Enhancing sensitivity to facial expression of pain

Pain 71 (1997) 279–284

Enhancing sensitivity to facial expression of pain Patricia E. Solomon a ,*, Kenneth M. Prkachin b, Vern Farewell c a

School of Rehabilitation Science, McMaster University, 1280 Main St. W. Bldg. T-16, Hamilton, ON, L8N 4K1 Canada b University of Northern British Columbia, Psychology Program, P.O. Bag 1950, Prince George, BC, V21 5P2 Canada c Department of Statistical Science, University College, London, Gower Street, London, WCLE 6BT, UK Received 4 March 1996; revised version received 26 August 1996; accepted 6 February 1997

Abstract Clinicians have long appreciated the information communicated by a patient’s facial expression. Advances in the measurement of facial movements, using the Facial Action Coding System (FACS) have allowed for identification of a universal expression of pain, which is primarily encoded in four facial movements. While the FACS provides a rigorous assessment of facial expression, the time required to learn the system and to analyze the facial expression by use of slow motion video recording, makes its use impractical in the clinical setting. The purpose of this research was to examine whether exposure to a brief training procedure, based on orienting subjects to the four facial movements, would increase sensitivity to pain communicated by facial expression. Seventy-five occupational and physical therapy student volunteers were randomly assigned to training or control groups. The trained group was exposed to a 30-min training session. Both groups were then asked to rate a videotape of patients undergoing assessment of a painful shoulder and rate the amount of discomfort the patients appeared to be experiencing. Analyses indicated that the trained group was significantly more sensitive to subtle facial movements associated with low levels of pain. Relative to the patients’ ratings, there was a tendency for raters to underestimate pain particularly when these were at a high level. The findings lend hope to the feasibility of developing a tool which would be clinically useful though this may be more difficult for observers judging more complex facial expressions associated with high levels of pain.  1997 International Association for the Study of Pain. Published by Elsevier Science B.V. Keywords: Facial expression; FACS; Facial movements

1. Introduction Clinicians have long appreciated the importance of observing a patient’s facial expression to gain insight into the pain experience. Typically clinicians use behavioral indicators of pain to augment self-report measures. However, there are certain populations for whom self-report is not possible where facial expression of pain becomes the primary means of communicating distress. The development of the Facial Action Coding System (FACS; Ekman and Friesen, 1978) a fine-grained measure of facial expression, which is anatomically based, allowed researchers to identify specific facial movements associated with pain. Studies using the FACS have demonstrated that useful information about a pain state can be obtained by

* Corresponding author. Tel.: +1 905 5259140, ext. 27820; fax: +1 905 5240069.

observation of the face. Observers are able to distinguish facial expressions of pain from expressions associated with other emotions (LeResche and Dworkin, 1984) and can discriminate between varying degrees of facial expression of pain (Prkachin and Craig, 1985; Patrick et al., 1986; Craig et al., 1988). Observers also appear to be sensitive to differences between true and faked pain expressions (Prkachin, 1992a). Measures of facial expression are only modestly correlated with other measures of pain suggesting that expressive behavior probably carries unique information. A potentially distressing finding is that observers tend to markedly underestimate pain when forced to make judgments based solely on facial expression. Prkachin et al. (1994) examined the degree to which untrained observers’ ratings of shoulder pain corresponded to FACS measurements. Results showed that observers were sensitive to gross variations in pain states. However, there was a tendency to underestimate the patients’ pain on a systematic basis anywhere from 50% to 80%.

