Autonomic dysregulation and impairments in the recognition of facial emotional expressions in patients with chronic musculoskeletal pain

Ethical considerations and sample characteristics

This cross-sectional study was approved by the Ethical Board Committee of the Federal Institute of Rio de Janeiro (CAAE 53993616.4.0000.5268). Study participants were assigned to symptomatic and asymptomatic groups. The study protocol followed the recommendations of the strengthening the reporting of observational studies in epidemiology (STROBE) Statement [29]. Data was collected from October 2018 to April 2019.

Participants

We recruited participants of both genders and aged between 18 and 60 years. The symptomatic group included participants with chronic musculoskeletal pain [30] (discomfort stemming from muscles, ligaments, tendons, and bones for more than three months) with moderate and severe pain intensity. Participants were recruited consecutively during their medical appointment at the Orthopedics and Traumatology Service of the Clementino Fraga Filho University Hospital (HUCFF). The asymptomatic group was composed of participants who did not have pain in any part of the body recruited by advertisement in the University and on the social media (Facebook and Instagram). Exclusion criteria were: regular intake of psychiatric medication (for example antidepressants, anticonvulsants), radiculopathies, diagnosis of psychiatric disorders, major cognitive impairment, heart diseases, history of heart infarction, hypertension, and diabetes.

Instruments

Initially, demographic and health-related data was collected using an assessment form. The following instruments were used to characterize the group with chronic musculoskeletal pain:

1. Pain intensity and pain interference: The brief pain inventory (BPI) was used to measure mean pain intensity, current pain, and pain interference. BPI is an 11-point numerical rating scale, where zero indicates ‘no pain’ and 10 indicates ‘worst pain possible’. The pain interference subscale consists of seven items to measure to what extent pain interferes in various aspects of life, including general activity, mood, mobility, normal work, relationships with others, sleep, and life pleasure. Items are also rated on an 11-point numerical rating scale where zero indicates ‘does not interfere’ and ten indicates ‘completely interferes’ [31].

2. Symptoms of anxiety and depression: The symptoms of anxiety and depression were investigated using the hospital anxiety and depression scale (HADS). This scale has 14 items, of which seven are anxiety related statements, and seven are depression related statements. Responses can be scored from zero to three points for a total of 21 points for each scale division and a total of 42 points [32].

3. Alexithymia: Alexithymia is a personality trait characterized by difficulty in experiencing and expressing emotions. The Alexithymia Toronto Scale (TAS) is a self-report instrument composed of 26 items and is designed to measure the degree of alexithymia. Items are answered on a five-point Likert scale ranging from 1 (strongly disagree), 2 (disagree), 3 (don’t know), 4 (agree), and 5 (strongly agree). Total scores range from 26 to 130. For scores above 75 the participant is considered to have alexithymia, and below 62 is considered as not having alexithymia [33].

4. Kinesiophobia: The Tampa Kinesiophobia scale is a self-reported 17-item scale that is used to measure fear of movement/re-injury. All items are rated on a four-point Likert scale (1 = strongly disagree, 4 = strongly agree). The total score ranges from 17 to 68. Higher scores indicate higher levels of kinesiophobia [34].

5. Catastrophizing: The pain catastrophizing scale (PCS) was used to measure pain catastrophizing thoughts. Catastrophizing pain can be defined as an exaggerated negative perception of pain and is well established as an important psychological risk factor for pain-related outcomes. The PCS consists of 13 items that refer to an expanded perception of pain-related threats, ruminant thoughts about pain, and feelings of helplessness. The frequency of thoughts associated with each item is rated on a five-point Likert scale, where zero indicates ‘not all’ and four indicates ‘all the time’. The total score is calculated as the sum of all items. Higher scores in the PCS indicate higher levels of catastrophizing. The PCS is the most common measure used for pain catastrophizing and has been shown to have strong reliability and predictive value among people with musculoskeletal pain [35].

