We assessed three groups of participants: (1) an experimental group of CP, (2) a control group of healthy individuals (HC), and (3) a control group of individuals with depressive symptoms (DC) but without pain.

We recruited 17 female and male CP according to DSM-IV (24) from the inpatient, day patient, and outpatient programs of the Department of Psychosomatic Medicine and Psychotherapy at the Medical University Hospital Tübingen and via announcements from the general population. In order to be included, patients had to report pain for at least 6 months and to report current pain intensity in the last 4 weeks of at least 4 on a visual analogue scale (VAS). According to the classification of pain proposed for ICD-11 (25), seven patients reported primary pain, seven patients suffered from musculoskeletal pain, two from headache, and one patient reported visceral pain. 40% of CP fulfilled criteria of at least one other mental disorder, mostly mood and anxiety disorders, according to DSM-IV.

To each of the CP, we selected two control participants who were matched according to age and sex, and one of these participants was additionally matched according to severity of depression symptoms (clinical control group DC). Participants of the HC and DC groups were predominantly recruited from the general population; some participants of the DC group were also recruited from clinical services of the Department of Psychosomatic Medicine and Psychotherapy at the Medical University Hospital Tübingen.

This results in a total sample of N = 51. The sex ratio slightly differs between groups for experimental data as single participants had to be excluded from data analysis (see below).

Exclusion criteria for all participants comprise impaired and non-corrected vision, psychosis, bipolar I disorder or substance addiction according to DSM-IV, heavy nicotine use, and psychotropic medication except for selective serotonin reuptake inhibitors in a stable dosage.

Participants of both control groups were free of clinically relevant pain. Participants of the healthy control group had no history of any mental or serious somatic disorder.

We used standardized stimulus material derived from the Karolinska Directed Emotional Faces (KDEF) database (26). The KDEF is a set of totally 4,900 pictures of human facial expressions of emotion. The set contains faces of 70 amateur actors (35 females and 35 males), each displaying 7 different emotional expressions, each expression being photographed (twice) from 5 different angles. The displayed emotions comprise neutral, happy, angry, afraid, disgusted, sad, and surprised. For the presents study, we selected picture pairs of a happy vs. neutral expression and sad vs. neutral expression (by the same actor, respectively). Pictures of different actors were selected for the happy condition and the sad condition. We used only frontal pictures. The picture set consisted of 26 picture pairs in the happy condition, displaying 13 female and 13 male faces, and 25 picture pairs in the sad condition, displaying 13 female and 12 male faces.

The individual pictures from each pair were presented in two opposing corners of the computer screen (top left/bottom right or top right/bottom left) against a gray background. The target and control stimuli locations were balanced across trials. All pictures covered 377 × 281 pixel.

We used a choice viewing paradigm (27), which has previously been used by our own group and others as a paradigm sensitive to individual differences in hedonic response and emotional behavior (Figure 1) (27–30).

In the present choice viewing paradigm, we presented picture pairs for 3 s each, consisting of an emotional picture paired with a neutral picture. The emotional picture depicted a facial expression of either happiness or sadness, while the control picture depicted a neutral facial expression of the same individual (see description of stimulus material above). Each picture pair was preceded by a central fixation cross presented for 1,500 ms. Participants were asked to fixate this cross. Within this cross, a small dot was presented in one of the four quadrants (top left/bottom right or top right/bottom left) as a cue, and this cue indicated with 100% reliability the position of the emotional picture in the next trial. Participants were told about this relationship, but there was no other instruction associated with this cue. We used a free viewing task, asking participants to explore the picture pairs freely as if they were watching television.

Our cued choice viewing paradigm thereby targets the goal directed attentional system as it allows for strategic attention deployment, especially in early processing stages. Gaze behavior in this paradigm provides information about how patients with chronic pain deal with the anticipated confrontation with positive vs. negative emotional information.

The emotion expressed by the facial stimuli was blocked, and a sequence of blocks (happy vs. sad) were randomized.

The study was conducted on two separate appointments: a first appointment for diagnostic purposes and a second appointment comprising the experimental task.

During the diagnostic assessment, we conducted the structured clinical interview for DSM-IV (SCID-I) (31) and assessed basic demographic characteristics. Onset, duration, localization, quality, and intensity of pain in CP were assessed using items of the German Pain Questionnaire. We additionally assessed strongest intensity of pain during the last 4 weeks using a VAS ranging from 0 (no pain) to 10 (strongest conceivable pain). All participants completed validated self-report instruments assessing depressive symptoms [Beck Depression Inventory (BDI)] (32) and general psychopathology (Patient Health Questionnaire) (33).

In order to assess difficulties recognizing and describing own emotions, we used the Toronto Alexithymia Scale 26 (TAS26) (34). The TAS is a self-report instrument very commonly used to assess alexithymia as a multidimensional concept, consisting of three subscales: difficulty identifying feelings, difficulty describing feelings, and externally oriented thinking.

