The facial expression of pain has attracted considerable interest in experimental and clinical research based on an increasing awareness that it supports the communication of pain as a second signal system besides the verbal one4,11 and thus can be used as another indicator of pain when self-report is missing (eg, in patients with dementia40). Right from the start of research on facial expressions of pain, researchers tried to characterize how facial activity during the experience of pain looks like. The vision was to define a prototypical facial expression of pain, similarly to prototypical facial expressions having been suggested for different emotional states.6 Groundbreaking research was conducted by Prkachin,51 who analyzed in a sample of 41 healthy students, which facial responses are displayed consistently across different types of experimental pain stimulation (pressure, temperature, electrical current, and ischemia). Facial responses were analyzed using the Facial Action Coding System (FACS7), the gold standard for facial expression research. The FACS is a fine-grained, objective, and anatomically based coding system that differentiates between 44 facial movements (action units [AUs]). Coders are trained to apply specific operational criteria to determine the onset and offset as well as the intensity of the AUs. Using the FACS, Prkachin51 suggested that there are 4 facial movements that are more steadily displayed across experimental pain modalities than other AUs, namely lowering the brows (AU4), cheek raise/lid tightening (AUs6_7), nose wrinkling/raising the upper lip (AUs9_10), and eye closure longer than 0.5 seconds (AU43). Prkachin and Salomon52 further suggested that this set of facial movements is not only indicative for experimental pain but also for clinical pain. When studying facial responses in a group of 129 shoulder pain patients undergoing a range of painful movement exercises, the authors found that the same set of facial movements was displayed as has been previously found for experimental pain.51 Mainly based on these 2 studies, this subset is regarded as presenting the key components of the facial expression of pain.9,28,50
Meanwhile, a substantial number of further studies have been conducted, investigating facial expressions of pain in various groups of individuals (eg, young, old,31 patients with depression,41 individuals with intellectual disabilities38) and during various types of pain conditions (low back pain,17 chest pain,5 and experimental pain19). At least parts of the above-described set of facial responses51 have also been found to be associated with pain in these further studies. Nevertheless, there is also considerable variability between studies, with other facial movements also having been found to be pain-related. For example, “raising the chin” (AU17)53 or even “oblique lip raising” (AU12, smiling)34,35 have been recurrently found to occur while individuals are experiencing pain. Indeed, some studies even include up to 17 AUs as a set of pain-associated AUs.13 One reason for the variability between studies is the difference in how studies defined whether an AU is pain-related. Overall, there are 2 main approaches. Approach one is to define an AU as pain-related when it occurs during pain above a critical frequency level (“frequency of occurrence” criterion), which is often set to 5%.16 Approach 2 is to define an AU as pain-related when it occurs (statistically) more frequently during pain compared with a nonpainful baseline condition or more frequently in pain patients compared with pain-free controls (“pain > baseline” criterion).51 Often, approach 2 is not conducted on all possible 44 AUs of the FACS system, but instead, authors use approach 2 consecutively after having used approach 1 to preselect AUs that fulfil the “frequency of occurrence” criterion, and then in a second step, the “pain > baseline” criterion is used to define which of these preselected AUs are really pain-related.20
The aim of this systematic review article is to examine the question of which facial movements are indeed pain-related by making use of the substantial number of primary studies that have analyzed facial responses during pain. Although it has been assumed that the above-described subset51 does include the most relevant pain-related facial movements, the meanwhile substantial empirical evidence being available has not yet been systematically used to scrutinize this assumption. We do so and take into consideration (1) the different criteria used to define whether an AU is pain-related. Moreover, given repeated doubts about the comparability of facial responses to clinical and experimental pain, we also consider (2) different types of pain (clinical vs experimental pain). Furthermore, given the increasing awareness of how important facial expressions are for pain assessment in individuals with cognitive impairments (eg, dementia40), we also consider (3) the cognitive status of the individuals being examined. Given that FACS is the most often used and best operationalized method to analyze facial expressions of pain, we limited our review to those studies using FACS, although other methods can also be used to assess facial communication of pain (eg, not FACS-based automatic systems and observational pain scales).
The systematic review was performed following the “Preferred reporting items for systematic review and meta-analysis protocols.”46
2.1. Search strategy and study selection
2.1.1. Literature search
An extensive search of literature published until April 2018 was conducted using the databases PubMed and PsycINFO. We set no restrictions with regard to the earliest year of publication. In our search, we combined with a logical AND keywords for pain (pain, nociception; with a logical OR) with keywords for facial expression (facial expression, facial display, facial activity, facial expressiveness, facial response, FACS; connected with a logical OR) (Precise search terms and combinations are available from the authors upon request). Given that we were interested in facial activity during pain in human adults, we excluded the following keywords by setting a NOT qualification: child, neonat*, animal. In addition, reference lists from identified articles as well as reviews59 and book chapters on facial expression of pain4,23 were screened for missing articles. The systematic search was limited to articles published in English or German.
2.1.2. Eligibility criteria
We selected only those studies (1) that analysed facial responses using the FACS, (2) that provide results on single AUs, (3) that include a minimum sample size of N = 20, and (4) that provided a clear description of statistics. We excluded nonoriginal research, conference proceedings, and doctoral theses. Two independent reviewers (the authors D.M. and M.K.) screened the titles and abstracts for the eligibility criteria. We retrieved full texts of all studies that were potentially relevant or could not be excluded based on the study title or abstract. In case of discrepancies/disagreement between the 2 reviewers, a third reviewer (author S.L.) was consulted and discrepancies/disagreements were resolved. The study selection process is displayed in Figure 1.