0304-3959/97/$17.00  1997 International Association for the Study of Pain. Published by Elsevier Science B.V. PII S0304-3959 (97 )0 3377-0

280

P.E. Solomon et al. / Pain 71 (1997) 279–284

Other work also suggests that clinicians are not particularly sensitive to pain in others. Zalon (1993) found that the accuracy of nurses’ assessment of the pain experienced by abdominal surgery patients post-operatively was related to the degree of pain experienced by the patients. Relative to the patient’s rating the nurses underestimated the higher levels of pain and overestimated the lower levels of pain using a visual analogue scale (VAS). Choiniere et al. (1990) compared nurses’ VAS ratings with those of burn patients and found that 43% of the nurses underestimated the patients’ pain and 27% overestimated the patients’ pain. When a physical cause of pain is less apparent there may be an increased likelihood that the patient’s pain will be underestimated. For example, Teske et al. (1983) asked nurses to rate patients experiencing both acute and chronic pain. There was a tendency for the nurses to underestimate pain in both samples but the discrepancy between patient and nurse ratings was greater for the chronic pain patients. Behavioral measures of pain are often more complex than self-report measures and may focus on gross indicators of pain. The FACS represents an attempt to provide more accurate and finer measures. However, what may be feasible in an experimental setting may be impractical in a clinical environment. Extensive training in identification of 44 facial movements is required to gain competence in using the FACS. Typically the facial movements are identified and analyzed on slow motion video recordings. The time taken to analyze the videotapes would be excessive in a clinical environment. To have clinical utility an assessment procedure must be easy to use, relatively inexpensive and efficient. While FACS may have dubious clinical practicality its use has been instrumental in the identification of a facial expression of pain. A number of studies (e.g., LeResche, 1982; Craig et al., 1991; Prkachin, 1992b) have attempted to use the FACS to identify whether there is a consistent facial expression associated with pain. The ability to identify a general facial expression of pain and to isolate specific facial movements could lead to the development of an abbreviated tool that could be used clinically. Early studies focused on facial movements associated with acute episodic pain. Prkachin (1992b) was able to identify a general facial expression of pain that was consistent across four pain inducing modalities. While a variety of movements occurred it appeared that there were four movements that were consistently associated with pain. Would increased awareness of the four movements improve the accuracy of judgments of pain by observation of facial expression? Is it possible, given the speed of occurrence and the complexity of facial movements, to improve an individual’s accuracy of assessment of facial expression of pain? The main purpose of this study was to examine whether exposure to a brief training procedure, designed to orient subjects to facial movements which are indicative of pain, would improve the sensitivity of their judgment of a patients’ pain.

2. Methods Seventy-five physiotherapy and occupational therapy student volunteers from McMaster University participated in this study. Students were randomly assigned to either a training or control group. There were 31 females and seven males in the training group and 31 females and six males in the control group. The mean age of the trained group was 27.15 years (SD = 3.43) while the mean age of the control group was 26.89 years (SD = 2.93). The training procedure was based on the research findings that indicate pain is encoded in four main facial movements (Prkachin, 1992b). The goal was to provide a brief training procedure which would focus on increasing the awareness of the four FACS movements in an easily understood and efficient manner. The training made use of an acronym, FENS (Frown, Eyes close, Nose wrinkle and Squint), which represented simple descriptions of facial movements adapted from criteria in the FACS manual (Ekman and Friesen, 1978). It was anticipated that the use of an acronym would assist in recall of the salient movements. To be clinically useful and feasible the training procedure needed to be delivered in a short time frame and simulate the rapidity with which the facial expressions would occur in a true clinical environment. The sequence of the training proceeded as follows: (i) the subjects were given written materials introducing the four movements and the FENS acronym; (ii) still slides depicting the individual movements were shown; (iii) these were followed by videotapes which isolated each of the four movements; and (iv) finally the subjects were exposed to brief facial expression scenarios depicting a variety of more complex movements. The task for all subjects was to view a videotape showing the facial displays of patients undergoing range of motion testing of a painful shoulder. There were 10 patients undergoing a total of 88 tests, for each subject to rate. All patients were volunteers who attended physiotherapy at the Waterloo Sports Medicine Centre for assessment and treatment of unilateral glenohumeral joint pain. Facial behavior was recorded as patients underwent a standardized procedure to assess active and passive range of motion of a painful glenohumeral joint (see Prkachin and Mercer, 1989, for details). Active movements were actions performed by the patient without assistance. Passive movements were actions in which the therapist guided the limb through its range of motion. Each test was followed by an interval of 5 s during which the subject was asked to make his or her judgment of the amount of pain the patients were experiencing. The subjects were asked to make their rating of pain using an affective pain descriptor scale developed by Gracely (Gracely et al., 1978; Heft et al., 1980). The patients had provided their rating of pain using the same scale. The videotape was constructed from a subset of 24 patients used in a previous study (Prkachin and Mercer, 1989). The selection of the patients was based on prior analyses of their self-report and facial expressions (Prkachin et al.,