For the experimental session, participants arrived at the laboratory during an afternoon appointment to control for circadian variation in the sample. Participants were asked to refrain from smoking and caffeine 3 h before testing. The eye-tracking system was adjusted to each individual and we performed a 13-point-calibration before starting the eye-tracking task. After completion of the task, participants rated all stimuli for pleasantness using a VAS ranging from −4 (maximum unpleasant) to +4 (maximum pleasant). Finally, we assessed current pain intensity using again a VAS ranging from 0 to 10 and emotional burden with a VAS ranging from 0 (no burden) to 10 (strongest conceivable burden).

Data were assessed using the remote Eye Tracking System iView X Hi-Speed (SensoMotoric Instruments GmbH, Berlin, Germany) with a sampling rate of 500 Hz.

The study was approved by the Ethics Committee of the Medical Faculty and the University Hospital Tübingen at the Eberhard Karls University Tübingen. This study was carried out in accordance with the recommendations of the Ethics Committee of the Medical Faculty and the University Hospital Tübingen at the Eberhard Karls University Tübingen with written informed consent from all subjects. All subjects gave written informed consent in accordance with the Declaration of Helsinki.

Raw gaze data were analyzed using BeGaze 2.0 (SensoMotoric Instruments, Berlin, Germany).

We excluded trials from data analysis where participants had not fixated on the fixation cross or data with insufficient data quality. We excluded participants with less than 50% of trials analyzable. This resulted in the exclusion of six participants from the block with happy facial expressions and of three participants from the block with sad facial expressions.

We defined areas of interests (AOIs) which comprised the face on each of the depicted photographs and analyzed gaze data exclusively for these AOIs.

We defined the following two gaze measures which were previously used in eye-tracking research [e.g., Ref. (22, 29)]: (1) the position of the initial fixation after the trial’s onset as a measure of attentional orienting and (2) percentage of total gaze duration on the AOI during a trial as a measure of continuous attentional engagement or maintenance. In order to perform exploratory correlation analyses, we calculated attentional bias scores: (1) a gaze direction bias score by subtracting the percentage of initial fixations on neutral stimuli from the percentage of initial fixations on emotional stimuli and (2) a gaze duration bias score by subtracting the total gaze duration on neutral stimuli from the total gaze duration on emotional stimuli. Positive scores reflect a bias toward emotional stimuli.

Data were analyzed using SPSS version 21.0 (IBM). All tests were two-tailed. We chose an α-level of 0.05 as threshold of statistical significance.

Based on the reported results of one primary outcome measure (i.e., initial orienting to negative emotion), a post hoc power analysis was performed, indicating that the group sizes of n = 17 obtained 76.3% power to detect differences between the CP and the HC group in this variable using a two-tailed statistical significance criterion of 5%.

To analyze potential baseline group differences in demographic and clinical variables, we used univariate analysis of variance (ANOVA) with group as between subject factor. Fisher’s least significant difference was used to identify post hoc group differences. To analyze gaze data, we used repeated measure ANOVAs with stimulus type (emotional facial expression vs. neutral facial expression) as within subject factor and group (CP, DC, and HC) as between subject factor. We performed Scheffé tests to identify post hoc group differences as this test procedure is considered a more conservative post hoc analysis. In order to assess within group biases, we used one-sample t-tests. We report effect sizes for group differences using η², with η² < 0.06 representing a small effect, 0.06 < η² < 0.14 representing a medium effect, and η² ≥ 0.14 representing a large effect.

We performed exploratory correlational analysis using Pearson’s correlations, assessing potential relationships between attention deployment (i.e., the gaze direction bias score and the gaze duration bias score) on the one hand and self-report continuous measures of alexithymia (TAS26 total score), depression (BDI score), and pain (current pain intensity and strongest pain intensity in the last 4 weeks).

Table 1 gives and overview on demographic and clinical characteristics of the study sample. The matching was successful: the three groups are comparable with respect to age, and the group CP and DC are comparable with respect to depressive symptoms. Level of depressive symptoms in CP and DC participants was classified as “mild” according to BDI classification.

Patients with chronic pain disorder and DC patients showed significantly higher scores on the TAS26 than HC (Table 1), indicating that both groups have more overall difficulties identifying and communicating own feelings.

Figure 2 shows results of the valence rating. We found a main effect of stimulus type [F (2) = 348.709; p < 0.001; η² = 0.879] with a large effect size. Sad faces were rated as significantly less pleasant than neutral and happy faces. We also found a significant stimulus × group interaction [F (4) = 4.078; p = 0.015; η² = 0.145] with a large effect size. Participants were not different with respect to their valence rating of sad and neutral faces [F (2, 48) = 1.963; p = 0.152], however, there was a significant effect for happy faces [F (2, 48) = 6.654; p = 0.003]. Post hoc-analyses revealed that the CP group rated the happy faces as significantly less pleasant than the HC group (p = 0.004).

Figure 3 shows the pattern of attentional orienting to emotional stimuli.