2.2. Information extraction
From each included study, we extracted the following information:
- (1) sample: patients or healthy participants, number of participants, age, sex, and cognitive status
- (2) type of pain: experimental pain (pressure, thermal, electrical, other [procedures like “blood sampling” or “injections” were added to the experimental category, given that the short invasive procedure shares more similarities to experimental pain induction than to clinical pain states]) and clinical pain
- (3) Facial Action Coding System coding: duration of sampling, how many and which AUs were FACS coded, and AU information being coded (intensity, frequency, duration, and apex)
- (4) approach used to determine pain-related AUs: selecting AUs as being pain-related based on a “frequency of occurrence” criterion or on a “pain > baseline” criterion (see the Introduction section for further explanation).
The information was extracted by one reviewer (D.M.) and documented in a data extraction form. All the extracted data were independently counterchecked by a second reviewer (M.K.). To control for bias caused by the inclusion of multiple reports of the same study, authors were contacted in cases where an overlap of the sample was suspected and the duplicate sample was excluded (eg, a healthy control sample29 was greatly overlapping with the sample of another publication31 and was, thus only included once). All ambiguities in data extraction (6% reviewer discrepancies) were double-checked and resolved.
2.3. Assessing the quality of studies
To assess the quality of the studies and the risk of bias, we graded the studies based on the following criteria (adopted from the Newcastle Ottowa criteria58), which were (1) reported sex distribution and age of the participants, (2) specification of the type of pain and in case of experimental pain on the pain induction procedure, (3) specification of the video recording (position of the camera and instruction for head positions), (4) FACS coding (duration of video samples, software used, and type of AUs being coded), (5) reliability of FACS coding, and (6) the extent to which the study sample represents the true population under investigation (eg, with regard to sex, education, severity, and duration of chronic pain). Each criterion was judged as either “successfully fulfilled” (1), “partially fulfilled” (0.5), or “not fulfilled” (0). The total possible quality score was 6.0.
Our main aim is to find out which AUs prove to be pain-related across studies. Given that studies differ with regard to how they defined whether an AU is pain-related, we separately report findings for (1) “frequency of occurrence” criterion (% occurrence during pain has to surpass a certain threshold [often 5%]) and for (2) the stricter criterion, “pain > baseline,” or “pain patients > pain-free controls” comparisons (based on significant P-values or moderate effect sizes), respectively. Moreover, given the possibility that facial responses to pain might be affected by the “type of pain” being induced/experienced or by the “cognitive status” of the person, we compiled the AU findings separately for these 2 domains. In some studies, more than one sample was investigated (eg, patients with dementia and healthy controls1). In these cases, AU outcomes are reported separately for each sample (Tables 1–3). Likewise, if studies used different types of experimental pain (eg, pressure and heat pain20), the outcomes are also reported separately for each type of pain (Tables 1–3).
Action unit findings are presented as descriptive frequency statistics.
3.1. Characteristics of included studies
The initial literature search identified 2304 studies with 4 additional studies found through manual searching of reference lists. The study selection process is displayed in Figure 1. After excluding duplicates and screening the remaining abstracts and titles, 97 studies remained. After reviewing the full texts of these remaining articles, 60 articles were excluded. The reasons for exclusion are listed in Figure 1. Altogether 37 articles were retained for analyses, with 27 studies assessing facial responses during experimental pain (Table 1) and 10 studies assessing facial responses during clinical pain (Table 2). Most of the included studies (78%) reached a high-quality score (≥5.0 of 6.0), and the remaining studies (22%) showed a good-quality score (4.0 to <5 of 6.0). Thus, we are confident that the reported outcomes are not biased by a lack of quality of the included studies.
3.1.1. Sample characteristics
Altogether, facial responses during pain were investigated in 2237 individuals. Most often, experimental pain models were used to study facial responses. Indeed, facial responses during experimental pain were assessed in 1578 individuals (847 females and 668 males [for 63 participants, gender information was missing]). Facial responses during clinical pain were assessed in 659 individuals (366 females and 293 males). Among the experimental pain models, thermal heat pain was used most often to elicit facial responses, followed by pressure pain (Table 1). The sex distribution across studies was quite balanced, with a slight tilt toward more female participants (56% of the participants were female).
3.1.2. Facial Action Coding System coding
With regard to the FACS coding, most studies coded the whole set of 44 AUs (84%), with only a few studies limiting the FACS coding to a set of AUs that has previously been found to be associated with pain (eg, 2 studies9,28 only coded those AUs reported to be pain-related by Prkachin51). Moreover, in most studies, AU frequency (87%) and AU intensity (93%) were coded, whereas only 25% of the studies coded AU duration. Interestingly, coding of AU duration was more common in clinical pain studies (50% of clinical pain studies coded the duration of an AU) and in experimental studies that used somewhat longer stimulation times (>5 seconds). Thus, the duration of an AU was supposed to hold more meaningful information when the painful stimulus or the pain experience is not limited to a few seconds. For analyses purpose, most studies combined those AUs that represent very similar facial movements into one aggregate AU, namely AU1 and AU2 were combined into AU1_2, AU6, & AU7 into AU6_7, AU9, & AU10 into AU9_10 and AU25 & AU26 & AU27 into AU25_26_27.