281

P.E. Solomon et al. / Pain 71 (1997) 279–284

1994). Factor analyses had revealed that the facial expressions contain two dimensions: one that reflects the intensity of pain experienced on active movements and one that reflects the intensity of pain experienced on passive movements. Three criteria were used to select the patients for the final videotape. First the full range of tests including patient’s reactions to four active and five passive tests were included. Second, patients were selected if their selfreport of pain was consistent with their facial expressions as measured by the FACS. Third, the patients chosen must have demonstrated a range of facial expressions varying in intensity and duration. This resulted in four patients who were defined as expressive in that their self-report and facial expressions indicated high levels of pain. Six patients were defined as unexpressive in that their self-report and facial expressions indicated they were experiencing low levels of pain (Prkachin et al., 1994). The facial expressions of each subject were measured using a composite FACS score consisting of the intensity × duration of brow lowering, orbital tightening and upper lip raise/nose wrinkling, summed and added to the duration of the eye closure. This index which has been reported in prior studies (Prkachin, 1992b, 1994) appears to provide a valid quantification of pain expression.

3. Results Initially the estimation accuracy of all raters was examined to determine the extent to which raters underestimated and overestimated pain. Fig. 1 illustrates the percentage of raters who underestimated pain for both active and passive movements in expressive and unexpressive patients. Estimates where the rater and patient scores were equal are provided for comparison. Underestimation of pain was most evident for the expressive patients particularly for the passive movements where 75% of the ratings were less than those of the patient. Increased sensitivity to pain was defined as being comparatively closer to the patient’s scores regardless of

Table 1 Absolute mean difference scores

Expressive Unexpressive

Active Passive Active Passive

Trained

Untrained

Improvement due to training (%)

5.27 13.48 1.30 2.60

5.37 14.98 1.56 2.93

2.0 10.7 17.0 11.0

Entries represent the absolute value of the difference between the patient’s rating and the observer’s rating averaged for expressive and unexpressive patients and active and passive trials.

whether this was due to an overestimation or underestimation of the patient’s pain. Therefore, the primary analysis was performed using the absolute value of the mean difference between the patient’s score and the rater’s scores as the response variable. To examine the influence of training on sensitivity to facial expression of pain the absolute mean difference scores on the affect scale were averaged separately by rater group for expressive and unexpressive patients and active and passive movements. These averages were entered into a 2 (expressive/unexpressive) × 2 (trained/untrained) × 2 (active/passive movements) split plot analysis of variance (ANOVA). The main effect of movement was significant, F(1,746) = 1508.33, P = 0.0001, as was the main effect of expressiveness, F(1,8) = 12.15, P = 0.008. The main effect of primary interest, training, was also significant, F(1,584) = 12.92, P = 0.006. Two-way interactions with training were significant; expressiveness × training F(1,584) = 8.69, P = 0.003 and movement × training F(1,746) = 7.37, P = 0.006. The 3-way interaction of expressiveness × rater group × movement type was also significant (F(1,746) = 8.51;P = 0.003). Table 1 helps to clarify the effects found in the ANOVA. In this table values that are closer to zero represent a more sensitive rating. In all conditions the trained raters’ scores are closer to zero. Although there was little training effect on the active movements of the expressive patients, there was a training effect in the other conditions. The improvement due to training is reflected by the difference between the groups given on the right of the table. The greatest effect of training is seen on the active movements of unexpressive patients.