With respect to happy vs. neutral faces, we found a main effect of stimulus type [F (1, 42) = 21.604; p < 0.001; η² = 0.34] with a large effect size, which means, there was a general bias in attentional orienting toward happy faces as compared to neutral faces. When looking at single participant groups, this tendency to orient attention first to happy faces was significant in the CP group [t(14) = 2.380; p = 0.032] and in the HC group [t(14) = 3.868; p = 0.002], while the DC group showed no preference for happy faces in attentional orienting [t(14) = 1.885; p > 0.05]. We found no interaction effect.

With respect to sad vs. neutral faces, we found no main effect of stimulus type, but a significant stimulus × group interaction [F (2, 45) = 3.611; p < 0.05; η² = 0.138] with a medium effect size. Post hoc-analyses revealed that the HC group significantly differed in attentional orienting to sad faces from the DC group (p = 0.036), while the CP lied in between. When looking at single participant groups, the tendency to orient attention first to sad faces was significant in the HC group [t(15) = 4.068; p = 0.001], however, both other groups showed no significant preference for sad faces in attentional orienting (p > 0.05).

Figure 4 shows the pattern of attentional maintenance on emotional stimuli.

With respect to happy vs. neutral faces, we found a main effect of stimulus type [F (1, 42) = 30.080; p < 0.001; η² = 0.417] with a large effect size, which means, there was a general bias in attentional maintenance on happy faces as compared to neutral faces in all three groups. We found no interaction effect.

With respect to sad vs. neutral faces, we found neither a main effect of stimulus type nor an interaction effect. This means, the gaze behavior of all groups stayed statistically on the 50% level of equal distribution, and no attentional bias was detectable.

No significant correlations were found for both attentional bias scores and the TAS26 total score as a measure of alexithymia or the BDI total score as a measure of depression in the total sample. We also analyzed potential relationships between both attentional bias scores and current pain intensity and strongest pain intensity in the last 4 weeks, respectively, for the CP group only and found no significant correlations.

This is to the best of our knowledge the first study to investigate attentional processing of emotional information in chronic pain using eye tracking. Recently, the importance to use eye-tracking methodology to investigate attentional processes in pain has been emphasized (7, 22). We have used standardized and well-validated stimulus material, depicting facial expressions of happiness or sadness vs. neutral expressions, hence being able to contrast processing of positive vs. negative primary emotions. A further strength of our design is the inclusion of a second control group of individuals who were carefully matched to the chronic pain patients according to their level of depressive symptoms using the BDI. Our sample of chronic pain patients was a clinical group characterized by long-lasting and intensive pain and heterogeneous with respect to pain localization.

The sample of the present study is not large; however, the effect sizes of all main and interaction effects are medium to large, indicating that the study was not underpowered. The cross-sectional design of our study limits the interpretation of results as being cause or consequence of chronic pain. We have not assessed arousal, therefore, we could not control for this variable and hence, it is difficult to distinguish the respective effects of valence and arousal in the obtained results. Moreover, we have not contrasted a positive vs. a negative emotion as a stimulus pair. It has to be considered that the CP sample was characterized by heterogeneity, which is illustrated by the fact that 40% of these patients fulfilled at least one further diagnosis of a mental disorder.

We found that patients with chronic pain as compared to healthy participants display different strategic attention deployment to emotional vs. neutral stimuli in early processing stages in a cued eye-tracking paradigm. Patients with chronic pain showed facilitated early attention to positive emotional stimuli (i.e., happy facial expressions), but no facilitated early attention to negative emotional stimuli (i.e., sad facial expressions), while we found no group differences in attentional maintenance on emotional stimuli.

This pattern of results was very similar to that of a second control group of individuals who were matched to the chronic pain patients according to level of depressive symptoms. We found few effects that were unique for chronic pain, rather there were many similarities between CP and DC in all measures of emotional processing in our study. Our study suggests that patients with chronic pain predominantly show alterations in early processing of negative emotions, while these difficulties also generalize to positive emotions in individuals with depressive symptoms. This is in line with the clinical large overlap between pain and depression (2) and points to potential common maintenance mechanisms. The blunted initial reaction to sadness could be interpreted in terms of a failure of the attentional system to attend to evolutionary salient emotional stimuli (i.e., an anhedonia bias) (1, 16) or as an attempt to suppress negative emotions that are difficult to identify, express, or regulate (2). However, it is often hard to disentangle sequence and causality of pain and depression (2). Difficulties in emotion processing in patients with chronic pain could be predominantly due to comorbid depressive symptoms, however, nevertheless, they might represent an important factor contributing to chronification and maintenance of pain and should therefore be addressed by targeted interventions.

Larger studies investigating emotion processing in chronic pain disorder are needed to expand our knowledge on mechanisms and potential consequences for treatment. These studies should include paradigms with longer presentation times of stimuli, tasks addressing different processing stages and competencies of the attentional system and could combine emotional stimuli with stimuli expressing pain, similarly as previously done by Liossi et al. (22). Potential interventions targeting emotion processing in chronic pain could comprise attentional bias modification, similarly as piloted with pain-related stimuli (40) or a range of other non-invasive neuromodulatory interventions, e.g., including mindfulness interventions (41) or emotional exposure techniques (1).