3.1.3. Definition of pain-related action units
As mentioned above, the studies differ in their approach of how to define whether an AU is pain-related or not. Overall, 5 studies based their selection of pain-related AUs solely on their “frequency of occurrence” (see column “% occurrence” in Tables 1 and 2). As soon as an AU was displayed in more than 5% (sometimes 1%) of the painful segments (or of the participants), it was classified as pain-related. Most studies (N = 32) chose the more stricter criterion, namely that an AU had to be displayed more frequently during pain compared with a baseline condition or more frequently in pain patients compared with healthy controls, respectively, to be chosen as pain-related (see column “pain > baseline/pain patients > controls” in Tables 1 and 2). To determine the fulfillment of this criterion, T-tests (P-values) or effect sizes (Cohen's d) were computed and presented comparing AU occurrences between pain vs baseline or pain patients vs healthy controls, respectively. Interestingly, 23 of these 32 studies used the stricter “pain > baseline/pain patients > controls” criterion as a second step, after preselecting AUs, which fulfilled the “frequency of occurrence” criterion in a first step and then computing which of these preselected AUs are really pain-related based on the stricter “pain > baseline/pain patients > controls” criterion.
3.2. Pain-related facial responses
To give a better overview on which AUs are found to be pain-related across studies, we calculated separately for each AU in how many studies the given AU met the “frequency of occurrence” criterion as well as the “pain > baseline”/“pain patients > pain-free controls” criterion. These data are presented in Table 3. Of the existing 44 AUs from the FACS system, we only included those AUs in Table 3 that fulfilled either the “frequency of occurrence” criterion or the “pain > baseline”/“pain patients > pain-free controls” criterion in at least one of the studies.
3.2.1. Pain-related action units: “frequency of occurrence” criterion
As can be seen in Tables 1 and 2 (column “% of occurrence”) as well as in Table 3, A, selecting AUs as pain-related based on their “frequency of occurrence” results in a large number of AUs, which meet this criterion.
Across all samples and across all types of pain, there are 10 AUs, which meet the “frequency of occurrence” criterion in at least 50% of the studies, namely AUs 1_2, 4, 6_7, 9_10, 12, 14, 17, 25_26_27, 43, 45 (Table 3, A, left column).
22.214.171.124. Clinical pain
When looking at the outcomes separately for clinical pain, the “frequency of occurrence” criterion was applied to select pain-related AUs in only 4 studies. Across these studies, the list of AUs meeting the “frequency of occurrence” criterion is quite extensive and includes 12 AUs (Table 3, A).
126.96.36.199. Experimental pain
When looking at the outcomes for experimental pain paradigms, the “frequency of occurrence” criterion was applied in 35 samples/paradigms. When comparing the overall experimental pain outcomes with the outcomes found for the different types of experimental pain, it becomes apparent that there are no systematic variations. Similar lists of AUs meet the “frequency of occurrence” criterion across experimental heat, pressure, and electrical pain induction. The only difference seems to be that some of the lower face movements (AU12 [lip corner pull], AU14 [dimple], and AU17 [chin raise]) are observed in fewer studies using pressure stimulation compared with those using heat or electrical stimulation.
188.8.131.52. Clinical vs experimental pain
There is a great overlap in AUs, which meet the “frequency of occurrence” criterion in at least 50% of the studies using clinical pain and those using experimental pain (Table 3, A). The greatest differences are that more lip movements (AU18 [lip pucker], AU20 [lip stretch], and AU24 [lip press]) are observed in clinical pain conditions compared with experimental pain, and that closing of the eyes for longer than half a second (AU 43) seems more prevalent in clinical pain conditions.
184.108.40.206. Cognitive status of the individual
Comparing the AUs outcomes between individuals with and without cognitive impairments, it becomes apparent that the AU percentage numbers tend to be lower for individuals with cognitive impairments (Table 3, A, right column). Only 6 AUs meet the “frequency of occurrence” criterion in at least 50% of the studies that included individuals with cognitive impairment (compared with 10 AUs in individuals without cognitive impairments).
3.2.2. Pain-related action units: “pain > baseline,” respectively, “pain patients > pain-free controls” criterion
As can be seen in Table 3, B, there are far fewer AUs that meet this stricter criterion compared with the “frequency of occurrence” criterion.
Across all samples and across all types of pain, there were only 4 AUs, which meet the “pain > baseline” criterion in at least 50% of studies/samples, namely AUs 4, 6_7, 9_10, and 25_26_27 (Table 3, B, left column).
220.127.116.11. Clinical pain
When looking at the outcomes separately for clinical pain, the list of AUs, which meet the “pain > baseline” criterion or the “pain patients > pain-free controls” criterion, respectively, is very comparable with the overall results, with the addition of 1 AU, namely closing of the eyes for longer than half a second (AU43).
18.104.22.168. Experimental pain
The findings for experimental pain are also very comparable with the overall results. Moreover, the same AUs meet the “pain > baseline” criterion when applying heat and pressure pain stimulation. Only the findings for electrical pain seem to differ, with more studies finding blinking (AU45) to be pain-related, which might be due to the sudden nature of this type of experimental pain stimulation eliciting startle responses. In the “others” category (eg, venepuncture and injection), only the brow lower movement (AU4) is consistently found to occur more often during pain compared with baseline.