4. The relationship between patient’s pain scores and training

Fig. 1. Estimation accuracy

To further investigate the relationship between the patient’s pain score and training a series of hierarchical regressions were performed using the patient’s pain rating on the affect scale for each of the 88 tests as the response variable. The average pain rating across trained and untrained raters for each test was the predictor variable.

282

P.E. Solomon et al. / Pain 71 (1997) 279–284

Table 2 Hierarchical regression analyses of the relationship between trained and untrained observers’ ratings of pain and the patients’ rating of pain Variable

r2

r2 change

B

F

Condition 1, expressive patients and active movements Patient 0.13 Trained scores 0.17 0.04 0.56 0.75 Patient 0.13 Untrained scores 0.16 0.03 0.58 0.69 Condition 2, expressive patients and passive movements Patient 0.29 Trained scores 0.52 0.23 1.90 6.13 Patient 0.29 Untrained scores 0.51 0.22 1.97 5.97 Condition 3, unexpressive patients and active movements Patient 0.31 Trained scores 0.49 0.18 1.39 3.19 Patient 0.31 Untrained scores 0.39 0.08 0.93 2.03 Condition 4, unexpressive patients and passive movements Patient 0.12 Trained scores 0.82 0.70 1.90 21.46 Patient 0.12 Untrained scores 0.78 0.66 1.42 16.59

P

0.54 0.57

0.0005 0.005

0.03 0.12

0.00001 0.00001

The strategy for the analysis was to run two regressions, one using the trained raters’ scores as the predictor variable and one using the untrained raters’ scores as the predictor variable. A simple comparison of the r2 values for the observer variable in each model would then indicate which explained more of the variation in patient scores. To allow for the correlation between observations on the same patient variable, indicator variables for each patient were introduced into the equation first. Separate analyses were performed for four conditions: (i) expressive patients, active movements; (ii) expressive patients, passive movements; (iii) unexpressive patients, active movements; and (iv) unexpressive patients, passive movements. Results of the analyses are summarized in Table 2. Though it is not possible to perform a test of statistical significance comparing the two values of r2 for trained and untrained raters, one can directly compare the values to assess which analysis explains more of the variation in patient pain ratings. In all conditions the r2 value for the trained group exceeds that for the untrained group though only an additional 1% of the variance was explained by training in both expressive conditions. The overall model for the expressive, active condition did not reach statistical significance for either trained or untrained groups. Consistent with the findings in the ANOVA, the largest difference between the r2 values was for the unexpressive patients on active movements where 18% of the variation in patient scores was accounted for by training compared with 8% in the untrained group.

5. The relationship between the FACS scores and training As the training procedure was based on a modification of the FACS training protocol one would expect the relationship of the raters’ scores with the FACS scores to be stronger in the trained group. To examine this relationship hierarchical regressions of a similar format to those described above were performed substituting the FACS scores for the patients’ ratings as the response variable. As in the previous analyses a series of hierarchical regressions were performed using the FACS score for each of the 88 tests as the response variable and the average pain rating across raters for each test as the predictor variable. Analyses were performed for the same four conditions noted above. Results of these analyses are summarized in Table 3. In comparing the r2 values, the largest difference between trained and untrained values is with the unexpressive patients on the active movements followed by the unexpressive patients on passive movements. For the unexpressive active condition, 13% of the variance in FACS scores was explained by the scores of the trained raters compared to only 2% of the variation explained by the untrained raters’ scores. There was a difference in r2 values of 0.05 in favor of the untrained group when judging expressive patients in the active condition. There was essentially no difference between trained and untrained raters judging the expressive patients in the passive condition. Table 3 Hierarchical regression analyses of the relationship between trained and untrained observers’ ratings of pain and the composite index of pain expression Variable

r2

r2 change

B

F

P

Condition 1, expressive patients and active movements Patient 0.03 Trained scores 0.73 0.70 1.50 10.14 Patient 0.03 Untrained scores 0.78 0.75 1.90 13.44

