unpleasant sounds were matched on the basis of physiological arousal (a proxy for salience), brain responses in the posterior parietal operculum distinguished painful stimuli from salient auditory stimuli [105]. Thus, even though pain is highly salient and activates the salience network, some brain responses are unique to pain.

Do we see similar specificity when it comes to placebo analgesia? To determine whether neural and psychological mechanisms of pain modulation are specific to placebo analgesia or whether they support placebo effects more generally, we must compare placebo analgesia with other modalities. Unfortunately, few studies have directly compared placebo effects across domains. In two studies that measured whether placebo analgesia can transfer across domains, participants were instructed that a treatment would not only reduce pain, but also that it would have effects on a second modality – either to improve motor performance [106] or to reduce anxiety [65]. Participants in the key conditions then underwent a standard placebo analgesia conditioning manipulation in which noxious stimuli were surreptitiously reduced. The researchers measured subjective and objective outcomes for both pain and the secondary modality. In each study, the researchers observed placebo analgesia as well as placebo effects in the secondary modality even though subjects were never conditioned in the secondary domain. Interestingly, placebo effects on objective outcomes for the secondary domain (i.e. improved motor performance [106] or altered P2 and N2 amplitude [65]) were only observed when instructions were paired with placebo conditioning, whereas placebo effects on subjective outcomes (fatigue [106] and unpleasantness ratings [65]) were observed whether or not subjects underwent conditioning.

Thus placebo-related expectations can transfer across modalities with expectations for cross-modal effects, although experiential learning may be necessary to reinforce expectations based on instructions. Interestingly, instructions elicited cross modal effects even without conditioning when researchers tested whether placebos could modulate both physical pain and so-called “social pain”, or feelings of rejection [107]: Relative to a control condition, individuals who received a placebo that was described as an “effective painkiller, reducing emotional distress” reported less pain in response to noxious heat and less negative affect when viewing images of partners who had rejected them. However, even within a modality, individuals show different responses to different placebo interventions [108], which might relate to individuals’ distinct expectations about different treatments. Together, studies of placebo generalization and transfer suggest that placebo effects can generalize across modalities when instructions induce expectations about this type of cross-modal reaction. However, under basic conditions individuals have different expectations about different outcomes, which may impact the extent to which expectations transfer in natural situations.

To what extent are the brain mechanisms of placebo conserved across modalities? Petrovic and colleagues [42] compared brain imaging data from a study of placebo analgesia [72] with a study of placebo anxiolytics [109] in which participants viewed emotional images when they believed they had received an anxiolytic treatment. In both studies, the magnitude of the placebo effect on subjective outcomes was correlated with the magnitude of the placebo effect on the right lateral OFC, a region that has been implicated in value processing and affective regulation in other studies. Though the lateral OFC was not identified in placebo analgesia meta-analyses [75,95], it is an important candidate to consider in placebo studies in other domains, as this region and the adjacent right VLPFC have been implicated in placebo analgesia [73,110], emotion regulation [111], and social exclusion [112], and may play a general role in regulating negative affect [113] and defensive responses [114] based on lesion models. More recently, Koban and colleagues directly compared placebo effects on pain and negative affect when participants viewed images of partners who had rejected them [107]. Activation in the right DLPFC mediated placebo effects in each domain, although the responses were distinct both in location and when tested using a classifier. The DLPFC has been identified in meta-analyses of placebo analgesia [75] and plays a causal role in placebo based on studies using transcranial magnetic stimulation [115] or measuring prefrontal degradation in Alzheimer’s Disease [116]. These findings are in line with a large body of work on the DLPFC’s general role in higher order knowledge and goal directed behavior [117], and thus it is perhaps the strongest candidate to play a modulatory role in placebo effects across domains.

Studies of cue-based expectations provide further insight on the brain mechanisms of cross-modal expectancy effects. Several recent studies [103,104] compared cue-based expectancy effects on painful and non-painful stimuli and measured the specificity of brain responses. Researchers compared pain with a different aversive outcome (disgusting odors [104] or aversive images [103]) that was matched in perceived unpleasantness. To test whether predictive cues induce cross-modal modulation, modality-specific predictive cues were presented prior to stimulation and occasionally followed by outcomes from the opposite modality (i.e. pain cues were followed by odors or images). In these studies, predictive cues modulated perception in a domain-specific manner (i.e. predictions about pain only modulated subjective pain and predictions about odors or images only modulated responses to the non-painful stimuli) and the anterior insula mediated domain-specific expectancy effects on perception in each domain. Both studies also found support for pain-specific processing in the posterior insula, similar to results mentioned earlier with respect to predictive coding [25]. As with the DLPFC’s role in placebo effects on physical and social pain [107], these studies suggest that even if the same brain region (i.e. the anterior insula) is associated with unpleasantness across modalities, it still functions in a largely modality-specific manner. Extrapolating from these findings, placebo effects may indeed engage the same circuits across modalities to support expectations, learning, and affect, but ultimately downstream mechanisms are likely to differ for distinct clinical outcomes.