22.214.171.124. Clinical vs experimental pain
When comparing outcomes for clinical vs experimental pain, there is only one difference, namely that closing of the eyes for longer than half a second (AU43) is found to be pain-related in 50% of the studies looking at clinical pain responses, whereas only 22% of the studies using experimental pain find this facial movement to occur more frequently during pain compared with baseline.
126.96.36.199. Cognitive status of the individual
As can be seen in Table 3, B (right column), the same AUs meet the “pain > baseline” criterion in more than half of the studies investigating facial responses during pain in individuals with as well as without cognitive impairments.
The stricter criterion “pain > baseline” resulted not only in smaller numbers of AUs to meet this criterion, compared with the “frequency of occurrence” criterion, but also in much more consistent results. The same set of AUs proved to be pain-related in at least 50% of the studies, regardless of observing facial responses during clinical or experimental pain and regardless of the cognitive status of the individual being observed. This subset is illustrated in Figure 2 and is composed of lowering the brows (AU4), cheek raise and lid tightening (AUs6_7), nose wrinkling and raising the upper lip (AUs9_10), and opening of the mouth (AUs25_26_27). There is only one substantial variation between clinical and experimental pain conditions, namely that half of the studies looking at clinical pain conditions found that individuals also show an increase in closing their eyes for longer than half a second (AU43, Fig. 2) when they are experiencing pain.
However, one has to keep in mind that this small subset of pain-related AUs (Fig. 2) does not occur consistently in all studies. As can be seen in Table 3, B, not a single AU is found to be pain-related in all studies. Moreover, even if a study finds an AU to be pain-related on a group level, this does not mean that every individual displayed this AU more frequently during the experience of pain. Therefore, even if Figure 2 suggests that the combination of AUs is very stable and uniform, the actual combinations of pain-related AUs vary substantially between individuals and across episodes.24
The aim of this article was to examine the question of which facial movements are indeed indicative of pain by conducting a systematic review of the available empirical evidence. Thirty-seven studies, investigating facial responses during pain by use of the FACS and separately reporting findings on single AUs, were included. The findings on pain-related AUs were synthesized across studies by taking into consideration (1) the different criteria used to define whether an AU is pain-related, (2) the different types of pain (clinical vs experimental pain), and (3) the cognitive status of the individuals being examined.
4.1. The role of criterion used to define whether a facial response is pain-related
Across the studies on facial responses during pain, there are 2 main approaches used when deciding which AUs to include as pain-related in the analyses. One approach is to include all AUs that were displayed above a critical frequency level during pain. Another, more stricter approach is to classify only those AUs as pain-related that were displayed more frequently or more intensely during pain compared with a baseline condition or observed in pain patients compared with pain-free persons by statistical threshold criteria (eg, certain effect sizes), which helps to define what “more” means. In the included studies, the baseline condition was most often a nonpainful stimulation procedure (in case of experimental pain stimulation), a resting phase, or a comparison with pain-free individuals (in case of clinical pain). Most often, authors combined these approaches, classifying AUs as pain-related if they fulfil the “frequency of occurrence” (step 1) and the “pain > baseline” (step 2) criteria.
As this review demonstrates, selecting AUs as pain-related only based on their “frequency of occurrence” results in a rather large, fuzzy subset of AUs that lack consistency across studies, across types of pain, and across individuals with and without cognitive impairments. By contrast, when using the stricter criterion and defining AUs as pain-related only if they increase in intensity or frequency during pain, a much smaller and quite stable subset of facial responses was found across studies. Most agreement overall could be found for brow lowering (AU4) and cheek raise and lid tightening (AUs6_7). These facial movements were found to increase during pain in around 80% of the reviewed studies. Similarly high agreement across studies was also found for nose wrinkling and raising the upper lip (AUs9_10), with more than 70% of all studies finding this facial movement to increase during pain. The agreement for the facial movement “opening of the mouth” (AUs25_26_27) was a bit lower, with approximately 60% of the studies finding this movement to increase during pain. To reverse perspective, even the most frequent facial signals of pain could not be found in all studies. Thus, there is commonality between studies but not to a perfect degree, which also excludes the notion of a strict uniformity of facial expressions.
Given that the stricter criterion (pain > baseline) resulted in a much smaller and much more consistent subset of facial responses, this strongly suggests to always include a baseline or control group condition when conducting research on facial responses to pain, especially in those studies that look for group-specific patterns in facial expressions of pain (eg, patients with migraine and patients with schizophrenia). Including a baseline or control group allows for defining which facial responses are pain-indicative for the given type of pain and for the given sample of individuals being studied.