HRV measurement and analysis

Volunteers attended a single visit in the laboratory between 8:00 a.m. and 10:00 a.m. for anamnesis, HRV recordings and facial emotion recognition task measurements. The experimental protocol was carried out in a quiet room with noise, temperature, and humidity controlled. Before starting the HRV measurement, participants were instructed to remain seated on a chair with armrests in front of a computer screen, and to breath normally for 5 min. During the HRV acquisition, participants were instructed to look at a white fixation cross in a black background, displayed on a 15-in. computer screen. For HRV acquisition, a portable heart rate monitor, Polar H7 Bluetooth (Bethpage, NY, USA), fitted just below the sternum, was used to register the RR intervals. Portable HRV methods are reliable and consistent with an electrocardiogram [36]. Participants were asked to avoid talking, performing large and sudden movements, and sleep. After the 5 min resting phase, HRV was measured for 10 min. The collected HRV was reviewed by visual inspection and data analysis was performed using the open-source software Kubios (http://www.kubios.com). HRV was analyzed in frequency domain including low frequency [LF (ms²), 0.04–0.15 Hz] (reflects a mix between sympathetic and vagal influences that shows an influence of both sympathetic and parasympathetic branches), high frequency [HF, (ms²) 0.15–0.4 Hz] (reflects vagal tone), and low frequency-to-high frequency ratio [LF (ms²)/HF (ms² 130)] components of the RR interval time series.

Recognition of facial expressions of emotion

The stimuli were composed of high-resolution monochrome digital photographs selected from the standardized set pictures of facial affect [37]. Male and female faces of emotional expressions (neutral, anger, disgust, fear, happiness, and sadness) were selected for the task. The morphing technique (i.e. gradual transition from one image to another until reaching the final image) was used to convert pictures into videos. The morphing was generated in Facemorpher lite (Luxand Development) and then transferred to E-prime 2.0 software (Psychology Software Tools USA). For all stimuli, the initial face consisted of a neutral face that changed to another face of emotional expression as follows: neutral–happiness; neutral–fear; neutral–disgust; neutral–sadness; neutral–anger. Faces of neutral emotional expression were used to neutralize the stimulus [38]. Videos were presented for ten seconds on a 15-in. computer screen in a randomized sequence, followed by a white fixation cross (1 s) and then a screen with response options with six options (anger, disgust, fear, happiness, sadness, and ‘not recognized’). Participants were instructed to press the left mouse button quickly (response time) and accurately as soon as they had identified the emotion during the video presentation. Figure 1 represents the timeline of the recognition of facial expressions task.

Figure 1:

A timeline showing the recognition of facial expressions of emotion task. Pictures were converted into videos using a morphing technique.

Data analysis

An a priori sample size calculation showed that a total of 74 individuals would be necessary to detect 30% difference on LF and on facial expressions of emotion task with a statistical power of 80%, alpha of 5%. The statistical analysis was performed using the Statistical Package for Social Sciences (SPSS) version 22 for Windows (IBM Corporation, Armonk, New York, USA). The demographic and clinical variables of the study population were presented as means and standard deviations for continuous variables. Categorical variables were presented as absolute values and frequencies. Data normality was tested through visual inspection of histograms and Shapiro–Wilk Test. For the HRV means within-group difference, we used the paired Student’s t-test and the Student’s t-test for independent samples for the mean difference between groups. Mean difference and confidence interval (95%) were presented. For facial emotion recognition tasks, the number of correct responses (hit rate) was computed for each volunteer from percentage correct classification out of the total number of stimuli for each emotional category (percentage relative to the videos presented). A non-parametric test, Friedman’s analysis of variance (ANOVA) was performed to test the hit rate for each emotion category, and the Kruskal–Wallis test was applied to assess the effect of the groups. Additionally, the unbiased hit rates (Hu) were calculated for each emotion using the formula proposed by Wagner et al. [39], and each response option was added to the calculation (five emotions and the ‘not recognized’ option). For each volunteer, the median and standard deviation of the valid response time (RT) for each emotional category were calculated. The valid RT considered the correct classification of emotional faces, and the anticipatory (below 100 ms), slow response (above 10 s), and incorrect or ‘not recognized’ classifications were excluded from RT calculation. These data were analyzed using repeated-measures ANOVA: the first factor was a group of volunteers with two sublevels (symptomatic and asymptomatic) and the second factor was the emotional category of the video with five sublevels (anger, disgust, fear, happiness, and sadness). The significance level adopted in the study was 95%.