0.0017

Condition 2, expressive patients and passive movements Patient 0.37 Trained scores 0.51 0.14 2.79 5.62 Patient 0.37 Untrained scores 0.51 0.14 2.90 5.55

0.0079

Condition 3, unexpressive patients and active movements Patient 0.43 Trained scores 0.56 0.13 0.97 4.08 Patient 0.43 Untrained scores 0.45 0.02 0.30 2.56

0.01

Condition 4, unexpressive patients and passive movements Patient 0.07 Trained scores 0.35 0.28 0.94 2.48 Patient 0.07 Untrained scores 0.26 0.19 0.76 1.63

0.061

0.0005

0.0083

0.06

0.19

P.E. Solomon et al. / Pain 71 (1997) 279–284

6. Discussion The present findings provide modest support for the hypothesis that a brief training program can improve observers’ sensitivity to facial evidence of pain. The effect was dependent on patient expressiveness and type of movement, both of which are related to the level of pain experienced by the sufferer. The greatest effect of training occurred when the observers were judging the active movements of unexpressive patients, conditions associated with the patients’ lowest pain reports. It appears that in both rater groups, the more painful the movement, the less sensitive were the raters to the patient’s pain. However, across all conditions the trained raters’ scores represented a more sensitive rating than those of the untrained scores in the sense that the differences between the ratings of the patients and those of the trained observers were smaller than those between the untrained observers and the patients. Though this might also seem paradoxical, it may also be the outcome one would seek in training observers. The results suggest that training is more effective for subtle facial expressions which indicate low levels of pain. The unexpressive active condition was associated with the least painful movements (Prkachin and Mercer, 1989). It is also the condition where the greatest training effect occurred. Regression analyses did not find a significant relationship between the patient and rater scores in the untrained group in this condition, whereas in the trained group 18% of the variation in the patient scores was explained by the judges’ ratings. A comparison of the regression analyses examining the relationship between the FACS scores and the rater’s scores, also showed the greatest difference between trained and untrained observers in the amount of variation explained in the FACS scores to be in the unexpressive active condition. This convergence of findings suggests that, in this condition, the trained observers were using the training information to make their judgments. Training alone appears to be insufficient to overcome the underestimation bias which occurred when observers rated the expressive patients. Other literature has shown that the tendency to underestimate pain is greater at higher levels of pain (Zalon, 1993). If this is true then it may be more difficult to remediate this problem than by exposure to a simple training procedure. A previous study, using methodology similar to the present research, found that observers were more sensitive to gross variations in facial expression of pain and missed more subtle indicators of pain (Prkachin et al., 1994). As a consequence, the training protocol of the present study encouraged the raters to identify small movements that could be easily missed or interpreted as signaling a state other than pain. This could account for the fact that the training had a greater effect with the unexpressive patients on the active, less painful shoulder movements, where the

283

facial expressions were more likely to demonstrate simple movements or less complex facial configurations. There are several other explanations for the findings of this study. The use of the FACS allows one to repeatedly analyze videotape in slow motion, a task that would be impractical in the clinical setting. The task that the raters were asked to perform is in some ways more complex than use of the FACS. The raters have to make decisions quickly and in rapid succession. The possibility exists that in some of the patients the movements occurred too rapidly for the raters to perform this multi-step rating procedure. There may be other elements related to facial expression such as the temporal relationship of the movements, the number of movements that occurred or observer variations in the weighting of facial information that swayed observers’ judgments but were not addressed in training. The effects of training found in this study were small. At the outset this study did not attempt to define what would constitute a clinically important difference in training. Clinical decision making is a complicated process which is dependent on many interacting variables. To be clinically important the difference must be of such magnitude that clinical decisions surrounding the diagnosis or treatment are affected. From an ethical viewpoint one could argue that even a very small change in practice would be important if the end result was less needless suffering. Ultimately one has to weigh the cost of training large groups of clinicians against the benefits of small changes in clinical practice. There are several limitation to the generalizability of the findings of this study. One cannot assume that findings from one health professional group are applicable to another. The sample chosen was representative of relatively inexperienced clinicians who have had previous opportunities to assess pain. The majority of the volunteers for this study were female. Future work needs to include greater numbers of males as there are differences between men and women in their sensitivity to non-verbal communication (Hall, 1978). Given that the findings of this study suggest that it may be more difficult for observers to judge complex facial movements, is it possible to train individuals to be more sensitive to the movements associated with higher levels of pain? It is not clear from the findings of this research whether the complexity of the movements associated with high levels of pain makes the task too difficult to assess in real time. The possibility exists that the complexity of the movements at high levels of pain may limit the improvement one is able to attain with a brief training procedure. Enhancements to the training protocol will be necessary to determine whether further improvements in sensitivity could result from additional training. Clearly the 100 h required to learn the FACS is not conducive to clinical practice. However, it also appears that the 30 min of training that occurred in this study allowed for only small improvements. Enhancements to the training procedure could poten-