4.2. Clinical vs experimental pain
This review corroborates previous assumptions, namely that facial responses elicited by experimental pain stimulation are very comparable with facial responses displayed during clinical pain conditions.52 Especially when applying the stricter criterion (pain > baseline), it becomes apparent that the core subset of pain-related facial responses was similarly displayed both during experimental and clinical pain conditions. There was only one variation, namely with regard to closing of the eyes for longer than half a second (AU43) (Fig. 2). Although half of clinical pain studies found this facial response to be pain-related, only 20% of the studies using experimental pain corroborated this. Thus, closing of the eyes for longer than half a second might be especially indicative for clinical pain, and, thus, for pain states that might be of longer duration and of greater severity than experimental pain. In line with this, closing of the eyes (AU43) is based on activity of the orbicularis oculi muscle, the same muscle that underlies the pain-related cheek raise and lid tightening (AU6_7).7 Although contraction of the orbital part of the muscle results in AU6_7 (narrowing of the eye aperture), activity of the palpebral part results in AU43 (complete closing the eyes). Thus, in the context of pain, AU43 might occur as an intensification of AU6_7, signalling more severe or prolonged levels of pain that are more likely in clinical pain than in experimental pain settings.50
With regard to differences between different types of experimental pain, the most variance occurred for electrical stimulation. Here, blinking (AU45) was found to increase during pain in 75% of the studies. It seems likely that this is due to the sudden, startling nature of this type of pain stimulation, resulting in more startle responses (the blink component of the startle reflex39) compared with other types of pain. Thus, when being interested in relevant facial responses during clinically ongoing pain, choosing an experimental pain protocol that uses electrical pain induction methods seems less ideal (with the exception of cases with attack-like clinical pain).
4.3. The role of cognitive status
One major reason for the increased interest in facial responses during pain is the notion that facial responses could serve as a substitute to self-report in individuals who are not capable to provide pain self-report because of cognitive impairments.12,40 However, to use facial responses to assess pain in individuals with cognitive impairments, one must first investigate whether the facial encoding of pain might be altered because of the cognitive impairment. For this review, we could include 9 studies investigating facial responses in individuals with cognitive impairments. The cognitive impairment was mostly due to dementia-related cognitive decline in samples of older individuals.1,14,15,29,32,36,45 Across all 9 studies, the same subset of facial responses proved to be pain-related (pain > baseline) in most of the studies as was found for cognitively unimpaired individuals. Thus, this review gives clear evidence that the type of facial responses being displayed during pain is unaffected by the cognitive status of the individual (Fig. 2). This is in line with those studies which directly compared facial responses with pain between individuals with and without dementia.1,29,36 In all 3 studies, the authors found that individuals with dementia display the same AUs in response to experimental pain stimulation as individuals without dementia do. Even those individuals with more advanced stages of dementia, who were not able to provide a self-report of pain, displayed the same subset of pain-related facial responses.36 The only difference found between groups was that individuals with dementia displayed this subset of pain-related facial responses more intensely or more vigorously compared with individuals without dementia.1,29,36
4.4. Comparing the findings with the “prototypical facial expression of pain”
As stated in the introduction, Prkachin et al. could show in 2 studies that there is a core subset of pain-related facial responses, which occurs across clinical and different types of experimental pain51,52 and which has sometimes been referred to as the prototypical facial expression of pain.9,26,56 Comparing this prototypical facial expression of pain to the subset of AUs that showed to be pain-related in at least half of the included studies of this review, it becomes apparent that the findings are very comparable. As demonstrated in Figure 2, 3 facial movements (brow lowering [AU4]; cheek raise and lid tightening [AUs6_7]; nose wrinkling and raising the upper lip [AUs9_10]) were found to be pain-related in most of the included studies. These 3 facial movements are identical to the core movements of the facial expression of pain as reported by Prkachin et al.51,52 However, there is also at least one crucial divergent finding. Although Prkachin et al. did not include the opening of the mouth (AUs25_26_27) in the subset of pain-related facial responses, our findings clearly suggest that this movement is one of the key facial movements because it was found to increase or become more frequent when individuals are experiencing pain. Both during experimental and clinical pain at least half of the studies found “mouth opening” to be pain-related. Opening the mouth during pain could be a preparatory movement for pain vocalizations (“ouch,” “ooh,” and “aah”). Based on this review, opening of the mouth should be included in the subset of pain-related facial responses. Another variation between the present review and Prkachins' findings is that one of the key movements of pain described by Prkachin et al., namely closing of the eyes for longer than half a second (AU43), only proved to be pain-related in clinical pain conditions.
4.5. Variability despite a core subset
To avoid any erroneous ideas of a strong uniformity of facial expressions of pain, which might be suggested by postulating a core subset of facial responses to pain, the following arguments have to be considered. The facial responses of the core subset are more often displayed during pain than other facial responses and are more frequently displayed during pain compared with baseline conditions, but they are far from being consistently displayed during each pain episode in each individual. Indeed, most often, individuals do not show the whole subset of pain-related facial responses when experiencing pain but may only display a single facial movement or combine 2 or 3 of them.24 One reason for this variability between individuals is due to people varying in their degree to which they facially express pain, with expressive vs stoic variants. We learn to inhibit the facial display of negative affective states, including pain, following different social display rules,4 which in turn results into individually different learning histories. The degree to which we inhibit the facial expression of pain is—besides this learning history—also dependent on intraindividual factors (eg, familiarity of social situations19) as well as on further interindividual factors (eg, general ability to inhibit automatic motor movements18); these factors can differentially affect the various facial muscles, with upper face muscles being more under automatic motor control compared with lower face muscles.54
This intraindividual and interindividual variability of facial expressions of pain does not contradict the assumption of a core subset of facial responses during pain, given that this core subset provides a limited number of facial signals characteristic of pain, which can be individually and situationally combined and aggregated.