Concluding remarks

In this paper I have reviewed the four main areas of research in placebo analgesia: expectancy/learning, affect, social influences, and pain specificity. I argue that we can leverage this active area of research to make headway in addressing the urgent crises surrounding health disparities in pain (see Box 2) and opioid addiction (see Box 3). While decades of research have isolated neural circuits that support pain and introduced potent pharmacological interventions that reduce pain, we are still building our knowledge base regarding the mechanisms of placebo (see Outstanding Questions). Understanding how expectations, emotion, and the social context surrounding treatment shape pain and clinical outcomes will allow us to directly target these psychological processes to reduce pain and related conditions and improve well-being. When pain researchers and clinicians join forces with social and affective neuroscientists, we may make true strides in our ability to address the opioid epidemic, reduce health disparities in pain, and improve health by reducing the burden of pain.

OUTSTANDING QUESTIONS:

• Experimental studies of placebo separately manipulate and measure cognitive processes and neural mechanisms that contribute to clinical placebo effects, including learning, instruction, affect, and social interactions, but rarely consider interactions between these different aspects of placebo. For example, do caring providers alter patients’ expectations, or do they reduce anxiety? How do expectations about pain shape emotion, mood, and affect?

• How reliable is the placebo effect across contexts or treatments? If biological factors determine an individual’s propensity to respond to placebo, responses should be stable across time within individuals, and we may be able to predict whether someone will be a “placebo responder”. Alternatively, since placebo effects depend on contextual factors, person × situation interactions might be more likely to determine responses to placebo.

• How well do experimental models of placebo analgesia generalize to clinical populations and clinical trials? Although most experimental models have manipulated acute noxious stimuli and measured placebo and mechanisms of expectancy-based pain modulation, acute experimental pain is a poor model for chronic pain. Basic researchers and clinicians should continue to combine efforts to understand whether and how placebo mechanisms shape outcomes for specific patients, specific pain conditions, and specific treatment contexts.

Highlights.

• The study of placebo analgesia provides insights on how pain and clinical outcomes are shaped by social, cognitive, and affective processes.

• Treatments for pain are highly influenced by the clinical context and the patient-provider relationship, which modulate pain by altering expectations, shifting mood, or providing social support.

• Placebo effects on pain depend on similar types of psychological mechanisms as other placebo effects, and underlying brain mechanisms contain both shared and distinct processes.

• The study of placebo has implications for mitigating health disparities and addressing the opioid crisis.

• To improve human health, pain researchers should work together with social and affective neuroscientists to understand links between pain, affect, and social processing.

Glossary

Acute pain

Brief pain that is caused by something specific (e.g. an injury or brief noxious stimulus), in contrast to chronic pain. Pain administered in the lab is usually acute pain, as it is controlled by a device that administers brief noxious stimuli such as heat, laser, or electrical pulses. Post-surgical pain or short-term pain due to injury are other examples of acute pain.

Nocebo effect

Negative health outcomes that reflect from the psychosocial treatment context, such as side effects or enhanced pain (see nocebo hyperalgesia).

Nocebo hyperalgesia

A phenomenon in which pain is exacerbated in response to an inert treatment due to expectations, emotions, and the context surrounding treatment.

Nociception

The neurobiological response to a noxious (i.e. potentially damaging) stimulus, regardless of how it is ultimately perceived.

Pain

A subjective feeling that is associated with discomfort, suffering, and either actual injury or the potential for physical damage.

Placebo

An inert substance that is administered instead of an active treatment. May be used in clinical trials to differentiate between active effects of a treatment and so-called “non-specific” factors (expectations and psychological effects, regression to the mean, and natural history).

Placebo analgesia

Pain reduction in response to an inert treatment that is due to expectations, emotions, and the context surrounding treatment.

Placebo effect

A positive treatment outcome that is attributable to the psychosocial context surrounding treatment, rather than pharmacological or biological properties of the treatment itself. Can be distinguished from placebo response in clinical trials by comparing placebo arm with waitlist control group.

Placebo response

Health outcomes in response to an inert treatment, including regression to the mean and natural fluctuations in clinical outcomes.