Comparison of resting HRV

In the HRV frequency domain, the mean of LF and LF/HF were higher for the symptomatic group when compared to the asymptomatic group. The symptomatic group had a mean HF lower (mean = 34.14; SD = 16.95; p<0.001) than the asymptomatic group (mean = 51.11; SD = 13.01; p<0.001) (Table 2).

Recognition of facial expressions of emotion

The results of video classification by Friedman’s ANOVA showed a significant effect of emotion category (Chi-squared (80, 4)=142.86; p<0.01). For both groups a lower hit rate was found for disgust (mean = 68.20 sd = 18.04), anger (mean = 72.75 sd = 13.63) and sadness (mean = 78.42 sd = 16.13), and a higher hit rate was found to happiness (mean = 97.33 sd = 4.70), and fearful (mean = 86.25 sd = 14.50) facial expressions (Figure 2). The results of the classification of each emotion by group are presented in a confusion matrix (Figure 3).

Figure 2:

Hit rate (%) responses for each facial expressions of emotion by group.

Figure 3:

Confusion matrix presenting the results of the recognition of facial expressions of emotion task by group.

The Kruskal–Wallis ANOVA revealed that the emotional facial expressions of disgust (H (1, 80)=7.82; p<0.01), anger (H (1, 80)=13.56; p<0.01), sadness (H (1, 80)=6.58; p=0.01), and happiness (H (1, 80)=12.68; p<0.01) were those for which volunteers from the symptomatic group had a lower hit rate of correct answers compared to the asymptomatic group. No significant group differences were observed to fearful faces expression (H (1, 80)=2.04; p=0.15). The participants in the symptomatic group presented a lower number of correct responses when compared to the asymptomatic group (Table 3).

The RT to corrected answers showed a major group effect (F (1.77) = 21.11; p<0.001) and emotional category (F (4.308) = 174.21; p<0.001), without presenting any interaction between the factors (F (4.308) = 0.446; p=0.775). In general, the symptomatic group was slower to perform the task of identifying facial emotional expression (7.066 s; SD = 1.188) than the participants in the asymptomatic group (6.298 s; SD = 1.203) for all emotional categories. The emotional categories that presented the highest differences in RT between the symptomatic and asymptomatic groups were the facial expressions of sadness (919.70 ms), fearful (866.98 ms), and disgust (818.69 ms) (Table 3).

Limitations

This study presents some limitations that need to be noted. First, other psychological comorbidities may influence HRV and also in the face recognition task. However, the patients included in this study represent those assessed in clinical practice. Second, we did not classify the patients according to the pain mechanism (nociceptive, neuropathic, or nociplastic). Participants with nociplastic pain may have different results regarding HRV and the recognition of faces of emotional expressions, since this population may have deficits in descending pain modulation. Clinical studies based on diagnostic criteria, incorporating an extensive pain examination, are needed to address this issue. Third, women were predominant in our sample. We tried to reduce the gender influence in the results by matching the samples by gender and age. The present results should be addressed with caution in terms of sex differences in autonomic measures of emotion reactivity [66]. Fourth, no information regarding the participants’ perception of social isolation was collected to verify the association with the deficit of face recognition of emotional expression. Besides, the cross-sectional design of our study does not allow attributing causal relationships or temporal associations between HRV, face recognition, and pain. Longitudinal studies to investigate the potential temporal sequence between reductions in vagal control, changes in face recognition, and changes in pain processing are strongly encouraged. Finally, we did not analyze other domains of HRV such as time domain and non-linear methods since they need a longer acquisition duration. Future studies are required to address these limitations as longitudinal studies in acute/subacute pain conditions, 24-h HRV recording, and the relationship of clinical improvement with HRV and face recognition are recommended.