What does the recognition of variability mean for clinical practice, when relying on facial expression to assess pain in nonverbal individuals (eg, individuals with dementia)? It is crucial that health care professionals become aware that facial expressions of pain vary between individuals and situations. Thus, when choosing an observational pain scale to assess pain in nonverbal individuals, which is clinically the necessary alternative to the time-consuming manual application of FACS, one should choose a scale that does not only include the general description of a prototypical facial expression of pain but instead include separate specific facial items that cover the facial signals characteristic of pain (eg, PACSLAC,8 PAIC-1542).
Moreover, given that these facial signals are not only truly specific to pain states, but also occur in other emotional states, the risk of false-positive pain judgements is quite high. Indeed none of the 4 to 5 pain-related facial movements is exclusively related to pain. The greatest overlap to other emotional states can be found with the facial expressions of disgust (sharing brow lowering [AU4], cheek raise and lid tightening [AUs6_7], and nose wrinkling and raising the upper lip [AUs9_10]8,33) and anger (sharing nose brow lowering [AU4], and cheek raise and lid tightening [AUs6_7]).8 This overlapping facial phenomenology makes the consideration of the combination and aggregation of single facial signals necessary for successful distinction of emotional and pain states. Furthermore, the observations of facial expressions in clinical settings do not occur in isolation but are embedded in a context, which favors the assumption of certain emotional and pain states relative to others. In addition, the facial expression is accompanied by other types of state-indicative behaviors (eg, body posture and vocalizations), the consideration of which surely helps to improve the specificity of observations. The final perspective is the use of multisensor data recording with the facial responses being among the key variables as basis of automatic pain recognition, which can be individualized by machine learning algorithms.37,55
4.6. Strengths and weaknesses
The review included studies with varying sample sizes, different sample characteristics, different intensities and different types of pain, different social settings, different stimulation protocols, and different protocols for FACS coding. These variations have surely affected the outcomes (eg, depending on the social setting, individuals tend to more or less inhibit their facial expression of pain19,21) and make it difficult to directly compare the studies. This high heterogeneity between studies at first glance was one of the main reasons why we decided to “only” conduct a systematic review instead of also performing a meta-analysis. To compile data into a meta-analysis, the data have to fulfil stricter homogeneity requirements. Our aim was to give a first broad and comprehensive overview of the empirical evidence on facial responses during pain without being constrained to the methodological requirements of meta-analyses. The next step would be to perform a meta-analysis on a homogenous subgroup of the included studies. It is noteworthy, that despite the heterogeneity in methodology between studies, a quite stable subset of pain-related facial responses was found across studies.
However, the results are limited to the measurement of facial expressions by the FACS, and it is not clear that other methods would produce the same results. Although FACS is the gold standard and the most widely used method in facial expression research, this method does have several limitations. Besides the enormous time effort, it takes to train somebody in FACS coding (approximately 100 hours), performing the FACS coding itself is also very time-consuming, thus, limiting its usefulness for clinical practice. Moreover, although FACS coding is generally viewed as an objective description of facial activity (given its anatomical base),8 it is based on human judgments and thus has elements of subjectivity in it, despite intrarater reliability values being quite high (usually above 0.8). Furthermore, given that FACS coding is based on observable movements in the face, more subtle facial activity remains unnoticed. The FACS coding is also limited in its possibility to capture the complex dynamics of temporal patterns in facial expressions. Some of these limitations can be overcome by alternative methods to analyse facial expressions of pain. Using surface electromyography, for example, allows to assess even very subtle changes in facial muscle activity. However, electromyography performs poorly compared with FACS coding with regard to pinpointing the exact location of the facial muscle activity, given that it captures activity from neighbouring muscles.57 More recent progress in computer vision technology has led to the development of automatic analyses of facial expressions, which are partially based on AU detection and partially use other forms of facial mapping. These approaches seem to promise an objective assessment of facial expressions of pain. However, they are more affected by illumination conditions, variation in head pose, errors in face mapping, wrinkles in the face, etc., compared with manual FACS coding.37 Therefore, they cannot be used as valid alternatives (clinically or experimentally) for the time being, but they hold great promise for the future, asking for further interdisciplinary cooperation between medicine, nurses, psychology, engineers, and computer sciences. The present review may help to inform the necessary classification algorithms for pain recognition by providing knowledge about the critical elements of pain-relevant facial responses.
When reviewing the research on facial responses to pain (based on FACS coding), our semiquantitative analyses revealed that there is a small subset of facial responses that are consistently found to be associated with pain. Corroborating previous findings, this subset is unaffected by the cognitive status of the individual and is very comparable between clinical and experimental pain states. However, despite this stable subset of pain-related facial responses, one has to keep in mind that this subset does not represent one uniform facial expression of pain that can—at all time and in each individual—be observed in the presence of pain.24 Instead, this subset of pain-related facial responses seems to convey—as already stated by Prkachin51—“the bulk of information about pain that is available in facial expression” but not a uniform facial expression of pain. Thus, both for clinical and experimental pain assessment, a more individualized approach should be preferred, which allows for determining the pain-related facial responses an individual combines and aggregates to express pain instead of erroneously searching for an uniform expression of pain in each sufferer's face.
Conflict of interest statement
The authors have no conflict of interest to declare.
The authors thank Dominik Seuss for the support in creating Figure 2.
. Beach PA, Huck JT, Miranda MM, Foley KT, Bozoki AC. Effects of Alzheimer disease on the facial expression of pain
. Clin J Pain 2016;32:478–87.