284

P.E. Solomon et al. / Pain 71 (1997) 279–284

tially affect the clinical utility that was a consideration in development of the current training procedure. The original intent was to design a brief training procedure that would have wide appeal. A lengthy training procedure is unlikely to have broad based clinical utility. Further work will need to carefully consider the balance between comprehensiveness and precision and clinical utility.

References Choiniere, M., Melzack, R., Girard, N., Rondeau, J. and Paquin, M.J., Comparisons between patients’ and nurses’ assessment of pain and medication efficacy in severe burn injuries, Pain, 20 (1990) 143– 152. Craig, K., Grunau, R. and Aquan-Assee, J., Judgement of pain in newborns; facial activity and cry as determinants, Can. J. Behav. Sci., 20 (1988) 442–451. Craig, K., Hyde, S. and Patrick, C., Genuine, suppressed and faked facial behaviour during exacerbation of chronic low back pain, Pain, 46 (1991) 161–171. Ekman, P. and Friesen, W., Facial Action Coding System, Consulting Psychologists’ Press, Palo Alto, CA, 1978. Gracely, R., McGrath, P. and Dubner, R., Ratio scales of sensory and affective verbal pain descriptors, Pain, 5 (1978) 5–18.

Hall, J., Gender effects in decoding non-verbal cues, Psychol. Bull., 85 (1978) 845–857. Heft, M., Gracely, R., Dubner, R. and McGrath, P., A validation model for verbal descriptor scaling of human clinical pain, Pain, 9 (1980) 363– 373. LeResche, L., Facial expression in pain: a study of candid photographs, J. Nonverbal Behav., 7 (1982) 46–56. LeResche, L. and Dworkin, S., Facial expression accompanying pain, Soc. Sci. Med., 19 (1984) 1325–1330. Patrick, C., Craig, K. and Prkachin, K., Observer judgements of acute pain: facial action determinants, J. Personality Soc. Psychol., 50 (1986) 1291–1298. Prkachin, K., Dissociating spontaneous and deliberate expressions of pain: signal detection analyses, Pain, 51 (1992a) 57–65. Prkachin, K., The consistency of facial expression of pain: a comparison across modalities, Pain, 51 (1992b) 297–306. Prkachin, K. and Craig, K., Influencing non-verbal expression of pain: signal detection analyses, Pain, 21 (1985) 399–409. Prkachin, K. and Mercer, S., Pain expression in patients with shoulder pathology: validity, properties and relationship to sickness impact, Pain, 39 (1989) 257–265. Prkachin, K., Berzins, S. and Mercer, S., Relationships between facial action and apparent pain: a judgement study, Pain, 58 (1994) 253–259. Teske, K., Daut, R. and Cleeland, C., Relationships between nurses’ observations and patients’ self-reports of pain, Pain, 16 (1983) 289–296. Zalon, M., Nurses assessment of postoperative patients’ pain, Pain, 54 (1993) 329–334.