. Craig KD, Patrick CJ. Facial expression during induced pain. J Pers Soc Psychol 1985;48:1080–91.
. Craig KD, Hyde SA, Patrick CJ. Genuine, suppressed and faked facial behavior during exacerbation of chronic low back pain. PAIN 1991;46:161–71.
. Craig KD, Prkachin KM, Gruneau RE. In: Turk DC, Melzack R, editors. Handbook of pain. New York: The Guilford Press, 2011.
. Dalton JA, Brown L, Carlson J, McNutt R, Greer SM. An evaluation of facial expression displayed by patients with chest pain. Heart Lung 1999;28:168–74.
. Ekman P, Freisen WV, Ancoli S. Facial signs of emotional experience. J Pers Soc Psychol 1980;39:1125.
. Ekman PE, Friesen WV. Facial action coding system. Palo Alto: Consulting Psychologists Press, 1987.
. Fuchs-Lacelle S, Hadjistavropoulos T. Development and preliminary validation of the pain assessment checklist for seniors with limited ability to communicate (PACSLAC). Pain Manag Nurs 2004;5:37–49.
. Hadjistavropoulos T, Browne ME, Prkachin KM, Taati B, Ashraf A, Mihailidis A. Pain in severe dementia: a comparison of a fine-grained assessment approach to an observational checklist designed for clinical settings. Eur J Pain 2018;22:915–25.
. Hadjistavropoulos T, Craig KD, Martin N, Hadjistavropoulos H, McMurtry B. Toward a research outcome measure of pain in frail elderly in chronic care. Pain Clin 1997;10:71–9.
. Hadjistavropoulos T, Craig KD, Duck S, Cano A, Goubert L, Jackson PL, Mogil JS, Rainville P, Sullivan MJ, de C Williams AC, Vervoort T, Fitzgerald TD. A biopsychosocial formulation of pain communication. Psychol Bull 2011;137:910–39.
. Hadjistavropoulos T, Herr K, Prkachin KM, Craig KD, Gibson SJ, Lukas A, Smith JH. Pain assessment in elderly adults with dementia. Lancet Neurol 2014;13:1216–27.
. Hadjistavropoulos T, LaChapelle DL, Hadjistavropoulos HD, Green S, Asmundson JG. Using facial expressions to assess musculoskeletal pain in older persons. Eur J Pain 2002;6:179–87.
. Hadjistavropoulos T, LaChapelle DL, MacLeod FK, Snider B, Craig KD. Measuring movement-exacerbated pain in cognitively impaired frail elders. Clin J Pain 2000;16:54–63.
. Hadjistavropoulos T, LaChapelle D, MacLeod F, Hale C, O'Rourke N, Craig KD. Cognitive functioning and pain reactions in hospitalized elders. Pain Res Manag 1998;3:145–51.
. Hampton AJD, Hadjistavropoulos T, Gagnon MM, Williams J, Clark D. The effects of emotion regulation strategies on the pain experience: a structured laboratory investigation. PAIN 2015;156:868–79.
. Hill ML, Craig KD. Detecting deception in pain expressions: the structure of genuine and deceptive facial displays. PAIN 2002;98:135–44.
. Karmann AJ, Lautenbacher S, Kunz M. The role of inhibitory mechanisms in the regulation of facial expressiveness during pain. Biol Psychol 2015;104:82–9.
. Karmann AJ, Lautenbacher S, Bauer F, Kunz M. The influence of communicative relations on facial responses to pain: does it matter who is watching? Pain Res Manag 2014;19:15–22.
. Karmann AJ, Maihöfner C, Lautenbacher S, Sperling W, Kornhuber J, Kunz M. The role of prefrontal inhibition in regulating facial expressions of pain: a Repetitive Transcranial Magnetic Stimulation Study. J Pain 2016;17:383–91.
. Krahé C, Springer A, Weinman JA, Fotopoulou AK. The social modulation of pain: others as predictive signals of salience—a systematic review. Front Hum Neurosci 2013;7:386.
. Kunz M, Hennig J, Karmann AJ, Lautenbacher S. Relationship of 5-HTTLPR polymorphism with various factors of pain processing: subjective experience, motor responsiveness and catastrophizing. PLoS One 2016;11:e0153089.
. Kunz M. Behavioural and facial markers of pain, emotion, cognition. In: Pickering G, Gibson S, editors. Pain, emotion and cognition. Cham, Switzerland: Springer International Publishing, 2015.
. Kunz M, Lautenbacher S. The faces of pain: a cluster analysis of individual differences in facial activity patterns of pain. Eur J Pain 2014;18:813–23.
. Kunz M, Chatelle C, Lautenbacher S, Rainville P. The relation between catastrophizing and facial responsiveness to pain. PAIN 2008;140:127–34.
. Kunz M, Chen JI, Lautenbacher S, Vachon-Presseau E, Rainville P. Cerebral regulation of facial expressions of pain. J Neurosci 2011;31:8730–8.
. Kunz M, Faltermeier N, Lautenbacher S. Impact of visual learning on facial expressions of physical distress: a study on voluntary and evoked expressions of pain in congenitally blind and sighted individuals. Biol Psychol 2012;89:467–76.
. Kunz M, Gruber A, Lautenbacher S. Sex differences in facial encoding of pain. J Pain 2006;7:915–28.
. Kunz M, Mylius V, Scharmann S, Schepelman K, Lautenbacher S. Influence of dementia on multiple components of pain. Eur J Pain 2009;13:317–25.
. Kunz M, Mylius V, Schepelmann K, Lautenbacher S. On the relationship between self-report and facial expression of pain
. J Pain 2004;5:368–76.
. Kunz M, Mylius V, Schepelmann K, Lautenbacher S. Impact of age on the facial expression of pain
. J Psychosom Res 2008;64:311–18.
. Kunz M, Mylius V, Schepelmann K, Lautenbacher S. Effects of age and mild cognitive impairment on the pain response system. Gerontology 2009;55:674–82.
. Kunz M, Peter J, Huster S, Lautenbacher S. Pain and disgust: the facial signaling of two aversive bodily experiences. PLoS One 2013;8:e83277.
. Kunz M, Prkachin K, Lautenbacher S. The smile of pain. PAIN 2009;145:273–5.
. Kunz M, Prkachin K, Lautenbacher S. Smiling in pain: explorations of its social motives. Pain Res Treat 2013;2013:128093.
. Kunz M, Scharmann S, Hemmeter U, Schepelmann K, Lautenbacher S. The facial expression of pain
in patients with dementia. PAIN 2007;133:221–8.
. Kunz M, Seuss D, Hassan T, Garbas JU, Siebers M, Schmid U, Lautenbacher S. Problems of video-based pain detection in patients with dementia: a road map to an interdisciplinary solution. BMC Geriatr 2017;17:33.
. LaChapelle DL, Hadjistavropoulos T, Craig KD. Pain measurement in persons with intellectual disabilities. Clin J Pain 1999;15:13–23.
. Lang P. The emotion probe: studies of motivation and attention. Am Psychol 1995;50:372–85.
. Lautenbacher S, Kunz M. Facial pain expression in dementia: a review of the experimental and clinical evidence. Curr Alzheimer Res 2017;14:501–5.
. Lautenbacher S, Bär KJ, Eisold P, Kunz M. Understanding facial expressions of pain in patients with depression. J Pain 2017;18:376–84.
. Lautenbacher S, Walz AL, Kunz M. Using observational facial descriptors to infer pain in persons with and without dementia. BMC Geriatr 2015;18:88.
. LeResche L, Dworkin SF. Facial expressions of pain and emotions in chronic TMD patients. PAIN 1988;35:71–8.
. Limbrecht-Ecklundt K, Werner P, Traue HC, Al-Hamadi A, Walter S. Mimic activity of differentiated pain intensities: correlation of characteristics of Facial Action Coding System and electromyography. Schmerz 2016;30:248–56.
. Lints-Martindale AC, Hadjistavropoulos T, Barber B, Gibson SJ. A psychophysical investigation of the facial action coding system as an index of pain variability among older adults with and without Alzheimer's disease. Pain Med 2007;8:678–89.
. Moher D, Shamseer L, Clarke M, Ghersi D, Liberati A, Petticrew M, Stewart LA. Preferred reporting items for systematic review and meta-analysis protocols (PRISMA-P) 2015 statement. Syst Rev 2015;4:1.
. Patrick CJ, Craig KD, Prkachin KM. Observer judgments of acute pain: facial action determinants. J Pers Soc Psychol 1986;50:1291–8.
. Priebe JA, Kunz M, Morcinek C, Rieckmann P, Lautenbacher S. Electrophysiological assessment of nociception in patients with Parkinson's disease: a multi-methods approach. J Neurol Sci 2016;368:59–69.
. Prkachin KM. Effects of deliberate control on verbal and facial expressions of pain. PAIN 2005;114:328–38.
. Prkachin KM, Mercer SR. Pain expression in patients with shoulder pathology: validity, properties and relationship to sickness impact. PAIN 1989;39:257–65.
. Prkachin KM. The consistency of facial expressions of pain: a comparison across modalities. PAIN 1992;51:297–306.
. Prkachin KM, Solomon PE. The structure, reliability and validity of pain expression: evidence from patients with shoulder pain. PAIN 2008;139:267–74.
. Rahu MA, Grap MJ, Cohn JF, Munro CL, Lyon DE, Sessler CN. Facial expression as an indicator of pain in critically ill intubated adults during endotracheal suctioning. Am J Crit Care 2013;22:412–22.
. Rinn WE. The neuropsychology of facial expression: a review of the neurological and psychological mechanisms for producing facial expressions. Psychol Bull 1984;95:52.
. Sangineto E, Zen G, Ricci E, Sebe N. We are not all equal: personalizing models for facial expression analysis with transductive parameter transfer. Proceedings of the 22nd ACM International Conference on Multimedia 2014. ACM, Orlando, FL. p. 357–366.
. Simon D, Craig KD, Miltner WHR, Rainville P. Brain responses to dynamic facial expressions of pain. PAIN 2006;126:309–18.
. Tassinary LG, Cacioppo JT, Vanman EJ. The skeletomotor system: surface electromyography. In: Cacioppo JT, Tassinary LG, Berntson GG, editors. Handbook of psychophysiology. 3rd ed. Cambridge: Cambridge University Press, 2009. p. 267–300.
. Williams AC. Facial expression of pain
: an evolutionary account. Behav Brain Sci 2002;25:439–55.